Computational Contact Mechanics - Konyukhov

Lecture Notes in Applied
and Computational Mechanics 67
Series Editors
Prof. Dr.-Ing. Friedrich Pfeiffer
Lehrstuhl B für Mechanik
Technische Universität München
Boltzmannstraße 15
85748 Garching
Germany
E-mail: [email protected]
Prof. Dr. Peter Wriggers

FB Bauingenieur- und Vermessungswesen
Inst. Baumechanik und Numer. Mechanik
Universität Hannover
Appelstr. 9 A
30167 Hannover
Germany
E-mail: [email protected]
For further volumes:

http://www.springer.com/series/4623
Computational Contact
Mechanics
Geometrically Exact Theory for Arbitrary

Shaped Bodies
Alexander Konyukhov and Karl Schweizerhof
ABC
Authors
PD Dr. habil. Alexander Konyukhov Prof. Dr.-Ing. Karl Schweizerhof
Institute of Mechanics Institut für Mechanik
Karlsruhe Institute of Technology Karlsruher Institut für Technologie
Karlsruhe Karlsruhe
Germany Deutschland
ISSN 1613-7736 e-ISSN 1860-0816

ISBN 978-3-642-31530-5 e-ISBN 978-3-642-31531-2
DOI 10.1007/978-3-642-31531-2
Springer Heidelberg New York Dordrecht London
Library of Congress Control Number: 2012940862
c Springer-Verlag Berlin Heidelberg 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of pub-
lication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.
Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)
Preface
Who understands geometry understands anything in this world.
1564-1642 Galileo Galilei
Geometry is founded on mechanical practice and is nothing other

than that part of universal mechanics which reduces the art of mea-
suring to exact propositions and demonstrations.
The Prinicipia: Mathematical Principles of Natural Philosophy.

May 1686 Isaac Newton
Geometrical ideas correspond to more or less exact objects in nature,

and these last are undoubtedly the exclusive cause of the genesis of
those ideas.
Relativity: The Special and General Theory.

December 1916 Albert Einstein
The beauty and the rigour of geometry was always coming along with the ra-
tionality of mechanics. All scientists starting from the ancient time have been
highly motivated to combine the contemporary knowledge in geometry with
propositions of natural philosophy in order to obtain the best description
of nature. Thus, the combination of Euclidean geometry with contemporary
knowledge in mechanics has led to Newton’s classical mechanics and the com-
bination of differential geometry originated from Gauss with knowledge in
mechanics and physics led to Einstein’s relativity theory. This fusion process
VIII Preface
is undoubtedly observed for many branches of mechanics: the combination

of elasticity theory and differential geometry has led to modern nonlinear
continuum mechanics formulated in a form independent of the selection of
coordinate systems - in the so-called covariant form; the combination of plate
theory originated and developed by many scientists – Jakob Bernoulli, Leon-
hard Euler, Gustav Kirchhoff, Augustus Love and others – together with
differential geometry has led to the modern non-linear shell theory etc. The
main motivation of this process is to make the description more universal and
applicable independently of unnecessary hypotheses, such as the selection of
a certain coordinate system.
The geometry is playing a major role for the analytical approach in contact
mechanics as the complexification of geometry can be observed through the
famous analytical solutions in contact mechanics. Newton’s law of impact
is formulated as a model for two contacting material points, the Coulomb
friction law is formulated as a model for a material point and a rigid line,
the Euler-Eytelwein problem is resolved for a rope sliding on a circular rigid
arch, the Bussinesque solution is given for a force acting on the elastic half-
space and the Hertz solution is given for two in general elliptically shaped
deformable bodies.
The evolution of computational mechanics also follows the way of the de-
velopment of analytically methods in mechanics: it has been motivated mostly
by the possibility to model objects of arbitrary geometry. All methods initially
formulated in the simplest Cartesian coordinate system are starting to gain
experience with more advanced geometrical methods. Thus, advanced finite
element algorithms for shell treatment should be formulated in a covariant
form. Although some more algorithms in computational mechanics are de-
pending on the geometrical description, computational contact mechanics is
by far depending on the geometry most. Though, the intuitive understanding
of contact between bodies is based on the geometry of adjoining bodies, the
geometrical aspect of contact modeling is often hidden inside a certain numer-
ical method. Many numerical methods developed into a separated branch of
the computational mechanics within the last decades and described in mono-
graphs such as Contact problems in elasticity of professors Noboru Kikuchi
and John Tinsley Oden Computational contact mechanics of professor Peter
Wriggers and Computational contact and impact mechanics of professor Tod
Laursen became standard tools to model contact problems in many engineer-
ing simulation packages. A more powerful approach including the application
of various numerical methods is to take advantage of the geometry of an an-
alyzed object and describe the problem in the best coordinate system. The
best coordinate system to describe the contact interaction in all its geometri-
cal details is a coordinate system attached to the geometrical features of the
contacting bodies.
This book is the result of the fusion of differential geometry into com-
putational contact mechanics. It gives the necessary introductory treatment
of differential geometry for curves and surfaces and contains also some new
Preface IX
analytical results for the verification of contact algorithms. A systematic anal-

ysis of various geometrical situations leading to contact pairs – surface-to-
surface, curve-to-surface, point-to-surface, curve-to-curve, point-to-curve is
presented. Each contact pair is inherited with a special coordinate system
based on its geometrical properties such as a Gaussian surface coordinate
system, or a Serret-Frenet curve coordinate system. This allows to formulate
all algorithms known in computational contact mechanics in a covariant form
which is the core of the geometrically exact theory of contact interactions.
Another achievement of the developed theory is the possibility to construct
various constitutive relations for contact pairs shown as the generalization
of Coulomb friction into a coupled anisotropic adhesion and friction law for
surface-to-surface and curve-to-curve contact pairs.
The book is conceptualized as a research monograph on computational
contact mechanics formulated in a covariant form with the final goal to give
the reader closed form algorithms for finite element implementations indepen-
dently of the type of approximation involved into the discretization process as
well as for any isogeometric analysis. These aspects – the authors hope – are
enabling this book a full embedment into the field of modern computational
contact mechanics.
Karlsruhe Institute of Technology, Alexander Konyukhov

May 2012 Karl Schweizerhof
Personal Acknowledgments
I owed my geometrical ideas to my education as well as PhD study at Kazan

State University where the geometrical methods in natural science were al-
ways favored since the time of its foundation in 1804 when the first rector and
the developer of non-Euclidean hyperbolic geometry, Nikolai Lobachevsky –
his research has been highly appreciated by Carl Friedrich Gauss, the German
developer of non-Euclidean hyperbolic geometry and differential geometry –
started his lectures in mechanics. Within the beginning of the research activ-
ity I was lucky to take advantage of many fruitful discussions with professor
Karl Schweizerhof and was motivated to come with geometrical ideas into
computational contact mechanics similar to the nonlinear shell theory. This
book is appearing as a summary of this research activity in the Institute of
Mechanics of the Karlsruhe Institute of Technology (former named as Uni-
versity of Karlsruhe (TH)) inspired by the founder of the continuum contact
mechanics, Professor Heinrich Hertz, who had worked here. My research ac-
tivity converted to the written manuscript I would like to dedicate to my
children.
I was also lucky to discuss various topics on computational contact me-
chanics on numerous conferences on computational mechanics directly with
Professor Peter Wriggers and Professor Tod Laursen. Namely their devel-
opments as well as friendly discussions were motivating me to describe all
computational algorithms by using the language of geometry.
The research mainly presented in the book, has been supported by the Ger-
man Research Foundation (Deutsche Forschungsgemeinschaft (DFG)). This
help is gratefully acknowledge.
I would like especially to thank my colleagues: PhD student, Dipl.-Ing.
Ridvan Izi with whom we discussed the development of several algorithms
presented in the book as well as the structure of the Course on Contact
XII Personal Acknowledgments
Mechanics arising from this book and being taught at KIT over years; Halil
Yeniavci who has provided the multimedia support of my presentations at
numerous international conferences resulting to colorful figures in this book
and Oana Mrenes for her careful work with the text and figure editing of the
manuscript.
Karlsruhe Institute of Technology, Alexander Konyukhov

May 2012
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Overview of Approaches to Model Contact Problems . . . . . . . 1
1.2 Discussion – Why Covariant Approach? . . . . . . . . . . . . . . . . . . 13
1.3 On Geometrical Approaches in Contact Mechanics . . . . . . . . . 17
1.4 Goals and Structure of the Book . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.1 Structure of the Current Book . . . . . . . . . . . . . . . . . . . . 18
2 Differential Geometry of Surfaces and Curves . . . . . . . . . . . 25

2.1 Definition of the Surface and Its Geometrical
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 The Fundamental Tensors and Property of the
Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.1.2 Study the Local Surface Structure . . . . . . . . . . . . . . . . 28
2.1.3 Differential Operations in the Surface Coordinate
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Definition of the Curve in 3D and Its Geometrical
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.1 Definition of a Local Coordinate (Serret-Frenet)
System. Serret-Frenet Formulas . . . . . . . . . . . . . . . . . . . 34
3 Closest Point Projection Procedure and Corresponding

Curvilinear Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1 Closest Point Projection Procedure for Arbitrary
Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.1 Formulation of the Closest Point Projection
Procedure in Geometrical Terms . . . . . . . . . . . . . . . . . . 38
3.1.2 Proximity Criteria for Different Surfaces . . . . . . . . . . . . 41
3.2 Solvability of the CPP Procedure for Surfaces – Allowable
and Non-Allowable Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.1 Reduction to 2D Plane Geometry – Solvability
Criteria and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . 44
XIV Contents
3.2.2 Proximity Domain for Globally C 0 -Continuous

Surface in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.3 Reference Example: Projection Domain for a
Hyperbolical Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3 Closest Point Projection Procedure for Point-To-Curve
Projection and Corresponding Projection Domain . . . . . . . . . 50
3.4 Closest Point Projection Procedure for Curve-To-Curve
Contact and Definition of Local Coordinate Systems . . . . . . . 53
3.4.1 Definition of a Local Coordinate System Attached
to a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4.2 Analysis of Uniqueness and Existence of Solutions
for the CPP: Definition of a Projection Domain . . . . . 56
3.4.3 Computational Issues of the CPP Procedure . . . . . . . 62
4 Geometry and Kinematics of Contact . . . . . . . . . . . . . . . . . . . . 63

4.1 Kinematics of the Surface-To-Surface Interaction . . . . . . . . . . 63
4.1.1 Local Surface Coordinate System . . . . . . . . . . . . . . . . . . 64
4.1.2 Spatial Curvilinear Local Coordinate System and
Its Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.1.3 Measure of Contact Interaction
for Surface-To-Surface Contact . . . . . . . . . . . . . . . . . . . . 66
4.1.4 Spatial Basis Vectors and Metric Tensor . . . . . . . . . . . 66
4.1.5 Motion of a Slave Point . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1.6 Geometrical Interpretation of Covariant Derivative
and Numerical Realization . . . . . . . . . . . . . . . . . . . . . . . . 74
4.1.7 Variation δξ 3 and Its Linearization . . . . . . . . . . . . . . . . 77
4.1.8 Variation δξ i and Its Linearization . . . . . . . . . . . . . . . . 79
4.2 Kinematics of 2D Contact Interaction . . . . . . . . . . . . . . . . . . . 83
4.2.1 Closest Point Projection Procedure and
Corresponding Coordinate System . . . . . . . . . . . . . . . . . 84
4.2.2 2D Contact Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2.3 Linearization of Variations δζ and δξ . . . . . . . . . . . . . . 92
4.3 Kinematics of Segment (Deformable)-To-Analytical
(Rigid) Surfaces Contact (STAS) – Two Strategies . . . . . . . . . 95
4.3.1 Rigid Surface Is a “Slave” Surface . . . . . . . . . . . . . . . . . 95
4.3.2 Rigid Surface Is a “Master” Surface . . . . . . . . . . . . . . . 97
4.3.3 Surfaces Allowing a Closed Form Solution for the
Penetration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.4 Kinematics of Point-To-Curve Interaction . . . . . . . . . . . . . . . . 105
4.5 Kinematics of Curve-To-Curve Interaction . . . . . . . . . . . . . . . 107
4.5.1 Development of Beam-To-Beam and Edge-To-Edge
Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.5.2 Measures of Contact Interaction . . . . . . . . . . . . . . . . . . . 109
4.5.3 Rates and Variations of Measures for Contact
Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Contents XV
4.5.4 Linearization in a Covariant Form of Variations for

Contact Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.6 Kinematics of the Curve-To-Rigid-Surface Interaction . . . . . 115
5 Weak Formulation of Contact Conditions . . . . . . . . . . . . . . . . 119

5.1 Weak Formulation of the Surface-To-Surface Contact . . . . . . 119
5.2 Weak Formulation of 2D Contact Interaction . . . . . . . . . . . . . 122
5.3 Weak Formulation for Point-To-Curve Contact . . . . . . . . . . . . 122
5.4 Weak Formulation for Curve-To-Curve Contact . . . . . . . . . . . 123
5.5 Weak Formulation for Curve-To-Rigid Surface Contact . . . . . 127
5.6 Nitsche and Lagrangian Formulations of Non-Frictional
Contact Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.6.1 Choice of the Lagrange Multiplier Set μ . . . . . . . . . . . . 130
5.6.2 Physical Meaning of the Non-penetration Terms . . . . . 130
5.6.3 Types of the Nitsche Approach . . . . . . . . . . . . . . . . . . . . 132
6 Contact Constraints and Constitutive Equations for

Contact Tractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.1 Surface-To-Surface Contact – Non-Frictional Case and
Isotropic Coulomb’s Frictional Case . . . . . . . . . . . . . . . . . . . . . 136
6.1.1 Contact Constraints for Normal Contact
Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.1.2 Tangential Contact Conditions. Evolution
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.1.3 Return-Mapping Scheme for the Real Tangential
Traction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.1.4 Integration of Evolution Equations. Geometrical
Interpretation of the Return-Mapping Scheme . . . . . . . 142
6.2 Generalization of Coulomb Friction Law into Complex
Contact Interface Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.2.1 Vector Form of the Isotropic Equations . . . . . . . . . . . . . 143
6.2.2 General Interface Model . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.2.3 Anisotropic Yield Function . . . . . . . . . . . . . . . . . . . . . . . 146
6.2.4 Tensor Representations for Anisotropy . . . . . . . . . . . . . 148
6.3 Derivation of the Anisotropic Adhesion-Friction Contact
Problem via the Principle of Maximum Dissipation . . . . . . . . 156
6.3.1 Continuous Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.3.2 Incremental Formulation . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.3.3 Specification of Initial Conditions for the
Return-Mapping Scheme . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.3.4 Derivation of the Sliding Incremental Displacement
Δξ sl and Update Scheme for the History
Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.3.5 Computational Aspects for Further Implementation
Considering Nonlinear and Constant Tensors . . . . . . . . 163
XVI Contents
6.3.6 Geometrical Interpretation of the Solution Process . . . 164

6.3.7 Rheological Model of the Orthotropic Adhesion-
Friction Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.4 Analysis of Various Models for Anisotropic Friction and
Adhesion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.4.1 Orthotropic Coulomb Friction Law . . . . . . . . . . . . . . . . 169
6.4.2 Model for Orthotropic Contact Interfaces Including
Both Adhesion and Friction . . . . . . . . . . . . . . . . . . . . . . 171
6.4.3 Recovering Circular Motion for Polar Orthotropy . . . . 172
6.5 2D Contact Problems – Evolution Equations for Contact
Tractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.6 Constitutive Equations for Point-To-Curve Contact
Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
6.7 Curve-To-Curve Contact – Constitutive Equations for
Contact Tractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.7.1 Normal Contact. Specification of Constitutive
Laws for the Traction N Coupled with Contact
Constraints for the Variable r . . . . . . . . . . . . . . . . . . . . . 176
6.7.2 Tangential Contact. Specification of Constitutive
Laws for Tractions TI Coupled with Contact
Constraints for the Variables sI . . . . . . . . . . . . . . . . . . . 178
6.7.3 Rotational Contact. Specification of a Constitutive
Law for the Rotational Moment MI Coupled with
Contact Constraints for the Variables ϕI . . . . . . . . . . . 180
6.8 Curve-To-Curve Contact: Rate of Contact Forces in a
Covariant Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.8.1 Covariant Form for Sticking . . . . . . . . . . . . . . . . . . . . . . 182
6.8.2 Covariant Form for Sliding . . . . . . . . . . . . . . . . . . . . . . . 182
6.9 Curve-To-Rigid Surface Contact – Transformation of the
Contact Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7 Linearization of the Weak Forms – Tangent Matrices

in a Covariant Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.1 Linearization of the Weak Form for the Surface-To-Surface
Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.1.1 Linearization of the Normal Contact Part δWcN . . . . . 187
7.1.2 Tangent Matrices for the Normal Contact
Part δWcN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
7.1.3 Linearization of the Tangential Contact Part δWcT
– Case of Sticking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
7.1.4 Linearization of the Tangential Contact Part δWcT
– Case of Sliding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.2 Linearization of the Weak Form for the 2D Case . . . . . . . . . . 194
7.2.1 Tangent Matrix for the Normal Part . . . . . . . . . . . . . . 195
7.2.2 Tangent Matrix for the Tangential Part . . . . . . . . . . . . 196
Contents XVII
7.3 Linearization of the Weak Form for the Surface-To-Surface

Contact – Case of Coupled Anisotropic Adhesion-Friction
Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.1 Linearization of the Tangential Part δWcT . . . . . . . . . . 198
7.4 Linearization of the Weak Form for Point-To-Curve
Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
7.5 Linearization of the Weak Form for Curve-To-Curve
Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
7.5.1 First Part, Representing Geometrical Nonlinearity . . . 203
7.5.2 Constitutive Part for Sticking . . . . . . . . . . . . . . . . . . . . 204
7.5.3 Constitutive Part for Tangential Sliding . . . . . . . . . . . 205
7.5.4 Linearized Part for Rotational Sliding . . . . . . . . . . . . . 206
7.6 Linearization of the Weak Form for Curve-To-Rigid
Surface Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
8 Surface-To-Surface Contact – Various Aspects for

Implementations within the Finite Element Method . . . . . 209
8.1 Finite Element Discretization for Various Contact
Approaches – NTS, STS and STAS Contact Elements . . . . . . 209
8.1.1 Node-To-Segment (NTS) Contact Approach . . . . . . . . 210
8.1.2 Segment-To-Segment (STS) Contact Approach . . . . . . 211
8.1.3 Segment-To-Analytical Surface (STAS) Contact
Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8.2 Various Approximations of Contact Surfaces Defined by
Finite Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
8.2.1 Quadrilateral Segment with Linear Approximation . . . 214
8.2.2 Closed Form Solution for CPP Procedure for
Linear Segment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
8.2.3 Quadrilateral Segment with Quadratic Lagrangian
Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
8.2.4 Surface Smoothing Techniques in a Covariant
Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
8.3 Non Frictional Contact Analysis: Node-To-Segment
Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
8.3.1 Bending of a Beam over a Rigid Cylinder . . . . . . . . . . . 231
8.3.2 Bending of a Beam over a Rigid Sphere . . . . . . . . . . . . 234
8.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
8.4 Non Frictional Contact Analysis: Large Penetration
Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
8.4.1 Large Penetration Algorithm – General
Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
8.4.2 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
8.4.3 Wedge Plate Indented into Cantilever Plate . . . . . . . . . 239
8.4.4 Semicircular Plate Indented into Rectangular
Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
XVIII Contents
8.4.5 Bending of a Beam over a Rigid Cylinder . . . . . . . . . . . 243

8.4.6 Bending of a Beam over a Rigid Sphere . . . . . . . . . . . . 246
8.4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
8.5 Frictional Contact Analysis: Node-To-Segment Approach . . . 249
8.5.1 Global Solution Scheme. Summary of the Results . . . . 250
8.5.2 Sliding of a Block. Linear Approximation of the
Contact Surfaces. Two Types of a Frictional
Contact Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
8.5.3 Sliding of a Block. Quadratical Approximation of
the Contact Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
8.5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
8.6 Computation of Contact Integrals – Mortar Type
Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
8.6.1 Convergence Test for the Integration Algorithm:
Computation of the Energy Associated with the
Penalty Functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
8.6.2 Integration Schemes with Subdomains . . . . . . . . . . . . . 265
8.7 Segment-To-Segment (Mortar) Approach: Analysis of the
Patch Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
8.7.1 Classical Contact Patch Test – Linear
Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
8.7.2 Contact Patch Test with Smooth Surfaces . . . . . . . . . . 271
8.8 Segment-To-Analytical Surface (STAS) Approach: Various
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
8.8.1 Large Sliding on a Rigid Parabolic Cylinder . . . . . . . . 273
8.8.2 Free Bending of a Metal Sheet on Two Cylinders . . . . 275
8.8.3 Deep Drawing of a Cylindrical Pot – Combination
of STAS Contact Elements . . . . . . . . . . . . . . . . . . . . . . . 281
8.8.4 Deep Drawing – Test for the Quality of Shell
Elements as Well as for the Quality of the Contact
Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
8.8.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
8.9 Implementation of the Nitsche Approache . . . . . . . . . . . . . . . . 287
8.9.1 Gauss Point-Wise Substituted Formulation . . . . . . . . . 287
8.9.2 Bubnov-Galerkin-Wise Partial Substituted
Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
8.9.3 Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
9 Special Case of Implementation – Reduction into

2D Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
9.1 Finite Element Implementation of 2D Contact
Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
9.1.1 Linear NTS Contact Element . . . . . . . . . . . . . . . . . . . . . 294
9.1.2 Closed Form Solution for CPP Procedure . . . . . . . . . . . 295
9.1.3 Return-Mapping Scheme – 2D Case . . . . . . . . . . . . . . . . 296
Contents XIX
9.1.4 Treatment of Special Cases . . . . . . . . . . . . . . . . . . . . . . . 297

9.1.5 Update of the Sliding Displacements in the Case of
Reversible Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
9.1.6 Crossing an Element Boundary – Continuous
Integration Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
9.1.7 Remarks on Additional Developments . . . . . . . . . . . . . . 300
9.2 Frictional Contact Analysis: Requirements of Special
Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
Contact Surfaces. Reversible Loading Process . . . . . . . 304
9.2.2 Drawing of an Elastic Strip into a Channel with
Sharp Corners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
9.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
10 Implementation of Contact Algorithms with High

Order FE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
10.2 Finite Element Implementation of High Order
Contact FE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
10.2.1 Computation of Contact Integrals . . . . . . . . . . . . . . . . . 319
10.2.2 Contact Layer - Rigid Surface (CLRS) Finite
Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
10.2.3 Contact Layer - Contact Layer (CLCL) Finite
Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
10.2.4 Lagrange Multiplier Method for Normal Traction . . . . 322
10.2.5 Global Solution Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . 323
10.3 Numerical Examples for High Order Contact Layer Finite
Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
10.3.1 Loading Case 1. Contact Zone within One
Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
10.3.2 Loading Case 2. Contact Zone within Several
Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
10.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
11 Anisotropic Adhesion-Friction Models – Some

Particular Details of Implementation and Numerical
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
11.1.1 Various Models for Friction . . . . . . . . . . . . . . . . . . . . . . . 331
11.1.2 Available Finite Element Models . . . . . . . . . . . . . . . . . . 333
11.2 Examples of the Finite Element Implementation of the
Coupled Anisotropic Adhesion-Friction Model . . . . . . . . . . . . . 334
11.2.1 Point-to-Analytical Surface Contact Element.
Linear Surface Approximation of a
Deformable Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
XX Contents

Arbitrary Surface Approximation of the Deformable
Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
11.2.3 Node-to-Segment Approach. Deformable
Anisotropic Contact Surface . . . . . . . . . . . . . . . . . . . . . . 337
11.3 Numerical Examples Illustrating Effects of the Coupled
Anisotropic Adhesion-Friction Model . . . . . . . . . . . . . . . . . . . . . 339
11.3.1 Linear Constant Orthotropy on the Plane . . . . . . . . . . 339
11.3.2 Polar Orthotropy on a Plane. Large Displacement
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
11.3.3 Spiral Orthotropy on the Cylinder . . . . . . . . . . . . . . . . . 351
11.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
11.4 Symmetrization of Various Friction Models Based on an
Augmented Lagrangian Approach . . . . . . . . . . . . . . . . . . . . . . . 355
11.4.1 Structure of Matrices after the Linearization
Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
11.4.2 Augmented Lagrangian Method and Symmetric
Uzawa Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
11.5 Numerical Examples for Augmented Lagrangian Method . . . 362
11.5.1 Small Sliding Problem. Constant Orthotropy . . . . . . . . 363
11.5.2 Large Sliding Problem. Polar Orthotropy . . . . . . . . . . . 363
11.5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
12 Experimental Validations of the Coupled Anistropic

Adhesion-Friction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
12.2 Experimental Investigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
12.2.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
12.2.2 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
12.3 Calibration of Parameters for Different Models . . . . . . . . . . . . 372
12.3.1 Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
12.3.2 Case 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
12.3.3 Case 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
12.3.4 Case 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
12.3.5 Case 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
12.3.6 Calibration of the Theoretical Curve by Extremal
Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
12.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
13 Various Aspects of Implementation of the

Curve-To-Curve Contact Model . . . . . . . . . . . . . . . . . . . . . . . . . 381
13.1 General Structure of Tangent Matrices for CTC Contact . . . 381
13.1.1 Tangent Matrix for the Normal Force N . . . . . . . . . . . . 382
13.1.2 Tangent Matrices for Tangential Sticking . . . . . . . . . . . 383
13.1.3 Tangent Matrices for Tangential Sliding . . . . . . . . . . . . 383
Contents XXI
13.1.4 Tangent Matrices for Rotational Sticking . . . . . . . . . . . 383

13.1.5 Tangent Matrices for Rotational Sliding . . . . . . . . . . . . 383
13.2 Residual Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
13.2.1 Part for Normal Interaction . . . . . . . . . . . . . . . . . . . . . . . 384
13.2.2 Part for Tangential Interaction . . . . . . . . . . . . . . . . . . . . 384
13.2.3 Part for Moment (Rotational) Interaction . . . . . . . . . . 385
13.3 Combination with Various Finite Element Models of the
Continuum for Curve-To-Curve Contact . . . . . . . . . . . . . . . . . . 385
13.3.1 Implementation of Curve-To-Curve Contact
Algorithm – Linear Element for Edge-to-Edge
Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
13.3.2 Combination of Finite Beam Elements
with the Curve-To-Curve Contact Algorithm . . . . . . . . 389
13.3.3 Development of Special “Solid-Beam” Elements for
Beam-to-Beam Contact . . . . . . . . . . . . . . . . . . . . . . . . . . 391
13.4 Numerical Examples for Curve-To-Curve Contact . . . . . . . . . . 393
13.4.1 Bending of a Flexible Beam by a Rigid Beam.
Non Frictional Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
13.4.2 Analysis of Contact for Intersecting Beams . . . . . . . . . 398
13.4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
13.4.4 Bending of a Flexible Beam by a Rigid Beam –
Comparison of Three Finite Element Models . . . . . . . . 405
13.4.5 Contact between Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
13.4.6 Tying of a Knot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
13.4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
14 Frictional Interaction of a Spiral Rope and a Cylinder

– 3D-Generalization of the Euler-Eytelwein Formula
Considering Pitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
14.1 Equilibrium Equations for the Rope in Space . . . . . . . . . . . . . 413
14.2 Solution of the Equilibrium Equation for
a Spiral Line (Helix) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
14.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
List of Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
1
Introduction
Computational contact mechanics has been developed into a separated

branch of computational mechanics during the last decades. A fairly large
number of publications including several monographs on computational con-
tact mechanics has been published since then. Modeling of contact interac-
tions became fairly standard in numerous finite element software packages
available for engineers. Various aspects of the numerical solution has been
polished by many researchers. One of the important aspects, even though
being obvious for everyone, – the geometrical treatment of the contact – is
often remains hidden inside the computational algorithm. Contact interac-
tion from a geometrical point of view can be seen as interaction between
deformable surfaces possessing various geometrical features such as surfaces,
edges and vertexes. This monograph aims on the development of the unified
geometrically exact formulation of contact algorithm for various possible ge-
ometrical situations of contacting bodies: surface-to-surface, curve-to-surface
etc. Known computational contact algorithms will be reconsidered in accor-
dance with the geometry of contacting bodies in a form independent of their
approximation, in the so-called covariant form. This combination forms the
geometrically exact theory of contact interactions .
In order to locate this development in computational contact mechanics
focusing on its computational issues and formulate the goals of the current
book, first, we try to classify these developments.
1.1 Overview of Approaches to Model Contact

Problems
The overview of the current state of the art in computational contact mechan-
ics is arranged in such a way that the publications are classified by different
approaches used for modeling, by different solution methods etc., therefore,
a separated article can be mentioned in different subsections of the overview.
Limits and drawbacks are then discussed in a special subsection.
A. Konyukhov and K. Schweizerhof: Computational Contact Mechanics, LNACM 67, pp. 1–24.
springerlink.com c Springer-Verlag Berlin Heidelberg 2013
2 1 Introduction
In computational contact mechanics the modeling process depends gener-

ally on assumptions concerning the geometry of objects – 3D bodies, shells,
beams, ropes and other combinations – leading to appropriate computational
models. The robustness and tolerance of the computations depends on the
numerical method involved in contact modeling.
In order to classify known approaches the modeling process can be split into
several steps:
1. kinematics of contact interactions
2. constitutive relations for contact conditions
3. methods to enforce contact conditions
4. formulation of basic mechanical principles (conservation of momentum,
equilibrium etc.) on contact boundaries
5. numerical solution and analysis of the proposed model.
Here we are discussing these steps classifying various numerical methods
known in literature focusing mostly on the recent developments.
Step 1: Kinematics of Contact Interactions

In this step types of contact interactions together with measures of contact
interaction are defined similar to deformation measures in continuum me-
chanics. Regarding the relative motion of contacting bodies it is important
to distinguish the following interactions: a) tied contact and b) arbitrary con-
tact with b1) small and with b2) large relative displacements.
a) Tied contact requires simple “sewing” of two contacting bodies, and,
therefore, a point-wise enforcement of contact as
r = rA − rB = 0, (1.1)
assuming that vectors from a body A and a body B describe the same geo-
metrical point at the beginning of a deformation process, e.g. rA − rB = 0.
Enforcement of a tied contact can be regarded as additional constraints on
interface surfaces and has been introduced for the finite element method in
earlier developments as mesh tying algorithms. This has been solved by the
Lagrange multiplier method for dominantly straight contact boundaries al-
ready by Francavilla and Zienkiewicz [42] (1975). More recently mesh tying
algorithms are developed for the case of curved contact interfaces. Tan [172]
(2003) considered a special construction of contact matrices in 2D for mesh
matching and for the contact patch test. Various techniques based on the
construction of special volume elements after the projection procedure were
developed by Heinstein and Laursen [108] (2003), [65] (2003) for mesh tying
of arbitrary curved interfaces. Puso and Laursen [150] (2003) used the Mortar
method for mesh tying of curved interfaces in 3D.
b) Both normal and tangential displacements should be distinguished in
the case of arbitrary contact. This leads to a splitting of displacements into a
normal penetration and into tangential displacements. In early publications,
1.1 Overview of Approaches to Model Contact Problems 3
see Kikuchi and Oden [84] (1988), a normal penetration has been introduced as
a result of the linearization of nonlinear non-penetration conditions eqn. (1.1)
in the direction of the normal to the surface of a contacting body. Hallquist
et.al. [52] (1985) introduced this split, and therefore, a penetration via the
projection operation. Thus, a, so-called, “master-slave” algorithm within an
explicit time integration scheme for dynamic problems has been introduced.
For applications with implicit schemes the “master-slave” algorithm has then
been modified in Wriggers and Simo [194] (1985) and in Simo et.al. [163]
(1986). The projection can be formulated in extremal form as
rs − ρ −→ min . (1.2)
The “master-slave” approach is treating the normal contact such that a

“slave” node rs of one body penetrates into a “master” segment of another
body parametrized with ρ. The value of penetration p is measured as a clos-
est distance to the master body and, therefore, in the direction of the normal
n of the “master” surface
p = (rs − ρ) · n. (1.3)
This measure can be defined differently depending either on the geometry of

contacting bodies, or on further application of various numerical methods.
a) For contact between surfaces the penetration can be defined in the fol-
lowing ways:
i in the direction of the normal n to a rigid undeformed body for Sig-
norini problems as e.g. in Kikuchi and Oden [84] (1988);
ii in the direction of the normal n to an initially undeformed surface
for small displacement problems, see the “node-to-node” approach in
Wriggers [190] (2002).
iii in the directions of normals to spheres around certain nodes repre-
senting the contacting body, see the “pinball algorithm” in Belytschko
and Neal [10] (1991).
iv in the directions of the normal to a master segment with the as-
sumption that the segment is updated during the deformation. This
is called “node-to-segment” approach, first published in Hallquist
et.al. [52] (1985) for explicit time integration and in Wriggers and
Simo [194] (1985) for the implicit scheme. An alternative is the
“segment-to-segment” approach with the projection of integration
points from a slave segment, see Zavarise and Wriggers [202] (1998)
and Harnau et.al. [56] (2005); for a Mortar approach with segment-
wisely defined normals, see Puso and Laursen [151], [152] (2004); for a
Mortar approach with point-wisely defined normals, see Fischer and
Wriggers [40] (2006).
b) For contact between beams the penetration is defined in the direction of
mutual normals assuming a shape of a cross sections which may vary.
4 1 Introduction
c) For contact between particles for discrete models, see Cundall and Strack
[171] (1979), the penetration is defined as distance between the centers of
the particles subtracting the distance between corresponding centers and
boundaries of the particles. Similar models for contact are used in the
“Smoothed particle hydrodynamics” (SPH) method, see e.g. in Libersky
and Petschek [110] (1991) and Campbell et.al. [25] (2000). For a specially
shaped geometry, however, an exact measure is necessary, thus, a special
projection procedure has been developed in Wellmann et.al. [181] (2008)
in order to compute the distance between particles shaped as superellip-
soids.
d) The measure of contact can be defined as a mutual volume of overlapping
contacting bodies arising from inequality constraints. This approach is
often used for contact between rigid bodies, see Kane et.al. [83] (1999)
and Pandolfi et.al. [137] (2002).
e) A measure of the tangent interaction Δρ can be formally introduced
via a projector operator constructed by subtracting a dyadic product of
normals from the unit tensor:
Δρ = (I − n ⊗ n)(rA − rB ). (1.4)
This approach is widely used e.g. in Wriggers and Simo [194] (1985),
Parisch and Lübbing [139], (1989) Peric and Owen [140] (1992) together
with a penalty regularization, in Heegaard and Curnier [62] (1993), Heege
and Alart [64] (1996) for Augmented Lagrangian methods and in McDe-
vitt and Laursen [124] (2000), Puso [148] (2004), Puso and Laursen
[151, 152] (2004) for the Mortar method.
The main difficulties of such a representation are the following:
• The absence of a geometrical interpretation leads to difficulties for
the transformation of history variables.
• Some difficulties in the linearization – transformations are required
to show that it would lead to a relative velocity vector. The linearized
results are very complex expressions for the tangent matrices, see the
book of Wriggers [190] (2002) for surface-to-surface contact, Litewka
and Wriggers [114, 115] (2002) for beam-to-beam contacts.
f) Another approach to describe a tangential interaction is to consider con-
vective variables arising from the surface approximations: see Simo et.al.
[163] (1986), Wriggers at.al. [195] (1990), Laursen and Simo [109] (1993),
Laursen [105] (1994) and recently in Konyukhov and Schweizerhof [87]
(2004), [88] (2005), [91] (2006) with
Δρ = Δξ i ρi . (1.5)
Then two convective variables Δξ 1 , Δξ 2 in a surface covariant basis ρ1 , ρ2

are introduced as tangential measure.
This approach has many advantages:

• objectivity is straightforwardly observed because the surface coordi-
nates ξ i are used.
• geometrical interpretation of a tangential contact measure is easily
possible – line on a surface; geometrical interpretation of a linearized
measure – relative tangent velocity of a contact point.
• the number of history variables is minimal (two for surface interac-
tion).
• a complex constitutive law for the tangent interaction can be easily
formulated in a robust form for computation.
• expressions for contact tangent matrices are by far less complex
within the fully covariant approach than in schemes as in eqn. (1.4).
The last method, however, requires a lot of preliminary transformations based
on the differential geometry of contacting objects (surfaces or even curves)
and extensive application of the tensor analysis especially for differential op-
eration and linearization. The complete transformation in a covariant form is,
however, absolutely necessary – the lack of such transformation leads again
to a very complex form of tangent matrices, see in Laursen [104] (1992), in
Laursen and Simo [109] (1993) and in the monograph of Laursen [106]. More-
over, it can even lead to the paradox result such as “the lack of symmetry in the
stick linearization” for 3D cases reported in Laursen and Simo [109] [P. 3468]
(1993).
Step 2: Constitutive Relations for Contact Conditions

A term “constitutive relations” here is understood in a continuum mechan-
ics sense, however, being formulated as a constitutive law only between con-
tacting surfaces. Thus, the relations are discussed separately for normal and
tangential tractions. They can be formulated in several ways, but essentially
two types can be distinguished: a) when a surface has randomly distributed
asperities, and b) when asperities have algorithmic structure with prescribed
geometrical and mechanical properties, e.g. the surface shows different macro
properties in different directions. Statistical analysis is applied for the model
of the first type in order to introduce and describe mechanical characteristics
of rough surfaces observed in experiments. Various theoretical generalizations
based on phenomenological observations are the basis for the model of rough
surfaces of the second type. We outline here both approaches with the focus
on computational analysis.
Statistical Friction Models
Statistical analysis of a real rough surface and experimental aspects of its

measurements have been developed in a series of publications: Longuet-
Higgins [116], [117] (1957), Greenwood and Williamson [49] (1966), White-
house and Archard [182] (1970) and more recently Whitehouse and Phillips
6 1 Introduction
[183] (1978), [184] (1982) and Greenwood [47] (1984). A comparative analysis
of such surface models is presented in McCool and John [123] (1986). These
statistical models formed a basis to construct a set of non-linear laws for nor-
mal compliance which has been then used for normal contact in the context of
finite element methods: see Zavarise et.al. [205] (1992), Wriggers and Zavarise
[197] (1993), [196] (1993) for the contact with asperities distributed by the
Gaussian normal distribution law. Numerical tests for particular statistical
models have been considered in Willner and Gaul [186] (1995), by Willner
[185] (1997), [187] (1997) with experimental verification in Willner and Görke
[188] (2006). Buczkowski and Kleiber [23] (1999) considered first an isotropi-
cal statistical distribution of asperities, and then in [24] (2000) an anisotropic
statistical distribution of asperities. The novel combined approach is based
on a multi-scale technique. In this case macro characteristics for friction are
obtained via a micro model including randomly generated surfaces with cer-
tain statistical characteristics, see also in Bandeira, Wriggers and Pimenta
[9] (2004) and in [8] (2005).
Phenomenological Friction Models
For this type of models the asperities possess an algorithmic structure and
various generalizations of the isotropic macro characteristics have been pro-
posed. A generalization can be derived separately for normal and tangential
tractions.
Various interface non-linear constitutive relations for normal traction have
been studied in Paggi et.al. [136] (2005). A generalization of a Coulomb fric-
tion law with regard to Maxwell and Kelvin viscoelastic models was con-
sidered in Araki and Hjelmstad [5] (2003). In the context of the discrete
element method various models have been discussed in Luding [118] (2007)
and models including normal adhesion can be found in Tomas [177] (2007).
A set of frictional models is obtained when various constitutive models are
applied for tangential traction. One of the first anisotropic friction models has
been proposed by Michalowski and Mroz [128] (1978) considering the sliding
of a rigid block on an inclined surface. A model with periodically inclined
asperities has been used in Mroz and Stupkiewicz [130] (1994). Zmitrowicz in
[213] (1981) described various anisotropic structures of the friction tensor for
the generalization of the Coulomb friction law and then developed a classi-
fication of anisotropic surfaces in [214] (1989). A rate independent theory of
anisotropic friction is developed in Curnier [31] (1984) and thermodynamical
restrictions for the friction tensor are discussed in He and Curnier [61] (1993).
Finite element models of anisotropic friction are presented as follows: with
an application to hot rolling processes these have been discussed in Mont-
mitonnet and Hasquin [129] (1995) and also in Alart and Heege [4] (1995).
Buczkowski and Kleiber [22] (1997), [23] (1999), [24] (2000) created an in-
terface element containing an orthotropic sliding law. Hjiaj et.al. [67] (2004)
formulated an anisotropic friction problem via the bi-potential and applied
Lagrangian multiplier methods. Parametric quadratic programming was used

in Zhang et.al. [206] (2004) to solve the almost identical problem. Jones and
Papadopoulos [82] (2006) developed a finite element model for anisotropic
friction, where the stick-slip condition is enforced via Lagrange multipliers
for small relative sliding.
Adhesion Models
In the earliest publications adhesion phenomena have been introduced in nor-

mal direction as force needed to detach one body from the other where lubri-
cation plays an essential role, see Hardy and Nottage [54, 55] (1926). Exper-
imental investigations of tangential adhesion for different pairs of materials
are described in McFarlane and Tabor [125] (1950). The relation between fric-
tion force and normal adhesion is experimentally investigated in McFarlane
and Tabor [126] (1950). The normal adhesion between clean metal surfaces
is experimentally investigated in Bowden and Rowe [20] (1956). Here tan-
gential adhesion forces have been observed in addition to the frictional force.
Also a simplified theoretical model was proposed which describes this elastic
force as a reaction of micro spherical asperities. Fuller and Tabor [43] (1975)
made experiments with rubber surfaces with spherical asperities to show the
influence of the surface roughness on normal adhesion. Surface adhesion on
elastic spherical asperities has been introduced to describe global adhesion
forces in Johnson, Kendall and Roberts [78] (1971). This model (so-called,
JKR-model) has been developed further describing the coupling of normal
adhesion and friction forces, see Johnson [76] (1997) and also Greenwood
[48] (1997). The influence of the roughness of a different scale on normal
adhesion is studied by Persson and Tosatti in [142] (2001). Concerning appli-
cations on different scales, a method of non-direct determination of adhesive
and elastic properties of contacting materials is proposed for various adhe-
sion models (JKR and others) for nano-indentation, Borodich and Keer [18]
(2004), Borodich and Galanov [17] (2008).
Coupled Adhesion-Friction Models
A constitutive model of tangential adhesion as a reversible elastic force has

been discussed in Curnier [31] (1984). Complex models where adhesion and
friction forces are un-coupled appear recently in computational mechanics.
For statics Raous, Cangemi and Cocou in [154] (1999) introduce a model
including a coupling of a regularization term called tangential adhesion and
friction. Also an adhesion term representing elastic effects is considered in
normal direction and introduced via additional internal variables. The fric-
tional force and the normal adhesion are coupled via a yield friction func-
tion of Coulomb type. The model has been developed further in Cocou,
Raous and Schryve [29] (2006) including the coupling of these adhesion terms
and viscous forces in dynamics. A coupled model including sliding forces in
8 1 Introduction
accordance with the Coulomb friction model and viscoelastic tangential forces
analogous to the tangential adhesion above in accordance with the Maxwell
and Kelvin model was considered in Araki and Hjelmstad [5] (2003).
The discussed coupled models include only isotropic constitutive relations.
Thus, a coupled interface contact model has been developed in Konyukhov
and Schweizerhof [89], [90]. This model includes anisotropy for both the fric-
tion and the tangential adhesion region and is derived via the application of
the covariant approach. A general approach for the combination of a complex
friction law together with a complex adhesion law has been proposed. The
approach necessarily leads to a computational model which allows a straight-
forward implementation into a finite element code. The important point is
the correlation of such a model to experimentally observed phenomena. A
first simple experimental test for a corrugated rubber mat has been reported
in [97]. The importance of the coupled model has been shown experimen-
tally in the so-called, “geometrically isotropic case”, where the combination
of both anisotropy for adhesion and anisotropy for friction can even lead to
geometrically isotropic behavior of the sliding body. In [98] this effect was
shown experimentally for a rubber surface possessing a periodical waviness.
Motivation of Interface Surface Models from Computational Tribology
In experimental tribology the physical properties between contacting bod-

ies are described via, a so-called, third body filling the space between the
contacting bodies and possessing specific properties, though, differing from
the properties of the contacting bodies. Recent developments in computa-
tional Molecular Dynamic (MD) simulations, see for 2D-simulations at the
nano-scale [85] (2007), have shown that the behavior of such a third body
(tribomaterial) during contact is equivalent to models known in computa-
tional mechanics, besides of course, the nano-effects. Another simplified MD
model described as movable cellular automata is also used for simulations
of third body behavior see Popov and Psakhie [147] (2007). A discussion of
several models for interfaces, classical friction laws and dislocations can be
found in Merkle and Marks [127] (2007).
Thus, the interface surface law can be also considered as an upper level
in the chain of models: nanotribology, ... , micro-structural model, interface
model. It also serves as a theoretical basis for the treatment of various inter-
face surface laws such as a constitutive law reduced to the surface.
The advantages of the covariant approach is especially pronounced for com-
plex contact interface laws. The coupled model including anisotropic friction
and tangential adhesion, proposed in Konyukhov and Schweizerhof [89], [90]
(2006), is directly formulated in a covariant form as a computational model
for an iterative Newton type solver. The approach allows the straightforward
coupling of various interface laws (plasticity, elasticity, viscosity etc.) for-
mulated in an arbitrary curvilinear coordinate system given on the contact
surface.
General discussions and reviews on constitutive models used in contact

mechanics can be found in the monographs of Laursen [106] (2002), Persson
[141] (2000), Sextro [157] (2002) and Wriggers [190] (2002).
Step 3: Methods to Enforce Contact Conditions

Several numerical methods are known in literature in order to enforce contact
conditions. These method are used to resolve the contact conditions formu-
lated as inequalities in the form of Kuhn-Tucker conditions. The terminology
is initially found in the optimization methods, see e.g. the monographs of
Bertsekas [13] (2003) and Borwein and Lewis [19] (2000). These methods can
be classified by the structure of the functionals enforcing these conditions
in a weak form: the Lagrange multiplier method, the Penalty method, the
Nitsche method and the Augmented Lagrangian method.
Lagrange multipliers together with a trial-and-error procedure are intro-
duced to describe additional contact nodal forces in the some first finite
element models, see Francavilla and Zienkiewicz (1975) and Hughes at.al.
[71] (1976). Cescotto and Charlier [27] (1993) considered four mixed vari-
ational formulations of frictional contact leading to contact elements with
independent approximation of stress and displacement fields. Taylor and Pa-
padopoulos in [174] (1991), Solberg and Papadopoulos in [164] (2005) studied
this method in order to find out so-called Babuska-Brezzi stable cases of La-
grange multiplier schemes leading to the satisfaction of the contact patch
test. For various orders of approximations, the patch test has been studied
in Crisfield [30] (2000). A patch-test for the small displacement frictional
case is studied in Jones and Papadopoulos in [79] (2000). Harnau et.al. [56]
(2005) used various integration techniques for the contact integral within “a
segment-to-segment” approach to improve the patch-test.
A special technique for the approximation of the Lagrange multipliers is
the so-called Mortar method with dual multipliers, see Hüeber and Wohlmuth
[69] (2005), though the Mortar method is not restricted to the enforcement
via the Lagrange multipliers only, another application will be discussed later.
For a small sliding problem, a perturbed Lagrangian method including both
penalty and Lagrangian functionals was discussed in Wriggers and Simo [194]
(1985).
Heintz and Hansbo [66] (2006) constructed a special functional including
not only penalized displacements, but also penalized normal stresses from
both contacting sides, the so-called Nitsche method, and obtained estimations
for the penalty parameter necessary for convergence. The method is proved to
be independent on the penalty parameter. However, due to the cumbersome
formulation its finite element implementation have been considered only for
2D problems, see in Wriggers and Zavarise [199] (2008) and in Oliver et.al.
(part 1) [133] and in Hartmann et.al. (part 2) [57] (2009).
The Penalty method theoretically discussed in detail in Kikuchi and
Oden [84] (1988) is among the most popular methods for finite element
10 1 Introduction
implementations. Both normal and tangential displacements have to be pe-

nalized with a large penalty parameter in order to enforce contact conditions.
This leads from a mechanical point of view to an “additional spring” acting
at the contact point as proposed in Hallquist et.al. [52] (1985) for an explicit
time integration scheme. The essential stage of the implicit solution scheme
is an iterative solution of e.g. Newton’s type, which requires, in due course,
consistent linearization. This procedure has been introduced for the 2D non-
frictional case in Wriggers and Simo [194] (1985). The penalty enforcement for
non-frictional cases is discussed also in Parisch [138] (1989) within a covariant
approach. A special linearization approach for the penalty term in the form
of covariant derivatives in the local coordinate system has been introduced in
and in Konyukhov and Schweizerhof [87] (2005). In order to compute correct
tangent forces for frictional problems Wriggers et.al. [195] (1990) used the
return-mapping algorithm to compute correct tangent forces for frictional
problems. The penalty formulation in a global coordinate system is used in
Peric and Owen [140] (1992) and Parisch and Lübbing [139] (1997). Starting
with Laursen and Simo [109] (1993) (also in Laursen [104] (1992) and [105]
(1994)) convective coordinates are directly used for regularization. The fully
covariant description in the coordinate system defined by the closest point
projection is developed in Konyukhov and Schweizerhof [88] (2005).
The penalty method is applied not only for contact between surfaces, but
also for contact between other objects such as beams. Thus, contact be-
tween straight beams with circular cross sections is considered in Wriggers
and Zavarise [198] (1997) for non-frictional contact and then in Zavarise and
Wriggers [203] (2000) for frictional contact. Straight beams with rectangu-
lar cross-sections have been studied in Litewka and Wriggers [114] (2002)
and in [115] (2002) for both frictional and non-frictional cases respectively.
The Lagrange multiplier method for a beam-to-beam contact is discussed in
Litewka and Wriggers [113] (2003). A special treatment (so-called “slide-line
contact”) also based on the penalty method for the straight rod/continuum
contact is developed in Maker and Laursen [121] (1994).
The Mortar method together with penalty regularization with a specially
defined Mortar projection is used in Puso and Laursen [151] (2004) for non-
frictional and in [152] (2004) for frictional contact. The Augmented La-
grangian scheme is exploited to enforce frictional forces in [152]. Various
approximations for contact surfaces concerning the order of the surface de-
scription within the Mortar method have been studied in Fischer and Wrig-
gers [40] (2006).
An Augmented Lagrangian method allowing to satisfy contact constraints
with prescribed tolerance, has been thoroughly studied in Pietrzak and
Curnier [145] (1999). The problem has been formulated as a saddle point
optimization. Convergence of various Augmented Lagrangian formulations in
contact is studied by Stadler [168] (2007). A Mortar method together with
Augmented Lagrangian enforcement of tangent forces is used in Hüeber et.al.
[68] (2006).
A symmetrization procedure based on a nested Uzawa algorithm has been

used in Simo and Laursen [162] (1992) in order to obtain symmetric tan-
gent matrices for the frictional case. The Uzawa symmetrization technique
has been extended also for a coupled anisotropic adhesion-friction model in
Konyukhov and Schweizerhof [93] (2007).
Remark 1.1. The application of various numerical methods requires a-priory

knowledge about existence and uniqueness of the solutions of the contact
mechanics problems. There is a large number of publications on this sub-
ject: more recent results are summarized in monographs of Han and So-
fonea [53] (2002) and Shillor, Sofonea and Telega [158] (2004), see also refer-
ences therein.
Remark 1.2. In a geometrically exact theory the discussed solution methods

depending on a geometrical feature (surface, curve or point) should be for-
mulated in the covariant form according to the corresponding geometrical
feature (e.g. in a Frenet coordinate system for a beam-to-beam contact etc.).
Step 4: Formulation of Basic Principles (Conservation of

Momentum, Equilibrium etc.) on Contact Boundaries
In this step fundamental equations such as balance of momentum etc. are for-
mulated either in strong, or in weak forms. The form of differential equations
(the strong form) is used directly in computations only in discrete models,
see in Strack and Cundall [171] (1979), while the weak form is a key issue
for the application of the finite element method as well as so-called various
meshless methods. The current contribution is aimed at the development
of computational algorithms in applications with the finite element method,
therefore, we mention here only a specific problem in dynamic computations,
namely, energy and momentum conserving algorithms for contact problems.
Various integrations schemes on contact boundaries have been proposed in
Laursen and Chawla [107] (1997) and in Armero and Petöcz [6] (1998) for
non-frictional problems, and then, so-called, entropy dissipative schemes in
Chawla and Laursen [28] (1998) and in Armero and Petöcz [6] (1999) for fric-
tional problems. The discrete null space method developed earlier in Betsch
[14] (2005) for continuum cases is extended for contact problems in Betsch
and Hesch [15] (2007).
Step 5: Numerical Solution and Analysis of the Proposed Model

In this step the equations defined for the discretized model are solved nu-
merically. Among numerous problems such as selection of methods, storage
of data, parallel algorithms we consider only two particular problems typical
for contact algorithms: the contact searching algorithm and the consistent
linearization necessary for implicit solvers.
12 1 Introduction
Global Searching Algorithms
The computation of contact-impact problems for structures undergoing large

deformations requires extensive searching of contact pairs, that is why robust
searching algorithms are necessary. Benson and Hallquist [11] (1990) proposed
a single surface contact algorithm effectively describing self-contact. Zhong
and Nilsson [209] (1989), [210] (1996) developed a hierarchical algorithm care-
fully treating contact cases with edges, see also the book of Zhong [208]
(1993). The Pinball algorithm proposed by Belytschko and Neal [10] (1991)
is based on searching in spherical regions around certain nodes. Schweizerhof
and Hallquist [156] (1992) described a searching strategy based on an addi-
tional subdivision of the searching space. Various searching algorithms based
on hierarchical subdivisions are discussed also in Feng and Owen [39] (2002),
Bruneel and Rycke [21] (2002). Especially for blow molding processes sev-
eral algorithms have been developed: the inside-outside algorithm by Wang
and Nakamachi [180] (1997); the moving box algorithm by Wang and Maki-
nouchi [179] (2000); the Free-Form-Surface (FFS) algorithm for modeling the
contact surface with C 1 -continuity by Wang et.al. [178] (2001). Numerical
aspects of searching especially for parallel computation are discussed in the
thesis of Persson [143] (2000). Various global searching techniques in combi-
nation with both finite element and discrete element method are considered
in monograph of Munjiza [131] (2004).
Consistent Linearization
Two approaches for linearization of the final functional representing the work
of contact tractions can be distinguished in order to obtain consistent tangent
matrices. The direct approach follows the following sequence: functional –
discretization – linearization and the covariant approach follows the rule:
functional – linearization – discretization.
The direct approach assumes that the discretization is then involved in
the process and the linearization is provided with regard to the displacement
vector u and, therefore, of the discretized system. This leads to the final
results containing a set of approximation matrices. Various aspects of this
approach are considered in the following publications:
a) for surface-to-surface contact in Wriggers and Simo [194] (1985), Wrig-
gers et.al. [195] (1990), Wriggers [189] (1995), Parisch [138] (1989),
Parisch and Luebbing [139] (1997), Peric and Owen [140] (1992), Simo
and Laursen [162] (1992), Laursen [104] (1992), Laursen and Simo [109]
(1993);
b) for anisotropic friction between surfaces in Alart and Heege [4] (1995);
c) for beam-to-beam contact in Wriggers and Zavarise [198] (1997), Zavarise
and Wriggers [203] (2000), Litewka and Wriggers [114] (2002), Litewka
and Wriggers [115] (2002).
1.2 Discussion – Why Covariant Approach? 13
The complexity in the derivation for curved contact interfaces led to the com-
bination of a finite element code with a mathematical software for automatic
derivations, see for surface-to-surface contact in Heege and Alart [64] (1996),
Stadler, Holzapfel and Korelc [170] (2003), Krstulovic-Opara and Wrig-
gers [100] (2001), Krstulovic-Opara and Wriggers [101] (2002), Krstulovic-
Opara, Wriggers and Korelc [102] (2002), for anisotropic friction in Mont-
mitonnet and Hasquin [129] (1995) and for beam-to-beam contact in Litewka
[112] (2007) and in [111] (2007).
The fully covariant approach, however, assumes only a local coordinate sys-
tem associated with the deformed continuum (convective coordinates) and re-
quires extensive application of covariant operations (derivatives etc.). Though
the convective coordinates were partially involved in some publications men-
tioned above, the full description has been first derived in Konyukhov and
Schweizerhof [87] (2004) for non-frictional; in [88] (2005) for frictional contact;
a specific consideration for 2D cases in [91] (2006); for the coupled adhesion-
friction anisotropic contact in [89] (2006) and in [90] (2006); in an applica-
tion of the Augmented Lagrangian approach for the coupled adhesion-friction
anisotropic contact in [93] (2007); for arbitrary contact between curves rep-
resenting beam and cable type structures in [95] (2009). Concerning the final
complexity of the derived algorithms the covariant approach leads to a simple
structure of the contact matrices. Other advantages of the covariant approach
are discussed in detail in the Section 1.4.1 describing the structure of the cur-
rent work.
1.2 Discussion – Why Covariant Approach?

Through all considered steps from 1 to 5 two general approaches can be dis-
tinguished: a) covariant approach when a contact description is fully based on
local surface coordinate systems and b) direct approach (non-covariant one)
when all parameters are described with the assumed finite element approx-
imations in a global coordinate system. The latter is historically motivated
by the development of the finite element method, while the fully covariant
approach, though, is intended for the finite element method, but does not
assume approximations from the beginning.
The covariant approach serves to describe all parameters necessary for
the solution based on the geometry of the contacting bodies in the local
coordinate system.
Open questions and drawbacks of the direct approach can be summarized
as follows:
• A closed form for tangent matrices is available only for linear approx-
imations of surfaces. For curved interfaces, either a form depending on
approximations (mathematical software), or a form of simplified matrices
(taken for linear approximations) is reported.
14 1 Introduction
knife
banana
Fig. 1.1 Various geometrical situations in contact lead to different contact al-
gorithms: Surface-To-Surface, Curve-To-Surface, Curve-To-Curve, Point-To-Curve
and Point-To-Point
Vertex banana
Vertex knife
Fig. 1.2 Point-To-Point (PTP) contact algorithm

1.2 Discussion – Why Covariant Approach? 15
Knife
Banana
Fig. 1.3 Curve-To-Curve (CTC) contact algorithm
Fig. 1.4 Surface-To-Surface (STS) contact algorithm
Fig. 1.5 Surface-To-Surface (STS) contact algorithm: transfer of variables between

elements.
16 1 Introduction
Edge of knife
Fig. 1.6 Curve-To-Surface contact as Surface-To-Surface algorithm
Edge of knife
Fig. 1.7 Curve-To-Surface contact as Point-To-Curve algorithm
• The structure of the derived matrices is very complicated and often in-
transparent. There is no clear interpretation of each part possible. Thus,
simplifications are hardly possible.
• A specification of complex contact interface laws with properties explic-
itly depending on the surface geometry (e.g. arbitrary anisotropy) is not
possible.
• A contact description of many geometrical features (curve-to-curve, curve-
to-surface) is almost not possible because of the necessity of convective
surface coordinates.
• Geometrically motivated measures of contact interaction are coupled
with convective variables in a specially defined coordinate system. This
straightforwardly leads to a description via convective variables. How-
ever, in the direct approach, they are not defined separately for various
1.4 Goals and Structure of the Book 17
geometrical objects such as surfaces, edges etc. and can be handled –

mostly improperly – with great efforts.
1.3 On Geometrical Approaches in Contact Mechanics

Contact interaction from a geometrical point of view can be seen as an inter-
action between deformable surfaces and, therefore, geometrical approaches
can be exploited. However, there are only a few publications uncovering
geometrical issues to some extent. Gurtin et.al. [50] (1998) considered sur-
face tractions on curvilinear interfaces describing them from a geometrical
point of view. Jones and Papadopoulos [81] (2006) considered contact de-
scribing various mappings from the reference configuration employing the Lie
derivative. Laursen and Simo [109] (1993) and Laursen [105] (1994) described
some contact parameters via geometrical surface parameters. Heegaard and
Curnier [63] (1996) considered geometrical properties of slip operators. The
geometrical approach is obvious for such applications as isogeometric analy-
sis, various aspects in contact applications for Surface-To-Surface applications
recently considered by Temizer, Wriggers and Hughes in [175], [176].
1.4 Goals and Structure of the Book

In order to formulate goals and describe the complete structure of the book we
consider a model contact problem with two bodies possessing not only smooth
surfaces, but also various geometrical features such as edges and vertexes –
an example of this is a banana and a knife shown in Fig. 1.1. Considering all
possible geometrical situations in which knife and banana can contact each
other, the following hierarchical sequence of contact pairs is appearing.
Possible contact pairs:

1. Point-to-Point (PTP) contact pair, see Fig. 1.2
2. Point-to-Curve (PTC) contact pair, see Fig. 1.7
3. Point-to-Surface (PTS) contact pairs, Figs. 1.4 and 1.5.
4. Curve-to-Curve (CTC) contact pair, see Fig. 1.3
5. Curve-to-Surface (CTS) contact pair, see Figs. 1.6 and 1.7
6. Surface-to-Surface (STS) contact pair, see Figs. 1.4 and 1.5.
Summarizing the literature review and the discussion in the previous sections
the main goal of the current work can be formulated as follows:
Development of the unified geometrical formulation of contact conditions
in a covariant form for various geometrical situations of contacting bodies
leading to contact pairs: Surface-To-Surface (STS), Curve-To-Surface (CTS),
Point-To-Surface (PTS), Curve-To-Curve (CTC), Point-To-Curve (PTC) and
Point-To-Point (PTP) and formulating corresponding computational algo-
rithms in a covariant form for those pairs. The description will be fully based
18 1 Introduction
on the differential geometry of specific features forming a continuum, because

it is carried out in the local coordinate systems attached to this feature: in
the case of a surface – in the Gaussian surface coordinate system; in the case
of a curve – in the Serret-Frenet basis; in the case of a point – in the coor-
dinate system standard for rigid body rotation problems (e.g. via the Euler
angles). This general description is forming a geometrically exact
theory for contact interaction.
1.4.1 Structure of the Current Book

The current work is appearing as a result of the several years research at
Karlsruhe Institute of Technology (formerly known as Karlsruhe University
(TH)) which has led also to the cumulative habilitation [86].
The current section gives to the reader the general overview of the geo-
metrical structure of contact pairs and describes the complete structure of
particular details written in the following chapters.
A consistent development of the geometrically exact theory of contact in-
teraction requires the knowledge of the differential geometry. Thus, the book
starts with a short but necessary overview of the differential geometry of sur-
faces and curves in Chapter 2. The reader familiar with this mathematical
apparatus can easily skip this chapter.
As mentioned in the literature overview, the closest distance has become
a natural measure of the contact interaction. This measure is introduced via
the closest point projection (CPP) procedure, for which the solution requires
the differentiability of the function representing the parametrization of the
surface of contacting bodies. An analysis of the solvability for the CPP pro-
cedure allows then to classify all types of all possible contact pairs given
above. Chapter 3 starts, first, with a consideration of the CPP procedure
for C 2 -continuous surfaces. The concept of the projection domain is intro-
duced as a domain from which any point can be uniquely projected, and
therefore, the contact algorithm can be further constructed. This domain
can be constructed for at the utmost C 1 -continuous surfaces. If the surfaces
contain edges and vertices’s then the CPP procedure should be generalized
in order to include the projection onto edges and onto vertices’s. The criteria
of uniqueness and existence of these projection routines and corresponding
domains are studied in detail.
The main idea for application for contact is then straightforward – the
CPP procedure corresponding to a certain geometrical feature gives rise to a
special, in general, curvilinear 3D coordinate system. This coordinate system
is attached to a geometrical feature and its convective coordinates are directly
used for further definition of the contact measures. Thus, all contact pairs
listed above can be described in the corresponding local coordinate system.
The existence requirement for the generalized CPP procedure leads to the
transformation rule between types of contact pairs according to which the
corresponding coordinate system is chosen. Thus, all contact pairs can be

uniquely described in most situations.
A Surface-To-Surface contact pair, see Fig. 1.4, is described via the well
known “master-slave” contact algorithm based on the CPP procedure onto
a surface. This projection allows to define a coordinate system as follows:
r(ξ 1 , ξ 2 , ξ 3 ) = ρ(ξ 1 , ξ 2 ) + ξ 3 n(ξ 1 , ξ 2 ). (1.6)
The vector r is a vector for the “slave” point, ρ is a parametrization of the

“master” surface, n is a normal to the surface. The slave point is taken to
be a node from the finite element mesh for the well-known Node-To-Segment
algorithm, however, the algorithm is applicable mainly to linear approxima-
tions due to failing the patch test. For arbitrary approximations Segment-
To-Segment, or Surface-To-Surface algorithms are taken, in which an inte-
gration point is taken as a slave point. Such algorithms will be discussed later,
however, it is obvious that the kinematics for both the Surface-To-Surface,
and the Point-To-Surface contact algorithms are described in a coordinate
system given in eqn. (1.6). Thus, Point-To-Surface and Surface-To-Surface
contact pairs are equivalent. Eqn. (1.6) describes, in fact, a curvilinear coor-
dinate transformation where convective coordinates are used for a measure
of contact interaction: ξ 3 is a penetration, Δξ 1 , Δξ 2 are measures for tangent
interaction. The algorithm is applied only in the existence domain for the
surface CPP procedure.
Consideration of the existence of the CPP procedure for edges allows to
define then the Point-To-Curve contact algorithm used for the Point-to-Curve
contact pair directly as well as for the Curve-To-Surface contact pair, see
Fig. 1.7. The local coordinate system is constructed using as follows:
r(s, r, ϕ) = ρ(s) + re(s, ϕ); e = ν cos ϕ + β sin ϕ. (1.7)
For the Curve-To-Surface algorithm, the vector r is describing a “slave” point

from the surface, ρ(s) is a parametrization of the “master” curve edge; e is
a unit vector describing the shortest distance between the point and the
curve written via the unit normal ν and bi-normal β of the curve ρ(s). The
convective coordinates are used as measures: r – for normal interaction; s –
for tangential interaction; ϕ – for rotational interaction.
The Curve-To-Curve contact pair requires the projection on both curves,
therefore, there is no classical “master” and “slave” parts and both curves
are equivalent. For the description both coordinate systems assigned to each
contacting I-th curve are used:
ρ2 (s1 , r, ϕ1 ) = ρ1 (s1 )+re1 (s1 , ϕ1 ); e1 = ν 1 cos ϕ1 +β1 sin ϕ1 1 2. (1.8)
Here, the vector ρ2 is a vector describing a contact point of the second curve,
ρ1 (s1 ) is a parametrization of the first curve; a unit vector describing the
shortest distance e1 is written via the unit normal ν 1 and bi-normal β 1 of the
20 1 Introduction
first curve. Eqn. (1.8) describes the motion of the second contact point in the
coordinate system attached to the first curve. The description is symmetric
with respect to the choice of the curve 1 2. Convective coordinates are
used as measures: r – is mutual for both curve and a measure for normal
interaction; sI – for tangential interaction and ϕI – for rotational interaction
for the I-th curve.
The Curve-To-Surface contact pair, however, can be described in a dual
fashion via the Surface-To-Surface contact algorithm if we consider a “slave”
point on the edge and project it onto the “master” surface, see Fig. 1.6
(Curve-To-Surface contact as Surface-To-Surface algorithm). The contact is
described then in the surface coordinate system (1.6). This special dual con-
sideration of contact both in the surface coordinate system in eqn. (1.6) and
in the curve coordinate system in eqn. (1.8) allows to build later the Curve-
To-Rigid Surface (CTRS) contact algorithm. In this algorithm all contact
parameters are defined, first, in the local surface coordinate system eqn. (1.9)
attached to the rigid surface, after fulfilling the surface CPP procedure, and
then they should be projected into the curve coordinate system eqn. (1.10),
see Fig. 1.8, assuming that in the contact condition n = −e
rs (α1 , α2 , p) = ρ(α1 , α2 ) + pn(α1 , α2 ) with basis vectors ρ1 , ρ2 , n. (1.9)

ρ(s, r, ϕ) = rs (s) + re1 (s, ϕ)
= rs (s) + r(ν cos ϕ + β sin ϕ) (1.10)
The Point-To-Point (PTP) contact pair is described then in a coordinate

system standard for rigid body rotation problems (e.g. via the Euler angles),
however in contact situations it is a very rare case, and in computations it
Fig. 1.8 Curve-To-Rigid Surface contact algorithm

is rather improbable unless specially treated, and therefore, because of the

numerical error would fall into other contact pair types.
Geometry and kinematics for considered contact pairs in the correspond-
ing coordinate systems are developed in Chapter 4. We study there how
to define measures for normal and tangential contact interaction in various
geometrical situations as well as how to compute kinematic and variational re-
lations such as relative motion for all contacting pairs. Separately, kinematics
of 2D contact interaction is treated dually based on 2D CPP procedure and
based on the reduction of 3D formulation. Kinematics of contact between de-
formable and rigid surfaces is studied carefully as the Segment-To-Analytical
Surface (STAS) approach – a closed form solution of CPP procedure are
presented for some surfaces. The chapter contains all necessary information
for further formulation of the weak forms as well as for their linearization.
Thus, this is the most important information to build all considered contact
algorithms: 3D Surface-To-Surface (STS), 2D Surface-To-Surface (or in this
case Plane Curve-To-Plane Curve), 3D Segment-To-Analytical (Rigid) Sur-
face (STAS), 3D Point-To-Curve (PTC), 3D Curve-To-Curve (CTC) and 3D
Curve-To-Rigid Surface (CTRS) contact pairs.
Variational formulations (or weak forms) for considered geometrical con-
tact pairs are built in the Chapter 5 based on the developed kinematics.
In addition, a variationally consistent inclusion of contact constraints lead-
ing either to the full Lagrangian formulation or to Nitsche formulation (as a
consistent penalty formulation) is discussed.
Chapter 6 defines constitutive relations for contact tractions appearing
in various geometrical contact pairs and gives the rule how to compute them
for further implementations. This chapter gives then a systematic general-
ization of a contact interface law from the Coulomb friction law into the
anisotropic region in a covariant form including various known visco-elasto-
plastic mechanical models for the Surface-To-Surface contact. Thus, the de-
veloped coupled anisotropic adhesion-friction model is studied in detail for
various type of curvilinear anisotropy. This model is formulated via the princi-
ple of maximum dissipation in a rate form. Finally, the computational model
is derived via the application of the return-mapping scheme to the incremen-
tal form. As a result a frictional force is derived in a closed form including
both, the adhesion and the friction tensors. The geometrical interpretation
of the return-mapping and the update algorithm is considered via an ellipse
on the tangent plane. The new constitutive relations are also given in the
case of Curve-To-Curve contact with various elastic and friction properties
along each curve. The Coulomb friction law for tangential interaction and the
Tresca friction law for rotational interaction are considered as examples for
constitutive relations for Curve-To-Curve contact allowing specification for
each curve separately. The Curve-To-Curve contact model is applied then for
the beam-to-beam contact, see Fig. 1.9. In this case constitutive relations are
formulated with regards to a cross section, while for the Edge-To-Edge con-
tact they are supplied for the sector in the orthogonal plane spanned by vector
22 1 Introduction
ν and β), see Fig. 1.10. However, not only circular cross sections are possible
– any arbitrary cross section can be supplied in the orthogonal plane. For
the Curve-To-Rigid-Surface (CTRS) contact pair, the projection algorithm
for contact parameters from the local surface coordinate system eqn. (1.9)
into the curve coordinate system eqn. (1.10), see Fig. 1.8, is developed.
Chapter 7 is devoted to the consistent linearization of the weak form
for all geometrical contact pairs. The consistent linearization is performed in
the form of covariant derivatives in the corresponding coordinate systems.
This chapter is then the basis for further finite element implementation of
the tangent matrices.
The Chapter 8 is devoted to the various computational aspects and
implementation details of the developed theory in the case of Surface-To-
Surface geometrical contact pair. All known contact approaches such “Node-
To-Segment” (NTS) and, Mortar type “Segment-To-Segment” (STS) are re-
considered within the covariant description. A smoothing technique based
on the application of NURBS surfaces is considered. Integration of contact
integral using a subdivision scheme into subdomains is shown to be effective
to satisfy the Patch Test within the Mortar approach. A special development
within the covariant approach is derived for the contact with rigid surfaces
described analytically. “The rigid surface is a slave” and “the rigid
surface is a master” approaches are analyzed. Finally, various implemen-
tations of the Nitsche method are considered.
A fast implementation of frictional contact in 2D is proposed in Chap-
ter 9. The algorithm to transfer the history variables, see Fig. 1.5, in con-
tact problems overcoming the discontinuity of history variables on element
boundaries and the algorithm to update the history variables in the case of
reversible loading are studied and illustrated in detail. The contents of this
Chapter became a core for the course of “Contact Mechanics” for Diploma,
Master and PhD students at the Institute of Mechanics of the Karlsruhe
Institute of Technology.
A special development of the covariant approach in combination with high-
order finite element methods is given in Chapter 10. Both the penalty, and
the Lagrange multiplier method are considered. Approximation and selection
of the integration schemes for the Lagrange multipliers and for other terms
are chosen in order to satisfy the discrete Babuska-Brezzi (BB) stability con-
dition. The finite element discretization procedure is enhanced for cases with
the exact geometry of contact boundaries being represented by the blending
function method. As a result a contact layer element allowing anisotropic
p−refinement is created. The layer contact element is applied then to ini-
tially linear meshes. A good correlation with the analytical Hertz problem is
achieved even within a single contact layer element.
Chapter 11 is specially devoted to the extensive numerical analysis of
the coupled anisotropic adhesion-friction model developed in Chapter 6. The
cases including a uniform orthotropy of a plane given by the spectral de-
composition, a nonuniform orthotropy of a plane inherited with the polar
coordinate system and a spiral orthotropy of a cylindrical surface. As an

interesting result, a geometrically isotropic behavior of the block has been
found: in this case the combination of both, anisotropy for adhesion and
anisotropy for friction leads to a trajectory which can be normally observed
only for isotropic surfaces. It is shown that the application of the spiral or-
thotropy on a cylindrical surface allows to simulate the kinematics of the bolt
connection on relatively rough mesh. The tangent matrices for sliding in the
case of the anisotropic coupled adhesion-friction model are non-symmetric,
due to the non-associativity and due to the coupled anisotropy. Thus, a sum-
marized algorithm based on the Augmented Lagrangian method for coupled
anisotropic friction is developed to overcome this asymmetry.
As a key for the practical application, the developed model is experimen-
tally investigated. It is shown in Chapter 12, that the coupled anisotropic
adhesion-friction model can successfully describe a set of trajectories of a
block on a rubber mat with a periodical wavy profile, while the classical
anisotropic friction model fails. Special attention is given to the analysis of
geometrically isotropic behavior.
A geometrically exact description in a covariant form for Curve-To-Curve
contact is studied numerically in Chapter 13 for various types of finite ele-
ment models for beams: with known “solid-shell” finite elements, with linear
and curvilinear finite element beam model and, finally, with the new de-
veloped “solid-beam” finite elements. The verification section contains the
Fig. 1.9 Contact between curvilinear beams with circular cross-sections

24 1 Introduction
Fig. 1.10 Contact between curvilinear beams: arbitrary cross-section can be defined
in (ν 1 − β)-plane
comparison between the beam-to-beam and the edge-to-edge finite element

models as well as the verification with a famous “Equilibrium of Euler elas-
tica problem” computed via a finite difference scheme. The further numerical
examples are illustrating the ability to describe various kinematics for curve-
to-curve contact situations e.g. partial sticking of a single curve. Special at-
tention is devoted to the applications with curvilinear beams: C 1 -continuous
spline finite beam elements for the beam allowing large rotations are enriched
with the possibility of contact and C 1 -continuous “solid-beam” finite element
defined only with displacement degree of freedom and enriched with covariant
contact description is the alternative for the beam element. The element is
shown to be effective to describe the “stiff” behavior of Bernoulli beam and
the “soft” behavior of ropes. Thus, the isogeometric approach for curvilinear
beams together with contact is realized. A new application area for the tying
of knots (mechanics of knots) is considered in some numerical examples.
The solution of the problem of frictional interaction between a rope and
a rigid cylinder known since many years as 2D Euler-Eytelwein problem or
belt friction problem problem describing the interaction between a rope and
a circle is generalized in Chapter 14 into the 3D case under the restriction
that the rope is represented by the spiral line. The result is assumed for
verification of the Curve-To-Rigid Surface (CTRS) contact algorithm.
2
Differential Geometry of Surfaces and
Curves
In the current chapter we are recalling the necessary information about differ-
ential geometry of surfaces and curves: all metrics and curvature parameters
which are necessary to describe properties of the surfaces and curves. We
also study differential operations in the corresponding curvilinear coordinate
systems in a covariant form, such as Weingarten and Gauss-Codazzi formulas
for surfaces and Serret-Frenet formula for 3D curves. This information will
be fully used further to build all contact kinematics, weak contact forms as
well as linearization of operations. The reader who familiar with differential
geometry can however easily skip this chapter containing an overview of the
formulas from differential geometry of surfaces and curves.
2.1 Definition of the Surface and Its Geometrical

Characteristics
A surface is 2D manifold in 3D Cartesian space, and therefore, can be
parametrized by the surface coordinates ξ 1 , ξ 2 . Let ρ be a surface vector,
describing any point on the surface, see Fig. 2.1
ρ = ρ(ξ 1 , ξ 2 ), (2.1)
or via Cartesian coordinates X ≡ x1 , Y ≡ x2 , Z ≡ x3
xj = xj (ξ 1 , ξ 2 ), j = 1, 2, 3. (2.2)
The surface is representing a surface boundary of a solid in 3D space. In a

finite element discretization the general representation of the surface can be
written in the following form

ρ= Nk (ξ 1 , ξ 2 )x(k) , (2.3)
k
springerlink.com
26 2 Differential Geometry of Surfaces and Curves
Fig. 2.1 A surface in 3D Cartesian space. Definition of a Local Coordinate (Gaus-

sian) System.
where Nk (ξ 1 , ξ 2 ) are shape functions and x(k) are nodal coordinates, or knots
coordinates depending on the type of approximation of the surface.
2.1.1 The Fundamental Tensors and Property of the

Surface
Two fundamental tensors of the surface: the metrics tensor, or the first fun-
damental tensor, and the curvature tensor, or the second fundamental tensor,
fully contain the properties of a surface. The metrics tensor is responsible for
any metric operation on the surface (length, area or angle). The curvature
tensor is responsible for the inclusion of a surface as a 2D manifold into the
3D space R 3 (Cartesian space) respectively for the local structure of the
surface in the 3D space.
First, two surface base vectors ρi , i = 1, 2 in the tangent plane of a surface
are introduced
∂ρ ∂ρ
ρ1 = 1 , ρ2 = 2 , (2.4)
∂ξ ∂ξ
then the normal unit surface vector is given as the normalized cross product
of the basis vectors
ρ × ρ2
n= 1 . (2.5)
|ρ1 × ρ2 |
These three vectors ρ1 , ρ2 , n define a local surface coordinate system and
they are used to obtain the two fundamental tensors of the surface [99], [46],
[155]: metric and curvature tensors.
2.1 Definition of the Surface and Its Geometrical Characteristics 27
The Metric Tensor

The covariant components of the metric tensor on the surface are defined as
the dot product of the base surface vectors (2.4)
aij = ρi · ρj , i, j = 1, 2 (2.6)
The contravariant components of the metric tensor aij are obtained via the
equation
aik akj = a.j j
i. = δi , (2.7)
i. e. as the inverse matrix is given in the following form:

1 a22 −a12
ij
a : , a = det(aij ) = a11 a22 − (a12 )2 , (2.8)
a −a12 a11
the mixed components a.j j

i. are in fact identical to the Kronecker delta δi .
An adjacent basis of the surface is defined by the contravariant base vectors
ρi , which are obtained via a linear form of the covariant base vectors:
ρi = aij ρj . (2.9)
Thus, the metric tensor can be defined either in the covariant, or the con-
travariant basis, or the mixed basis
A = aij ρi ⊗ ρj = aij ρi ⊗ ρj = a.j

i. ρ ⊗ ρj
i
(2.10)
The metrical characteristics, which are necessary for the further description,
are length and area. The differential dl of the length is obtained as

dl = (ρi · ρj )dξ i dξ j = aij dξ i dξ j . (2.11)
The differential ds of the area can be obtained either via the determinant of
the matrix of the metric tensor eqn. (2.6), or via the absolute value of the
cross product of the surface vectors eqn. (2.5):

ds = | det(aij )|dξ 1 dξ 2 = |ρ1 × ρ2 |dξ 1 dξ 2 . (2.12)
The angle θ between coordinate vectors can be measured as

ρ1 · ρ2 a12
θ= =√ (2.13)
ρ1 ρ2 a11 a22
The Curvature Tensor
In differential geometry the curvature tensor is used to describe a local surface

structure via e.g. main curvatures, for more information see [99], [46], [155].
The tensor is defined by its covariant components hij , which are computed
as the dot product of the second derivative of the vector ρ and the normal n
hij = ρij · n. (2.14)
The mixed components are obtained as a linear combination of the covariant

components with the contravariant metrics components
h.j kj
i. = hik a . (2.15)
The contravariant components are obtained as a bi-linear combination of the

covariant components with the contravariant metrics components
hij = hkn aik anj . (2.16)
Equation (2.15) resp. (2.16) gives also a general rule how to compute mixed
resp. contravariant components of any second order tensor via covariant com-
ponents (the lowing and rising indexes rule).
2.1.2 Study the Local Surface Structure

Both the metrics and curvature tensors allows to study the local structure of
a surface. To emphasize this, we present here main formulas necessary for the
further developments. For more information including the specific derivations
of the formulas see e.g. in [99], [103].
The geometrical analysis of the surface structure is given by the generalized
eigenvalue problem :
(hij − kaij )ej = 0, (2.17)
which leads to the real roots k1 , k2 of the equation
det[(hij − kaij )] = 0. (2.18)
These roots are called principal curvatures and correspond to the orthogo-
nal principal directions e1 , e2 . The principal curvature k1 (resp. k2 ) is the
curvature of a line arising from the intersection of the surface with the plane
containing both, normal vector n and principal vector e1 (e2 ). Their inverse
values are principal radii Ri = 1/ki . The geometrical meaning is that the
osculating circle has the same radius as the radius of the curve. The local
structure of the surface in the vicinity of the computed contact point can be
then classified by the Gaussian curvature K
K = k1 · k2 (2.19)
into four cases as follows:

1) K = k1 · k2 > 0 — an elliptic point , i.e. a surface in vicinity of the point

looks like an elliptic paraboloid, see Fig. 2.2.
2) K = k1 · k2 < 0 — a hyperbolic point, i.e. a surface looks like a hyperbolic
paraboloid , see Fig. 2.3.
For the case with zero Gaussian curvature K = 0 a more careful analysis
is required.
3) Either k1 = 0, k2 = 0, or k2 = 0, k1 = 0 — a parabolic point. A
surface looks then like a parabolic cylinder , see Fig. 2.4. If these conditions
Fig. 2.2 Structure of a surface in the vicinity of the elliptic point.
Fig. 2.3 Structure of a surface in the vicinity of the hyperbolic point.

Fig. 2.4 Structure of a surface in the vicinity of the parabolic point.
are fulfilled for each surface point of the surface, then a flat surface is
obtained. This surface (e.g. a cylinder, or a cone) can be unwrapped on
a plane.
4) Both k1 = 0 and k2 = 0 — a planar point . The local structure can
not be identified without the analysis of the higher order derivatives.
Nevertheless, the case with k1 = 0, k2 = 0 for each point leads to a plane
in 3D.
Eqn. (2.18) for a definition of the principal curvatures can be written in the
following form:

k 2 − 2Hk + K = 0, =⇒ k1,2 = H ± H 2 − K (2.20)
where the Gaussian curvature K is defined as

det[hij ] h11 h22 − h212
K = k1 · k2 = = , (2.21)
det[aij ] a11 a22 − a212
and the mean curvature H is defined as

1 1 a22 h11 − 2a12 h12 + a11 h22
H= (k1 + k2 ) = . (2.22)
2 2 a11 a22 − a212
Remark 2.1. Any other structure of the surface in the vicinity of a point dif-
ferent from the described cases 1)-3) (e.g. edge point etc.) can arise either
from a planar point (case 4), or from violation of the a-priory assumed C 2 -
continuity and a more advanced analysis would be necessary. Surfaces with
more complicated local structure leading e.g. to a so-called “star”-looking do-
main, self-contacting and self-penetrating surfaces etc. are relatively seldom
in practical applications and are out of the scope of the current contribution.
2.1.3 Differential Operations in the Surface

Coordinate System
For any further derivations the most important mathematical operations in
the surface coordinate system are frame independent differential operations.
They are defined in order to describe the kinematics from the local surface
coordinate system point of view. For this the derivatives of base vectors have
to be determined. The Weingarten formula and the Gauss-Codazzi formula
[99], [155] give us a complete set for derivatives of base vectors and are used
to define covariant derivatives on the surface.
The Weingarten Formula

The Weingarten gives directly derivatives of the unit normal n:
ni = −hij ajk ρk = −hki ρk . (2.23)
Proof. Having taken the derivative of the unity equation n·n = 1 with respect
to surface coordinates ξ i , we obtain n · ni = 0, from which follows that the
vectors ni are orthogonal to n and, therefore, lay on the tangent plane of the
surface. Thus, ni is expressed as a sum of the surface vectors ρi
i. ρk .
ni = c.k (2.24)
Computing a dot product with ρj
(ni · ρj ) = c.k
i. (ρk · ρj ) → (ni · ρj ) = ci. akj ,
.k
(2.25)
a derivative of the orthogonality condition ρi ·n = 0, gives ρij ·n+ ρi ·nj = 0.

Thus
hij ≡ (ρij · n) = −(ρi · nj ) (2.26)
Therefore, the c.k
i. can be defined as
i. = −hij a ,
c.k jk
(2.27)
from which Weingarten’s formula (2.23) is obtained.
The Gauss-Codazzi Formula

The Gauss-Codazzi formula allows directly the computation of derivatives of
the basis vectors ρi .
ρij = Γijk ρk + hij n, (2.28)
where Γijk are Christoffel symbols [99], [155], [122], defined on the surface as
follows
Γijk = ρij · ρk = ρij · ρn ank . (2.29)
Proof. In general, the derivatives of the coordinate surface vectors ρi are

no longer on the surface, therefore, one should express them by the vectors
ρ1 , ρ2 , n
ρij = Γijk ρk + hij n. (2.30)
Expressions for Γijk and hij follow from the computation of the dot product
with the basis vectors ρj and the normal n.
Indifferent Covariant Derivative

The derivatives of the base vectors allow to evaluate a frame indifferent
derivative of any object defined on the surface. We now consider a vector
T, defined by its local contravariant coordinates T i in the surface coordinate
system
T = T i ρi . (2.31)
The full time derivative of T – with the assumption that the vector ρi is
implicitly time dependent via the coordinates ξ i – gives
d ∂T i ∂T i
T= ρi + j ξ˙j ρi + T i ξ˙j ρij . (2.32)
dt ∂t ∂ξ
Further applying the Gauss-Codazzi formula, we get
d ∂T i ∂T i i ˙j
T= ρi + ( j + T k Γjk )ξ ρi .
dt ∂t ∂ξ
Finally, the full material time derivative of the vector has the following form
d ∂T i
T=( + ∇j T i ξ˙j )ρi . (2.33)
dt ∂t
The term ∇j T i is a covariant derivative of the contravariant component T i
∂T i
∇j T i = + T k Γjk
i
. (2.34)
∂ξ j
A similar expression can be found for the covariant derivative of the covariant
components Ti :
∂Ti
∇j Ti = j − Tk Γijk . (2.35)
∂ξ
In the case of covariant components we need the derivative of a contravariant
∂ρ i
base vector instead of ρij , see eqn. (2.32). First, take the derivative of
∂ξ j
the mixed metric components:
∂aki ∂(ρi · ρ k ) ∂ρ k
= = (ρ ij · ρ k
) + (ρ i · ) = Γijk + (ρi · ρ,jk ) = 0 , (2.36)
∂ξ j ∂ξ j ∂ξ j
2.2 Definition of the Curve in 3D and Its Geometrical Characteristics 33
therefore,
(ρi · ρ,jk ) = −Γijk (2.37)
and the covariant derivative for the covariant component gets the following
form
∂Ti
∇j Ti = j − Tk Γijk . (2.38)
∂ξ
2.2 Definition of the Curve in 3D and Its Geometrical

Characteristics
Let us consider a curve in 3D Cartesian space arbitrarily parametrized with a
parameter ξ (1D manifold). For further developments – except special cases –
the curve is assumed to be continuously differentiable, namely C 1 -continuous
with resp. to the parameter ξ. The parametrization in vector notation can be
written as, see Fig. 2.5
ρ = ρ(ξ) (2.39)
or via coordinates X ≡ x1 , Y ≡ x2 , Z ≡ x3 in Cartesian space as
xj = xj (ξ). (2.40)
An arc-length s for the curve is then used in the differential geometry consid-
erations for the curves as a natural coordinate to describe all their geometrical
characteristics. It is defined via the differential of a vector ρ

ds = ds(ξ) = ρξ · ρξ dξ = Jdξ (2.41)
Fig. 2.5 A curve in 3D Cartesian space. Definition of a Local Coordinate (Serret-

Frenet) System.
Here the following short notations are introduced for derivatives:

• derivative with respect to a convective coordinate ξ
∂ρ
ρ̇ = ρξ ≡
∂ξ
• derivative with respect to the arc-length (natural coordinate) s gives us
a unit tangent vector τ
∂ρ
ρ ≡ = τ;
∂s
• Jacobian of the transformation s → ξ

3

ds ∂xj ∂xj
J= = ρξI · ρξI =
dξ j=1
∂ξ ∂ξ
2.2.1 Definition of a Local Coordinate (Serret-Frenet)

System. Serret-Frenet Formulas
A standard Serret-Frenet basis connected to the curve is constructed as fol-
lows. At each sufficiently smooth point of a continous curve one can define
three unit vectors: a unit tangent vector τ , a unit normal vector ν pointing
to a center of curvature of the curve and a unit bi-normal vector β defined
as a cross product
β = τ × ν. (2.42)
These three unit vectors are connected via the derivatives of the curve ρ with
respect to the arc-length s. The formulas are known in the standard literature
on differential geometry, see e.g. in [103], [33], [99] as Serret-Frenet formulas:
⎧
⎪ dτ
⎪
⎪ = kν
⎪
⎪ ds
⎨ dν
= −kτ + κβ . (2.43)
⎪
⎪ ds
⎪
⎪
⎪
⎩
dβ
= −κν
ds
Here, k is a curvature , and κ is a torsion of a spatial curve. Assuming an arbi-

trary parametrization in eqn. (2.39) and in the Serret-Frenet formulas (2.43)
the curvature k and the torsion κ of a spatial curve can be computed as
ρ̇ × ρ̈ |ρ̇ × ρ̈|
kβ = −→ k = , (2.44)
|ρ̇|3 |ρ̇|3
det(ρ̇, ρ̈, ∂ 3 ρ/∂ξ 3 ) (ρ̇ × ρ̈) · ∂ 3 ρ/∂ξ 3

κ= = . (2.45)
|ρ̇ × ρ̈|2 |ρ̇ × ρ̈|2
3
Closest Point Projection Procedure
and Corresponding Curvilinear
Coordinate System
In this chapter the Closest Point Projection (CPP) procedure as the first im-
portant tool to define contact measures is introduced for surfaces and curves.
The fundamental issues such as existence and uniqueness of the solution of
the Closest Point Projection procedure are intensively studied. In due course
the CPP procedure leads to a special curvilinear coordinate system in which
later all contact kinematics and weak forms are formulated separately for each
contact pair. This chapter creates a basis of the geometrically exact theory of
contact interactions for certain geometric pairs such as Surface-To-Surface,
Curve-To-Surface and Curve-To-Curve.
3.1 Closest Point Projection Procedure for Arbitrary

Surfaces
The so-called closest point projection procedure is often introduced as a nu-
merical scheme to compute coordinates of a point projected onto a surface.
In variational formulations for contact problems it appears due to the split of
contact displacements into a normal and a tangential part. In early publica-
tions for finite element models it was explained as a result of the linearization
of non-penetration conditions, Kikuchi and Oden [84]. Hallquist et al. [52]
considered the split into normal and tangential direction via the projection
operation for a so-called “master-slave” approach within the finite element
contact model. Such an approach is based on looking at normal contact where
a “slave” node of one body penetrates into a “master” segment of another
body. The value of penetration is measured as the closest distance to the
master body. In a next step of the contact algorithm, the penalty method
together with explicit time integration was used to enforce the contact condi-
tions. For an implicit solution Wriggers and Simo [194] proposed a consistent
linearization procedure of the penalty functional containing the penetration
function and obtained the corresponding tangent stiffness matrices within
the Newton iteration scheme. However, the penetration function based on
springerlink.com
36 3 Closest Point Projection Procedure
the closest point projection is not only used for formulations in combination
with a penalty functional, but also in other computational schemes to en-
force the condition of non-penetration in contact. The concept of the closest
distance is also important for contact searching routines in order to define
potentially contacting points.
In order to show the importance of the projection operation, we briefly
focus here on methods in contact mechanics, where the closest point pro-
cedure occurs. Contact elements for bilinear surface approximations based
on the penalty method for non-frictional problems are discussed in Parisch
[138]. Due to decoupling of normal and tangential forces the projection pro-
cedure necessarily appears for 2D frictional problems in Giannakopoulos [44]
and Wriggers et al. [195], then also for 3D problems in Peric and Owen
[140] and in Parisch and Lübbing [139]. All mentioned references combine
the penalty method together with the regularization of the Coulomb fric-
tion model and a radial-return algorithm within the friction algorithm. This
algorithm is predominantly used in computational plasticity, see Simo and
Hughes [161], and also exploits the closest point procedure; however, then the
projection is given onto the yield surface in the force space. Full Lagrange
multipliers methods are standard for enforcing constraints in computational
mechanics, but cannot be easily transported into frictional problems using
the same penetration, or distance function as mentioned above, see the review
in Zhong [208]. This is due to the nature of friction forces depending on ve-
locity and, therefore, requiring an incremental update procedure. For a small
sliding problem, a perturbed Lagrangian method including both penalty and
Lagrangian functionals was discussed in Wriggers and Simo [194]. Then aug-
mented Lagrangian methods appeared to be more robust. Thus, a 2D finite
element algorithm based on the augmented Lagrangian method has been pro-
posed in Wriggers et al. [195]. A mixed penalty-duality approach based on the
augmented Lagrangian scheme is proposed in Alart and Curnier [3], where
different distance functions have been discussed. Laursen and Simo [109] for-
mulated contact conditions via convective coordinates on the contact surface
within both, the penalty method and the augmented Lagrangian method. For
various geometrical situations the method has been discussed in Heegaard and
Curnier [62] for large-slip problems, in Heege and Alart [64] for contact with
CAD surfaces and in Pietrzak and Curnier [145] for finite deformations and
large sliding. The full Lagrangian approach for frictional problems has been
discussed in Jones and Papadopoulos [79] utilizing stick-to-slide conditions in
one step, and therefore, covering only small sliding problems. More recently
the so-called mortar method, originally based on Lagrangian multipliers en-
forcing the non-penetration conditions is considered in Hüeber and Wohlmuth
[69] for non-frictional problems, where a special discretization technique has
been taken for the Lagrange multiplier functions. A mortar scheme based on
a special integration technique for constraint equations is proposed in Puso
and Laursen [151] for non-frictional problems and extended into the frictional
case via the augmented Lagrangian method by the same authors [152].
3.1 Closest Point Projection Procedure for Arbitrary Surfaces 37
The publications listed above cover the most applied methods in compu-
tational contact mechanics and all contain the closest point procedure as the
first necessary step. Despite the enormous number of publications on contact
mechanics, there are only a few publications covering to some completeness
the problem of uniqueness and existence of the closest point procedure for ar-
bitrary approximations of the contact surfaces as well as describing effective
numerical algorithms to overcome problems. The problem of non-uniqueness
and non-existence of the projection for e.g. bi-linear approximations of a sur-
face by finite elements is known since the first publications, see Hallquist et
al.[52], and mostly reported in theoretical manuals of popular commercial
codes, see [51], [1]. Heegaard and Curnier in [63] mentioned that geometri-
cal parameters of a surface, such as a focal point can be used to determine
existence and uniqueness of the projection for smooth surfaces. However, for
arbitrary surfaces described e.g. by CAD systems and containing combina-
tions of smooth surfaces, curved edges and corner points, the situation can
be far more complex. Some techniques dealing with non-existence of the pro-
jection in certain cases are well known, e.g. a situation when the slave point
is passing over the edge of a locally concave part of a body can be overcome
by taking an average vector from the normals of neighboring contact surface
segments. A description of this rather heuristic approach can be found in
the books of Wriggers [191] and Laursen [106] as well as in theory manuals
[51], [1]. This technique is used in a similar manner in multi-surface plasticity
when the C 1 -continuity of the yield surface is broken, see Simo and Hughes
[161]. Other techniques such as the projection onto the edge for such cases,
see e.g. in Heegaard and Curnier [62] and in Zhong and Nilsson [210], are
also reported to be implemented in commercial programs [51].
In this chapter we provide analytical tools allowing to create, a-priori, prox-
imity domains of contact surfaces from which a given contact point is always
uniquely projected for arbitrary surfaces. This approach is based on the geo-
metrical properties of contact surfaces exploiting the covariant description for
contact problems originally developed in [87], [88] and discussed in Chapter 2.
First, all C 2 -continuous surfaces are classified according to their differential
properties allowing a unique projection. Then, proximity domains are cre-
ated for C 1 -continuous surfaces. Finally, proximity domains are proposed for
globally C 0 -continuous surfaces covering all practical approximations. The
projection scheme in the latter case is further generalized including the pro-
jection onto geometrical objects of lower order (curved edges, corner points).
In such cases the corresponding proximity domains are created by a geomet-
rical analysis of those objects. Also the reduction into the 2D plane case is
discussed. The examples are chosen to show the necessity of the proposed
generalized projection procedure in certain mechanical problems as well as
the necessity of the proximity domain for searching techniques when a contact
surface is given by arbitrary discretization, e.g. described by CAD-systems,
or found in high order FEM analysis.
3.1.1 Formulation of the Closest Point Projection

Procedure in Geometrical Terms
The most important operation for the data transfer between contacting bod-
ies in contact mechanics is the closest point projection procedure, see Fig. 3.1,
where one seeks the projection of a given contact point from one body r, usu-
ally called a slave point, onto another contact body, usually called a master
body, parameterized as
ρ = ρ(ξ 1 , ξ 2 ). (3.1)
The parameterization (3.1) by Gaussian coordinates is arbitrary and ξ 1 , ξ 2

can be provided either by e.g. a finite element approximation, or by a spline
approximation or by a NURBS approximation. The closest point procedure
appears also in many other applications such as fluid-structure interactions,
computational plasticity and others. The projection problem is formulated
as an extremal problem
||r − ρ(ξ 1 , ξ 2 )|| → min, −→ (r − ρ) · (r − ρ) → min, (3.2)
which is solved then mostly numerically. However, then a fundamental prob-

lem arises: Does the solution of (3.2) exist? And, if it exists, then, is it unique
for any arbitrary surface approximation?
The direct and strict answer is fully covered by the application of the fa-
mous theorem from the convex analysis to problem (3.2), see e.g. [13].
Fig. 3.1 Closest point procedure and definition of the spatial coordinate system.
If the function
1
F(ξ 1 , ξ 2 ) = (r − ρ) · (r − ρ) (3.3)
2
is convex in a domain (ξ 1 , ξ 2 ) ∈ D, then the solution of problem (3.2) exists
and is unique in this domain.
This leads to the fact that the solution can be obtained numerically by
e.g. a Newton iteration procedure, which will converge from any initial point
1
inside the domain (ξ(0) 2
, ξ(0) ) ∈ D.
Using this criterion we can focus on the geometrical properties of the sur-
faces. Then the goal of this section is to create the classification of surfaces
from its differential geometry point of view onto which a point can be pro-
jected uniquely.
The analysis of the local geometrical structure of surfaces allows to create a
projection domain Ω in 3D from which any point can be uniquely projected
onto the given surface. Starting with the C 2 -continuous case, we will con-
sider also cases possessing parts with solely C 1 - and C 0 -continuity discussing
the solvability of problem (3.2) and in addition the possible multiplicity of
solutions.
Assuming that the function F is twice differentiable, i.e. for C 2 -continuous
surfaces, we can construct the Newton iterative process for the solution of
the minimization problem (3.2) as follows:
⎡ 1 ⎤
Δξ(n)
Δξ (n) = ⎣ ⎦ = −(F )−1 F(n)
(n) (3.4)
2
Δξ(n)
ξ (n+1) = ξ (n) + Δξ (n) ,
where the first derivative F and the second derivative F with respect to the
surface coordinates ξ 1 , ξ 2 are computed as:
⎡ ⎤
∂F
⎢ ∂ξ 1 ⎥
⎢ ⎥ ρ · (r − ρ)
F = ⎢ ⎥=− 1 (3.5)
⎣ ∂F ⎦ ρ2 · (r − ρ)
∂ξ 2

ρ1 · ρ1 − ρ11 · (r − ρ) ρ1 · ρ2 − ρ12 · (r − ρ)
F = . (3.6)
ρ2 · ρ1 − ρ21 · (r − ρ) ρ2 · ρ2 − ρ22 · (r − ρ)
Here a short notation for the partial derivatives has been introduced as
∂ρ ∂2ρ
ρi = i
, ρij = i j . (3.7)
∂ξ ∂ξ ∂ξ
Now, we introduce a 3D spatial coordinate system related to the surface

coordinate system , see Fig. 3.1, as follows:
r(ξ 1 , ξ 2 , ξ 3 ) = ρ(ξ 1 , ξ 2 ) + nξ 3 . (3.8)
All necessary parameters are defined in the covariant basis of the surface
ρ1 × ρ2
tangent vectors ρ1 , ρ2 and a unit normal vector n = . The solv-
ρ1 × ρ2
ability of the minimization problem (3.2) will be considered in this coordi-
nate system. The introduction of such a coordinate system is the basis of
the covariant description for contact problems, developed in Konyukhov and
Schweizerhof [87], [88] for isotropic frictional problems, and then in [89], [90]
for anisotropic frictional problems. In this coordinate system we will con-
sequently study kinematics, variational formulations, linearized variational
equations and, finally, formulation of numerical methods in the following sec-
tions.
The dominantly geometrical structure of all contact parameters is among
the main advantages of this description, e.g. a value of penetration is simply
the third convective coordinate ξ 3 .
The following surface tensors are used to describe the metric and the cur-
vature properties of surfaces in differential geometry [99], [103]:
a) metric tensor with components:
aij = ρi · ρj (3.9)
b) curvature tensor with components:
hij = ρij · n (3.10)

Inserting now eqn. (3.8) into eqn. (3.6) and using the notations given in
eqns. (3.9) and (3.10) we obtain the expression F in the introduced spatial
coordinate system as:

a − ξ 3 h11 a12 − ξ 3 h12
F = 11 . (3.11)
a21 − ξ 3 h21 a22 − ξ 3 h22
Let us start with a case assuming a unique projection: In this case the
function F must be convex. As is known for the convexity of F, the second
derivative F must then be a positive definite matrix . Thus, the Sylvester
criterion from the basic algebra is exploited to check the positivity of the
matrix given in eqn. (3.11):
(a11 − ξ 3 h11 ) > 0

(3.12)
det[(aij − ξ 3 hij )] > 0
3.1.2 Proximity Criteria for Different Surfaces

In this section we consider the structure of 3D domains (proximity domains)
surrounding a given surface, from which a point can be a-priori uniquely pro-
jected onto the surface. This structure depends on the classification of surface
points given in Sect. 2.1.2. Taking into account eqn. (2.20) for principal cur-
vatures, the Sylvester criteria in eqn. (3.12) can be written in the following
form:
(a11 − ξ 3 h11 ) > 0

2 (3.13)
1 1
− 2H + K > 0
ξ3 ξ3
This system of inequalities can be reformulated in terms of principal curva-
tures k1 , k2 as:

1
− k1 >0
ξ3

(3.14)
1 1
− k1 − k2 > 0.
ξ3 ξ3
Within the last transformation, it was assumed that the coordinate ξ 3 is a
positive value into the normal vector direction n, as chosen in eqn. (3.8). Thus,
zones with positive and negative coordinates ξ 3 should be distinguished.
Projection Domains for an Elliptic Point

According to the geometry presented in Fig. 3.2, a normal n is pointing into
a convex part, where ξ 3 > 0 and, both k1 > 0 and k2 > 0.
Domain for the Convex Part ξ 3 > 0
Using the geometrical interpretation of the principal curvatures as principal

radii Ri = 1/ki , eqn. (3.14) can be written as:
R1 − ξ 3 > 0
(3.15)
(R1 − ξ 3 )(R2 − ξ 3 ) > 0
The solution allows to create a projection domain Ω(ξ 1 , ξ 2 , ξ 3 ) with a pa-

rameter ξˆ3 = min{R1 , R2 } with an a-priory unique solution of the projection
problem, see Fig. 3.2:
Ω(ξ 1 , ξ 2 , ξ 3 ) := {r = ρ + ξ 3 n, where 0 < ξ 3 < ξ̂ 3 }. (3.16)
This domain is created in the local coordinate system and contains all points
between the original surface and shifted by nξ 3 surface.
Domain for the Concave Part ξ 3 < 0
In this case, criterion (3.15) is automatically fulfilled leading to an infinite

domain Ω above the surface:
Ω(ξ 1 , ξ 2 , ξ 3 ) := {r = ρ + ξ 3 n, with − ∞ < ξ 3 < 0}. (3.17)
ξ <0
3 ρ1
ρ2
R2
n
R1
ξ >0
3
Fig. 3.2 Elliptic point . Structure of the surface and projection domain assuming
C 2 -continuity.
Projection Domain for a Hyperbolic Point

Considering a situation as presented in Fig. 3.3, in the case with ξ 3 > 0,
where a projection domain is created as an overlapping of the semi-infinite
domain due to the line with negative curvature k1 and the finite domain
due to the line with positive curvature k2 similar to the domain as for a
convex part for an elliptic point. The case with ξ 3 < 0 leads to a projection
domain constructed in an identical fashion. Summarizing both cases, a finite
projection domain Ω(ξ 1 , ξ 2 , ξ 3 ) is created as a domain between two shifted
surfaces:
Ω(ξ 1 , ξ 2 , ξ 3 ) := {r = ρ + ξ 3 n, where − R1 < ξ 3 < R2 }. (3.18)
Projection Domain for a Parabolic Point

A projection domain for a parabolic point, see Fig. 3.4, consists of a finite
domain for the convex part with ξ 3 > 0, and a semi-infinite domain for
the concave part with ξ 3 < 0. Thus, the projection domain Ω(ξ 1 , ξ 2 , ξ 3 ) is
described as follows:
Ω(ξ 1 , ξ 2 , ξ 3 ) := {r = ρ + ξ 3 n, where − ∞ < ξ 3 < R2 }. (3.19)

R1
ξ <0
3
ρ1
ρ2
n
R2
ξ >0
3
Fig. 3.3 Hyperbolic point . Structure of the surface and projection domain assum-
ing C 2 -continuity.
k1 =0, R 1 =
8
ρ1
ρ2
n
R2
ξ >0
3
Fig. 3.4 Parabolic point . Structure of the surface and projection domain assuming
C 2 -continuity.
Discussion about Planar Points – Required Approximation for

the Plane
As mentioned in Sect. 2.1.2 case 4), the structure of a surface in the vicinity
of the planar point requires an analysis of the higher order derivatives with
respect to the surface coordinates. In this case, it is also difficult to discuss
the convergence of the Newton iterative process eqn. (3.4). However, for the
plane with zero curvature tensor hij = 0 criterion eqn. (3.12) becomes
det[aij ] > 0. (3.20)
Since the metric tensor aij is always positive, a trivial result is recovered:
The projection onto a plane always exists and is unique. Neverthe-
less, from the numerical point of view it is necessary to avoid cases with
det[aij ] ≈ 0. In such cases the angle between the convective coordinate lines
is close either to 0o or to 180o, which is normally avoided ab initio as e.g. in
finite element approximations this would be a sign for inadmissibly distorted

elements.
3.2 Solvability of the CPP Procedure for Surfaces –

Allowable and Non-Allowable Domains
In the previous section, the projection domains were constructed under the
main assumption of sufficient smoothness of the corresponding surfaces,
specifically C 2 -continuity was necessary to derive all curvature parameters.
In this section, we will obtain that C 1 -continuity of a surface is sufficient
for the construction of a continuous projection domain, while violation of
C 1 -continuity can cause either multiplicity of solutions, or non-existence of
solutions at all. Starting with a 2D case as a preliminary case, we then develop
“a remedy” for globally C 0 -continuous surfaces by constructing a continuous
projection domain allowing unique projections onto the surface. A general-
ized projection procedure includes then several projection procedures onto
curved edges and corners.
3.2.1 Reduction to 2D Plane Geometry – Solvability

Criteria and Uniqueness
A covariant approach for contact problems allows to look at a 2D case either
as a special reduction from a cylindrical geometry in 2D space, or as a case
based on a 2D plane curve geometry, see [91]. In the last case, the length of a
curved line s is used as a convective coordinate. The closest point projection
procedure eqn. (3.2) is then formulated as follows:
1
F = ||r − ρ(s)||2 → min . (3.21)
2
A curvilinear coordinate system in 2D is based on a plane curve geometry
and is introduced as follows:
r(s, ζ) = ρ(s) + ζν. (3.22)
A second derivative with respect to s in this coordinate system is computed

as:
∂ρ ∂ρ ∂ 2 ρ ∂τ
F = · − 2 · (r − ρ(s)) = τ · τ − · (r − ρ(s)) (3.23)
∂s ∂s ∂s ∂s
For further transformation it is necessary to introduce the Serret-Frenet for-
mulae
∂ρ ∂τ
τ = , = kν, (3.24)
∂s ∂s
3.2 Existence of the Projection – Allowable and Non-Allowable Domains 45
where τ is a unit tangent vector, ν is a unit normal vector and k is a curvature

1
of the curve (the radius of the curvature R = is also often used). Taking
k
into account the Serret-Frenet formulae together with eqn. (3.22) the second
derivative F is transformed as
F = 1 − kζ. (3.25)
Since the normal vector ν is always pointing to the center of curvature C1 ,

see Fig. 3.5, we obtain from the condition F = 1 − kζ > 0 a finite projection
domain O1 O2 C2 C1 for the convex part as follows:
1
Ω(s, ζ) := {r = ρ(s) + ζν, with 0 < ζ < = R}, (3.26)
k
and a semi-infinite projection domain for the concave part above the curve
O1 O2 :
Ω(s, ζ) := {r = ρ(s) + ζν, with − ∞ < ζ < 0}. (3.27)
O1 τ
ζ <0
ν
R= 1
k
ζ
C1
ζ >0
Ω(s ,ζ)
O2
C2
Fig. 3.5 2D case – plane curve. Projection domain.
The case with F = 0 can lead to a multiplicity of the solution. This is visible
in Fig. 3.6, where an infinite number of projections is possible from a point
O1 with the coordinate ζ = R onto an arc BC, and from a point O2 with
the coordinate ζ = −R onto another arc CD.
Violation of C 2 -Continuity
The violation of C 2 -continuity, but keeping C 1 -continuity leads to a dis-
continuous projection domain. In this case, however, it is easily possible
to construct a continuous projection domain, from where uniqueness of the
projection operation is automatically fulfilled. The idea is presented in Ta-
ble 3.1 and is illustrated in Fig. 3.6. Since discontinuity is resulting due to
the only piecewise continuous parameter ζ, a continuous projection domain is
Table 3.1 Violation of C 2 -continuity of a curve leads to piecewise continuous

curvature and consequently to a piecewise continuous parameter ζ. A continuous
projection domain Ω(s, ζ) := {r = ρ(s) + ζν} can be constructed by setting ζ to a
minimal value. An example is given for ζ > 0 in Fig.3.6.
Part of a curve Curvature k Computed ζ Minimal ζ
AB 0 +∞ R
BC R R R
CD -R +∞ R
A minimal projection domain for ζ >0
B ν O1
R
C
ζ<0 D
R O2
Fig. 3.6 Violation of C 2 -continuity results in changing the discontinuity of the
projection domain. A continuous domain is constructed by setting ζ to a minimal
value.
constructed by taking the minimal parameter ζ = min{R, ∞} = R along the

curve.
Violation of C 1 -Continuity
We consider here the most practical case which is standard in the trian-
gulation of surfaces with low order finite elements: piecewise differentiable
functions with finite jumps for the first derivative. From a geometrical point
of view this situation leads to an angular point for curves or an edge for
surfaces. The violation of C 1 -continuity for a surface parameterization leads
also to a discontinuity of the normal vector n causing either difficulty with
definition of the local coordinate system, or multiplicity of projection. The
situation in 2D is presented in Fig. 3.7. Now, we consider a point following
the path S1 S2 S3 S4 S5 in the local coordinate system (3.22). Since a normal
vector ν has jumps (points B and C), there are portions of the trajectory
which can not be described in this local coordinate system. They are located
in the non-allowable domain with regard to the projection onto the curve:
any point S2 laying in the non-allowable domain can not be described in the
local coordinate system given by eqn. (3.22). The term non allowable domain
with regard to the projection onto surface resp. curve resp. point is then used
for a domain where any point can not be described in the local coordinate
system corresponding to the projection onto surface resp. curve resp. point.
However, in such a situation it is possible to create a continuous mapping
of the path S1 S2 S3 S4 S5 onto the curve by introducing a new projection
operation in the non-allowable domain with regard to the projection onto the
curve. This is a projection of the point into an angular point of curves (e.g.
S2 into B and S4 into C in Fig. 3.7), or a closest point projection onto the
edge of surfaces. This method is mentioned for 2D examples in Heegaard and
Curnier [62] and in Zhong and Nilsson [210]; it is also reported to be done e.g.
in LS-DYNA [51]. A projection procedure keeping a continuous mapping of
any path on the curve will be called the generalized closest point procedure. In
2D cases this procedure includes both a projection onto the curve and onto
the corner point.
Another situation arising from the violation of C 1 -continuity is overlapping
of two or more projection domains causing consequently the multiplicity of
projection, see e.g. domain of multiple solutions in Fig. 3.7.
We already could observe the multiplicity of solutions in the case of the
violation of C 2 -continuity where the second derivative of the distance function
is zero F = 0, see point O1 in Fig. 3.6. In this case, a strong inequality in
the formulation of the projection domains allows to eliminate the multiplicity.
A case with violation of C 1 -continuity (e.g. domain of multiple solutions in
Fig. 3.7) can lead to a more severe situation, see point M passing over corner
point C. A natural remedy in this case is to define a minimum distance in the
sense of the generalized closest point procedure onto neighboring segments
and onto corner point C. However, one can find a line with points which are
equidistantly situated from both segments.
In this situation, though relatively seldom in numerical computations due
to round-off errors, a choice either with a random selection of projection sides,
or with an averaged normal from all projection sides can be applied, implicitly
assuming that the distance is fairly small. Within a contact algorithm the
storage of the “slave” point path as history variable would be then necessary.
3.2.2 Proximity Domain for Globally C 0 -Continuous

Surface in 3D
It is necessary to define additional projection procedures in order to create
a proximity domain for the 3D space allowing a continuous mapping of any
path laying inside, similar to that discussed for the 2D case in 3.2.1. These
projections include a projection onto an edge and onto a corner. The combi-
nation of proximity domains leads finally to a continuous domain. The idea
is illustrated in Fig. 3.8 for the contact surface of a hexaeder focusing on the
surfaces containing the corner point O. The continuous proximity domain
surrounding corner point O consists of
1. three domains for sides arising from the standard projection procedure:
S1 = {x, y, z |x > 0, y > 0, z < 0} for side BOCF ,
S2 = {x, y, z |x > 0, y < 0, z > 0} for side AOBE,
S3 = {x, y, z |x < 0, y < 0, z < 0} for side AOC;
2. three domains for edges arising from the point-to-edge projection proce-
dure:
E1 = {x, y, z |x < 0, y < 0, z > 0} for edge OA,
E2 = {x, y, z |x > 0, y > 0, z > 0} for edge OB,
E3 = {x, y, z |x < 0, y > 0, z < 0} for edge OC;
3. one domain arising from the point-to-corner point projection procedure:

P = {x, y, z|x < 0, y > 0, z > 0} for corner point O.
The last domain is added to fulfill the continuity, because from this domain a
projection onto the cube in general is possible only into a corner point. This
projection is trivial, always exists and can be defined simply as difference
between “slave” and corner point: rs − ρO .
The point-to-edge projection , in case of a curved edge, requires a numerical
solution and again the problem of existence and uniqueness arises.
non−allowable domain
S1 S2
ν
O
A P1 B S3
S4
P3 S5
R C
M P5
D
domain of multiple solutions
Fig. 3.7 Violation of C 1 -continuity leads to non-allowable domains with regard to
the projection onto curve ABCD. A special treatment is necessary to preserve a
continuous mapping.
A
Y
O
S
E
B
X C
D F
Fig. 3.8 Hexaedral contact surface in the vicinity of O. A continuous proximity
domain (i.e. allowing continuous projection onto the C 1 -surface) consist of domains
for sides, edges OA, OB, OC and, finally, of a domain for the corner point O.
3.2.3 Reference Example: Projection Domain for a

Hyperbolical Surface
Consider a quadratic surface of the form: z = (x2 − y 2 )/2, 0 ≤ x, y ≤ 1 see
Fig. 3.9. The upper projection domain is then a domain between the origi-
nal surface OABC and the surface O A B C created by shifting in normal
direction as follows:
r = ρ(x, y) + R(x, y)n(x, y), ρ(x, y) = {x, y, (x2 − y 2 )/2}T , (3.28)
where R(x, y) is a function of the corresponding radius of curvature. Exem-

plarily, the position of point C is computed as follows. A point on the surface
C is defined by the vector ρ = {1.000, 0.000, 0.500}T , with the corresponding
normal vector n = {−0.707, 0.000, 0.707}T . The main curvatures at point C
are computed via eqn. (2.20), (2.21) and (2.22): k1 = 0.354, k2 = −0.700. A
positive value here defines a curvature of a corresponding line which is locally
convex with respect to the chosen direction of the normal n, i.e. a normal
is pointing into a center of curvature of the line. According to the rule for a
hyperbolic point discussed in Sect. 3.1.2, the value k1 is taken for the shift
as R1 = 1/k1 = 2.825. Thus, the upper boundary of the projection domain
at point C is computed as r = ρ + R1 n = {−0.997, 0.000, 2.497}T .
Remark 3.1. It is obvious, that the complete algorithm for the projection
for an arbitrarily composed CAD surface requires the application of some
Boolean operations with domains known from CAD applications. Thus, in
Fig. 3.9 Structure of the upper proximity domain for a given surface OABC with
z = (x2 − y 2 )/2, 0 ≤ x, y ≤ 1.
general, a fairly large number of cases has to be analyzed within a contact

search routine taking advantage of the derived proximity domains.
3.3 Closest Point Projection Procedure for

Point-To-Curve Projection and Corresponding
Projection Domain
An arbitrary curved edge as a result of the intersection of two surfaces is
represented by a curved line, see e.g. AOB in Fig. 3.10. This line can be
parameterized by the arc-length parameter s as well as by an arbitrary pa-
rameter ξ, e.g. a normalized coordinate in a finite element approximation:
ρ = ρ(s) = ρ(s(ξ)). (3.29)

The transformation of the parameters s ← ξ is provided by the formula:

∂ρ ∂ρ
ds = · dξ = ρξ · ρξ dξ. (3.30)
∂ξ ∂ξ
The projection routine, though it is looking similar to (3.21), is now formu-
lated in 3D as follows:
1
F (s(ξ)) = ||rs − ρ(s(ξ))||2 → min . (3.31)
2
3.3 Closest Point Projection Procedure for Point-to-Curve Projection 51
The necessary optimum condition leads to the projection operation onto the
curve
∂ρ
F = −(rs − ρ(s(ξ)) · = −(rs − ρ(s(ξ)) · τ = 0, (3.32)
∂s
showing the orthogonality of the vector (rs − ρ) and the tangent vector τ .
According to this, a new coordinate system τ , ν, β is introduced as a natural
coordinate system of the curved line, see Fig. 3.10, where τ is a unit tangent
vector, ν is a unit main normal vector and β is a unit binormal vector of the
curve, see [99], [103]. This results in orthogonal planes I, II, III. The given
point S (“slave” point) is defined as follows:
rs = ρ(s) + re, (3.33)
where r is the shortest distance between the point S and the curve AB. e
is a vector giving a director of the shortest distance in the plane ν0β (plane
II), defined via ν and β as follows:
e = ν cos ϕ + β sin ϕ. (3.34)
The second derivative F taking into account (3.33) is transformed in the

curvilinear coordinate system as follows:
F = τ · τ − (r − ρ(s)) · kν = 1 − kr(e · ν) (3.35)
and finally using representation (3.34) for the vector e as
F = 1 − kr cos ϕ. (3.36)

Considering now F > 0, the projection domain is constructed from two
parts 1 and 2, see Fig. 3.10:
1: a semi-infinite domain with negative ν, or with ϕ ∈ [π/2, 3π/2]. Here the
projection of the vector re onto the normal ν is negative leading to the
automatic satisfaction of F > 0,
and,
2: a layer with positive ν, or with ϕ ∈ (−π/2, π/2). Here the projection
of the vector re on the normal ν must be less than the radius of the
curvature R, i.e.
1
r cos ϕ < = R. (3.37)
k
As one can see, two domains surround a curve densely, e.g. without any void.
A problem can occur if vector ρ(s) looses C 1 -continuity, however in this case
we obtain a corner point as discussed above.
Remark 3.2. If the edge is a straight line then the coordinate system τ , ν, β
should not be derived via the curved properties of the line according to the
Serret-Frenet formulae , however, it can be defined arbitrarily.
In the case of an arbitrary parameterization of a line, the projection routine

is fulfilled via the Newton iterative process in (3.4) for the parameter ξ, where
Δξ is computed as:
F (rs − ρ) · ρξ
Δξ = − =− , ξ(n+1) = ξ(n) + Δξ. (3.38)
F (ρξ · ρξ ) − (rs − ρ) · ρξξ
R
s ϕ<
B
r co
I
C
R
II ν
S
ϕ
τ
rs re O
A
s(ξ )
β III
Z ρ (s(ξ ))
Y
X
Fig. 3.10 Point-to-Curve projection. Proximity criteria and projection domain.
Summary. A continuous projection domain for a globally C 0 -continuous sur-

face allowing unique continuous projections of any path laying inside this
domain can be constructed via a generalized projection procedure including
three projection operations: a) onto a surface; b) onto an edge; c) onto a
corner point.
Remark 3.3. An algorithm for the continuous projection domain can be used
also as a preparation stage for the global searching routine. Thus, in fol-
lowing computations within an analysis of a contact problem any possible
“slave” point can be uniquely projected onto the “master” surface via the
corresponding projection onto a surface, an edge or a point.
3.4 Closest Point Projection Procedure for Curve-To-Curve Contact 53
3.4 Closest Point Projection Procedure for

Curve-To-Curve Contact and Definition of Local
Coordinate Systems
The closest disctance between two curves is a natural measure of any con-
tact interaction between two curves. This distance can be defined in any
convenient coordinate system. The standard coordinate system for the final
discretization and thus for the numerical computation is a global Cartesian
coordinate system, but in order to consider some local geometrical properties
it is far more convenient to consider the kinematics of deformation and to
describe the differential properties of the curves in coordinate systems ge-
ometrically attached to the curves. Known in differential geometry such a
system is the Serret-Frenet coordinate system to be defined in any point of
the curve e.g. in a point which gives rise to a contact measure between curves.
This allows to exploit standard notations known in differential geometry
([33], [99], [103]) throughout the article.
Fig. 3.11 Closest point projection (CPP) procedure. Definition of the Serret-
Frenet coordinate system (τ I , ν I , β I ) and a local coordinate system (τ I , eI , gI )
based on CPP procedure.
Let us consider, see Fig. 3.11, two curves in 3D Cartesian space arbitrarily
parameterized with parameters ξ I , I = 1, 2, where I is a number assigned to
the curve. For further developments – except special cases – the curves are
assumed to be continuously differentiable, namely C 1 -continuous with resp.
to the parameter ξ I . The parameterization in vector notation can be written

as
ρI = ρ(ξI ), I = 1, 2 number of curve (3.39)
or via coordinates x1I , x2I , x3I in Cartesian space as
xjI = xjI (ξI ), I = 1, 2, j = 1, 2, 3. (3.40)
An arc-length sI for the I-th curve is then used in the differential geometry
considerations for the curves as a natural coordinate to describe all their
geometrical characteristics. It is defined via the differential of a vector ρI

dsI = ds(ξI ) = ρI,ξI · ρI,ξI dξI = J(I) dξI
(no summation) I = 1, 2.
(3.41)
Here the following short notations are introduced for derivatives:
• derivative with respect to a convective coordinate ξI
∂ρI
ρ̇I = ρI,ξI ≡
∂ξ I
• derivative with respect to the arc-length (natural coordinate) sI
∂ρI
ρI ≡ ;
∂s
• Jacobian of the transformation sI → ξI

3

dsI ∂xj ∂xj
J(I) = = ρI,ξI · ρI,ξI =
dξI j=1
∂ξI ∂ξI
The coordinates s1 , s2 will be used then to define measures of the tangential

interaction.
−−−→
The closest distance |C1 C2 | = |ρ12 | between two curves can be chosen
as a natural measure of the contact interaction (see Fig. 3.11). However, the
definition of this distance requires the knowledge about the points C1 and C2 ,
and, therefore, its arc-lengths s1 and s2 as local coordinates on both curves
respectively. It leads to the auxiliary minimization problem for the following
distance function F(s1 , s2 ) :
1
ρ (s1 ) − ρ2 (s2 )2 −→ min
F(s1 , s2 ) = (3.42)
2 1
This mathematical problem is called Closest Point Projection (CPP), proce-
dure however, defined for curves. The solution gives us two coordinates sp1
and sp2 . Then the closest distance is defined as
r = |ρ12 | = |ρ1 (sp1 ) − ρ2 (sp2 )|. (3.43)

The necessary conditions minimizing F(s1 , s2 ) in (3.42) lead to derivatives
with respect to the arc-lengths s1 and s2 as local coordinates being zero:

(ρ1 − ρ2 ) · ρ1 = 0
F =
−(ρ1 − ρ2 ) · ρ2 = 0

(ρ1 − ρ2 ) · τ 1 = 0
= , (3.44)
−(ρ1 − ρ2 ) · τ 2 = 0
with unit tangent vectors introduced as derivatives with respect to the arc-
length τ I = ρI .
In general, equation system (3.44) is non-linear and an iterative method
should be applied for its solution in a numerical realization. For the nonlinear
system the question of solvability becomes then important. Such an analysis
is possible in a special coordinate system defined in the following section.
3.4.1 Definition of a Local Coordinate System

Attached to a Curve
First, we start with a standard Serret-Frenet basis connected to the I-
th curve. As is known, at each sufficiently smooth point of a curve one can
define three unit vectors: a unit tangent vector τ , a unit normal vector ν
pointing to a center of curvature of the curve and a unit bi-normal vector β
defined as a cross product
β = τ × ν. (3.45)
These three unit vectors τ , ν and β are connected via the derivatives of
the curve ρ with respect to the arc-length s. The formulas are defining this
relation are the Serret-Frenet formulas defined in eqn. (2.43)
Now, we attach these Serret-Frenet frames τ I (sI ), ν I (sI ), βI (sI ) to the
projection points (C1 and C2 ), see Fig. 3.11, as described in Sect. 2.2.
The closest distance between the curves denoted as C1 C2 becomes a sec-
ond coordinate r in our specially defined coordinate system, while the arc-
length will be the first coordinate sI . A vector corresponding to the closest
distance C1 C2 and defined as ρ12 is orthogonal to both vectors τ 1 and τ 2
and, therefore, is representing an intersection line of two planes ν 1 , C1 , β 1
and ν 2 , C2 , β 2 . The unit vector of ρ12 is denoted as e1 . Introducing the pa-
rameterization in the plane ν 1 , C1 , β 1 for the vector e1 via an angle ϕ1 , see
again Fig. 3.11 we obtain
ρ12 = ρ2 − ρ1 = re1 (ϕ1 ) ∈ ν I CI βI − plane. (3.46)
Here, the angle ϕ1 between the unit normal ν I and the closest distance vector
ρ12 becomes then a third coordinate ϕ1 in our new coordinate system. Thus,
we can redefine a new coordinate system based on
a) a unit tangent vector τ I ,
b) a unit vector of the closest distance vector eI ,
c) a unit vector gI orthogonal to both τ I and eI .
The last two vectors are defined via the basic unit vectors ν I , β I as follows:
e1 (ϕ1 ) ≡ ν 1 cos ϕ1 + β 1 sin ϕ1 (3.47a)
∂e1
g1 (ϕ1 ) = = −ν 1 sin ϕ1 + β 1 cos ϕ1 . (3.47b)
∂ϕ1
In fact e1 and g1 are defining a polar coordinate system in the plane or-
thogonal to τ I . The new introduced coordinate system defined by the unit
vectors τ I , eI , gI is orthogonal, but changing in space with regard to a given
curve. Summarizing, a relative motion of a point C2 and thus the motion of
a second curve can now be considered in the following coordinate system:
ρ2 (s1 , r, ϕ1 ) = ρ1 (s1 ) + re1 (ϕ1 ), 1 ↔ 2. (3.48)
Thus, a 3D-local curvilinear coordinate system with convective coordinates

s1 , r, ϕ1 is defined.
Remark 3.4. the description is symmetric with regard to any choice of a curve
1 ↔ 2.
Remark 3.5. the coordinate r is always positive and identical for both curves.
3.4.2 Analysis of Uniqueness and Existence of

Solutions for the CPP: Definition of a
Projection Domain
In this section we discuss a choice in which solutions of the CPP procedure
exist and are unique a-priori. As is known, the convexity of the function
F in (3.42) would lead to the fulfillment of uniqueness and existence. Then

the positivity of a second derivative F > 0 follows. The second derivative
exploiting the Serret-Frenet formulas (2.43) can be written as

ρ1 · τ 1 + (ρ1 − ρ2 ) · τ 1 −ρ2 · τ 1
F =
−ρ1 · τ 2 ρ2 · τ 2 − (ρ1 − ρ2 ) · τ 2

τ 1 · τ 1 + k1 (ρ1 − ρ2 ) · ν 1 −τ 1 · τ 2
=
−τ 1 · τ 2 τ 2 · τ 2 − k2 (ρ1 − ρ2 ) · ν 2

1 + k1 (ρ1 − ρ2 ) · ν 1 −τ 1 · τ 2
= .
−τ 1 · τ 2 1 + k2 (ρ2 − ρ1 ) · ν 2
(3.49)
Defining then ψ as an angle between two tangent lines τ 1 and τ 2
τ 1 · τ 2 = cos ψ (3.50)
and taking into account the coordinate system in (3.48) we obtain

1 − k1 r cos ϕ1 − cos ψ
F = (3.51)
− cos ψ 1 − k2 r cos ϕ2
with the determinant as

det F = (1 − k1 r cos ϕ1 )(1 − k2 r cos ϕ2 ) − cos2 ψ. (3.52)
A Sylvester criterion can be now applied to enforce the positivity of the

matrix F . This leads to both, a positive first term of the matrix F11 > 0

and a positive determinant det F > 0:

1 − kI r cos ϕI > 0
(3.53)
sin2 ψ + r(rk1 k2 cos ϕ1 cos ϕ2 − (k1 cos ϕ1 + k2 cos ϕ2 )) > 0.
We shall now study step-by-step several particular cases in order to under-

stand the geometrical properties of the inequality conditions in (3.53).
Case 1: Both Lines Are Straight Lines

In this – simplest – case both curvatures are zero k1 ≡ 0, k2 ≡ 0 and (3.52),
resp. the Sylvester criterion in (3.53), is transformed as
sin2 ψ = 0 −→ ψ = πn, n = 0, 1, 2, ... (3.54)
This has a simple geometrical interpretation – for the existence and unique-
ness of the CPP procedure for straight lines the angle defined between tangent
vectors τ 1 and τ 2 in (3.50) should not be zero and, therefore, the lines should
not be parallel.
Case 2: Two Circles in a Plane

As a next step for the complexification we consider first two overlapping
circles laying in a plane, see Figs. 3.12, 3.13 and 3.14. In this case an angle
between the tangent vectors τ 1 and τ 2 is zero ψ = 0 (the critical case for the
straight lines). The situation is also rather difficult for overlapping circles with
the angle between the normal ν 1 and the vector e1 being zero ϕ1 = ϕ2 = 0.
1
Denoting the radii of curvature as as RI = , the first equation in the
kI
Sylvester criterion (3.53) leads to r < RI , however the determinant gives
us the requirement r > R1 + R2 . This inconsistency for an enforcement

of positivity means that the second derivative F is neither positive, nor

negative, making it impossible to use the second derivative F only. One
can see the difficulties in Fig. 3.12 where the multiplicity of the solution
is depicted: the shortest distance can be defined as C1 S2 , while the initial
procedure assumes C1 C2 . We note that even in this case the determinant
is not zero which makes it possible to exploit the Newton scheme for the
iterative solution.
However, it is possible to analyze the uniqueness for the solution if we con-
sider projection domains for both curves separately. The projection domain
for a curve, see Sect. 3.3 for PTE (originated to [94]), is defined as a 3D
domain from which any point can be projected uniquely.
Projection Domain for a Spatial Curve:
1. a semi-infinite domain with negative ν, or with ϕ ∈ [π/2, 3π/2]. The

projection of the vector re onto the normal ν is negative.
2. a layer with positive ν, or with ϕ ∈ (−π/2, π/2). The projection of the
vector re onto the normal ν must be less than the radius of curvature R:
1
r cos ϕ < = R. (3.55)
k
In a case of overlapping circles we are falling into case (2), see Fig. 3.13.
Summarizing, we can derive the domain of existence and uniqueness of the
CPP procedure as an overlapping of two projection domains for the corre-
sponding curves. Thus for the existence and uniqueness of the CPP procedure
for two overlapping circles the distance between two points should be less than
the minimal radius of curvature.
For non-overlapping circles, see Fig. 3.14, the angle is ϕ1 = π, and the
Sylvester criterion (3.53) is always fulfilled leading to the conclusion that
the closest distance between two non-overlapping circles is uniquely defined.
τ2 τ1
O1 r O2 S2
C2 C1
Fig. 3.12 Two overlapping circles in a single plane. The CPP procedure can have
a multiple solution: C1 C2 or C1 S2 ?
τ1 τ2
O1 r O2
C2 C1
Fig. 3.13 Two overlapping circles in a single plane. The CPP procedure has a
unique solution in the overlapping region of two projection domains {r ∈ (0, R1 )} ∩
{r ∈ (0, R2 )} and, therefore, if r < min{R1 , R2 }
τ1
τ2
O1 ν 1 C1 r O2
C2
ϕ1
Fig. 3.14 Two non-overlapping circles in a single plane. The CPP procedure has
always a unique solution (Closest distance is uniquely defined).
This situation is also falling into case (1) as the overlapping of projection
domains for two separated circles.
Remark 3.6. The conclusions of the current case are still valid if we consider
arbitrary curves with the osculating circles laying in parallel planes.
Case 3: Osculating Circles in Orthogonal Planes

Let us consider now two arbitrary curved lines (k1 = 0, k2 = 0) which have
two osculating circles laying in orthogonal planes, but still possessing parallel
tangent vectors, e.g. ψ = πn, see Fig. 3.15.
O
ν2
2
C2
τ2 ϕ2
0 r Y
ϕ1
C1 P1
ν1 O1
τ1
K
Fig. 3.15 Two curves with osculating circles laying in orthogonal planes. The
closest distance is uniquely defined if two points are situated in the over-
lapping projection domains. C1 ∈ {(r, ϕ2 ) | 0 < r + ∞, ϕ2 ∈ [π/2, 3π/2]} C2 ∈
{(r, ϕ1 ) | r cos ϕ1 < R1 , ϕ1 ∈ [−π/2, π/2]}.
One can see that for the exterior projection with ϕ1 , ϕ2 ∈ [π/2, 3π/2]. the
Sylvester criterion is fulfilled.
For other situations, the concept of two overlapping projection domains is
effectively applied. Fig. 3.15 illustrates the analysis. The test points C1 and
C2 are situated in the overlapping domain because,
• a point C1 lays in the projection domain of type (1) of the second curve
with ϕ2 ∈ [π/2, 3π/2].
• C2 lays in the projection domain of type (2) of the second curve, with
ϕ1 ∈ [−π/2, π/2], satisfying r cos ϕ1 < R1 .
Case 4: A Singular Determinant for Curved Lines

The example of an extremal situation with singular determinant F in (3.52)
is illustrated in Fig. 3.16. In this case the tangent vectors τ 1 and τ 2 are
orthogonal leading to ψ = π, and therefore, cos ψ = 0. Moreover, the closest
point C2 is projected exactly into a center of curvature O1 leading to the

fulfillment of R1 − r cos ϕ1 = 0 and finally to F = 0.
Oz
O2
C2
τ2
0 Oy
ϕ
C1
ν1 R1 O1
τ1
K
Ox

Fig. 3.16 Singular determinant det F = 0 for arbitrary curves. The tangent vec-
tors are orthogonal and the closest point C2 is projected into a center of curvature.
Case 5: Arbitrary Curves

The geometrical interpretation of an arbitrary general case shown in Fig. 3.17
requires the consideration of osculating circles in the 3D space with regard
to the Sylvester criterion (3.53). Consideration of osculating circles will allow
directly to construct projection domains (1) and (2) for separated curves
and, finally obtain, the overlapping domain in which the CPP procedure has
a unique solution. Many geometrical situations can be constructed then which
is left to the interested reader.
Fig. 3.17 Two arbitrary curves . The solution depends on the positions of oscu-
lating circles which give rise to corresponding projection domains. The region for
uniqueness of the solution of a CPP procedure is constructed as overlapping of
projection domains for two curves.
3.4.3 Computational Issues of the CPP Procedure

The CPP procedure written in terms of the arc-length parameter in (3.42)
is efficiently solved numerically via a Newton-type solver. The procedure can
be outlined as follows. With arbitrary parameterizations via ξ1 and ξ2 , the

second derivative F in (3.49) becomes then a matrix

∂ 2F ρ̇1 · ρ̇1 + (ρ1 − ρ2 ) · ρ̈1 −ρ̇1 · ρ̇2
= F̈ = (3.56)
∂ξ2 −ρ̇1 · ρ̇2 ρ̇2 · ρ̇2 + (ρ2 − ρ1 ) · ρ̈2

while the first derivative F is defined similar to (3.44)

(ρ1 − ρ2 ) · ρ̇1
Ḟ = . (3.57)
−(ρ1 − ρ2 ) · ρ̇2
The tangential coordinates ξ1p and ξ2p along both lines are computed then via
a Newton iteration procedure as
ξ (n+1) = ξ (n) + Δξ with an initial guess ξ (0)

Δξ1
and increments Δξ = = −(F̈)−1 Ḟ. (3.58)
Δξ2
In the section devoted to the finite element implementation of Curve-To-

Curve contact algorithm we specify this procedure for a particular contact
element.
4
Geometry and Kinematics of Contact
In Chapter 3 is studied the existence and the uniqueness of solutions for

the Closest Point Projection (CPP) procedure depending on the geometry
of contacting objects. In due course the corresponding CPP procedure leads
to a specific curvilinear coordinate system. In those coordinate systems we
are going to describe geometry and kinematics for various contact pairs de-
pending on their geometrical description: Point-To-Surface, Point-To-Curve,
Curve-To-Curve etc. These cases later on give rise to contact algorithms de-
scribing contact between surfaces, contact between curves and surfaces and,
finally, contact between curves. In those coordinate systems we are going to
build the variational formulation of contact and further on to continue with
finite element discretization and formulation of the solution algorithm. Spe-
cial cases including contact between rigid surfaces and deformable solids are
separately studied. The chapter also includes linearization of the necessary
kinematic parameters for corresponding contact pairs.
4.1 Kinematics of the Surface-To-Surface Interaction

With frictional contact a specific interaction between bodies contacting each
other along surfaces of those bodies is described. Differential geometry pro-
vides a powerful mathematical tool to capture the change of these surfaces
in the covariant form. Another essential feature to model frictional contact
problems is the formulation of the contact conditions as kinematic constraints
which leads to a nonlinear problem and, therefore, in the correct description
of the solution process, to a problem of consistent linearization.
The current section gives the kinematics of the Surface-To-Surface con-
tact interaction which is obtained in the curvilinear Gaussian surface coordi-
nate system inherited with CPP procedure for STS interaction described in
Sect. 3.1.
Despite the large amount of contributions the fully covariant description
of contact is still not available in literature. In this contribution we employ
64 4 Geometry and Kinematics of Contact
the highly developed “apparatus” of differential geometry to reconsider the

contact conditions in a specially defined spatial local coordinate system which
corresponds to the well-known closest point procedure. All differential oper-
ations necessary for kinematics and linearization are considered as covariant
derivatives.
4.1.1 Local Surface Coordinate System

A surface 2D coordinate system is usually defined to describe the surface
geometry. In addition, a special 3D local coordinate system which is related
to this surface can be constructed to describe any spatial object. This system
will be used to define any characteristics that belongs to the surface as well
as to transfer the result of the linearization into the global coordinate system
for the purpose of a finite element implementation. All geometrical and kine-
matical characteristics of the contact are investigated in the local coordinate
system. First, all necessary operations in the surface coordinate system are
described. For this a surface element (Fig.1), the so-called “master” element,
is considered, which is parametrized by local coordinates ξ 1 , ξ 2 . ρ is a vector,
describing an arbitrary point on the surface. It has the following form

ρ= Nk (ξ 1 , ξ 2 )x(k) (4.1)
k
where Nk (ξ 1 , ξ 2 ), k = 1, 2, .., n are later the shape functions of e.g. finite

elements.
In order to describe the geometry of the contact surface element from the
internal differential geometry point of view, surface tangent vectors ρi , i =
1, 2 have to be specified
∂ρ ∂ρ
ρ1 = , ρ2 = . (4.2)
∂ξ 1 ∂ξ 2
The normal surface vector is computed as a cross product of the tangent
vectors
ρ × ρ2
n= 1 . (4.3)
|ρ1 × ρ2 |
The surface vectors ρ1 , ρ2 define a surface coordinate system, while the
normal n is used to describe geometrical properties of the surface and to
construct a local 3D coordinate system as well.
4.1.2 Spatial Curvilinear Local Coordinate System

and Its Characteristics
We consider two interacting bodies (Figure 4.1). One of them is chosen as
the contact body: its surface is called “master” surface, on which we will set
4.1 Kinematics of the Surface-To-Surface Interaction 65
up a coordinate system in which in due course we will observe all contact

conditions. On the surface of the second body, we consider a “slave” point S,
which is e.g. an integration point or a nodal point. Two bodies are coming
into contact, if a slave point of the second surface penetrates into the master
surface, where penetration is defined as the shortest distance between the
two surfaces of the contacting bodies.
Fig. 4.1 Surface-To-Surface contact. Definition of a local surface coordinate system

on the master surface.
As contact between two bodies is dominantly an interaction between these

two surfaces, the main aim of the following consideration is to take advan-
tage of the differential geometry of the contact surfaces in order to describe
the kinematics of the contact conditions. First, we consider the geometry of
the master surface and its characteristics and then define a special spatial
coordinate system attached to this surface.
Two bodies are coming into contact if a slave point penetrates at least
at the closest distance into the master surface. This point is computed via
the Closest Point Projection (CPP) procedure as discussed in Sect. 3.1. This
procedure can be included in the variational formulation, see also the theo-
retical details in Kikuchi and Oden [84]. One of the important aspects in the
consideration of kinematics is to construct a special spatial local coordinate
system on the master surface corresponding to the CPP procedure.
Now we define a special local coordinate system related to the master

surface at the penetration point C. Any spatial vector in space can be defined
as
r(ξ 1 , ξ 2 , ξ 3 ) = ρ(ξ 1 , ξ 2 ) + ξ 3 n(ξ 1 , ξ 2 ) (4.4)
By assuming the normal vector to be known, the projection procedure has al-
ready been taken into account into this consideration. The equilibrium equa-
tions for contact will now be formulated in the defined local coordinate sys-
tem, but since contact is an interaction between surfaces then each necessary
equation especially for the linearization will be considered on the tangent
plane, i.e. at ξ 3 = 0. For this, we define all the geometrical and differential
characteristics with special attention on their values on the tangent plane.
4.1.3 Measure of Contact Interaction for

Surface-To-Surface Contact
Equation (4.4) is defining a coordinate system inherited with the CPP
procedure onto the Surface. This coordinate system is defined such that
the convective coordinates ξ 1 , ξ 2 , ξ 3 are defining the measure of contact
interaction.
Measure of Normal Interaction – The Penetration
A value of the penetration ξ 3 , essential for formulation of the non-penetration

conditions in contact mechanics, see [190], [106], [84], is exactly the third
coordinate in our surface coordinate system:
ξ 3 = (rs − ρ) · n. (4.5)
Measure of Tangential Interaction – Surface Convective Coordinates
Increments of the surface convective coordinates Δξ 1 and Δξ 2 are used fur-

ther for measures, however, in order to formulate the rule of computation we
have to consider the motion of the slave point relative to the master surface.
4.1.4 Spatial Basis Vectors and Metric Tensor

In the spatial curvilinear coordinate system all the characteristics such as
metrics, covariant derivative etc. can be defined. Our focus is now only on
those which are necessary for the further development. The base vectors of
the system are given as
∂r
ri = = ρi + ni ξ 3 = (aki − hki ξ 3 )ρk , i = 1, 2, r3 = n, (4.6)
∂ξ i
where the Weingarten formula (2.23) and the first fundamental tensor in the
mixed formulation (2.7) have been used to obtain a more compact formula.
The coordinate vectors are defined in each point in 3D space belonging to
the allowable domain of the surface. In Fig. 4.1 they are depicted at the
slave point S. At the tangent plane they are coinciding with the surface
vectorsri |ξ3 =0 = ρi , i = 1, 2, r3 = n. The covariant components of the metric
tensor of the spatial coordinate system are defined via the dot product of
vectors (4.6).
gij = (ri ·rj ) = aij −2 ξ 3 hij +hik hkj (ξ 3 )2 , i = 1, 2 gi3 = 0, g33 = 1. (4.7)
Contravariant metric components g ij , as well as contravariant base vectors

ri are defined in a similar fashion, see eqns. (2.7), (2.9).
Time Derivative of the Covariant Metrics Components aij

During the forthcoming linearization it is essential to consider this procedure
as a 3D process in the spatial coordinate system (4.4), therefore, in general,
derivatives also with respect to the third coordinate ξ 3 should be considered.
Thus, time derivatives of the surface metric components aij are calculated as
values of the spatial metric components gij on the tangent plane at ξ 3 = 0,
namely

damn dgmn
= = (4.8)
dt dt ξ3 =0

∂ j ∂ 3 ∂

k 3 2
= + ξ̇ +ξ˙ amn − 2ξ hmn + hmk hn (ξ )
3
= (4.9)
∂t ∂ξ j ∂ξ 3 ξ 3 =0

∂ ∂
= + ξ˙j j (ρm · ρn ) − 2hmn ξ˙3 = (4.10)
∂t ∂ξ
= (vm · ρn ) + (ρm · vn ) +
l
+ Γmj (ρl · ρn ) + Γnj
l
(ρm · ρl ) ξ˙j − 2hmn ξ˙3 . (4.11)
The Christoffel symbols appear in eqn. (4.11) due to the usage of the Codazzi
formula. All indices are running from 1 to 2.
Time Derivative of the Contravariant Metrics Components aij

The time derivative of the contravariant component of the metric tensor aij
is obtained from the derivation of eqn. (2.7):
d ik
(a akj ) = 0
dt
daik dakj
akj + aik =0
dt dt
daik damn
= −aim ank . (4.12)
dt dt
Spatial Christoffel Symbols. Covariant Derivative on the Tangent

Plane
In order to distinguish in the summation agreement a spatial object from
the surface one, we will use capital letters, i.e. I, J, ... = {1, 2, 3}. Covariant
derivatives in the spatial coordinate system require the spatial Christoffel
K
symbols Γ̂IJ . They are defined, similar to eqn. (2.29) but with the spatial
K
base vectors rI , as Γ̃IJ = (rK · rIJ ). The full time derivative in the spatial
coordinate system in the form of eqn. (2.33), computed in convective coordi-
nates ξ I via covariant derivatives for contravariant components in eqn. (2.34)
or for covariant components in eqn. (2.35), is a frame indifferent derivative
and coincides with the Lie time derivative definition Lt in the form
d −1 d
Lt T := F (F T) = T (4.13)
dt dt
where F is a push-forward and F−1 is a pull-back operator, see more in Bonet
and Wood [16], Marsden and Hughes [122].
Proof. Proof that the full time derivative is a Lie time derivative.
In order to prove eqn. (4.13) we consider the vector r(ξ i ) in the reference
Cartesian frame:
r = X k ek (4.14)
where ek are unit vectors of the Cartesian reference frame. By definition of
the reference frame the vectors ek are time independent. Since here only the
spatial case is considered, all indices are running from 1 to 3. The vector r is
assumed to be time independent only for simplicity without loss of generality.
The coordinate vectors ri are defined as
∂X k
ri = ek = (F −1 )ki ek (4.15)
∂ξ i
where (F −1 )ki are components of the inverse gradient deformation tensor F

with components
∂ξ i
Fji = , (4.16)
∂X j
which are used for the vice versa transformation:
ek = Fki ri . (4.17)
Equations (4.17) and (4.15) give the push-forward operator F and the pull-
back operator F−1 respectively in a tensor form:
F = Fji ri ⊗ e j , F−1 = (F −1 )ij ei ⊗ r j

(4.18)
The Lie time derivative of the spatial vector T = T i ri is taken following

the rule: pull-back to the reference configuration, take time derivative, push-
forward to the current configuration:
d −1
Lt T = F (F T) = (4.19)
dt
d −1 j
= Fnk rk ⊗ e n · (F )i ej ⊗ r i · T m rm =
dt
d −1 j
= Fn rk ⊗ e ·
k n
(F )i ej δm i
Tm =
dt
d d((F −1 )j T i )
= Fnk rk δjn (F −1 )ji T i = i
Fjk rk .
dt dt
This is a full time derivative. It can be seen directly
d[(F −1 )ji T i ] k d[(F −1 )ji T i ]

Fj rk = ej =
dt dt
d[(F −1 )ji T i ej ] d[T i ri ] dT
= = . (4.20)
dt dt dt
For the proof only the time independence of the reference basis vectors ej
was used. Some algebraic manipulations of equation (4.19) are required to
show this for the components.
d[(F −1 )ji T i ] k
Lt T = Fj rk = (4.21)
dt
∂T i −1 j k ∂T i ∂(F −1 )ji i k ˙n
= (F )i Fj rk + n (F −1 )ji Fjk ξ˙n rk + T Fj ξ rk =
∂t ∂ξ ∂ξ n
∂T i k ∂T i ∂(F −1 )ji i k ˙n
= δi rk + n δik ξ̇ n rk + T Fj ξ rk .
∂t ∂ξ ∂ξ n
The last term contains Christoffel symbols . In order to elaborate this, their
determination in the reference frame has to be considered. Equation (2.29)
can be written as
Γijk akl = rij · rl (4.22)
∂X m ∂X m ∂X n ∂X n
Γijk =
∂ξ k ∂ξ l ∂ξ i ∂ξ j ∂ξ l
and, exploiting the chain rule,
∂X m ∂X m ∂ξ k ∂ξ l ∂X n ∂X n ∂ξ k ∂ξ l
Γijk k l p r
= i j −→
∂ξ ∂ξ ∂X ∂X ∂ξ ∂ξ ∂ξ l ∂X p ∂X r
∂X n n ∂ξ k
Γijk δpm δrm = δ
∂ξ i ∂ξ j r ∂X p
finally one obtains

∂X n ∂ξ k
Γijk = . (4.23)
∂ξ i ∂ξ j ∂X n
Now the Lie derivative (equation (4.21)) can be written as:
∂ai ∂ai ∂X j ∂ξ k i ˙n
Lt a = = ri + n ξ˙n ri + i n a ξ rk (4.24)
∂t ∂ξ ∂ξ ∂ξ ∂X j
∂ai ∂ai
= ri + n ξ˙n ri + Γijk ξ˙j ai rk .
∂t ∂ξ
This is a full vector derivative (2.33) including the covariant derivative (2.34).
The Lie time derivative is usually exploited for the linearization, therefore,
the computation of the covariant derivatives will be employed for further
linearization. In the further considerations we concentrate on the full time
derivative on the tangent plane. Then the values of the spatial Christoffel
symbols on the surfaces Γ̂IJ |ξ =0 , i.e. if ξ 3 = 0, define a value of covariant
K 3
derivatives for any spatial object on the tangent plane. It can be easily seen
from their definition and the Weingarten formula that the following relations
between the spatial and surface terms hold:
Γ̂ijk |ξ3 =0 = Γijk , i, j, k = 1, 2
Γ̂ij3 |ξ3 =0 = 0 (4.25)

k
Γ̂3j |ξ3 =0 = −hkj ,
where Γijk are the surface Christoffel symbols (2.29) and hkj are mixed com-
ponents of the curvature tensor.
With the vector T in the tangent plane in covariant components, i.e.
T = Ti ri |ξ3 =0 = Ti ρi ; (4.26)
its full time derivative is computed employing the rules given in (2.32) and
(2.35)

dTi ∂Ti ∂Ti
= + − Γ̂ K
IJ T K ξ̇ J −→
dt ∂t ∂ξ J 3
ξ =0

∂Ti ∂Ti
= + − Γijk Tk ξ˙j + hki Tk ξ˙3 (4.27)
∂t ∂ξ j
One should distinguish that the full time derivative with the surface Christof-
fel symbols in the form given in (2.32) and (2.35) can be applied to an object
that belongs to the internal geometry of the surface, e.g. for ξ̇ i ; for the full
time derivative of an arbitrary spatial object, however, positioned in the tan-
gent plane, the form in (4.27) must be used.
4.1.5 Motion of a Slave Point

The “slave” point S in the local coordinate system (4.4) has the local co-
1 1
ordinate ξC , ξC , ξ 3 , where ξC 1 1
, ξC are convective surface coordinates of the
projection point C on the surface, see Fig. 4.1. We now consider the mo-
tion of the “slave” point S in the local coordinate system, assuming that
the “master” surface is moving, i.e. the surface vector ρ(t, ξ 1 , ξ 2 , ξ 3 ) as well
as the normal n(t, ξ 1 , ξ 2 , ξ 3 ) are time dependent. Within a quasi-static and
static processes, the time t is an incremental load parameter. Then the full
time derivative becomes
d d d
rs (t, ξ 1 , ξ 2 , ξ 3 ) = ρ + (nξ 3 ) = (4.28)
dt dt dt
∂ρ ∂ρ ∂n 3 ∂n
= + j ξ˙j + ξ + nξ˙3 + j ξ 3 ξ˙j , j = 1, 2.
∂t ∂ξ ∂t ∂ξ
Let a point C be a projection point of the “slave” node onto the master
∂ρ
surface. Denote the translational velocity of the point C as v = and the
∂t
d
velocity of the “slave” point as vs = rs (t, ξ 1 , ξ 2 , ξ 3 ). Then equation (4.28)
dt
has the following form, using the Weingarten formula (2.23),
∂n
vs = v + ξ 3 + nξ˙3 + (ρj − ξ 3 hij ρi )ξ˙j , i, j = 1, 2, (4.29)
∂t
where hij are mixed components of the curvature tensor.
The difference between the velocity vs of “slave” point S and the velocity
of point C is a relative velocity vrel of the “slave” point, or in other words,
the velocity of point S as it can be seen from point C.
Convective Normal Velocity – Rate of Penetration ξ̇ 3

∂n
The vector is orthogonal to n due to the fact that n is a unit vector:
∂t
∂n
n · n = 1 −→ ·n= 0 (4.30)
∂t
Evaluating, then the dot product of eqn. (4.29) with r3 = n, and using the
last expression (4.30), we obtain the projection of the relative velocity on the
normal, or the full time derivative of the penetration:
ξ˙3 = (vs − v) · n. (4.31)
Convective Tangent Velocities ξ̇ i

The scalar values of convective tangent velocities ξ˙i can be obtained from
(4.29) either by evaluating the dot product with the surface vector ρi , or by
evaluating the dot product with the base vectors ri defined in eqn. (4.6).
Evaluation via the Surface Tangent Vectors ρi
In this case the dot product of the relative velocity vrel ≡ vs − v and the
coordinate vector ρi is taken leading to
∂n
(vs − v) · ρi = (aij − ξ 3 hij )ξ˙j + ξ 3 ( · ρi ), (4.32)
∂t
where aij are the components of the metric tensor. Therefore, the convective
velocities are defined as
∂n
ξ˙j = âij [(vs − v) · ρi − ξ 3 ( · ρi )], (4.33)
∂t
where âij are contravariant components of the tensor with components aij −
ξ 3 hij .
Evaluation via the Spatial Curvilinear Basis Vectors ri
A dot product of eqn. (4.29) with the base vectors ri gives the following
expression:
(vs − v) · (ρi − ξ 3 hki ρk ) =

∂n
= ξ3 · (ρi − ξ 3 hki ρk ) + (aij − 2ξ 3 hij + (ξ 3 )2 hki hjk )ξ˙j , (4.34)
∂t
from which an expression for the first two convective velocities is obtained:

˙ξ j = âij (vs − v) · ρ − ξ 3 ∂n · ρ + hk (vs − v) · ρ , (4.35)
i i i k
∂t
where âij are components of the inverse matrix (aij − 2ξ 3 hij + (ξ 3 )2 hki hjk ),
or in other words the contravariant components.
Evaluation on Tangent Plane with ξ 3 = 0
From kinematic point of view the evaluation with the basis vectors ri via
eqn. (4.35) gives the expression of the relative velocity vs − v if the observer
is positioned at the slave point C, while evaluation with the surface tangent
vectors ρi gives us rather the result of the projection of the relative velocity
vs − v onto the the surface tangent vectors ρi . Considering both equations
(4.33) and (4.35) at ξ 3 = 0 we obtain the result on the tangent plane:
ξ˙j = aij (vs − v) · ρi . (4.36)
Interpretation of this procedure is that the observer is positioned at the tan-

gent plane with ξ 3 = 0 where the contact namely is appearing and he is
observing the relative velocity of the slave point S.
This discussion is related to further considerations based on the follow-
ing assumption: Only the contact problem is considered, but not the motion
and deformation of the two body system connected by means of the normal
vector with coordinate ξ 3 . The penetration is assumed to be very small, as
usual during the solution of contact problems. Further, the global iteration
procedure for the solution leads to a decreasing value of the penetration ξ 3 .
This definition of the convective velocity (4.36) on the tangent plane allows
to consider the contact kinematics on the master surface only and to exploit
the differential geometry of the surface during further considerations. The
formula in (4.36) was mentioned in Wriggers [189] as a possible simplifica-
tion as well. Further we show that from a mathematical point of view this
simplification leads to a consistent expression of the contact integral and, as
it can be seen from numerical results, also leads to high numerical efficiency.
Estimation the Difference between Different Evaluations
In order to estimate the difference between the expression in equation (4.33)

and the evaluation on the tangent plane eqn. (4.36), a series expansion of
Aij assuming ξ 3 as a small parameter is performed. Suppose A is a metric
tensor with components aij , H is a curvature tensor with components hij
and tensor Â with components aij − ξ 3 hij
Â−1 = (A − ξ 3 H)−1 = A−1 + ξ 3 A−1 HA−1 + O((ξ 3 )2 ) (4.37)
where A−1 is the contravariant metric tensor with components aij . Then the
convective velocity has the following form:
ξ˙j = aij (vs − v)·ρi + ξ 3 [ aik amj hkm (vs − v)·ρi − aij ṅ·ρi ]+ O((ξ 3 )2 ) (4.38)
It is obvious that (4.36) describes the constant main part of the full equation.
4.1.6 Geometrical Interpretation of Covariant

Derivative and Numerical Realization
The covariant derivatives require C 1 continuity of the surface. Lack of the sur-
face continuity leads to oscillations in the characteristics, e.g. at the crossing
of element boundaries. Therefore, various approaches based on the usage of a
C 1 approximation of the surface with Hermite splines, NURBS etc. were de-
veloped e.g. in the following articles [192], [135], [37], [2], [149], [170]. Wriggers
et. al. [192] have shown that for C 1 continuous contact surfaces a continuity
problem of internal parameters on the element boundary arises and proposed
an algorithm for the 2D case, based on the usage of the path length of the
projection point. Puso and Laursen [149] proposed to determine increments
of convective coordinates in the geometric form for the 3D case. Here we
construct a numerical algorithm based on a geometrical interpretation of the
covariant derivative as a parallel translation, see Marsden and Hughes [122].
The result of this section will be used for the computation of the contact
tractions within the return-mapping algorithm.
Continuous Numerical Integration Algorithm for a Relative

Motion Vector Δρ
Consider a relative motion of the projection point C on the master surface.
The relative velocity vector of this motion is laying in the tangent plane, i.e.
vr = ξ˙i ρi . (4.39)
We are interested in the relative distance Δρ which was passed by point C

from step (n) to step (n + 1). For the C 1 -continuous surface and continuous
convective coordinates we can write the following
ρ(n+1) − ρ(n) = ρ(ξ i + Δξ i ) − ρ(ξ i ) = ρi Δξ i + O((Δξ i )2 ) (4.40)
We define the incremental vector Δρ at step (n + 1) as

(n+1)
Δρ = ρi Δξ i , (4.41)
from which the incremental components Δξ i are derived as
Δξ i = (Δρ · ρj ) aij
(n+1) . (4.42)
If the convective coordinates are no longer continuous then the incremental

vector Δρ can not be derived via (4.41), but it can be derived directly in the
3D space. For illustration, see Fig. 4.2, we consider at step (n) two adjacent
patches A(n) B (n) D(n) G(n) and G(n) D(n) E (n) F (n) for a C 1 -continuous sur-
face, i.e. a surface normal n being continuous while crossing the line D(n) G(n) ,
but with independently defined convective coordinates of the patches. Let
S (n) be a slave point and C (n) its projection onto the patch A(n) B (n) D(n) G(n)
at step (n). A pair of points S (n) and C (n) defines then a spatial coordinate
system, (4.4). Now we consider a case, when at the next step (n+ 1) the same
pair is shifted into a position S (n+1) and C (n+1) with the slave point pro-
jected onto the adjacent patch G(n+1) D(n+1) E (n+1) F (n+1) to obtain C (n+1) .
On the surface it can be interpreted as a motion of the projection point from
position C̃ (n) to position C (n+1) , where the projection point has been cross-
ing the line D(n+1) G(n+1) (see a vector Δρ = C̃(n) C(n+1) in Fig. 4.2). Since
a moving surface is considered, point C (n) is shifted in the 3D space to the
position C̃ (n) by the vector u. Thus, the increment vector Δρ is obtained in
the global reference Cartesian system as
Δρ = ρC (n+1) |ξ(n+1)
1 , 2
ξ(n+1) − (ρC (n) + uC (n) ) |ξ(n)
1 , 2
ξ(n) (4.43)
The computation in the global reference Cartesian system clearly defines the
increment vector and, therefore, allows to avoid jumps which would occur
with the local convective coordinates ξ i . It should be noted that vectors ρC (n)
and ρC (n+1) are defined after the closest point projection procedure, therefore
(n) (n)
the information about internal variables ξ1 , ξ2 must be stored. However,
within the “segment-to-analytical surface”–approach the value of penetration
is computed at the same integration points, i.e. ρC (n) ≡ ρC (n+1) . Then it is
only necessary to keep the information about the increment vector u from
the last load step in the global coordinate system. (4.42) is then reduced to
Δξ i = −(u · ρj )aij
(n+1) . (4.44)
Summarizing the result we obtain the rule for the continuous numerical al-
gorithm to compute the increment vector:
The increment vector Δρ is defined in the spatial coordinate system at step

(n + 1) by its projection in (4.41), where the increments Δξ i are computed
via (4.42) and (4.43), or in the case of the “segment-to-analytical surface”–
approach via (4.44).
Parallel Translation of a Vector T on the Tangent Plane

The full time derivative of a vector T in the covariant form in eqn. (4.27)
describes its change along the tangent plane. The geometrical interpretation
of the numerical increment analogy is to consider the evolution of the vector
T by enforcing its position on the tangent plane. This operation is called
“parallel translation” in differential geometry terminology, see e.g. Gray [46],
Schoen [155] and applications in mechanics in Marsden and Hughes [122].
This interpretation also allows to overcome a variation in the representation
of the vector T due to different local element coordinate systems.
(n)
F (n+1)
(n) S
G 00
11
(n) 11
00
S 00
11 (n+1)
0
1
1
0 F
0
1
11111
00000 (n)
00000
11111 00000 (n+1)
11111
00000
11111
00000
11111 E 00000
11111
00000
11111 ρ2(n) 00000ρ2
11111
00000
11111 00000
11111
(n) (n) 10
1
00000
11111
0
00000
11111 (n+1)11111
00000
0
1
00000
11111
0
1
000000
111111 00
11
0000000000000000000000
1111111111111111111111 00000
11111
A 111111 C
000000
111111
0
1
00
11
0000001111111111111111111111
0000000000000000000000
00
11
0000000000000000000000
1111111111111111111111
(n) C 1111111111
(n+1) 11111
00000
00000
11111
0000000000
00000
11111
0
1
0000000000
1111111111
00000
11111
00000
11111
000000
111111 00
11
0000000000000000000000
1111111111111111111111
T (n) G 00000
11111 00000
11111 (n+1)
000000
111111 00
11
0000000000000000000000
1111111111111111111111 D 00000
11111 00000
11111 T
0000001111111111111111111111
111111 00
11
0000000000000000000000 00000
11111 00000
11111
000000
111111 00
11
0000000000000000000000
1111111111111111111111 00000
11111 00000
11111
000000
111111 00
11
0000000000000000000000
1111111111111111111111 00000
11111 00000
11111 (n+1)
000000
111111
000000
111111 0000000000000000000000
1111111111111111111111
(n) 00000
11111 ρ1
0000001111111111111111111111
111111 0000000000000000000000
ρ1
0000000000000000000000
1111111111111111111111 00000
11111
Δ11111
ρ
00000
11111 (n+1)
000000
111111
0000001111111111111111111111
111111 0000000000000000000000 00000
000000
111111 0000000000000000000000
1111111111111111111111 u 00000
11111 E
000000
111111 0000000000000000000000
1111111111111111111111
0000000000000000000000
1111111111111111111111 00000
11111
00000
11111
000000
111111 0000000000000000000000
1111111111111111111111 00000
11111
000000
111111 0000000000000000000000
1111111111111111111111
(n) 00000
11111
000000
111111 0000000000000000000000
1111111111111111111111
B 00000
11111~(n)
000000
111111 0000000000000000000000
1111111111111111111111 00000
11111
000000
111111 0000000000000000000000
1111111111111111111111
0000001111111111111111111111
111111 00000
11111C (n+1)
0000000000000000000000
0000000000000000000000000000
1111111111111111111111111111
000000
111111 00
1100000
11111
rC(n)
0000000000000000000000000000
1111111111111111111111111111
000000
111111 00
11 D
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111
Z 1111111111111111111111111111
0000000000000000000000000000
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111 (n+1)
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111 A
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111 r C~ (n)
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111
0000000000000000000000000000
1111111111111111111111111111
000000
111111
O Y
0000000000000000000000000000
1111111111111111111111111111
000000
111111
(n+1)
0000000000000000000000000000
1111111111111111111111111111
000000
111111 B
0000000000000000000000000000
1111111111111111111111111111
000000
111111
Fig. 4.2 Contact point moving across segment boundaries. Covariant derivatives.
Sketch of integration scheme.
If T(n) is defined at the step (n), see Fig. 4.2, and eK are basis vectors of
the global Cartesian coordinate system, then the vector T can be written in
both local and global coordinate systems as
(n) (n) (n)

∂xK
(n)
T= Ti aij(n) ρj = Ti aij(n) eK . (4.45)
∂ξ j
Projections of this vector to the new basis at state (n + 1) gives us the vector
T(n+1) translated in parallel. This operation in the Cartesian coordinate
system leads to:
(n+1) (n+1)
Tl = T(n) · ρl =
(n)
∂xK
(n) ∂xM
(n+1)
= Ti aij(n) ∂ξ j eK · eM =
∂ξ l
(n) ij
∂xK K
(n) ∂x(n+1)
= Ti a(n) , (4.46)
∂ξ j ∂ξ l
or in compact form
(n+1) (n) ij (n) (n+1)

Tl = Ti a(n) (ρj · ρl ). (4.47)
In other words, this operation can be seen as a pull-back from the current
configuration at time (n) into the reference configuration and then a push-
forward into the current configuration at time (n + 1). This procedure allows
to keep continuity due to the use of the same reference configuration.
Remark 4.1. In the case of translation in a plane, the metric tensor is constant
and (4.47) defines a standard parallel shifting
(n+1) (n) ij (n) (n+1) (n) ij (n) i (n)
Tl = Ti a(n) (ρj · ρl ) = Ti a ajl = Ti δl = Tl . (4.48)
4.1.7 Variation δξ 3 and Its Linearization

Variation of the third coordinate δξ 3 is written according to the kinematic
analogy between velocity and variation similar to eqn. (4.31)
δξ 3 = (δrs − δρ) · n. (4.49)

Linearization is taken as a full time derivative assuming that both the projec-
tion point ρ and the normal vector n are functions of convective coordinates
ξ 1 , ξ 2 and time t.
d d
(δξ 3 ) = ((δrs − δρ) · n) = (4.50)
dt dt
∂ dξ i ∂
= (δrS − δρ) · n + (δrS − δρ) · n
∂t dt ∂ξ i

∂n ∂n dξ i
+ (δrS − δρ) · + i =
∂t ∂ξ dt
∂n
= −δρ,i · ξ˙i n + (δrs − δρ) · +
∂t
+ (−(δrs − δρ)hji · ξ˙i ρj )],
where the Gauss-Codazzi formula in eqn. (2.28) has been used. The variations
δrs δρ are time independent, thus
∂
(δrs − δρ) = 0 (4.51)
∂t
and the first term in eqn. (4.50) is disappearing.
∂
Operations δ and are commutative, thus we are writing as
∂ξ i
∂δρ ∂ρ
= δ i = δρ,i (4.52)
∂ξ i ∂ξ
We evaluate separately the following terms
Term A: −δρ,i · ξ˙i n (4.53)

∂n
Term B: (δrs − δρ) · (4.54)
∂t
Term C: −(δrs − δρ)hji · ξ˙i ρ (4.55)
Term A:

−δρ,i · ξ˙i ·n = −δρ,i · (vS − v) · ρj aij ·n = −(δρ,i ·n)aij ρj ·(vS −v) (4.56)
Term B:
∂n
(δrs − δρ) · = (4.57)
∂t
Any derivative of the unit vector n is orthogonal to this vector.
Proof. Take a derivative of the unit vector:
n·n=1 (4.58)
as
∂n
·n= 0 (4.59)
∂t
∂n
the last scalar product shows the orthogonality of and n vectors.
∂t
∂n
Due to orthogonality, the vector can be represented via the tangent vec-
∂t
tors ρi as
∂n
= ai ρi (4.60)
∂t
and multiplying by ρj :
∂n
· ρj = ai (ρi · ρj ) (4.61)
∂t
ai are defined as

∂n
i
a = · ρj aij (4.62)
∂t
where
ρi · ρj = aij (4.63)
aij are covariant components of the metric tensor and aij are contravariant
components of the metric tensor.
Finally we have:

∂n ∂n
= · ρj aij ρi . (4.64)
∂t ∂t
We can use the orthogonality of n and ρi
n · ρi = 0 , (4.65)
take the partial derivative with respect to t:

∂n ∂ρi
· ρi + n · =0
∂t ∂t
∂ρi ∂ ∂ρ ∂ ∂ρ ∂v
= = i = i = v,i . (4.66)
∂t ∂t ∂ξ i ∂ξ ∂t ∂ξ
Summarizing we have the following representation of the derivative:
∂n ∂ρi ij
= −n · a ρj = (n · v,i )aij ρj . (4.67)
∂t ∂t
Term B becomes finally
(δrS − δρ) · ρj (n · v,i aij ) . (4.68)
Term C is transformed as follows:
−(δrs ) · hji · ξ˙i ρj = −(δrs − δρ) · hji · (vs − v) · ρk aki ρj =
= −(δrs − δρ) · ρj hji aki ρk · (vs − v) = (δrs − δρ) · hjk ρj ⊗ ρk (vs − v) , (4.69)
where the formula for convective velocities in eqn. (4.36) is used
Terms A, B and C together are leading to
d
(δξ 3 ) = δρ,i · (n ⊗ ρj )aij (vs − v) + (δrs − δρ) · (ρi ⊗ n)aij v,i
dt
− (δrs − δρ) · hij ρi ⊗ ρj (vs − v) (4.70)
Remark 4.2. During transformation the following formula from the tensor
algebra has been exploited
(a · b)(c · d) = a · b ⊗ cd (4.71)
4.1.8 Variation δξ i and Its Linearization

The variation of the convective coordinates δξ i is written according to the
kinematic analogy between convective velocities and variations similar to
eqn. (4.36). We express them in the tangent plane:
δξ j = aij (δrs − δρ) · ρi . (4.72)

The linearization of the variation of the convective coordinates δξ i is one of
the important parts which requires the results about differential operations
in the spatial coordinate system from Section 4.1.5 together with the tensor
algebra operations on the tangent plane. The full time derivative gives

i ∂ ∂ ˙j
L(δξ ) = + ξ (δξ i )
∂t ∂ξ j
daik d
= (δrs − δρ) · ρk + aik [(δrs − δρ) · ρk )] . (4.73)
dt dt
Linearization of (δrs − δρ) · ρk
requires the application of the Gauss-Codazzi formula (2.28):

d
[(δrs − δρ) · ρk )] = (4.74)
dt
= ((δrs − δρ),j · ρk )ξ˙j + (δrs − δρ) · vk )+
l
+Γkj ((δrs − δρ) · ρl )ξ˙j + hkj ((δrs − δρ) · n)ξ˙j .
Linearization of the Contravariant Components aij
was already given in eqn. (4.12).
d i
Simplification of δξ
dt
The final formula is long, but can be simplified. In addition, the following
transformations are cumbersome but necessary to show the symmetry of the
tangent matrix in the case of sticking. Summarizing the results in one formula,
we obtain
d
(δξ i ) =
dt
= −aim ank (vm · ρn )((δrs − δρ) · ρk ) (4.75a)
−aim ank (ρm · vn )((δrs − δρ) · ρk ) (4.75b)
−aim ank Γmj

l
(ρl · ρn )ξ˙j ((δrs − δρ) · ρk ) (4.75c)
−aim ank Γnj

l
(ρm · ρl )ξ˙j ((δrs − δρ) · ρk ) (4.75d)
+2aim ank hmn ξ˙3 ((δrs − δρ) · ρk ) (4.75e)
+aik ((δrs − δρ)),j · ρk )ξ˙j (4.75f)
+aik (δrs − δρ) · vk (4.75g)

+aik Γkj
l
((δrs − δρ) · ρl )ξ˙j (4.75h)
+aik hkj ((δrs − δρ) · n)ξ˙j (4.75i)

The nine parts in eqns. (4.75a–4.75i) will be tremendously simplified, if we
take into account the expression for the convective velocities (4.36) and con-
sider tensor operations on the tangent plane. The following five transforma-
tions will lead to a simple structure:
a.
The sum of the terms (4.75a) and (4.75g) becomes zero on the surface:
−aim ank (vm · ρn )((δrs − δρ) · ρk ) + aik (δrs − δρ) · vk = 0. (4.76)
In order to show this the dot product in the second term is expressed on the
tangent plane, i.e. as double sum with the surface metric tensor components
aij :
aim (δrs − δρ) · vm = aim ((δrs − δρ) · ρk )ρ k

· (vm · ρn )ρ n
=
= aim akn ((δrs − δρ) · ρk )(vm · ρn ),

from which (4.76) is obtained.
b.
The sum of the terms (4.75c) and (4.75h) becomes zero on the surface:
−aim ank Γmj

l
(ρl · ρn )ξ˙j ((δrs − δρ) · ρk ) + aik Γkj
l
((δrs − δρ) · ρl )ξ˙j =

= −aim ank aln Γmj
l
((δrs − δρ) · ρk ) + aik Γkjl
((δrs − δρ) · ρl ) ξ˙j =

= −aim akl Γmj
l
((δrs − δρ) · ρk ) + aik Γkj
l
((δrs − δρ) · ρl ) ξ˙j =

= −aim Γmjk
((δrs − δρ) · ρk ) + aik Γkj
l
((δrs − δρ) · ρl ) ξ˙j = 0.
Here the properties of the covariant and contravariant components (2.6) and
(2.7) have been used.
c.
The sum of (4.75b) and (4.75f) leads to a symmetrical rotational part. We
start with using the expression for the convective velocities (4.36):
−aim ank (ρm · vn )((δrs − δρ) · ρk ) + aik ((δrs − δρ),j · ρk )ξ˙j =
= −ail ajk (ρl · vj )((δrs − δρ) · ρk ) − aik ajl (δρ,j · ρk )((vs − v) · ρl ) =
= −(δrs − δρ) ail ajk ρk ⊗ ρl vj − δρ,j aik ajl ρk ⊗ ρl (vs − v). (4.77)
The final expression is found via the tensor product.

d.
After grouping (4.75e) with (4.75i), we obtain
2aim ank hmn ξ˙3 ((δrs − δρ) · ρk ) + aik hkj ((δrs − δρ) · n)ξ˙j =
= aim ank hmn ξ˙3 ((δrs − δρ) · ρk )+ (4.78a)
aim ank hmn (vs − v) · n)((δrs − δρ) · ρk )
+aik ajm hkj ((δrs − δρ) · n)((vs − v) · ρm ) = (4.78b)
In order to show the symmetry of the part in eqn. (4.78b), the tensor product
and contravariant components of the curvature tensor eqn. (2.16) are used.
For a reduction of eqn. (4.78a) the equation for the variation of the convective
velocity (4.36) and mixed components of the curvature tensor are taken,
leading finally to
= hin ξ̇ 3 δξ n + (4.79a)

+hij (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v) (4.79b)
The last part (4.79b) defines the curvature part of the tangent matrix.
e.
The equation for the variation of the convective velocity (4.36) is used to
simplify (4.75d):
−aim ank Γnj

l
(ρm · ρl )ξ˙j ((δrs − δρ) · ρk )
i ˙j
= −Γkj ξ δξ k (4.80)
The resulting parts in eqn. (4.79a) and (4.80) remain untransformed, however
they will disappear in both sticking and sliding cases, after taking into account
the fully linearized contact integral together with the evolution equations for
tangential contact traction as shown in the Sections 7.1.3 and 7.1.4.
Summarizing the result of the complete transformation, we obtain
d
(δξ i ) =
dt
= −(δrs − δρ) ail ajk ρk ⊗ ρl vj − δρ,j aik ajl ρk ⊗ ρl (vs − v) (4.81a)

+hij (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v)+ (4.81b)
+hin ξ˙3 δξ n − Γkj

i ˙j
ξ δξ k (4.81c)
4.2 Kinematics of 2D Contact Interaction 83
Thus, the full time derivative consists of a symmetrical rotational part

(4.81a), a symmetrical pure curvature part (4.81b) and a connection part
with the Christoffel symbols (4.81c), describing the connection properties.
4.2 Kinematics of 2D Contact Interaction

In the literature various contact descriptions for an effective finite element
implementation are available, which can be basically characterized by the
following: from 2D to 3D formulations, from non-frictional to frictional con-
tact. Some major references are cited in the following: Wriggers et. al. [195]
used an elasto-plastic analogy and the penalty regularization for 2D frictional
problems restricted to piecewise linear contact elements. Parisch [138] consid-
ered non-frictional 3D contact and Parisch and Lübbing (1997) [139] revised
the procedure for frictional contact within the penalty method for piecewise
bilinear surface elements. Peric and Owen [140] used the penalty method for
3D frictional contact problems with small deformations. The main charac-
teristics of the cited investigations are that the penalty functional as well as
its linearization were considered in a global coordinate system and restricted
to linear resp. bilinear surface elements. Laursen and Simo [109], however,
formulated the penalty based contact conditions and the return mapping al-
gorithm via convective surface coordinates, but the following linearization
performed in the global coordinate system led to an artificial non-symmetry
of the tangent matrix in the case of sticking. Wriggers [190] could overcome
this artifact using the idea of mesh tying functionals. General overviews over
contact conditions and contact algorithms which are nowadays used in prac-
tice, are covered by the books of Wriggers [190] and Laursen [106], while the
theoretical aspects of the regularization methods in contact mechanics can be
found in Kikuchi and Oden [84]. Beyond that the covariant description pro-
posed in Konyukhov and Schweizerhof [87], [88] allows a unified description
of contact problems within the penalty method independently of the surface
discretization. The method contains rather complicated mathematical trans-
formations in the local 3D coordinate system, however, finally leading to the
consistent formulation of frictional contact. Some advantages of this approach
were shown, e.g. the sticking matrix preserves necessarily its symmetry.
In the current section, we aim to present the development in a more simple
comparative manner for both 2D and 3D formulations. We will show the unity
of 2D and 3D formulations, where the 2D case can be derived, from one hand,
as a simplified case of the particular 3D geometry of contact surfaces and,
from the other hand, can be constructed separately based on the differential
geometry of 2D plane curves. This consideration has additional advantages,
e.g. the subdivision of the contact tangent matrices into the ”main”, the
”rotational” part and the ”curvature” part has a pure geometrical meaning.
It is also possible to distinguish a-priori various cases, where some of the
parts are necessary or can be omitted.
We start with 2D kinematics based on a curved geometry in a plane. The

main results concerning the 3D covariant description can then be presented
without extensive involvement into mathematics. Two-dimensional contact is
considered separately in 2D space, in the coordinates system inherited with
CPP for 2D curve as described separately in Sect. 3.2.1 as well as a reduction
of the 3D developments.
Considering a special contact case – contact between two cylindrical infi-
nite bodies with plane strain deformations, see Fig. 4.3, leads to a definition
of a 2D contact. In this case a generatrix GH of the first cylindrical body is a
contact line and corresponds to a contact line G’H’ which is also a generatrix
but of the second cylindrical body. Thus, 3D contact which can be seen as
an interaction between two surfaces is reduced to an interaction between two
boundary curves in the 2D case, see Fig. 4.4. One of boundary curves is cho-
sen as the master curve. A coordinate system is considered on the boundary,
either for a surface in 3D or for a curve in 2D. Thus contact occurs or two
bodies are coming into contact, if a slave point belonging to the second body
S penetrates into the master body, where penetration is defined as the short-
est distance between the surfaces of the two bodies. For simplicity we assume
now that the parametrization of the boundaries is sufficiently smooth.
4.2.1 Closest Point Projection Procedure and

Corresponding Coordinate System
Let the boundary of the master body be a smooth curve, parametrized by the
parameter ξ: ρ = ρ(ξ). The vector rs describes the location of a slave point
S, see Fig. 4.4. Then the problem to find the shortest distance between the
curve ρ(ξ) and the slave point S is defined via the minimum of the function:
F := rs − ρ(ξ) −→ min . (4.82)

The necessary condition for the minimum is the requirement of the first
derivative to be zero:
dρ
F = (rs − ρ(ξ)) · = 0. (4.83)
dξ
The existence and uniqueness of CPP procedure we discussed already in
Sect. 3.2.1 after transformation into arc-length parameter s. Eqn. (4.83) is
identical to the orthogonality condition between the vector rs − ρ(ξ) and
dρ
the tangent vector ≡ ρξ , and serves to define a projection point C, see
dξ
Fig. 4.4. The solution can be obtained e.g. by an iterative Newton scheme.
For the latter the second derivative is necessary:
d2 ρ dρ dρ
F = (rs − ρ(ξ)) · − · , (4.84)
dξ 2 dξ dξ
which is finally shown to be positive to specify the minimum distance. The

iterative scheme is then defined as:
⎧
⎪
⎪ (rs − ρ) · ρξ
⎨ Δξ = −F /F = −
(rs − ρ) · ρξξ − (ρξ · ρξ ) . (4.85)
⎪
⎪
⎩ (n+1) (n)
ξ = ξ + Δξ
We will show in the finite element implementation section, that for a 2D
contact element with a linear approximation the general iterative scheme is
reduced to an exact definition of the projection point.
In the 2D case contact bodies are bounded by plane curves, therefore, one
can take advantage of their geometry. The geometrical properties of the con-
tact quantities can be defined in a very straightforward manner, if we use the
D H’
E
S slave body
G’ H
^r n= ν
3= r^2 =−k
r^1 =ρξ
A
G B
Z
ξ
Y
master body
X
Fig. 4.3 Two dimensional contact as a special case of three dimensional contact –
contact between cylindrical surfaces with parallel axes Z. Local surface coordinate
system on smooth master contact surface.
slave body Ε
D S (t+ t)
t)
S(t) τ (t+
t)
ζ τ
(t+
rs ν τ (t)
ν (t)
C (t) τ (t) ρ
Α ξ
B
y s( ξ)
ρ(ξ) master body
x
Fig. 4.4 Two dimensional contact. Definitions. Contact boundaries are smooth
curves in 2D.
natural parameter length, i.e. ρ = ρ(s) with s = s(ξ). On the plane we define
a curvilinear coordinate system associated with the curve by introducing two
∂ρ
principal vectors as a basis: the tangent vector ρξ = and the unit normal
∂ξ
vector ν
rs (ξ, ζ) = ρ(ξ) + ζν(ξ). (4.86)

Looking at the following implementation of the algorithm in a FE program,
the introduction of a natural parameter s would lead to additional numerical
effort, because the length of a boundary is changing during deformation.
Thus, we will only show the geometrical properties using the parameter s,
whereas for a finite element implementation we then turn to the Lagrangian
coordinate ξ. As the local coordinate system is associated with the slave point
S, then the closest point procedure (4.83) is already fulfilled by this definition.
The second coordinate ζ is an exact (not scaled!) value of penetration often
also known as gap and used for the formulation of the non-penetrability
condition.
The normal unit vector ν in the case of arbitrary Lagrangian parametriza-
tion with ξ can be defined via a cross product in a Cartesian coordinate
system as:
∂ρ [k × ρξ ]
ρξ = ; =⇒ ν=√ , (4.87)
∂ξ ρξ · ρξ
where k is the third unit vector in this Cartesian coordinate system. The
definition of the normal vector in (4.87) comes from the standard definition
of the surface unit normal for cylindrical surfaces, see the contact between
cylinders in Fig. 4.3, as
[r̂1 × r̂2 ]
ν= , r̂1 = ρξ , r̂2 = k, (4.88)
|[r̂1 × r̂2 ]|
where r̂2 is a unit vector of the cylinder generatrix. The definition in (4.87)
gives a set of covariant basis vectors ρξ , k for a cylindrical surface. The
surface metric tensor can then be defined by the following matrix aij :

(ρξ · ρξ ) 0
[aij ] = (4.89)
0 1
This matrix allows to define a contravariant basis for the cylindrical surface
ρ1 , ρ2 , where only the first vector is changing its length:

−1

ρ1 (ρξ · ρξ ) 0 ρξ ρξ
= =⇒ ρ1 = , ρ2 = ν. (4.90)
ρ2 0 1 ν (ρξ · ρξ )
We note here, that the dot product of the covariant basis with the contravari-
ant basis ρ1 , ρ2 leads to a unit matrix (a mixed metrics tensor):

j 10
[ai ] = [(ρi · ρ )] =
j
. (4.91)
01
Derivatives of the Basis Vectors

Derivatives of the covariant basis vectors ρξ and ν are necessary for the
further formulation and for the linearization. The derivative of the tangent
vector ρξ can be expressed via the covariant basis vectors
∂ρξ
ρξξ := = Γ ρξ + h11 ν, (4.92)
∂ξ
where Γ and h11 are defined after taking a scalar product of (4.92) with ρξ
and ν:
ρξξ · ρξ
Γ = , h11 = ρξξ · ν. (4.93)
ρξ · ρξ
In order to compute the derivative of the unit normal ν, first, we have to
take the derivative of the following identity:
∂ ∂ν
(ν · ν = 1) =⇒ · ν = 0, (4.94)
∂ξ ∂ξ
∂ν
which leads to the orthogonality condition of the vectors ν and . Thus,
∂ξ
this derivative can be expressed via the tangent vector ρξ as:
ν ξ = aρξ . (4.95)
Then the scalar a is obtained after taking the dot product of (4.95) and ρξ :
a = (ν ξ · ρξ )/(ρξ · ρξ ) = −(ρξξ · ν)/(ρξ · ρξ ). (4.96)
The last equation is obtained via the derivative of the equation ρξ · ν = 0.

Finally, we get
(ρξξ · ν) h11
νξ = − ρ =− ρ . (4.97)
(ρξ · ρξ ) ξ a11 ξ
In (4.93) and (4.97) the scalar value h11 = (ρξξ · ν) is a curvature coefficient
and a11 = (ρξ · ρξ ) a metric coefficient for a cylindrical surface. Eqn. (4.97)
represents the Weingarten formula and eqn. (4.92) represents resp. the Gauss-
Codazzi formula for a cylindrical surface see comparison with eqn. (2.23) and
resp. with eqn. (2.28) for surfaces in 3D. Now, we can obtain the basis vectors
for the coordinate system in (4.86):
∂r h11
r1 = = (1 − ζ)ρξ ,
∂ξ a11 (4.98)
r2 = ν.
The geometrical properties in 2D contact are easily found, if they are recon-
sidered from the plane curve geometry with a natural parametrization. This
parametrization is based on a length parameter s, i.e. ρ = ρ(s). The length
s represents then the full path length passed by the projection point C on
the master curve during contact interaction, see Fig. 4.4. The tangent vector
τ in this case has a unit length
dρ
τ = . (4.99)
ds
The relation between the parameter ξ and the length parameter s is defined
as:
ds = (ρξ · ρξ )dξ. (4.100)
For the plane curve in natural parametrization the Serret-Frenet formulas are
used to define derivatives of the basis vectors in analogy to (4.92) and (4.97):
dτ dν
= κν; = −κτ , (4.101)
ds ds
where κ is a curvature of the curve. In this case the normal unit vector ν is
defined from (4.101) and is pointing into the convex part of a body. Then the
curvature κ can be computed from (4.101) by multiplying the first equation
with ν and taking the chain rule into account:
dτ (ρξξ · ν) h11
κ= ·ν = = . (4.102)
ds (ρξ · ρξ ) a11
Covariant Derivative of a Tangent Vector T

We define a tangent vector field T – later taken as a friction force – as a
covariant vector field in the spatial coordinate system:
ρξ
T(ξ) = T (ξ)r1 |ζ=0 = T (ξ)ρ1 = T (ξ) . (4.103)
(ρξ · ρξ )
The definition in the form of (4.103) has the advantage that the weak form
– used later – becomes rather simple. The absolute value of the covariant
vector (4.103) is computed as:
|T (ξ)| |T (ξ)|
T(ξ) = = √ . (4.104)
ρξ a11
The full time derivative of this vector field is determined taking the changing
metric into account and then considering its value at ζ = 0:
⎡ ⎤
⎢ ⎥
⎢ ⎥
⎢

⎥
dT ⎢ ∂T ∂T ∂r1
∂r1 ⎥
=⎢
⎢ + ξ̇ r1 + T ξ̇ + ζ̇ ) ⎥
⎥ . (4.105)
dt ζ=0 ⎢ ∂t ∂ξ ∂ξ ∂ζ ⎥
⎢ ! ⎥
⎣ dT ⎦
dt ζ=0
The derivative is expressed then via the contravariant basis vectors as ρ1 , ρ2

dT D1 T 1 D2 T 2
= ρ + ρ .
dt ζ=0 dt dt
Here a full time derivative in the covariant form is introduced. Its value on
the tangent line is computed using the scalar product with ρξ as:

D1 T dT dT ρξξ · ρξ h11
:= · ρξ = − ξ˙ + ζ̇. (4.106)
dt dt ζ=0 dt (ρξ · ρξ ) a11
The second term on the right hand side contains the Christoffel symbol and
the last term contains the curvature for a cylindrical surface.
Proof. Proof of the formula 4.106.

The full time derivative of the tangent vector T is considered in the con-
travariant basis r1 , r2
dT d 1 dT 1 dr1
= Tr = r +T . (4.107)
dt dt dt dt
For the computation we assume, that the component T is a scalar function
of t, ξ, i.e. T = T (t, ξ), and the basis vector r1 depends implicitly on time via
the convective coordinates ξ and ζ, see the definition in eqn. (4.98). Then we
obtain:

dT ∂T ∂T ˙ 1 ∂r1 ∂r1
= + ξ r + T ξ̇ + ζ̇ . (4.108)
dt ∂t ∂ξ ∂ξ ∂ζ
The partial derivatives of the contravariant basis vectors are expressed via
the Christoffel symbols Γijk , see e.g. Marsden and Hughes [122],
∂r1 1 1 1 2
= Γ11 r + Γ12 r ,
∂ξ
(4.109)
∂r1 1 1 1 2
= Γ21 r + Γ22 r .
∂ζ
Using the orthogonality of the covariant and contravariant basis vectors ri
and ri , we can write the following:

dT dT dT
= · r1 r1 + · r2 r2 . (4.110)
dt dt dt
For the formulation of the evolution equations, it is necessary to obtain only a

covariant component of the full time derivative on the tangent line, therefore,
from the expansion (4.110) we need only the first term computed at ζ = 0.
We introduce this derivative as follows

D1 T dT dT
≡ · r1 = · ρξ =
dt dt ζ=0 dt
∂T ∂T 1 ˙ 1
= + ξ̇ + T (Γ11 ξ + Γ21 ζ̇)ζ=0 . (4.111)
∂t ∂ξ
One can recognize in eqn. (4.111) the full time derivative via the covariant
derivatives. We compute the value of the necessary Christoffel symbols in
brackets, according to their definition in eqn. (4.109) and to the definition of
r1 in (4.98).
1
∂r
1
(Γ11 )ζ=0 = · r1 ,
∂ξ ζ=0

1 (4.112)
∂r
1
(Γ21 )ζ=0 = · r1 .
∂ζ ζ=0
In order to avoid the derivative of the contravariant vector in (4.112), we take

a derivative of the following identity
∂
r1 · r1 = 1 =⇒ (r1 · r1 ) = 0
∂(...)
∂r1 ∂r1
=⇒ · r1 + r1 · =0
∂(...) ∂(...)
∂r1 ∂r1
=⇒ · r1 = −r1 · . (4.113)
∂(...) ∂(...)
Now, we can compute the Christoffel symbols directly

∂r1 1 ∂ h11
1
(Γ11 )ζ=0 = − ·r =− (1 − ζ)ρξ · ρ1 = (4.114)
∂ξ ζ=0 ∂ξ a 11 ζ=0
the contravariant basis vector ρ1 is expressed via the covariant one in

eqn. (4.90)

∂ h11 h11 ρξ ρξξ · ρξ
=− (1 − ζ)ρξ + (1 − ζ)ρξξ · =− = −Γ.
∂ξ a11 a11 ζ=0 (ρξ · ρξ ) ρξ · ρξ
From eqns. (4.92) and (4.93) the identity with −Γ is found.

The last necessary Christoffel symbol is computed analogously

∂r1 1 ∂ h11
1
(Γ21 )ζ=0 = − ·r =− (1 − ζ)ρξ · ρ1 = (4.115)
∂ζ ζ=0 ∂ζ a 11 ζ=0
h11 ρξ h11
= ρξ · = .
a11 (ρξ · ρξ ) a11
Finally, the tangent component of the full derivative in eqn. (4.111) gets the
following form:
D1 T ∂T ∂T ˙ 1 ˙ 1
:= + ξ + T (Γ11 ξ + Γ21 ζ̇)ζ=0 =
dt ∂t ∂ξ

dT ρξξ · ρξ h11
= −T ξ˙ − ζ̇ . (4.116)
dt ρξ · ρξ a11
4.2.2 2D Contact Kinematics

An important part for the formulation as well as for the linearization of the
weak form is a time derivative of the vector of a slave point S in eqn. (4.86)

drs ∂ρ ˙ ∂ν ∂ν
= + ξρξ + ζ + ξ̇ + ζ̇ν. (4.117)
dt ∂t ∂t ∂ξ
drs ∂ρ
With vs = as the absolute velocity of the slave point S resp. v =
dt ∂t
as the velocity of its projection on the master surface. The dot product with
the normal vector ν leads to the rate of the penetration
ζ̇ = (vs − v) · ν. (4.118)
Considering a value of the convective tangent velocity ξ˙ on the tangent line,
i.e. at ζ = 0, we need the dot product of (4.117) with ρξ :
(vs − v) · ρξ
ξ̇ = . (4.119)
(ρξ · ρξ )
In the case of a length parametrization with s = ξ, eqn. (4.119) leads to the

following convective velocity ṡ:
ṡ = (vs − v) · τ . (4.120)
From the kinematic equation (4.117) we can obtain an equation for the vari-
ations by changing the time derivative operator into the variation operator
δ. This equation is also considered on the tangent line, i.e. at ζ = 0:
δrs − δρ = δξρξ + δζν. (4.121)
Equation (4.121) gives a variation of the displacement field for the expression
of the virtual work of contact tractions on the contact surface.
4.2.3 Linearization of Variations δζ and δξ

Here, we show the derivation of the normal contact matrices using the geom-
etry of plane curves. For all other results, we give a sketch of the linearization
procedure with a comparative discussion of results available in the literature,
in order to avoid the repetition of complicated mathematics.
Since in the contact integral the linearization of the contact tractions is di-
rectly given by the evolution equations, it is only necessary to find derivatives
of the convective variations δζ and δξ to fulfill all steps in the preparation
for further forms.
Linearization of δζ
The result will be obtained assuming a natural parametrization of the corre-
sponding boundary curve.
d d
δζ = [(δrs − δρ) · ν] =
dt dt
∂(δrs − δρ) ∂ν ∂ν
= · ν ṡ + (δrs − δρ) · + (δrs − δρ) · ṡ. (4.122)
∂s ∂t ∂s
The first term can be rewritten, taking into account (4.120) for a convective
velocity in the case of a natural parametrization, as follows:
∂(rs − ρ)
δ ·ν ṡ = −(δτ · ν) ((vs − v) · τ ) = −δτ · (ν ⊗ τ )(vs − v). (4.123)
∂s
! !
−τ ṡ
In order to rewrite the second term, we have to take first a partial time
derivative of the orthogonality condition:
∂(τ · ν) ∂2ρ ∂ν
τ ·ν =0 =⇒ = ·ν+ · τ = 0, (4.124)
∂t ∂s∂t ∂t
leading to the expression
∂ν ∂v
·τ =− · ν. (4.125)
∂t ∂s
From the other side, using the unity condition of the vector ν, we can ex-
press the time derivative in terms of the tangent vector τ by analogy to
(4.95), (4.96), as
∂ν ∂ν
= aτ = (τ · )τ (4.126)
∂t ∂t
and substituting (4.125) we obtain

∂ν ∂v
=− · ν τ. (4.127)
∂t ∂s
Equation (4.127) allows to transform the second term in (4.122) as follows

∂ν ∂v
(δrs − δρ) · = −(δrs − δρ) · ·ν τ = (4.128)
∂t ∂s
introducing a tensor product τ ⊗ ν in order to transform a dot product

∂v ∂τ
= −(δrs − δρ) · (τ ⊗ ν) = −(δrs − δρ) · (τ ⊗ ν) . (4.129)
∂s ∂t
The last term in (4.129) is obtained reversing the order of differentiation as
∂v ∂ ∂ρ ∂ ∂ρ ∂τ
= = = . (4.130)
∂s ∂s ∂t ∂t ∂s ∂t
The third term in (4.122) is reorganized into a tensor form with a second
Serret-Frenet formula and with equation (4.120) for the convective velocity
ṡ:
∂ν
(δrs − δρ) · ṡ = −(δrs − δρ) · κτ ⊗ τ (vs − v). (4.131)
∂s
Therefore, combining (4.123), (4.128) and (4.131), we obtain a final formula

for the linearization of δζ:

d ∂τ
δζ = − δτ · ν ⊗ τ (vs − v) + (δrs − δρ) · τ ⊗ ν (4.132)
dt ∂t
− (δrs − δρ) · κτ ⊗ τ (vs − v).
Linearization of the Convective Variation δξ

The linearization of the convective variations δξ i in a 3D formulation is the
most complicated part of the process. First, see Parisch [138], Laursen and
Simo [109], Wriggers [189] the convective variations δξ i were introduced via
an iterative Newton scheme, see (4.85). A kinematical definition of δξ i can be
found in the books of Wriggers [190] and Laursen [106]. The full linearization
of these terms combined with the contact integral defined only on the surface
led to an artificial non-symmetry of the tangent matrix for the sticking case,
which was mentioned in Laursen and Simo [109]. Wriggers [190] could avoid
this by looking at it as a mesh tying procedure. In [88] the variations of
δξ i were defined kinematically and expressed on the tangent plane of the
contact surface. In addition, the linearization process was performed in the
covariant form on the tangent plane. For the sticking case this leads directly
to a symmetric matrix and allows to avoid the artificial non-symmetry.
Here only the main points of the linearization process are depicted, for the
full derivation is given for 3D in Sect. 4.1.8
1. The convective variations are defined on the tangent plane of the spatial
coordinate system via consideration of the slave point velocity as ξ˙j =
aij (vs − v) · ρi .
2. During the linearization of δξ i the derivative of the metric tensor is ob-
tained as derivative of the spatial metric tensor considering its value on
the tangent plane, see eqns. (4.81a),(4.81b),(4.81c).
The reduction into the specific plane 2D geometry leads to the following
equation:
d
(δξ) =
dt
(δrs − δρ) · ρξ ⊗ ρξ vj + δρξ · ρξ ⊗ ρξ (vs − v)
=− (4.133a)
(ρξ · ρξ )2
(ρξξ · ν)
+ (δrs − δρ) · ρξ ⊗ ν + ν ⊗ ρξ (vs − v)+ (4.133b)
(ρξ · ρξ )2
h11 ρξξ · ρξ
+ ζ̇δξ − ˙
ξδξ. (4.133c)
a11 (ρξ · ρξ )
4.3 Contact Kinematics for STAS – Two Contact Strategies 95
The non-symmetric part in (4.133c) is intentionally kept in untransformed

form, because it will give a zero in sum with similar terms in the evolution
equation (6.171) during the forthcoming linearization of the contact integral.
4.3 Kinematics of Segment (Deformable)-To-Analytical

(Rigid) Surfaces Contact (STAS) – Two Strategies
If one body is rigid then a stand-alone technique can be applied to de-
scribe this type of contact. In this case the standard CPP procedure is not
appearing – one has to develop a separate algorithm to define the penetration
point. In cases with a simple geometry it is even possible to obtain the closed
form solution. Two strategies can be applied if one of the contacting bodies
is rigid and can be analytically parametrized (directly by known analytical
functions, or by suitable NURB splines in a CAD system). These strategies
depend on the selection of the master or slave part in a surface coordinate
system.
4.3.1 Rigid Surface Is a “Slave” Surface

If the rigid surface is a “slave” surface then we can write the following coor-
dinate system :
rs (α1 , α2 ) = ρ(ξ 1 , ξ 2 ) + ξ 3 n(ξ 1 , ξ 2 ). (4.134)
The rigid surface rs is then parametrized by internal Gaussian coordinates
α1 , α2 . ρ(ξ 1 , ξ 2 ) is the parametrization of our “master” deformable segment
from the finite element mesh and n is the normal to the master segment.
In other words, a point rs of this surface is observed in the local coordinate
system of the contact master element. The standard Closest Point Projec-
tion procedure, which was necessary for the previous description with surface
segments, now turns into the determination of the surface point defined by
equation (4.134). Using a “Segment-To-Segment” type strategy for the com-
putation of the contact integral, the integration points ξI1 , ξJ2 are defined on
the “master” segment and then the corresponding internal coordinates α1 , α2
of the rigid surface as well as the penetration ξ 3 are computed e. g. by the
Newton method. For this algorithm we define a function F(α1 , α2 , ξ 3 ) with
the components given in eqn. (4.134)
⎡ ⎤
xs1 − x1 − n1 ξ 3
F(α1 , α2 , ξ 3 ) = ⎣ xs2 − x2 − n2 ξ 3 ⎦ with xi = xi (ξ 1 , ξ 2 ). (4.135)
xs3 − x3 − n3 ξ 3
Its derivative with respect to the coordinates (α1 , α2 , ξ 3 ) is:

⎡ ⎤
xs1,1 xs1,2 −n1
F = ⎣ xs2,1 xs2,2 −n2 ⎦ . (4.136)
xs3,1 xs3,2 −n3
Then, the Newton iteration procedure gives as follows for iteration step n:
⎡ ⎤
Δα1n
Δαn = ⎣ Δα2n ⎦ = −(F )−1n Fn , (4.137)
Δξn3
αn+1 = αn + Δαn .
Surface of Revolution
For a surface of revolution, given by an analytical function, or described by
NURBS, see [38], a matrix form solution for the coordinate increments in
eqn. (4.137) can be directly developed. In the simplest case f (r) can be a
plane curve uniquely projected onto the r axis, see Fig. 4.5. The revolution
of the curve about the axis OZ gives a surface of revolution. In a Cartesian
coordinate system it can be written as
⎡ ⎤ ⎡ ⎤
xs r cos φ
rs (r, φ) = ⎣ ys ⎦ = ⎣ r sin φ ⎦ . (4.138)
zs f (r)
slave
f(r)
S
ξ3
Y
n
r
φ C
X
master
Fig. 4.5 The surface of revolution

Then the iteration vector Δαn in eqn. (4.137) gets the following form:
⎡ ⎤
Δrn
Δαn = ⎣ Δφn ⎦ (4.139a)
Δξn3
where
1
Δrn = · ((x3 − f (r))(n1 cos φ + n2 sin φ) (4.139b)
D
+ n3 (r − x1 cos φ − x2 sin φ))
1
Δφn = · (f (r) − x3 − rf (r))(n1 sin φ − n2 cos φ)+ (4.139c)
Dr
+ f (r)(n1 x2 − n2 x1 ) + n3 (x1 sin φ − x2 cos φ)
1
Δξn3 = · f (r)(x1 cos φ + x2 sin φ − r) + f (r) − x3 (4.139d)
Dr
+ ξ 3 [f (r)(n1 cos φ + n2 sin φ) − n3 ]
and with the determinant
D = − n3 + f (r)(n1 cos φ + n2 sin φ). (4.139e)
4.3.2 Rigid Surface Is a “Master” Surface

Preserving variables ξ 1 , ξ 2 only for the finite element approximations for the
case the rigid surface is a “master” surface we can write the following coor-
dinate system:
rs (ξ 1 , ξ 2 ) = ρ(α1 , α2 ) + p n(α1 , α2 ). (4.140)
Now, an integration point rs (ξ 1 , ξ 2 ) from the deformable “slave” finite ele-
ment segment is found in the direction of the normal n(α1 , α2 ) to the rigid
“master” surface ρ(α1 , α2 ). This distance denoted as p plays role of the pene-
tration. It is important to note that the distance p between the master surface
and slave point is not coinciding with the penetration ξ 3 measured from the
finite element segment, because, the normals from the master and the slave,
in general, are not parallel. The situation is illustrated for the contact be-
tween the rigid sphere and a “slave” segment in Fig. 4.7. However, when
two bodies are close to contact then the normals are almost parallel, in fact,
they are enforced to be parallel by the contact algorithm. This leads to the
possibility to use the current approach for the contact mechanics. Another
important remark is that the normal for any further computational algorithm
(e.g. necessary for tangent matrices) must be computed from the deformable
“slave” segment, otherwise disconvergence will be obtained. The correspond-
ing penetration is taken as ξ 3 = p.
The Newton method is exploited in this approach in order to solve

eqn. (4.140) defining then a point with the coordinates α1 , α2 on the rigid
surface and the distance p between this surface and a selected integration
point on the “slave” segment rs (ξ 1 , ξ 2 ). In this case an integration point
rs (ξ 1 , ξ 2 ) is projected onto the rigid “master” surface, or in other words,
CPP procedure onto rigid surface is involved. In due course, the result of
Sect. 3.1, concerning however the rigid surfaces defined now analytically, is
fully applicable to analyze the existence and uniqueness of CPP procedure
and to build the proximity domains.
Remark 4.3. This approach allows to work with arbitrary finite element ap-
proximations including NURBs for deformable bodies as well as for rigid
surfaces.
4.3.3 Surfaces Allowing a Closed Form Solution for

the Penetration
Here some simple analytical surfaces in both strategies are chosen for which
it is not necessary to solve nonlinear equations (4.134) or (4.140) in order
to compute the value of the penetration. We consider here plane, cylinder,
sphere, torus and cone. For them it is possible to derive a closed form solution
for the penetration.
Contact with a Rigid Plane

The simplest example with a closed form solution for the rigid surface as a
“slave” strategy is a contact with a rigid plane. Consider a rigid “slave” plane
given in the in analytical form as:
(r − r0 ) · N = 0, (4.141)
where r0 is any point on the plane, N is a normal vector for the plane and r
is a vector with Cartesian coordinates {x, y, z}. Assuming now that from one
side the slave vector rs belongs to the plane eqn. (4.141) and from another
side it is observed from the master segment, i.e. it is satisfying eqn. (4.134)
we can write the following system:

(rs − r0 ) · N = 0
(4.142)
rs = ρ(ξ 1 , ξ 2 ) + ξ 3 n(ξ 1 , ξ 2 ).
Just inserting rs from the second equation into the first one, we can obtain
the value for the distance ξ 3 , and therefore for the penetration as:
1
(ρ(ξi.p. 2
, ξi.p. ) − r0 ) · N
ξ3 = − 1 2 )·N (4.143)
n(ξi.p. , ξi.p.
C
master
ρ N
ξ3
Z
rs
S
Y r0
slave
X
Fig. 4.6 Contact with a rigid plane given analytically
It is easy to see, that the distance is always resolved if

1
n(ξi.p. 2
, ξi.p. ) · N = 0, (4.144)
which means the slave plane is not orthogonal to the tangent plane of the
master segment.
For the further computations, as a vector ρ(ξi.p.1 2
, ξi.p. ) is given at integra-
1 2 1 2
tion points ξi.p. , ξi.p. of the master segment, the normal n(ξi.p. , ξi.p. ) is also
computed at the same points.
Remark 4.4. Since, the parametrization of the master segment is considered

arbitrary (the segment in Fig. 4.6 is shown intentionally curved), the corre-
sponding contact approach works with any kind of approximation for finite
elements.
Contact with a Rigid Sphere

The computation of the penetration for contact with a rigid sphere is the most
trivial case for the rigid surface is a “master” strategy, because the absolute
value of a vector, and therefore, a distance in the Cartesian coordinate system
is defined in a form of a sphere equation. However, we formally start with
the definition of a coordinate system assigned to the spherical surface:
rs = RC + ρsph + p nsph , (4.145)

where RC is a center of the sphere, ρsph is a vector of the sphere with radius
R, satisfying ρsph = R. Since ρsph is parallel to the unit normal nsph , see
Fig. 4.7, we can rewrite eqn. (4.145) as
rs − RC = R nsph + p nsph , (4.146)
Taking then the absolute value we obtain
rs − RC = |R + p|, (4.147)
and the distance p as

"
rs − RC − R – for an outward normal
p= (4.148)
R − rs − RC – for an inward normal .
The first part in equation (4.148) is describing the simple geometrical fact
AS = CS − CA in Fig. 4.7 as a positive distance from the sphere in the
outward direction of the sphere. A continuum body is the interior of the
sphere in this case. The second equation describes a positive distance in the
inward direction of the sphere. In this case the continuum is the exterior of
the sphere.
n ξ3
S
ρ sph A
Z nsph
RC p
C
O slave
Y
X
master
Fig. 4.7 Contact with a rigid sphere given by an analytical equation

Contact with a Rigid Cylinder

Another example of a closed form solution within the rigid surface is a “mas-
ter” strategy is a distance between a cylinder and a point, see Fig. 4.8. The
cylinder with a radius R can be given in the following form:
ρ(ϕ, z) = RC + z ez + R eϕ (ϕ) . (4.149)
where RC is an arbitrary point on the central axis of the cylinder CCz , ez is

a unit vector of the central axis, eϕ (ϕ) is a unit vector in the radial direction
of the polar coordinate system in the orthogonal plane, see Fig. 4.8. The unit
coordinate vectors are orthogonal (eϕ (ϕ) · ez ) = 0. Noting that the vector
eϕ is normal to the cylinder, the slave point S is observed in the coordinate
system assigned to the cylindrical surface as:
rs = ρ(ϕ, z) + p eϕ (ϕ), (4.150)
Substituting ρ in eqn. (4.150) from eqn. (4.149) we obtain:
RC − rs + z ez + (R + p) eϕ (ϕ) = 0. (4.151)
3
Cz ξ
n
eϕ ϕ
z A p
ez S
C
Z
RC rs
O Y
Fig. 4.8 Contact with a rigid cylinder given by an analytical equation

First, a coordinate z is defined after taking the scalar product with ez :
z = −(RC − rs ) · ez ; (4.152)
afterwards the absolute value in eqn. (4.151) is taken as
RC − rs + z ez = |R + p|, (4.153)
then the distance p is defined as

"
RC − rs − (RC − rs ) · ez ez − R – for an outward normal
p=
R − RC − rs − (RC − rs ) · ez ez – for an inward normal .
(4.154)
The distance p is defined to be positive in both outward and inward direction.
Contact with a Rigid Torus

Consider a torus as a result of rotating a circle given in the XOZ-plane along
the OZ-axis by increasing the angular coordinate ψ in the XOY -plane, see
Fig. 4.9. The torus can be given in the following form:
ρ(ϕ, ψ) = R eR (ψ) + r eϕ (ϕ), (4.155)
where eR and eϕ are radial unit vectors for the polar coordinate system
given by two orthogonal planes. First, we define a scalar product eR · ρ.
−→
This product defines a projection of the vector OA = ρ = (x, y, z) on the
plane XOY , therefore,
ρ · eR = x2 + y 2 . (4.156)
This product allows to get rid of eϕ in eqn. (4.155), by taking an absolute
value:
ρ − R eR 2 = r eϕ 2 . (4.157)
Transformation of the scalar products leads to a torus equation in the fol-
lowing form
ρ2 + R eR 2 − 2R (ρ · eR ) = r eϕ ;
or written by coordinates:

x2 + y 2 + z 2 + R2 − 2R x2 + y 2 = r2 . (4.158)
Now we can proceed with the observation of the slave point rs = (xs , ys , zs )
at the distance p in the toroidal coordinate system, remembering that eϕ is
a normal vector to the torus surface
rs = R eR (ψ) + (r + p) eϕ (ϕ). (4.159)

Similar transformations are leading to the following distance:

⎧
⎪
⎨ x2s + ys2 + zs2 + R2 − 2R x2s + ys2 − r – for an outward normal
p=
⎩ r − x2 + y 2 + z 2 + R2 − 2Rx2 + y 2 – for an inward normal .
⎪
s s s s s
(4.160)
The distance p is defined to be positive in both outward and inward directions.
Contact with a Rigid Cone

A cone can be considered also within the rigid surface is a “master” strategy.
The cone, see Fig. 4.10, has the OZ-rotation axis and is defined as:
ρ(r, ϕ) = r er (ϕ) + r tan α ez , (4.161)
where er and ez are unit vectors of the cylindrical coordinate system. The
outward normal vector n is simply defined from the geometry of a triangle,
see Fig. 4.10
n = sin α er − cos α ez . (4.162)
The slave point at the distance p from the cone surface is written first in the
cone coordinate system and then in the cylindrical coordinate system as:
rs = ρ(r, ϕ) + pncone (4.163)

= r tan α ez + r er + p(sin α er − cos α ez )
= (r tan α − p cos α) ez + (r + p sin α) er .
Definition of the distance p is tremendously simplified if we notice that the

vectors ρ and n are orthogonal
ρ · n = 0, (4.164)
Z r
3
R eR C ξ
O Y A n
eϕ ϕ
X rs S
ψ p
Fig. 4.9 Contact with a rigid torus given by an analytical equation

then from eqn. (4.163) we have
p = (rs · n) = sin α (er · rs ) − cos α (ez · rs ). (4.165)
It is more convenient to consider rs in a cylindrical coordinate system in

order to compute the scalar product in eqn. (4.165) as

rs = x2s + ys2 er (ϕ) + zs ez , (4.166)
then the penetration in eqn. (4.165) becomes:

p = x2s + ys2 sin α − zs cos α. (4.167)
C
z
r
master A
Z ncone
ez
p n
Y
O rs S
α
X er
ϕ slave
B
Fig. 4.10 Contact with a rigid “master” cone surface
Remark 4.5. The rigid surface is a “master” strategy does not give the answer
to the solvability of the contact algorithm. This criterion is appearing from
the geometry analysis of the contact segment. Thus, considering the contact
with a rigid cone described by the rigid surface is a “slave” strategy as a
surface of revolution in Sect. 4.3.1 it is necessary for the solvability that the
determinant in eqn. (4.139e) is not zero: D = 0. Taking f (r) = tan α, the
determinant is transformed as
4.4 Kinematics of Point-To-Curve Interaction 105
sin α(n1 cos φ + n2 sin φ) − n3 cos φ

D= . (4.168)
cos α
Using then the definition of the normal for the cone in eqn. (4.162), the
determinant can be written as
nξ · ncone
D= = 0. (4.169)
cos α
Now, the criterion of the solvability is clear – a segment normal nξ should
not be orthogonal to the cone normal ncone .
Contact with a Parabolic Cylinder

This case within the rigid surface as a “slave” surface approach, though
does not allow a closed form solution for the penetration, however, allows
to reduce the corresponding Newton iterative procedure in eqn. (4.137) to a
single equation. Consider a parabolic cylinder in the canonical form:
xs = α
(4.170)
ys = cα2
The Newton procedure in eqn. (4.137) in this case is reduced to the definition
of α from the following iterative expression:
c(α(n) )2 n1 + n1 x2 − n2 x1
α(n+1) = , (4.171)
2cα(n) n1 − n2
where an initial guess can be computed e.g. as
α(0) = x1 . (4.172)
The value of the penetration does not require an iterative procedure and can
be computed after the definition of α as
cα2 − 2cαx1 + x2 − n2 x1
ξ3 = . (4.173)
2cα(n) n1 − n2
4.4 Kinematics of Point-To-Curve Interaction

The introduction of the CPP procedure for Point-To-Curve (PTC) interaction
in Sect. 3.3 and a corresponding curvilinear coordinate system allows also to
consider the kinematics of Point-To-Curve contact, and therefore, to compute
all necessary contact parameters such as forces, measures of displacements
etc.
Consider a motion of a given point S (“a slave” point) in the moving

coordinate system τ , ν, β. The velocity vector vs is computed as a full time
derivative:
drs ∂ρ de
= + ṡτ + ṙe + r , (4.174)
dt ∂t dt
de
where the full time derivative = ω describes a rigid body rotation of the
dt
coordinate system τ , e, g both, due to the changing curvature along a curve,
and due to the rigid motion of a curve. For the further numerical model a
value of r must lay within an allowable small distance surrounding a curve,
see Fig. 4.11. This assumption necessarily will lead also to the corresponding
small load or time increments to solve the contact problem. Altogether these
de
assumptions allow to neglect the contribution of the last term (r ) similar
dt
to the covariant description for the 3D-dimensional case for kinematics of
Surface-To-Surface interaction, see Sect. 4.1, where a relative velocity vector
has been considered on the tangent plane. In this case a torsional contribu-
tion is neglected. This contribution including arbitrary, will be considered for
the Curve-To-Curve interaction in Section 4.5. With the dot product of the
vectors τ and e, the following components of the velocity vector are obtained:
ṡ = (vs − v) · τ , (4.175)
ṙ = (vs − v) · e, (4.176)
drs ∂ρ
where vs = is the absolute velocity of the slave points, and v = is
dt ∂t
the translational velocity of the projected point O.
Remark 4.6. According to the assumptions mentioned above, all variational

parameters for the further weak form are created in a similar fashion as the
kinematical relations (4.174), (4.175), (4.176), e.g.
δrs − δρ = δsτ + δre + rδe. (4.177)
If the torsional interaction is neglected for the following consideration then

we take r = 0 for the forthcoming weak formulation.
The convective velocities in eqns. (4.175), (4.176) define essential measures

for the contact interaction as: r – for the normal interaction, Δs – for the
tangential interaction. In addition for the normal interaction a penetration
area can be defined in the plane νOβ (plane II in Fig. 3.10) as a curve in
the polar coordinate system
r = r(ϕ), φ ∈ [ϕ0 , ϕn ]. (4.178)

4.5 Kinematics of Curve-To-Curve Interaction 107
The simplest case is a circular area with a radius Rε , see Fig. 4.11. Thus, a
measure can be taken as
ζ = r − Rε = rs − ρ − Rε . (4.179)
Remark 4.7. A case with artificial Rε is rather necessary for the point-to-
curve algorithm applicable for curve-to-curve contact. In the case of a globally
C 0 surface, e.g. in the case of an edge, the penetration is computed exactly
at the projection point: ζ = rs − ρ.
ν
S re
ϕ τ
O
ζ
A Rε
Δs
β
Fig. 4.11 Point-To-Curve contact element. Definition of measures of contact inter-

actions: ζ – for normal interaction defined according to a circular area with radius
Rε ; Δs – for tangential interaction.
4.5 Kinematics of Curve-To-Curve Interaction

In this Section a geometrically exact description for Curve-To-Curve contact
situation in a covariant form is developed. The contact kinematics which is
further necessary for the variational formulation and the constitutive relations
for contact tractions is described in a specially defined curvilinear coordinate
system based on the closest point projection (CPP) procedure developed in
Sect. 3.4. The introduced coordinate system is independent of the choice of
a contacting curve. This allows to describe all possible relative motions of
both curves including not only normal and tangential interactions, but also
rotational interaction between curves representing e.g. circular cross sections
of beams.
4.5.1 Development of Beam-To-Beam and

Edge-To-Edge Contact
Contact between beams as a branch of computational contact considerations
appears to be a “stand alone specific problem” due to its special geometrical
tasks and following difficulties. Thus only fairly recently, a first computational
algorithm for beam-to-beam contact was proposed by Wriggers and Zavarise
[198] (1997) for non-frictional contact between straight beams with circular
cross sections. The measure of contact has been introduced as the shortest
distance between the central lines. Then the specific task is that two projec-
tion points on both contact elements have to be found. Further, in order to
extend this into the frictional case in Zavarise and Wriggers [203] (2000) the
return-mapping algorithm is exploited. The solution algorithm is based on an
iterative solution and involves the linearization for the particular linear con-
tact elements. A generalization of this algorithm for the case of rectangular
cross-sections is given in Litewka and Wriggers [114] (2002) and [115] (2002)
for both frictional and non-frictional cases respectively. While the mentioned
publications were based on a penalty method to enforce the contact con-
straints, in Litewka and Wriggers [113] (2003) the Lagrange multiplier method
has been involved. One of the obvious – and up-to now not resolved – difficul-
ties is the linearization procedure for the case of arbitrarily curved geometry
of beams. As a first step thus, Litewka in [111] (2007) and in [112] (2007)
considered a smooth Hermite type element for contact formulations, how-
ever, the linearization was provided as the direct derivation of the obtained
functionals via mathematical software (Mathematica in the current case).
Another approach to contact between beams as an application for cable
contact is described in Zhou et.al.[211] (2004) in which, a so-called ”sliding
cable” element leading to an element with a ”broken line” is derived. This
model, however, became a standard for a model of ”seat belt connections”
since many years in commercial codes, such as LS-DYNA, see [51]. A special
problem of mooring cables has been considered in Souza de Cursi [166] (1992)
as a 2D contact problem of Signorini type.
These references reveal the main difficulties and limitations of the current
state of beam-to-beam contact models:
1. Existing finite elements models are yet not serving for arbitrary order of
approximations due to problems with the linearization, though, specific
applications of high order finite elements are known e.g. in [112], [111].
2. kinematics of contact considered in the existing models are rather limited;
by far not all possible variances of the relative motion and corresponding
relative constitutive equations are fully involved.
Thus, the main goal of the current section is to develop a general theory
for curve-to-curve contact from a geometrical point of view considering a
particular case of contact between bodies when the contacting bodies are
interacting along two curves. It can be viewed as either contact between
two sharp edges for full 3D bodies, or between two beams as a mechanical
model in the latter, if center lines of beams or edges of bodies are considered
in contact. Our aim is to describe all contact parameters such as relative
displacements and velocities, contact forces in the specific coordinate system
which is most suited to describe the various geometrical diversities of both two
curves. This concept includes internal geometrical characteristics of curves
such as curvature and torsion as well as their relative position in 3D space.
The basics of such a concept, originated to [95], have been developed for
contact between surfaces in previous Sections.
The important feature for any curve-to-curve contact is a suitable defini-
tion of the coordinate system which will allow to distinguish all kinemati-
cal details of the mutual interaction. Thus all contact measures for relative
motions should be directly obtained in this coordinate system. Obviously a
distance between two curves provides a definition of a measure of normal
contact interaction. This leads to the well known idea of the closest point
projection (CPP) procedure developed in Sect. 3.4. This specific CPP pro-
cedure allows to set up two local curvilinear coordinate systems attached to
the each curve. These coordinate systems are given in terms of Serret-Frenet
formulas for 3D curved lines. This allows to consider any motion of one curve
relative to another. A good observation of the motion gives us then a choice
of proper measures for such an interaction between two curves. Afterward, all
possible parameters such as sliding distances, variations, forces as well as the
necessary operations such as derivations, formulation of the return-mapping
algorithm are considered in a covariant form using the curvilinear metrics of
the specific two coordinate systems attached to the curves. A s a result, the
complete diversity of contact interactions including even a possible moment
interaction are described. Finally, a compact form of the computational algo-
rithm – a-priori independent on the type and order of approximation of the
curves – will be further established.
4.5.2 Measures of Contact Interaction

Differential properties of for the Curve-To-Curve contact interaction are stud-
ied in the coordinate system defined in Sect. 3.4.1, namely
ρ2 (s1 , r, ϕ1 ) = ρ1 (s1 ) + re1 (ϕ1 ), 1 ↔ 2. (4.180)
Since the coordinate system is arbitrarily curvilinear it is necessary to define

the differential operations in a covariant form. As a preparation, we compute
partial derivatives of the unit vectors e1 , g1 introduced in eqns. (3.47a, 3.47b),
see Fig. 4.12.
The derivatives of the vector e1 with respect to s1 and with respect to ϕ1
are
Fig. 4.12 Geometry of the Curve-To-Curve contact pair: definition of basis vectors
τ i , ei , gi on each curve and corresponding measures: si , r, ϕi
∂e1
= cos ϕ1 (−k1 τ 1 + κ1 β1 ) + sin ϕ1 (−κ1 )ν 1 =
∂s1
= −k1 τ 1 cos ϕ1 + κ1 g1 (4.181a)
∂e1
= g1 . (4.181b)
∂ϕ1
The derivatives of the vector g1 with respect to s1 and with respect to ϕ1
are
∂g1
= − sin ϕ1 [−k1 τ 1 + κ1 β1 ] − cos ϕ1 κ1 ν 1 = k1 τ 1 sin ϕ1 − κ1 e1 (4.182a)
∂s1
∂g1
= −e1 . (4.182b)
∂ϕ1
Now, we can define coordinate vectors r1 , r2 , r3 (a subscript defining the first
curve is omitted) of the curvilinear coordinate system as partial derivatives
with respect to the coordinates s1 , r, ϕ1 :
∂ρ2 ∂ρ1 ∂e1

r1 = = +r = τ 1 (1 − k1 r cos ϕ1 ) + κrg1 (4.183a)
∂s1 ∂s1 ∂s1
∂ρ2
r2 = = e1 (4.183b)
∂r
∂ρ2
r3 = = rg1 . (4.183c)
∂ϕ
The corresponding covariant metric tensor is given by components (subscript
(.)1 is omitted)
g11 = (r1 · r1 ) = (1 − kr cos ϕ)2 + κr2

g12 = (r1 · r2 ) = 0
g13 = (r1 · r3 ) = κr2
g22 = 1
g33 = r2
leading to a matrix
⎡ ⎤
(1 − kr cos ϕ)2 + κ 2 r2 0 κr2
[gij ] = ⎣ 0 1 0 ⎦. (4.184)
κr2 0 r2
The system is, in general, non-orthogonal (see g13 = 0) and is changing in

space reflecting the geometry of a given curve. However, at the given point
on a curve with r = 0 the system is orthogonal.
For the case of edge-to-edge contact the distance r between curves is a
small value and the terms containing the torsion κ1 and the curvature k1 are
also usually small at least within a single finite element. It motivates us to
consider all parameters in such a case in a coordinate system exactly attached
to the curve at r = 0, removing all small terms containing combinations of
either a curvature or a torsion together with the r-coordinate, see the analogy
for a tangent plane for the surface-to-surface contact in [87], [88] and [91].
Thus, we consider also, an orthogonal basis as
r1 = τ 1 , r2 = e1 , r3 = rg1 (4.185)
with the corresponding metric tensor defined by a matrix
⎡ ⎤
10 0
[gij ] = ⎣ 0 1 0 ⎦ . (4.186)
0 0 r2
Exactly for this situation we define the contravariant components of the met-
ric tensor as inverse components of the matrix in eqn. (4.186). It leads to the
standard definition of the contravariant basis ri , i = 1, 2, 3 for the curve with
r = 0 as
ri = g ij rj (4.187a)
1
r1 = τ 1 , r2 = e1 , r3 = g1 . (4.187b)
r
4.5.3 Rates and Variations of Measures for Contact

Interaction
Now we can compute a full time derivative for a vector r2 (s1 , r, ϕ1 ) describing
a full velocity vector in our coordinate system
dρ2 ∂ρ1
= + (1 − rk1 cos ϕ1 )ṡ1 τ 1 + ṙe1 + r(κ1 ṡ1 + ϕ̇1 )g1 . (4.188)
dt ∂t
Redefining now the full velocity vector of the second point C2 as
dρ2
v2 = (4.189)
dt
and the translational velocity of the first point C1 as
∂ρt1
v1t = (4.190)
∂t
we can rewrite the relative velocity vector of a point C2 in the coordinate
system of the first curve as
v2 − v1t = (1 − rk1 cos ϕ1 )ṡ1 τ 1 + ṙe1 + r(κ1 ṡ1 + ϕ̇1 )g1 . (4.191)
Keeping in mind the edge-to-edge contact applications it is more convenient

to consider the relative velocity vector in the coordinate system attached
exactly to the contact point. Thus, we obtain the expression of the relative
velocity vector (4.191) via the vectors τ 1 , e1 , g1 as
v2 − v1t = ṡ1 τ 1 + ṙe1 + (κ1 ṡ1 + ϕ̇1 )rg1 . (4.192)
Here, the last term remains reflecting the geometry of a 3D spatial curve,
because rg1 is a covariant basis vector r3 , but not a small value, however a
term rk1 cos ϕ1 is assumed to be small.
Computing a scalar product with the coordinate vectors rI at r = 0 and,
therefore, with τ 1 , e1 , g1 we obtain the rate of convective coordinates as
⎧
⎪
⎨ ṡ1 = (v2 − v1t ) · τ 1
ṙ1 = (v2 − v1t ) · e1 1 ↔ 2. (4.193)
⎪
⎩ ϕ̇1 + κ1 ṡ1 = (v2 − v1t ) · g1
r
This gives us the rate of measures for the contact interaction between curves
• ṡ1 – for tangential interaction;
• ṙ – for normal interaction;
• ϕ̇1 – for angular interaction.
The last coordinate is describing the rotation of the first curve (plane e1 , g1 )
along the vector τ 1 .
For arbitrary large motions we also derive the exact – not simplified –
expression for the rates of measures from eqn. (4.191):
⎧
⎪ (v2 − v1t ) · τ 1
⎪
⎪ ṡ1 =
⎪
⎪ (1 − rk1 cos ϕ1 )
⎪
⎪
⎨ ṙ1 = (v2 − v1t ) · e1
(v2 − v1t ) · g1 1↔2 (4.194)
⎪
⎪ ϕ̇1 = − κ1 ṡ1
⎪
⎪ rt
⎪
⎪ (v2 − v1 ) · g1 (v2 − v1t ) · τ 1
⎪
⎩ = − κ1 .
r (1 − rk1 cos ϕ1 )
Remark 4.8. Using the analogy between the kinematical values and the vari-
ations we can write the variations of the relative displacements in full form
(see eqn. (4.191)) as
δρ2 − δρ1 = (1 − rk1 cos ϕ1 )δs1 τ 1 + δre1 + r(κ1 δs1 + δϕ1 )g1 , (4.195)
or taking into account small values at r = 0 (see eqn. (4.192)) as
δρ2 − δρ1 = δs1 τ 1 + δre1 + r(κ1 δs1 + δϕ1 )g1 (4.196)
as well as the variations of convective coordinates at r = 0 as

⎧
⎨ δs1 = (δρ2 − δρ1 ) · τ 1
δr1 = (δρ2 − δρ1 ) · e1 1 ↔ 2. (4.197)
⎩
r(κ1 δs1 + δϕ1 ) = (δρ2 − δρ1 ) · g1
4.5.4 Linearization in a Covariant Form of

Variations for Contact Measures
For the forthcoming linearization we need also the covariant derivatives of
the variation of the relative displacement vector in eqn (4.195). The covariant
derivation of vector components is the derivation operation taking into ac-
count the change of a basis vector in space. Following the standard derivation
the Christoffel symbols Γijk , see [33], are required
∂ρi
= ρij = Γijk ρk , (4.198)
∂ξ j
but instead of computing the Christoffel symbols, we will directly compute
derivatives of coordinate vectors using their definitions in eqns. (3.47a-3.47b)
and the Serret-Frenet formulas (2.43). For the derivation of the relative vir-
tual displacements we can take the full values given in eqn. (4.195) without
any simplification. The following derivatives are obtained
with respect to s1
∂
(δρ2 − δρ1 ) = (4.199)
∂s1
= k1 (1 − k1 r cos ϕ1 )δs1 ν 1 + (−k1 cos ϕ1 τ 1 + κ1 g1 )δr +
+r(κ1 δs1 + δϕ1 )(k1 sin ϕ1 τ 1 − κ1 e1 )
= [−k1 cos ϕ1 δr + (κ1 δs1 + δϕ1 )rk1 sin ϕ1 ]τ 1 +
+k1 (1 − k1 r cos ϕ1 )δs1 ν 1 − rκ1 (κ1 δs1 + δϕ1 )e1 + κ1 δrg1 ;
with respect to r
∂
(δρ2 − δρ1 ) = −k1 cos ϕ1 δs1 τ 1 + (κ1 δs1 + δϕ1 )g1 (4.200)
∂r
and with respect to ϕ1
∂
(δρ2 − δρ1 ) = k1 r sin ϕ1 δs1 τ 1 + δrg1 − (κ1 δs1 + δϕ1 )re1 . (4.201)
∂ϕ1
The full covariant derivative operator Ls,r,ϕ [...] has the following form (taking
into account the derivation of coordinate vectors)
∂ ∂ ∂
Ls,r,ϕ [...] ≡ ṡ [...] + ṙ [...] + ϕ̇ [...]. (4.202)
∂s ∂r ∂ϕ
Summarizing all partial derivatives in eqns. (4.200), (4.200), (4.201) in one

operator (4.202) and expressing a normal ν via e and g (eqns. (3.47a, 3.47b))
as
ν 1 = e1 cos ϕ1 − g1 sin ϕ1 (4.203)
we obtain the derivative of the variation of the relative displacement vector
, or in fact the second derivative of the relative displacement vector, as
Ls,r,ϕ [(δρ2 − δρ1 )] = (4.204a)

= [rk1 κ1 sin ϕṡ1 δs1 − (4.204b)
− k1 cos ϕ1 (ṡ1 δr + ṙδs1 ) + rk1 sin ϕ1 (ṡ1 δϕ1 + ϕ̇1 δs1 )] τ 1 (4.204c)
+ [k1 cos ϕ1 (1 − rk1 cos ϕ1 )ṡ1 δs1
−r(κ1 ṡ1 + ϕ̇1 )(κ1 δs1 + δϕ1 )] e1 (4.204d)
+ [−k1 sin ϕ1 (1 − rk1 cos ϕ1 )ṡ1 δs1
+(ṙδϕ1 + ϕ̇1 δr) + κ1 (ṡ1 δr + ṙδs1 )] g1 (4.204e)
Remark 4.9. The long expressions in eqns. (4.204c), (4.204d), (4.204e) repre-
sent the exact linearization in a covariant form for the case of an arbitrary
motion (even for a large distance r between curves). For simplification the
derivatives of the curvature k and the torsion κ are not presented, however,
4.6 Kinematics of the Curve-To-Rigid-Surface Interaction 115
due to the derivation with respect to the arc-length s, the expression is exact
for a line of constant curvature and torsion, e.g. for circular and spiral lines.
We have to note that the linearization is given by a symmetric bilinear form

with respect to convective variables s, r, ϕ.
4.6 Kinematics of the Curve-To-Rigid-Surface

Interaction
The kinematics of the Curve-To-Rigid-Surface contact interaction is based
on the combination of the Surface-To-Analytical-Surface and the Curve-To-
Curve contact kinematics. Thus all parameters are formulated dually in the
surface coordinate system (via Gaussian coordinates ξ 1 , ξ 2 ) and in the Serret-
Frenet curve coordinate system (via coordinate ζ). The curve rs (η), see Fig-
ures 1.8 and 4.13, is assumed to be deformable, thus, all parameters for the
curve are considered as developed for the Curve-To-Curve contact in Sect. 4.5.
A rigid surface ρ(ξ 1 , ξ 2 ) is assumed to have arbitrary analytical description,
e.g. via NURBS surfaces. In order to describe contact between deformable
curves and rigid surfaces a Segment-To-Analytical-Surface (STAS) algorithm,
discussed mainly for surfaces in 4.3, is modified as follows:
• A set of contact points (integration points) ηi is set on the curve segment
AB: all kinematical parameters are considered then in the Serret-Frenet
curve coordinate system τ , ν, β;
• The contact point (integration points) is projected onto the rigid surface.
At each point all kinematical parameters are considered in the surface
coordinate system ρ1 , ρ2 , n via convective coordinates ξ 2 , ξ 2 .
The combination of both CTC and STAS strategies leads to the Curve-To-
Rigid (analytical) Surface contact algorithm and to the following definition
of the coordinate system on the master rigid surface:
rs (η) = ρ(ξ 1 , ξ 2 ) + ξ 3 n(ξ 1 , ξ 2 ), (4.205)
where rs (η) is defining an integration point positioned on the curve, or in the

case of the beam contact on the mid-line of the curvilinear beam element,
ρ(ξ 1 , ξ 2 ) is the parametrization of the rigid “master” surface. The integration
point ρs (η) is found in the direction of the normal n(ξ 1 , ξ 2 ) to the rigid
“master” surface. The shortest distance between integration points and the
surface denoted as ξ 3 is considered as penetration. Now the Closest Point
Projection (CPP) procedure as the projection onto the rigid surface plays
the main role. In general, Newtons method is exploited to solve the CPP
procedure defining then a point on the rigid surface and the penetration ξ 3
between this surface and the selected integration point S. A set of analytical
solutions is possible for some surfaces as discussed in Sect. 4.3.
Kinematical relations during the contact can be obtained dually consider-

ing the relative velocity of the contact point during contact:
• The relative velocity on the tangent plane from the surface point of view,
i.e. in the surface coordinate system with ξ 3 = 0 in eqn. (4.29) is:
vs − v = ξ˙3 n + ξ˙j ρi , i, j = 1, 2. (4.206)

• The relative velocity on the curve from the curve point of view, i.e. in
the Serret-Frenet coordinate system at r = 0, attached to the deformed
curve, see eqn. (4.191):
v − vs = vτ τ + ve e + vg g (4.207)
Since, in this contact approach the point is fixed on the curve, we intro-
duce here notations vτ , ve , vg for projections of the relative velocity vector
on the curvilinear coordinate system τ, e, g instead of using the convective
velocities ṡ, ṙ and ϕ̇.
Summing up both equations (4.206) and (4.207), we obtain:
ξ˙j ρi + ξ˙3 n + vτ τ + ve e + vg g = 0. (4.208)
Vector e, defined in eqns. (3.33) and (3.46), is re-represented in the surface

coordinate system, see eqn. (4.4), at the contact rs − ρ = |ξ 3 | = 0 on the
tangent plane
ρ − rs
e= = −n (4.209)
ρ − rs
From eqn. (4.208), we can define the following components of the relative
velocity vector:
• normal relative velocity during contact
ve = ξ˙3 = ṙ = (vs − v) · n = (v − vs ) · e (4.210)
• pulling relative velocity
vτ = −(ρi · τ )ξ˙i (4.211)
• dragging relative velocity
vg = −(ρi · g)ξ˙i (4.212)
Remark 4.10. From eqn. (4.211) one can see, that during the contact a tan-
gent vector of the curve τ is laying in the tangent plane of the surface (because
it is expressed via ρi )
Remark 4.11. The dragging velocity vg is not zero if r = 0 and can be repre-
sented as a limit vg = limr→0 (rϕ̇), i.e. a velocity represented by the rolling of
4.6 Kinematics of the Curve-To-Rigid-Surface Interaction 117
the single wire on the surface if the radius of the wire goes to zero, compare
with eqn. (4.191).
Since, the slave contact point rs is fixed in the curve coordinate system, all
kinematical parameters for the CTRS contact should be defined, first, in the
surface coordinate system ρ1 , ρ2 , n as follows:
• Measure of the normal interaction (in the case of a curve edge)
ξ 3 = (rs − ρ) · n, (4.213)
or in the case of a curvilinear beam
ξ 3 = (rs − ρ) · n − R, (4.214)
where R is a radius of the curvilinear beam (rope, cable etc.) as shown

e.g. in Fig. 4.13.
• Measure of the tangential interaction (similar for both STS and STAS
contact) in the rate form as :
ξ̇ i = (vs − v) · ρj aij . (4.215)

Further computation of tangential measures e.g. in the incremental form
Δξ i follows the STS algorithm:
Δξ i = ξ (n+1), i − ξ (n), i , (4.216)
where ξ (n), i are projection point coordinates onto the master rigid surface
at the load (time) step (n).
These measures will be used to compute the contact tractions (both normal
and tangential ones) in the form of Surface-To-Surface contact. However, the
weak form should be formulated in the coordinate system of the deformed
curve, thus, in the form of Curve-To-Curve contact. In order to fulfill this,
the contact tractions should be projected into the Serret-Frenet coordinate
system, associated with a curve rs (η), see further Sect. 6.9. Combining the
result for the relative velocity vector in eqn. (4.207) with equations for normal,
pulling and dragging velocities in eqns. (4.210, 4.211, 4.212) we obtain the
following variations:
• variation of normal displacement
δue = δξ 3 = δr = (δρ − δrs ) · e (4.217)
• variation of pulling displacement
δuτ = −(ρi · τ )δξ i (4.218)

Fig. 4.13 Kinematics of the Curve-To-Rigid Surface (CTRS) Contact – definition

of both the Gaussian surface and the Serret-Frenet curve coordinate systems.
• variation of dragging displacement
δug = −(ρi · g)δξ i (4.219)
The relative displacement vector is then written in the curve coordinate

system as:
δρ − δrs = δuτ τ + δue e + δug g. (4.220)
5
Weak Formulation of Contact
Conditions
All expressions for contact kinematics developed for various contact pairs
in Chapter 4 allow us, using the analogy between kinematical and varia-
tional values, to formulate the weak forms for all considered contact sit-
uations: Surface-To-Surface, Point-To-Curve, Curve-To-Curve etc. in corre-
sponding curvilinear coordinate systems. In addition various formulations of
the Nitsche approach are discussed.
5.1 Weak Formulation of the Surface-To-Surface

Contact
In order to formulate the variations of the contact displacements the kine-
matic relations developed for the Surface-To-Surface case in Section 4.1 will
be used directly. We consider contact tractions T1 , T2 acting on the infinitesi-
mally small surfaces ds1 and ds2 in the current deformed configuration. δui is
then the variation of the displacement field on the corresponding surface dsi .
The virtual work δWc of the contact tractions is obtained by the following
surface integral
# #
δWc = T1 · δu1 ds1 + T2 · δu2 ds2 (5.1)
s1 s2
which must be added to the global virtual work of the internal and external
forces. Due to the equilibrium equation at the contact boundary, T1 ds1 =
−T2 ds2 , equation (5.1) can be written as
#
δWc = T1 · (δu1 − δu2 )ds1 (5.2)
s1
Up to now, one surface has to be specified as the ”master” and the other
one as the ”slave” surface, see Fig. 5.1 The contact integral is computed over
the ”slave” surface. It can be computed using quadrature formula [106] or, in
120 5 Weak Formulation of Contact Conditions
Fig. 5.1 Contact tractions acting on the infinitesimally small master surface ds
and ds2 .
simple cases, nodal quadrature [195]. With s1 as the ”slave” surface ss , the
redefined previous notation leads to δu1 = δrs as the variation of the ”slave”
point and δu2 = δρ as a projection of the the variation of ”slave” point onto
the ”master” surface. Thus, we obtain
#
δWc = Ts · (δr1 − δρ)dss (5.3)
ss
This selection allows us later on to exclude expensive linearization of the

dss . This happens because the slave point is assumed to be fixed during
linearization due to the CPP procedure involved as the first procedure to
define a point of contact on the master surface.
The traction vector from the slave is expressed, however, in the local co-
ordinate system set up on the master surface, see Fig. 5.1 . It can be split
into a normal and into a tangential part
Tslave ≡ T = N n + T i ρi . (5.4)
Here the traction vector is defined as a contravariant vector. An equation

for the variation is derived following the kinematic equation (4.29) using an
analogy between the derivative operation and the variation operation:
δrs − δρ = (ρj − ξ 3 hij ρi )δξ j + nδξ 3 + ξ 3 δn (5.5)

5.1 Weak Formulation of the Surface-To-Surface Contact 121
It should be mentioned, that the variation itself is time independent. Then

the contact integral (5.2) can be written in the following form:
# #
δWc = N δξ ds + [aij T i δξ j + ξ 3 T i (δn · ρi − hkj aik δξ j )]ds
3
(5.6)
s s
The full integral in eqn. (5.6) must be considered with the variation of the
convective coordinates which are obtained from eqn. (4.31) for the penetra-
tion as the third coordinate ξ 3 in the form
δξ 3 = (δrs − δρ) · n, (5.7)
and from eqn. (4.35) for the convective coordinate ξ j in the form
$ %
δξ j = âij (δrs − δρ) · ρi − ξ 3 δn · ρi + hki (δrs − δρ) · ρk , (5.8)
with âij as inverse of the matrix with components (aij − 2ξ 3 hij + (ξ 3 )2 hki hjk )
as discussed in Section 4.1.5.
The full formulation with eqns. (5.6), (5.7) and (5.8) in the local coordinate
system is very cumbersome. Now the most important requirement for further
consideration: we consider the motion of the slave point S as if the observer
is placed on contact point C (projection of the slave point onto the master
surface), see Fig. 4.1 and 5.1. Thus, we are taking the third coordinate as
ξ 3 = 0. That means that the contact integral in eqn. (5.6) is considered on
the tangent plane. We obtain the following form:
# #
δWc = N δgds + Tj δξ j ds = (5.9)
s s
# #
= N (δrs − δρ) · nds + Tj aij (δrs − δρ) · ρi ds,
s s
which is accompanied with the variation of the convective coordinates on the
tangent plane in the form:
δξ j = aij (δrs − δρ) · ρi . (5.10)
The formulation of the contact integral in the form presented in (5.9) is

mostly used in contact mechanics (see Wriggers [190] and Laursen [106]).
Remark 5.1. The weak formulation for both the Point-To-Surface and the
Surface-To-Surface contact pairs is identical. Within the Point-To-Surface
algorithm within the further discretization a single slave node is taken leading
to the famous Node-To-Segment (NTS) approach. For the contact integral
within the Surface-To-Surface algorithm an integration point from the slave
surface is involved into the Closest-Point-Projection (CPP) procedure leading
to the Mortar method.
5.2 Weak Formulation of 2D Contact Interaction

Here we are formulating a weak form for 2D contact interaction, kinematics
of which has been considered in Section 4.2. Let us consider the contact
tractions Rs and Rm on both the slave and the master contact curves with
corresponding lengths ls and lm in the current configuration. Let δus resp.
δum be variations of the displacement field on the curves ls resp. lm , then
the work of the contact forces is determined in the following integral
# #
δWc = Rs · δus dls + Rm · δum dlm , (5.11)
ls lm
which must be added to the global work of the internal and external forces.
Due to equilibrium at the contact boundary Rs dls = −Rm dlm , equation
(5.11) can be also written as
#
δWc = Rs · (δus − δum )dls . (5.12)
ls
The integral in (5.12) is considered in the local coordinate system. We redefine

now the variations δus = δrs for a slave point and δum = δρ for a projection
of the slave point onto the master curve.
Substituting the variation δus − δum = δrs − δρ from eqn. (4.121) and
also the full contact traction vector eqn. (6.166) into the integral (5.12) we
obtain
#
δWc = (N δζ + T δξ)dl. (5.13)
l
A closer look reveals that the contact integral (5.13) contains the work of
the contact tractions T and N defined on the master contact curve and is
computed along the slave curve l ≡ ls .
5.3 Weak Formulation for Point-To-Curve Contact

In order to define the components of the contact force vector R in the curvi-
linear curve coordinate system as conjugate variables with regard to the work
W we introduce a coordinate system via τ , e, g, see Fig. 5.2, where
g =τ ×e (5.14)
Thus, vector R is decomposed as:
R = N e + T τ + Bg. (5.15)
A point-wise equilibrium contact condition RS + RO = 0 is formulated in

variational form as:
5.4 Weak Formulation for Curve-To-Curve Contact 123
ν
S
ϕ
O Tτ
Ne
A Gg
Fig. 5.2 Point-To-Curve contact element. Contact forces acting on the contact
point 0.
δW = RS · δrs + RS · δρ, (5.16)

and is transformed exploiting eqns. (5.15) and (4.177) into
δW = (δrs − δρ) · FS = N δr + T δs + rδe · (T τ + Bg). (5.17)
with the assumption that r is small (applications only to the edge contact)
we obtain
δW = (δrs − δρ) · FS = N δr + T δs (5.18)
Further developments for finite values of r as well as application to the beam-
to-beam contact see the next Section 5.4.
5.4 Weak Formulation for Curve-To-Curve Contact

By taking into account equilibrium between two interacting curves in the
corresponding coordinate system, developed for kinematics of CTC contact
in Sect. 4.5, a weak formulation of contact, moreover, a symmetric form is
obtained. Since a curve-to-curve interaction is resulting from various geomet-
rical situations constitutive equations for contact traction have to be consid-
ered separately for the cases of beams or cables and for edges of solid bodies.
The full formulation is derived for arbitrary large motions and an arbitrary
structure of the curves. A weak form, or variational form, is representing the
equilibrium conditions between two curves via the principle of virtual work.
D C
R1 ds1
R2 ds2
A B
Fig. 5.3 Curve-To-Curve. Acting forces on infinitesimal segments dsI .
Let us consider RI dsI as a force vector distributed along a curve segment

dsI , I = 1, 2, see Fig. 5.3
For an infinitesimal curve segment dsI a virtual displacement vector δρI
is defined. Then a virtual work δW for two contacting curves is computed as
integral along the mutual contact area
# #
δW = δρ1 · R1 ds1 + δρ2 · R2 ds2 . (5.19)
s1 s2
Considering that two curves are contacting along the mutual boundary the
equilibrium conditions should be point-wisely fulfilled
R1 ds1 + R2 ds2 = 0. (5.20)
Using eqn. (5.20) in eqn. (5.19) the virtual work can be equivalently written
as an integral either along the first curve, or along the second curve
# #
δW = R1 · (δρ1 − δρ2 )ds1 = R2 · (δρ2 − δρ1 )ds2 . (5.21)
s1 s2
For symmetry reasons we can use the following:

# #
1
δW = R1 · (δρ1 − δρ2 )ds1 + R2 · (δρ2 − δρ1 )ds2 . (5.22)
2 s1 s2
Now taking into account the developed contact kinematics between curves –
only point-wise contact between curves with regard to the CPP procedure – we
can define forces acting point-wisely via the Dirac delta function D(sI − spI )
along curve sI at the contact (projection) point spI
RI (sI ) = RI D(sI − spI ). (5.23)

5.4 Weak Formulation for Curve-To-Curve Contact 125
Thus, the virtual work in eqn. (5.22) is immediately transformed into

δW = Sym {R1 · (δρ1 − δρ2 )}
1
= {R1 · (δρ1 − δρ2 ) + R2 · (δρ2 − δρ1 )} . (5.24)
2
The force vector R1 should be defined as the energy conjugate vector, i.e.
force components and variation of coordinates (4.196) are conjugated pairs
representing that the scalar product in eqn. (5.24) leads to the virtual work.
For arbitrary large motions of curves the force vector should be expressed
via the vectors ri in eqns. (4.183a), (4.183b), (4.183c) exploiting then the full
metric tensor in eqn. (4.184). Thus, we can define R1 in a contravariant form
as
ri
R1 = Ri ri = Rphys
i
√ , (5.25)
gii
i
where the physical components Rphys are introduced as
1
Rphys = T (s1 ) − tangential force along τ 1 ,
2
Rphys = N (r) − normal force along e1 , (5.26)
3
Rphys = Rϕ (ϕ) − rotational (torsional) force along g1 .
Thus, the full force in the basis of a first curve, see Fig. 6.9, becomes

(1 − rk1 cos ϕ1 ) τ 1 + κ1 r g1
R1 = T1 √ + N e1 + Rϕ1 g1 =
g11

(1 − rk1 cos ϕ1 ) κ1 r
= T1 √ τ 1 + N e1 + Rϕ1 + T1 √ g1 . (5.27)
g11 g11
Since the coordinate vectors ri are changing, the tangential component T on

a curve has then a contribution projected on both τ 1 and g1 vectors at r = 0.
However, the force at the projection point of the first curve with r = 0 is
expressed as follows
M1
R 1 = T 1 τ 1 + N e1 + g1 1 ↔ 2. (5.28)
r
The force N is a mutual normal force between two curves acting along
the common normal vector e1 = −e2 (thus, the subscript is omitted) and
M1
Rϕ = represents a force acting at a distance r. This force results in an
r
equivalent rotational moment M1 at the projection point acting along the
τ 1 axis of the first curve (plane e1 , g1 ), see Fig. 6.9. The moment is appear-
ing due to the dimension reduction and can be interpreted as a moment in
a 1D-Cosserat continuum. Using symmetry the corresponding components
with the index (...)2 are acting on the second curve.
Now, the virtual work in eqn. (5.24) taking into account eqn. (4.195) for
the variation of the relative displacement vector is transformed into
δW = Sym{R1 · (δρ1 − δρ2 )} = −Sym{Ri δξ j gij }

" &
i
Rphys
= −Sym √ j
gij δξ , (5.29)
gii
where the scalar product is expressed via the full metric tensor gij in
eqn. (4.184) and, for a while, convective coordinates denoted as s = ξ 1 , r =
ξ 2 , ϕ = ξ 3 . Expressing eqn. (5.29) in the basis of a curve via the physical
components we obtain:
'
δW = −Sym T1 (1 − rk1 cos ϕ1 )2 + (rκ1 )2 δs1 + N δr + M1 δϕ1
(5.30a)
&
κ1 r2
+T1 δϕ1 + M1 κ1 δs1 . (5.30b)
(1 − rk1 cos ϕ1 )2 + (rκ1 )2
The minus sign appears as the expression in eqn. (4.195) is given for (δρ2 −
δρ1 ). The weak form expressed in the basis ri with arbitrary r contains – due
to the non-orthogonality of the coordinate vectors ri – the last mixed terms
in eqn. (5.30b). One can also obtain this result using eqn. (4.195) together
with eqn. (5.27) directly. At the projection point on a curve (r = 0) the
variational equation becomes
⎧ ⎫
⎪
⎨ ⎪
⎬
δW = Sym T1 δs1 + N δr +
! M (δϕ + κδs ) . (5.31)
⎩ ! δW
1 1 1
⎪ !⎪⎭
δWT N δWM
Exploiting the variations of convective coordinates at r = 0 in eqn. (4.197)

the variational equation can be written in vector form as
⎧ ⎫
⎪
⎪ ⎪
⎪
⎨ M1 ⎬
δW = Sym T1 (δρ2 − δρ1 ) · τ 1 + N (δρ2 − δρ1 ) · e1 + (δρ2 − δρ1 ) · g1 .
⎪
⎪ ! ! r !⎪
⎪
⎩ δWT δWN
⎭
δWM
(5.32)
We obtain the weak form allowing to describe the:
1. normal interaction between curves (part δWN );
2. relative tangential interaction independently along each curve (part δWT );
3. relative moment interaction independently for each curve (part δWM ).
The last two positions are the obvious achievement of the current covariant
description, because
• it is not possible within the previously known models, see the overview
in Wriggers [190], to distinguish independently the relative motion along
each curve;
5.5 Weak Formulation for Curve-To-Rigid Surface Contact 127
• only the current covariant description allows to define the moment inter-
action between curves in an energy consistent form.
Remark 5.2. The weak form presented in eqns. (5.31), (5.32) is obtained with
the assumption of a small distance r and thus by construction is suitable
for the case of edge-to-edge contact. For the case of beam-to-beam contact
the value r includes the geometrical parameters of a beam and, therefore,
is not small for arbitrary beams. In such a case as well as in a case with
extremely curved geometry (e.g. for high-order FE with exact geometry) the
full variational equation (5.30a), (5.30b) should be taken.
The developed theory produces an open question:

How to define additional constitutive equations for the moments M1 and M2
?. Some proposals considered in Section 6.7 are devoted to constitutive equa-
tions, however, it is clear that a non-frictional problem can be characterized
only by the functional δWN , while the full sticking case will require all parts of
the functional δW in the corresponding variational equations (5.30a), (5.30b),
or in the case of small r in (5.31).
5.5 Weak Formulation for Curve-To-Rigid Surface

Contact
Considering the developed kinematics for the Curve-To-Rigid Surface inter-
action, see Sect. 4.6, we are formulating the weak form in the curve Serret-
Frenet coordinate system as an integral along the curve l. It is formulated
via parameters expressed either in the curve Serret-Frenet coordinate system
τ , e, g, or via parameters defined in the Gaussian surface coordinate system
ρi , n, thus using the projection of force defined in Sect. 6.9, eqn. (6.193).
Starting with the weak form expressed at the curve, we have:
#
δW = Rs · (δrs − δρ)dl (5.33)
l
#
= − (δue N + δuτ T + δug B)dl via curve parameters (5.34)
# l
= (T (δrs − δρ) · τ + N (δrs − δρ) · e + B(δrs − δρ) · g) dl(5.35)
l
or via surface parameters
#
$ %
= N (δrs − δρ) · e + T (ρi · τ )δξ i + B(ρi · g)δξ i dl (5.36)
#l
= [N (δrs − δρ) · e
l
%
− (T j ρi · (δrs − δρ)) ρi · (τ ⊗ τ + g ⊗ g)ρj dl (5.37)
For the linearization, however, the CTC form in eqns. (5.34-5.35) and corre-
sponding results should be used.
5.6 Nitsche and Lagrangian Formulations of

Non-Frictional Contact Problems
Contact problems within numerical approaches like the finite element method
are mainly dealing with well known contact formulations like the Penalty or
Lagrange multiplier method. In the current Chapter two different Nitsche
schemes applicable to contact problems derived in the sense of optimization
theory are presented for the Surface-To-Surface contact. The common basis
of both Nitsche schemes is that the Nitsche approach is taking two con-
tact constraints into account. Besides the well known non-penetration con-
straint (ξ 3 > 0 ≡ no contact) there is also the equilibrium of stresses between
both bodies considered. Starting with the minimization problem considering
both contact constraints the following statement for the contact problem is
obtained:
Minimize the potentials

WM + WS → min (5.38)
subjected to constraints on the contact boundary:
non-penetration ξ 3 = (rS − ρM ) · nM = 0 (5.39)
equilibrium of stresses σ M · nM + σ S · nS = 0. (5.40)
Here WM and WS denote the work of the structural part of the problem
for the master and slave body, respectively. ρM and rS define the position
vectors on the master and slave surface. Together with the normal vector
nM on the master body these position vectors are enabling the computation
of the penetration ξ 3 according to the Surface-To-Surface contact approach
(see Fig. 5.4) with the closest-point projection procedure. Eqn. (5.40) gives
the equilibrium on the mutual contact area on both bodies expressed by the
surface stress tensor σ and the normal vector n.
The surface stress vector σ ·n, thereby, can be presented in parts of normal
and tangential components leading to the formulation
σ · n = [(σ · n) · n]n + T = σn n + T. (5.41)
For the non-frictional problem considered here this decomposition of the

stress vector enables to set the component T representing sticking and slid-
ing cases to zero. Afterward, for contact the direction of the normals of both
bodies is equal (nM = −nS ).
Thus, the equilibrium of stresses in eqn. (5.40) can be expressed by
[(σ M · nM ) · nM ]nM + [(σ S · nS ) · nS ]nS = (σnM − σnS )nM = 0. (5.42)

5.6 Nitsche and Lagrangian Formulations 129
Fig. 5.4 Definition of a spatial coordinate system specific for closest-point

procedure
In order to enforce the considered constraints in eqn. (5.39) and (5.42) in

a fully variational sense – in accordance with the optimization theory – an
additional set of Lagrange multipliers on the master (μM ) and on the slave
surface (μS ) are defined. Finally, the Nitsche approach is gained as the min-
imization problem for the following functional
#
N = WM + WS + WD + μM · (σnM nM ) dSM −
#SM
− μS · (σnS nM ) dSS → min (5.43)
SS
"
μS if SS
with a set of Lagrange multipliers μ =
μM if SM .
The constraint of the equilibrium of stresses (eqn. (5.42)) between the con-
tacting surfaces of the master and slave body (SM and SS ) is expressed by the
+ functional WD represents
last two terms. The the non-penetration
+ constraint
1 3 2 3
either in Penalty SC 2 N (ξ ) dSC or Lagrange SC λξ dSC formalities
with SC denoting the contacting area. SC is equal to the contact area mea-
sured either on the master surface (SM ) or on the slave surface (SS ). The
parameter N within the Penalty expression is the penalty factor.
In the following sections some remarks have to be pointed out in order
to achieve a convenient formulation out of eqn. (5.43). The first remark is
regarding the choice of the second set of Lagrange multipliers μ. Further-
more, the physical meaning of the Lagrange multiplier λ and its counterpart
the Penalty term N ξ 3 for the standard enforcement of the non-penetration
constraint have to be depicted.
5.6.1 Choice of the Lagrange Multiplier Set μ

Obviously, the additional Lagrange multipliers μ are representing position
values due to their introduction in eqn. (5.43) for terms concerning stress
vectors.
The choice of μ as a specific position vector follows by taking the variation
in the sense of Gateaux derivative for eqn. (5.43) as
#
δWM + δWS + δWD + {δμM · (σnM nM ) + μM · (δσnM nM ) −
SC
−[δμS · (σnS nM ) + μS · (δσnS nM )]} dSC = 0.(5.44)
Transforming this functional then step by step into the strong formulation by
excluding all terms regarding the strong form of equilibrium of the bodies and
the non-penetration constraint (eqn. (5.39)) the following expression remains
#
{[(δρM − δrS ) + (δμM − δμS )] · (σnM nM ) +
SC
(μM − μS ) · (δσnM nM )} dSC = 0. (5.45)
The boundary terms arising within this transformation out of the work ex-
pressions WM and WS are expressed in contact kinematic formalities using
δρM and δrS .
In order to fulfill the requirement – first part in square brackets to become
zero – the additional set of Lagrange multipliers μ is chosen specifically as
μM = −ρM
μS = −rS (5.46)
leading to
# #
[(rS − ρM ) · (δσnM nM )] dSC = ξ 3 δσnM dSC = 0 (5.47)
SC SC
and resulting again in a strong formulation for the non-penetration constraint

(ξ 3 = 0).
5.6.2 Physical Meaning of the Non-penetration Terms

The second remark before dealing with the formulation given in eqn. (5.43) is
the physical meaning of the non-penetration terms. Thus, here the mechanical
interpretation of the non-penetration terms for both, the Penalty and the
Lagrange multiplier method, are repeated. The interpretation is necessary
due to the Nitsche method – as will be shown later on – being a Lagrange
multiplier method with specially chosen Lagrange multipliers.
Lagrange Multiplier λ
Taking the standard Lagrangian contact formulation
#
L = WM + WS + λξ 3 dSC → min (5.48)
SC
in variational form with the Gateaux derivative as

#
δWM + δWS + (ξ 3 δλ + λδξ 3 ) dSC = 0 (5.49)
SC
into account and removing with the help of the Gauss theorem all terms
regarding the strong form – equilibrium of the bodies and the non-penetration
constraint – the functional (5.49) can be written as
# # #
(σnM nM ) · δρM dSM − (σnS nM ) · δrS dSS + λδξ 3 dSC = 0.
SM SS SC
(5.50)
Considering the contact kinematics ξ 3 = (rS −ρM )·nM the following equation
#
(−σnM + λ) δξ 3 dSC = 0 (5.51)
SC
remains. Thus, leading in the strong formulation to the requirement that the
Lagrange multiplier defined on the contact surface is equal to the normal
!
stress on the master surface (σnM = λ).
Penalty Term N ξ 3
In order to gain the mechanical interpretation of the Penalty term the fol-
lowing functional is minimized
#
1
P = WM + WS + N (ξ 3 )2 dSC → min (5.52)
SC 2
by taking its variation into account

#
δWM + δWS + N ξ 3 δξ 3 dSC = 0. (5.53)
SC
Transforming the variation step by step into the strong formulation using the
Gauss theorem results in
#
[(σnM nM ) · δρM − (σnS nM ) · δrS + N ξ 3 δξ 3 ] dSC = 0. (5.54)
SC
!
Finally, the requirement σnM = N ξ 3 is obtained for the Penalty formulation.
Summarizing the considerations within the Lagrange multiplier and the
Penalty method the Lagrange multiplier λ used for the non-penetration con-
straint can be expressed physically as a combination of a normal stress σn
and a penalty term N ξ 3 .
5.6.3 Types of the Nitsche Approach

The necessity to obtain the minimum of the functional in eqn. (5.43) together
with the specific choice of the Lagrange multiplier set μ defined in Sect. 5.6.1
leads now to the following variational equation
δWM + δWS + δWD +

#
+ [(δrS − δρM ) · (nM σnM ) + (rS − ρM ) · (nM δσnM )] dSC = 0. (5.55)
SC
Considering the contact kinematics ξ 3 = (rS − ρM ) · nM and its variation

δξ 3 = (δrS − δρM ) · nM the following functional is achieved:
#
δWM + δWS + δWD + (δξ 3 σnM + ξ 3 δσnM ) dSC = 0. (5.56)
SC
Expressing then the functional WD representing the non-penetration con-

straint according to the Penalty formulation in Sect. 5.6.2 the Nitsche ap-
proach is gained as
# #
δWM + δWS + N ξ 3 δξ 3 dSC + (δξ 3 σnM + ξ 3 δσnM ) dSC = 0.
SC SC
(5.57)
This formulation is described by Wriggers and Zavarise in [199] and is also

obtainable by considering a standard Lagrangian contact formulation like
stated in (5.49) by using the physical meaning of the Lagrange multiplier λ
(see Sect. 5.6.2) and substituting it with σnM + 12 N ξ 3 . This can be interpreted
as a Gauss point-wise substitution of the Lagrange multiplier in eqn. (5.49)
and therefore is further on denoted as the Nitsche approach with Gauss point-
wise substituted Lagrange multipliers.
The representation of the Lagrange multiplier as
λ = σnM + N ξ 3 (5.58)
can also be substituted in eqn. (5.49) using the Bubnov-Galerkin procedure

with δλ as a weighting function. Thus, we obtain the following equation
system
#
δWM + δWS + (ξ 3 δλ + λδξ 3 ) dSC = 0
SC
#
δλ(λ − σnM − N ξ 3 )dSC = 0. (5.59)
SC
This approach is denoted as the Nitsche approach with Bubnov-Galerkin-wise

partial substituted Lagrange multipliers and is used by Heintz and Hansbo,
see [66].
6
Contact Constraints and Constitutive
Equations for Contact Tractions
It is required to define the contact conditions including both geometrical

conditions such as non-penetration and sticking, and constitutive relations
for contact forces and stresses. With regards to the introduced coordinate
system and kinematics the contact tractions are naturally split into normal
and tangential components. If for the normal components non-penetration
conditions can be formulated then for tangential components the constitutive
relations are required. Two major approaches are used in contact mechanics
to formulate non-penetration conditions and to enforce contact constraints :
• The Lagrange multiplier method allows to define the contact tractions
as separate unknowns and the contact constraints are enforced via ad-
ditional Lagrange multipliers. This leads to a formulation known in the
optimization theory as a saddle-point problem. Then all constraints are
exactly enforced.
• The penalty method allows to define the contact constraints via specifi-
cally defined and a-priory convex penalty functions. Thus, the method is
enforcing the contact constraint only approximately.
However, since the penalty method can be mechanically interpreted as a spec-
ification of an additional interface law acting on the contact area and coupled
with contact constraints then the question, especially in experimental contact
mechanics, still remains open whether to interpret this approach as an approx-
imate mathematical method to enforce contact constraints, or to interpret it
as a specification of an additional constitutive law for contact interfaces. The
following discussion is illustrating some difficulties. The normal contact con-
straints have a geometrical interpretation as non-penetration, and, so far, can
be satisfied via a stable numerical algorithm with Lagrange multipliers. How-
ever, the tangent constraints including e.g. exact satisfaction for the Coulomb
friction law have many numerical problems arising from the proved non-
uniqueness of the solution (see the discussion in Eck and Jarušek [35] (1998)).
Actually all stable numerical algorithms for Coulomb friction for large dis-
placement problems are mainly based either on the penalty method, or on
136 6 Contact Constraints and Constitutive Equations for Contact Tractions
the Augmented Lagrangian method. Moreover, there are many experimental

proofs for the existence of elastic tangent deformations and, therefore, are
motivating to work with the penalty method interpreting it as an additional
constitutive law for contact interfaces. In any case constitutive relation for
the tangential forces have to be defined in order to describe “sticking-sliding”
transition. The most standard and classical law is the Coulomb friction law.
The current Chapter deals with formulations of contact conditions for all
geometrical cases such as Surface-To-Surface and Curve-To-Curve contact
pairs and other particular cases kinematics of which is studied in Chapter 4.
A special attention is given to the generalization of the classical Coulomb
friction law for the Surface-To-Surface as well as for Curve-To-Curve contact
cases.
6.1 Surface-To-Surface Contact – Non-Frictional Case

and Isotropic Coulomb’s Frictional Case
In this section we start with non-frictional setup of contact problem for
Surface-To-Surface case, kinematics of which has been described in Sec-
tion 4.1, with the specification of contact constraints in the form of Karush-
Kuhn-Tucker (KKT) conditions. Then we discussed a frictional setup with
the classical Coulomb friction law. In both frictional and non-frictional cases
we introduce regularization leading to the penalty method.
The vector of contact tractions R is defined as a covariant vector in the
spatial coordinate system described Sect. 4.1 in eqn. (4.4), and, therefore,
is expressed via the contravariant basis vectors ρi and n as the sum of the
tangential and normal components on the tangent plane ξ 3 = 0.
R = T + N = Ti ρi + N n. (6.1)
Remark 6.1. For the non-frictional contact problems tangential traction is

zero T = 0 and only normal traction N should be defined.
6.1.1 Contact Constraints for Normal Contact

Conditions
Contact conditions in terms of For normal contact they can be formulated
as the Karush-Kuhn-Tucker complementary conditions for the variational
problem as a pure geometrical enforcement
1. Contact is not appearing and contact force is zero Nmaster = 0 for positive
value of penetration
ξ 3 > 0; Nmaster = 0; no contact (6.2)

6.1 Surface-To-Surface Contact 137
2. Contact occurs when a slave point penetrates into the tangent plane ξ 3 ≤
0, however for exact representation non-penetration condition should be
fulfilled
ξ 3 = 0; Nmaster < 0; non penetration (6.3)

3. The contact traction N exists only, if the slave point is on the tangent
plane, i.e. when ξ 3 = 0, N = 0 , otherwise N = 0 , ξ 3 = 0.
ξ 3 N = 0; complementary condition (6.4)
Remark 6.2. Constitutive equations should be generally supplied in the lo-

cal coordinate system separately for normal and tangent tractions leading
to a variety of models between contact interfaces (e.g. elastic, viscoelastic,
plastic, adhesive etc. laws in both normal and tangential directions). A pure
geometrical enforcement of the non-penetration condition does not require
additional constitutive relations for normal traction and is usually specified
via the Karush-Kuhn-Tucker conditions mentioned above.
If the full Lagrange multiplier method is further involved to satisfy the condi-
tions (6.2-6.4) then the normal traction N is independent variable and should
be independently approximated in finite element discretization. It leads to the
exact satisfaction of the non-penetration condition (6.3).
Penalty Method for Normal Contact Conditions

The penalty method from the mathematical point of view leads to an ap-
proximately satisfied non-penetration condition (regularization) and from a
mechanical point of view it is a constitutive equation for normal elastic com-
pliance. This condition is formulated as

0 if ξ 3 > 0 no contact
N= (6.5)
N ξ if ξ 3 ≤ 0 penetration.
3
The statement in eqn. (6.5) can be written via the Heaviside function H(x)
as follows:
N = H(−ξ 3 )N ξ 3 (6.6)
where the Heaviside function H(x) is defined as

0 if x < 0
H(x) = (6.7)
1 if x ≥ 0
Linearization of the Normal Contact Traction N for the Penalty

Method
In order to linearize the contact traction N in eqn. (6.5) in the local coordinate
system, the rate of penetration in eqn. (4.31), Sect. 4.1.5 is directly used. Thus
we obtain
Ṅ = N ξ˙3 H(−ξ 3 ) for penetration ξ 3 ≤ 0. (6.8)
6.1.2 Tangential Contact Conditions. Evolution

Equations
Additional constitutive equations are necessary for the tangential contact
tractions Tj . Traditionally the Coulomb friction law is taken to define the
transfer between the sticking and sliding situations. This kinematic approach
allows to describe stick and slide conditions in the surface spatial coordinate
system, see Section 4.1.
1. The slave point S sticks, if its projection point C is not moving on the
tangent plane, i.e. has zero relative velocity vr = 0, or equivalently via
the convective coordinates ξ˙j = 0, j = 1, 2.
2. The slave point S slides, if during loading there is a relative motion of
its projection point C, i.e. vr = 0, or equivalently via the convective
coordinates ξ˙j = 0, j = 1, 2.
These conditions for the simplest constitutive model as a Coulomb dry fric-
tion law can be specified as follows:
1. The slave point sticks as long as the Coulomb dry friction inequality holds
vr = 0 if Φ := T − μN ≤ 0 (6.9)
where μ is a friction coefficient, and T is the absolute value of the

tangential traction T, which is computed as

T = Ti Tj aij . (6.10)
2. Beyond the threshold defined by the friction condition (6.9) the slave
point starts to slide in the direction of the relative velocity vector; the
tangential tractions are then acting in the opposite direction.
T
if Φ > 0 then ∃ ζ > 0 vr = −ζ , (6.11)
T
where ζ is a consistency parameter.

3. Sliding happens only if Φ = 0, thus
ζΦ = 0. (6.12)
Again the contact conditions lead to a lack of differentiability and, there-

fore, numerical problems. In order to overcome this Kikuchi [84] considered a
penalty regularization for the contact functional assuming a small tangential
motion in the case of sticking; a penalty regularization based on the elasto-
plastic analogy was developed then in Wriggers et. al. [195], Laursen and
Simo [162]. In the last article the following regularization was proposed in
convective coordinates for the trial tractions :
Ti 1 ∂Ti
aij ξ˙j − ζ =− , (6.13)
T T ∂t
where T is a penalty parameter. Then a return-mapping algorithm known

from plasticity can be used to satisfy the stick-slide condition.
From a mathematical point of view (see Marsden and Hughes [122]), it
appears to be more correct to consider a parallel translation of the vector
field Ti (ξ i (t)) on the master surface. In this situation the relative velocity
vector vr of the projection point C on the master surface, see eqn. (4.39) must
be equal to the full time derivative in the covariant form (2.33) of the vector
T defined on the tangent plane in the spatial coordinate system. Thus, for
the corresponding regularization we propose the following form
T 1 dT
vr − ζ =− , (6.14)
T T dt
or employing the covariant derivative of T on the tangent plane in eqn. (4.27),

we obtain the following expression for the components

Ti 1 ∂Ti ∂Ti
aij ξ˙j − ζ =− + k ˙j k ˙
− Γij Tk ξ + hi Tk ξ 3
(6.15)
T T ∂t ∂ξ j
or finally, having denoted the time derivative of Ti as

dTi ∂Ti ∂Ti ˙j
= + jξ , (6.16)
dt ∂t ∂ξ
we obtain

Ti 1 dTi
aij ξ˙j − ζ =− − Γijk Tk ξ˙j + hki Tk ξ˙3 , (6.17)
T T dt
where Γijk are the Christoffel symbols and hki are mixed components of the
curvature tensor of the master contact surface.
6.1.3 Return-Mapping Scheme for the Real

Tangential Traction
In order to compute the real tangential traction, i.e. in accordance with the
Coulomb friction law, we employ a return-mapping algorithm based on the
backward Euler implicit scheme for the ordinary differential equations to
integrate the differential equation (6.17), see e.g. Simo and Hughes [161].
The trial step is assumed to be with sticking, therefore ζ = 0. The consistent
backward Euler scheme for eqn. (6.17) has the following form

j (n+1)
δki − Γijk |(n+1) ξ(n+1) + hki |(n+1) ξ(n+1)
3
(T trial )k =
(6.18)
(n) (n+1) j (n) j
= (T trial )i − T aij ξ(n+1) − aij ξ(n) −
(n)j (n)
−Γijk |(n) Tk ξ(n) + hki |(n) ξ(n)
3
Tk ,
which can be seen as a backward scheme for the following ordinary differential
equations
dTi
= (−T aij + Γijk Tk )ξ˙j − hki Tk ξ˙3 (6.19)
dt
The system of ordinary differential equations for the computation of the tan-
gential traction (6.19) is called the evolution equations . They are important
for the linearization process. Keeping the form with the covariant derivatives
(6.19) instead of the form in eqn. (6.13) leads to a symmetrical tangent ma-
trix for sticking, while as used in Laursen and Simo [109], [104], the form
(6.13) leads to a non-symmetrical tangent matrix for the arbitrary 3D case.
We will write for further generalization evolution equation (6.19) in the
vector form as:
dT
= −T ξ̇ i ρi , (6.20)
dt
dT
where a full time derivative is taken in covariant form as
dt
dT DTi i DTi dTi
= ρ, ≡ − Γijk Tk ξ˙j + hki Tk ξ˙3 , (6.21)
dt dt dt dt
Remark 6.3. Consider the backward Euler scheme (6.18) in the case with a
linear approximation of the master surface. Then, having taken all Christof-
fel symbols and components of the curvature tensor as zero, we obtain the
following equations:
(n+1) (n) j j
(T tr )i = Ti − T aij (ξ(n+1) − ξ(n) ). (6.22)
This algorithm can be found in Laursen [106] for the trial step solution of
equation (6.13).
Any analysis based on equation (6.18) becomes computationally rather ex-

pensive, because a full matrix appears on the left side and additional history
(n)
variables Γijk |(n) , aij have to be used. Moreover, the integration scheme (6.18)
as well as (6.22) suffers from jumps occurring at element boundaries due to
the different internal coordinates ξ i . Thus, we propose a discrete analog of
the evolution equations (6.19) for the numerical computation
ΔT = −T Δρ. (6.23)
The application of the results of Sect. 4.1.6 about Continuous Numerical

Integration Algorithm for a Relative Motion Vector to eqn. (6.23) together
with the sliding condition leads to the following return-mapping scheme , see
Fig. 4.2 for definition of the vector Δρ:
Trial Step
⎫
N (n+1) = N g (n+1) ⎪
⎪
⎪
⎪
⎪
⎪
⎪
j (n+1) ⎪
(n+1)
(T tr )i
(n)
= Tk akj
(n)
(ρj · ρi
(n+1)
) − T Δξ aij ⎪
⎪
(n) ⎬
, (6.24)
⎪
⎪
(n+1) := T(n+1) − μN
Φtr tr (n+1)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
Ttr
(n+1) (n+1) ij ⎭
(n+1) = (T tr )i (T tr )j a(n+1)
where Δξ j is obtained as
⎧
⎪
⎪ (Δρ · ρk ) ajk
(n+1) for NTS and STS approaches, where
⎪
⎨ Δρ = ρ (n+1) | 1
C 2
ξ(n+1) , ξ(n+1) − (ρC (n) + ΔuC (n) ) |ξ(n)
1 , ξ2
Δξ j = (n) . (6.25)
⎪
⎪
⎪
⎩ −(Δu · ρ )akj
k (n+1) for STAS approach
Return Mapping Scheme
The stick-slip condition is checked within the return mapping process:

⎧ (n+1)
⎪
⎪ (T tr )i (n+1) ≤ 0 (stick)
if Φtr
⎪
⎨
(n+1)
Ti = tr (n+1) . (6.26)
⎪
⎪ (n+1) (T )i
⎪
⎩ μN if Φ tr
> 0 (slide)
Ttr
(n+1)
(n+1)
Remark 6.4. The regularized frictional problem is strictly path-dependent: it

follows from the fact that the contact tractions Ti in the contact functional
in eqn. (5.9) must satisfy the evolution equations (6.19). The return-mapping
algorithm for the incremental solution, as is known, is unconditionally stable,
but a problem of choosing the displacement increments arises due to the

correct definition of sticking and sliding zones. A simple a-priory estimation
will be proposed further for the numerical example, see Sect. 8.5.2.
6.1.4 Integration of Evolution Equations. Geometrical

Interpretation of the Return-Mapping Scheme
As shown later in numerical examples, see Sect. 8.3 and 8.5, originated to
[87], [88], within the contact description for Surface-To-Surface algorithm the
curvature part can be omitted in numerous cases without loss of efficiency
leading to considerable simplifications and a major gain in numerical effort.
This is especially pronounced for such contact problems, where the develop-
ment of sticking-sliding zones is important. In such cases we can e.g. simplify
all equations, assuming constant metrics. The evolution equation is solved
then numerically via the implicit backward scheme with n indicating the
load step arising from a subdivision of the loads applied in sequential load
steps.
(n+1) (n) j j
Ti = Ti − T aij (ξ(n+1) − ξ(n) )= (6.27)
continuing recursively:
(n−1) j j j j
= Ti − T aij (ξ(n) − ξ(n−1) ) − T aij (ξ(n+1) − ξ(n) )
(n−1) j j
= Ti − T aij (ξ(n+1) − ξ(n−1) ) = ...
(0) j j
= Ti − T aij (ξ(n+1) − ξ(0) )
(0)
Assuming that the initial tangential forces are zero, Ti = 0, we get
(n+1) j j
Ti = −T aij Δξ j , with Δξ j = (ξ(n+1) − ξ(0) ). (6.28)
Eqn. (6.28) defines trial tractions Ti at load step (n + 1) at contact points on

1 2
the tangent plane ξ(n+1) , ξ(n+1) . The absolute value of the tangent traction
is computed as:
(n+1) (n+1) ij j j
T2 = Ti Tj a = 2T aij (ξ(n+1)
i
− ξ(0)
i
)(ξ(n+1) − ξ(0) ). (6.29)
With eqn. (6.29) the sticking zone in combination with the Coulomb law
becomes then:
j j
T2 ≤ μ|N |2 =⇒ 2T aij (ξ(n+1)
i
− ξ(0)
i
)(ξ(n+1) − ξ(0) ) ≤ μ2 N 2 . (6.30)
Eqn. (6.30) describes a circle in a Cartesian coordinate system via the con-
vective coordinates ξ1 , ξ2 . The circle is placed on the tangent plane with the
6.2 Generalization of Coulomb Friction Law 143
1 2
center at ξ(0) , ξ(0) . The inner part of it defines the allowable elastic region
for the projection of a slave point, the so-called adhesion domain. Thus, the
geometrical interpretation of the solution of the evolution equation is a tra-
jectory of the contact point which is allowed to be inside the circle in the case
of sticking. If eqn. (6.5) is taken for regularization of the normal traction N ,
then we obtain a cone equation in the spatial coordinate system:
j j
2T aij (ξ(n+1)
i
− ξ(0)
i
)(ξ(n+1) − ξ(0) ) ≤ μ2 2N (ξ 3 )2 . (6.31)
This interpretation can be found in Krstolovic-Opara and Wriggers [101],
where a so-called ”frictional cone description” was proposed.
6.2 Generalization of Coulomb Friction Law into

Complex Contact Interface Law
The regularization for a Coulomb type frictional law, discussed in Sect. 6.1.2
leads to a subdivision of the motion of the contact point on the master sur-
face into reversible and irreversible parts. The first, reversible part appears
due to the regularization and usually contains a penalty parameter. It de-
scribes the elastic tangent deformation, the so-called tangential adhesion, see
also discussion in [31]. The second, irreversible part is described by a flow
rule due to a specific yield function such as e.g. Coulomb friction Law. Both
parts can be generalized for anisotropic domains by taking proper tensors
and equations. In this section, we summarize all necessary equations in vec-
tor form convenient for the expansion into the anisotropic domain. For the
representations of the anisotropic tensor we will use the representation based
on the spectral decomposition in the simple case of constant orthotropy, and
in the general case, the representation based on the tensor product of unit
vectors of an arbitrary curvilinear surface coordinate system.
6.2.1 Vector Form of the Isotropic Equations

The generalization is based upon the consideration of anisotropic tensors
instead of isotropic tensors in the evolution equations. The structure of
the anisotropic tensors for contact, if one or both contact surfaces have
anisotropic structure is theoretically discussed by Zmitrowicz [213], [214] and
also by He and Curnier [61]. Here we assume the described properties for
tensors and will further discuss some restrictions for both the adhesion ten-
sor and the friction tensor. All tensors are defined in the basis of the tangent
plane of the contact master surface.
Elastic Part of the Contact Deformation

To start the development, we reorganize the evolution equations for tangent
tractions in scalar (6.19), or in vector form (6.20) and the evolution equation
for the normal traction (6.8), describing in fact the elastic reversible part of
the deformation in vector form in the local surface coordinate system.
dR
= Êvr , (6.32)
dt
where R is a traction vector acting on a slave point S; vr = vs −v = ρi ξ˙i +nξ˙3
is a relative velocity of a slave point and Ê is an isotropic tensor of penalty
parameters: ⎡ ⎤
−T 0 0
Ê = ⎣ 0 −T 0 ⎦ . (6.33)
0 0 −N
Eqn. (6.32) describes the force-displacement relationship in a rate form for
the reversible elastic part of the contact interaction. The irreversible part in
the case of a simple Coulomb friction law is correlated to the rigid plasticity
model.
Yield Function for the Isotropic Coulomb Friction Law

Remembering, that the scalar product is computed on the surface via the
metric tensor components aij , we can define a yield function for the Coulomb
friction law as
√
T·T Ti Tj aij
Φ := −1≡ − 1. (6.34)
μ|N | μ|N |
The sticking and sliding zones are now defined by the rule:
Φ ≤ 0 means sticking; Φ > 0 means sliding. (6.35)

Irreversible parts including the sliding forces etc. can be obtained via the flow
rule.
6.2.2 General Interface Model

In order to take into account a diversity of linear mechanical models including
viscoelasticity etc., we can generalize eqn. (6.32) as follows:
dR
+ ÂR = B̂vr + ĈΔξ, (6.36)
dt
where Â, B̂, Ĉ are tensors defined in the local surface coordinate system and
Δξ is a vector of the relative displacements
Δξ = Δξ i ρi + Δξ 3 n, Δξ i = ξ i − ξ(0)
i
, i = 1, 2, 3. (6.37)
i
A point with convective coordinates ξ(0) indicates an initial position where
the contact traction vector R equals zero, i.e.
R|ξi =ξi = 0. (6.38)

(0)
The equation for the elastic region in the form of eqn. (6.36) covers various vis-
coelastic models such as Maxwell and Kelvin models in arbitrary anisotropic
forms including adhesion. A generalization of the isotropic Coulomb friction
model according to Maxwell and Kelvin viscoelastic models was considered
in Araki and Hjelmstad [5]. A rate-dependent model with orthotropic fric-
tion coefficients in Cartesian coordinate system was considered in Oancea
and Laursen [132].
Assuming the decoupling of the third normal coordinate ξ 3 , we can rewrite
eqn. (6.36) for the surface components in the form:
dT
+ AT = Bvr + CΔξ, (6.39)
dt
where A, B, C are tensors defined on the tangent plane, T is a tangent force
vector, vr = vs − v is a relative tangent velocity vector expressed on the
d
tangent plane, Δξ = Δξ i ρi is a relative tangent displacement vector and
dt
is a full time derivative in the covariant form on the contact surface.
The third equation for the normal force N and for the penetration ξ 3 , is
treated separately
Ṅ + aN = bξ̇ 3 + cΔξ 3 . (6.40)

For the further development we consider only a rate independent motion,
i.e. assume a specific structure of the evolution equations (6.36) and (6.40)
excluding the direct time dependency of the contact tractions.
Anisotropic Evolution Equations. Rate-Independent Model

Considering a case of rate-independent motions by taking A = 0, C = 0 in
eqn. (6.39) and a = 0, c = 0 in eqn. (6.40) we define anisotropy for an elastic
part of the contact conditions. Therefore, from eqn. (6.39) the following rate
forms remain:
dT
= B(vs − v). (6.41)
dt
Expressing this in the tangent plane by coordinates, we get the following
evolution equation
∂Ti
+ ∇j Ti ξ˙j = bij ξ˙j , j = 1, 2. (6.42)
∂t
The evolution equations (6.41) resp. (6.42) describe the fact that the reaction
tangential forces are acting not in the opposite direction to the velocity vector,
but in the direction defined by tensor B. In other words, if a force T is acting
on point C on the surface, then this point is moved in a somewhat different
direction defined by the angle β, see Fig. 6.1, but, in general, not in the
direction of the force.
Remark 6.5. The evolution equations (6.41) describe the elastic deformations
of contact interaction in the rate form. This elastic tangential deformation is
known as tangential adhesion, see Curnier [31]. Thus, we call a tensor B the
elastic adhesion tensor, or simply the adhesion tensor.
Remark 6.6. The mechanical restriction for the elastic force T to act in op-
posite direction to the relative velocity can be formulated in an energy sense
according to the thermodynamical restriction: the power of the elastic force
T must be negative. This requires that the adhesion tensor taken with
a minus sign −B is a positively defined tensor.
Since we are working with a decoupled model regarding the normal and
tangential contact interactions, the evolution equation for the normal force
N and thus the parameter b in eqn. (6.40) is kept in the form given by
eqn. (6.19).
6.2.3 Anisotropic Yield Function

Several theoretical approaches are known in literature for formulations for the
yield function and for the sliding rule. Before presenting a particular structure
of the tensor we will briefly review these approaches as well as restrictions
for the tensor which are known in literature.
Various Approaches for Formulations of Yield Criteria and

Sliding Rules
In a first publication, Michalowski and Mroz [128] proposed to distinguish
functions for limit criteria and for sliding, introducing the so-called associ-
ated and non-associated sliding rules. These functions were built by analogy
looking at the solution for the sliding of a rigid block on an inclined surface.
Zmitrowicz [213] introduced the friction tensor F into the originally
isotropic Coulomb criteria (6.34) and described its properties based on the
groups of material symmetry. The sliding forces were formulated directly
without correspondence to the yield criteria. It was also obtained that ”There
is no restrictions for the arbitrary anisotropic friction tensor except its posi-
tive definition, but the orthotropic friction tensor is symmetric”. The super-
position of two friction tensors when two anisotropic surfaces are in contact
was also discussed. In a later publication Zmitrowicz [214] assumed that the
sliding force nonlinearly depends on the relative velocity introducing in ad-
dition a 4th -order friction tensor.
He and Curnier [61] applied the theory of tensor function representations

– the fundamental aspects of this theory see Zheng [207] – to obtain the
general irreducible structure for the nonlinear friction tensor in the case of
friction between two orthotropic surfaces. The structure at the contact point
is defined by [61] as follows:
• m is a unit vector of the preferable direction for the first surface,
• k is a unit vector of the preferable direction for the second surface,
• u is a unit vector of the relative sliding velocity.
Then the sliding force T is defined as
T = N [ α1 (I1 , I2 )E + α2 (I1 , I2 )(m ⊗ m) + α3 (I1 , I2 )(k ⊗ k) ] u , (6.43)
where N is a normal force, E is a unit tensor, αi are scalar functions of the

following invariants: I1 = u · (m ⊗ m)u, I2 = u · (k ⊗ k)u. Also the sym-
metric properties of the friction tensor for orthotropic surfaces were derived
and it was shown, that a yield function is a direct consequence of the rate-
independence condition. Other important results were ”the thermodynamical
restrictions for the friction tensor leading to its positivity and the formulation
of the convex energy dissipation function”.
Coulomb Type Yield Functions

Summarizing the results of the previous developments, we will later use a
generalization of the isotropic yield criterion in eqn. (6.34) by replacing the
scaled metric components aij /μ2 N 2 with tensor components f ij , assuming
the a-priori necessary properties as discussed above. Thus, we can obtain:

ij
ΦN = fN Ti Tj − 1 = (T · FN T) − 1. (6.44)
The standard assumption of proportionality of the sliding force Tsl to the

normal traction N , the so-called Coulomb’s form, see Curnier [31], leads to
F
FN = , (6.45)
N2
and the yield function can then be written as:
√
Φ = f ij Ti Tj − |N | = T · FT − |N |. (6.46)
The sliding criteria (6.34-6.35) become then
if Φ ≤ 0 then sticking, (6.47)

describing a contact point inside the adhesion domain,
if Φ > 0 then sliding, (6.48)
describing a contact point outside the adhesion domain.

The tensor F is called friction tensor and is defined by its components
f ij in the surface tensor basis as
F = f ij ρi ⊗ ρj . (6.49)
Comparing with the isotropic function, we introduce anisotropic friction co-

efficients :
aij
f ij = 2 , i, j = 1, 2, (6.50)
μij
where μij are coefficients of friction.
6.2.4 Tensor Representations for Anisotropy

Here we consider particular structures for the anisotropic tensors for the evo-
lution equations (6.41) as well as for the yield function (6.46), automatically
satisfying the necessary restrictions mentioned in the previous sections. We
start with the simplest case – a constant orthotropy on the plane. In this case
all tensors are symmetrical ones. The more general anisotropic case can be
defined then in an arbitrary coordinate system by setting different properties
along the coordinate lines.
Spectral Representation of the Tensor – Constant Orthotropy in

the Plane
As mentioned above, constant orthotropy in the plane is described by con-
stant symmetric tensors for which a spectral representation is chosen. Any
symmetric positive tensor A can be decomposed as:
A = QΛQT , (6.51)
where Λ is defined in the mixed tensorial basis as a diagonal matrix of eigen-
values:
λ1 0
Λ= ; (6.52)
0 λ2
and Q is an orthogonal tensor. Since it describes the rotation on the tangent
plane between the main axes ei and the axes ξ i , see Fig. 6.1, it becomes

cos α − sin α
Qα = . (6.53)
sin α cos α
v
e2 ε 2 ξ2
β
1
0
0
1 T
C
e1 ε1
α ξ
1
0 1
0
1
O
Fig. 6.1 Main orthotropy axes and local surface coordinate system.
The main axes are hereby defined as orthotropy axes. Then the spectral
representation of A is obtained as
A = Qα ΛQTα = [Aij ] =

λ1 cos2 α + λ2 sin2 α (λ1 − λ2 ) sin α cos α
= · (6.54)
(λ1 − λ2 ) sin α cos α λ1 sin2 α + λ2 cos2 α
Taking into account, that the model for constant orthotropic friction contains
the tensor in the evolution equations (6.41) and in the yield function (6.46)
in mixed form which allows the spectral decomposition given in eqn. (6.54),
the following derivations are possible:
First we focus on the evolution equation given by eqn. (6.42), where
the mixed components are introduced via the spectral decomposition (6.54).
Taking into account that the adhesion tensor B (see Remark 6.6 in Sect. 6.2.2)
is negative, we introduce positive values εi = −λi > 0 and an orthotropy
angle α. Thus, after a transformation into a local coordinate system the
following matrix description for the tensor is obtained:

ε cos2 α + ε2 sin2 α (ε1 − ε2 ) sin α cos α
B = [bij ] = − 1 . (6.55)
(ε1 − ε2 ) sin α cos α ε1 sin2 α + ε2 cos2 α
Matrix B in eqn. (6.55) describes the orthotropic properties in the reversible

part of the tangential interaction, where the parameters ε1 , ε2 can be seen
as orthotropic moduli of the tangential adhesion. Assuming ε1 = ε2 leads to
isotropic behavior and then the isotropic evolution equation is recovered.
Second, the yield function for the constant orthotropic friction is defined
via the orthotropic friction tensor F which allows a spectral decomposition.
After introducing the angle β between the local coordinate axes and the
orthotropy axes, and eigenvalues as λi = 1/μ2i , the following matrix [fki ] is
obtained:
⎡ 1 1 1 1 ⎤
cos2
β + sin2
β ( − ) sin β cos β
⎢ μ21 μ22 μ21 μ22 ⎥
⎢ ⎥
F = [fki ] = ⎢ ⎥. (6.56)
⎣ 1 1 1 1 ⎦
( 2 − 2 ) sin β cos β 2 sin β + 2 cos β
2 2
μ1 μ2 μ1 μ2
The restriction of positivity for the friction tensor F leads to the known
positive orthotropic friction coefficients μi > 0.
Structure of the Anisotropic Tensor Inherited from an Arbitrary

Surface Curvilinear Coordinate System
We assume that on the surface defined via the Gaussian coordinates ξ 1 , ξ 2 as
⎧ ⎫
⎨ x1 (ξ 1 , ξ 2 ) ⎬
ρ = ρ(ξ 1 , ξ 2 ), ρ = x2 (ξ 1 , ξ 2 ) (6.57)
⎩ ⎭
x3 (ξ 1 , ξ 2 )
another Gaussian coordinate system is defined. Thus, the Cartesian coordi-

nates of the same surface xi are defined by Gaussian convective coordinates
α1 , α2 :
xi = φi (α1 , α2 ), i = 1, 2, 3. (6.58)
One can say, that we have re-parameterization of the surface
αi = αi (ξ 1 , ξ 2 ), i = 1, 2, (6.59)
or, in another words, eqn. (6.59) defines a new coordinate system on the same
surface.
Arbitrary anisotropic properties of a surface can be defined as different
characteristics along these coordinate lines. The unit tangent vectors along
the coordinate lines are then defined as:
ri
ei = √ , i = 1, 2 no summation over i (6.60)
gii
∂r
where ri = are the basis vectors, and gii = ri · ri are diagonal coefficients
∂αi
of the covariant metrics tensor.
The tensor of anisotropy A can then be defined via the unit tensor basis
as:
r1 ⊗ r1 r2 ⊗ r2 ri ⊗ ri
A := λ1 e1 ⊗ e1 + λ2 e2 ⊗ e2 = λ1 + λ2 = λi . (6.61)
g11 g22 gii
Remark 6.7. From now on and further in the last representation in eqn. (6.61)
the summation convention is implied also for the sum over i-index, though
the index i is repeated four times.
It is obvious, that the tensor structure prescribed above, see eqn. (6.43), is
preserved. As a next step a transformation into the surface basis ρi (ξ 1 , ξ 2 ) is
necessary for the evolution equation as well as for the yield equation.
Remark 6.8. The tensor in eqn. (6.61) is defined, in general, via the coordi-
nates α1 , α2 . We introduce a notation AC , if the tensor A is defined in the
global reference Cartesian coordinates, and will keep the notation A if the
tensor is defined after the tensor transformation to the surface coordinate
system ξ 1 , ξ 2 , eqn. (6.57).
The evolution equation is transformed as follows. Assuming the stiffnesses

along the coordinate lines as λi = −i , we obtain

r1 ⊗ r1 r2 ⊗ r2
B := − 1
C
+ 2 . (6.62)
g11 g22
The tensor BC is a Cartesian tensor. Some steps are necessary to transform

it to surface coordinates ξ 1 , ξ 2 . Substitution into the right hand side of the
evolution equation (6.41) leads to:
ri ⊗ ri i (ri · ρj )ξ˙j
BC (vs − v) := −i · ρj ξ˙j = − ri . (6.63)
gii gii
The dot product of the evolution equation (6.41) with ρk , taking into account
eqn. (6.63), leads to the equation in components according to the surface
metrics:
dTk i (ri · ρk )(ri · ρj ) j
=− ξ˙ . (6.64)
dt gii
For the further implementation into a finite element code, it is more appro-
priate to introduce a tensor decomposition:
dTk i ri ⊗ ri
=− : (ρk ⊗ ρj )ξ˙j , (6.65)
dt gii
i ri ⊗ ri
where the components of the first tensor BC = − can be computed
gii
in a Cartesian coordinate system separately for the surface as:
i ∂φl ∂φn
ln = −
bC . (6.66)
gii ∂αi ∂αi
After this the evolution eqn. (6.65) is defined as:
dTk ∂xl ∂xn ˙j

= bC
ln ξ . (6.67)
dt ∂ξ k ∂ξ j
From the last equation the covariant components of the tensor B in the global
surface basis ρ1 , ρ2 are obtained as tensor transformations, i.e.
∂xl ∂xn
bij = bC
ln . (6.68)
∂ξ i ∂ξ j
Introducing the friction coefficients λi = 1/μ2i into the yield function and
then applying analogous tensor operations as for the evolution equation lead
to the following form:

ri ⊗ ri
Φ= : (Tk Tl ρk ⊗ ρl ) − |N |, (6.69)
μ2i gii
from which a similar definition of covariant components for the friction tensor
are obtained:
ri ⊗ ri
fkl = 2 : (ρk ⊗ ρl ). (6.70)
μi gii
In the following we consider particular structures for covariant components
in the Cartesian coordinate system bC ij in eqn. (6.66) as well as for bij in
eqn. (6.68) in the local surface coordinate system for the anisotropic plane,
for polar orthotropy on a plane and for spiral orthotropy on a cylinder.
Anisotropic Plane. Structure of the BC and B Tensors in

Cartesian Coordinates
On the plane x3 = 0, we consider anisotropic properties defined by two unit
vectors r1 and r2 :
⎧ ⎫ ⎧ ⎫
⎨ cos α ⎬ ⎨ cos β ⎬
r1 = sin α , r2 = sin β . (6.71)
⎩ ⎭ ⎩ ⎭
0 0
The Cartesian components of the adhesion tensor bC
ln are obtained as:
⎡ ⎤
ε1 cos2 α + ε2 cos2 β ε1 sin α cos α + ε2 sin β cos β 0
BC = − ⎣ ε1 sin α cos α + ε2 sin β cos β ε1 sin2 α + ε2 sin2 β 0⎦.
0 0 0
(6.72)
For an analysis, the approximation of the surface is necessary in order to
obtain the structure of the tensor in surface coordinates ξ 1 , ξ 2 in eqn. (6.68).
If the latter coincide with the global Cartesian coordinates ξ i = xi , we obtain

ε1 cos2 α + ε2 cos2 β ε1 sin α cos α + ε2 sin β cos β
B=− . (6.73)
ε1 sin α cos α + ε2 sin β cos β ε1 sin2 α + ε2 sin2 β
It is obvious, that isotropy is no longer recovered simply by taking ε1 = ε2 .

However, if we take the unit vectors to be orthogonal, i.e. β = π/2 + α,
the orthotropic matrix obtained previously via the spectral representation in
eqn. (6.55) is recovered.
Orthotropic Surface in Polar Coordinates. Structure of the BC

and B Tensors in Cartesian Coordinates
As an example with curvilinear orthotropy, a plane with orthotropic prop-
erties in polar coordinates, see Fig. 6.2, is considered. These properties are
defined by the elastic constants εr , εφ acting along the coordinate lines. The
definition in eqn. (6.58) is then a transformation to polar coordinates:
x = r cos φ, y = r sin φ, z = 0. (6.74)

The Cartesian components of the orthotropic tensor bC
kn are then defined by
the following matrix:
⎡ ⎤
εr cos2 φ + εφ sin2 φ (εr − εφ ) sin φ cos φ 0
BC = − ⎣ (εr − εφ ) sin φ cos φ εr sin2 φ + εφ cos2 φ 0 ⎦ . (6.75)
0 0 0
Applying the inverse mapping from polar coordinates to Cartesian coordi-
nates in eqn. (6.74) (see also eqn. (6.68), we obtain:

1 εr x2 + εφ y 2 (εr − εφ )xy
B=− 2 . (6.76)
x + y 2 (εr − εφ )xy εr y 2 + εφ x2
The adhesion tensor for polar orthotropy in eqn. (6.76) contains a typical
example of nonlinear surface properties in the reference Cartesian coordinate
system.
rφ, ε φ
rr, ε r
11
00
φ Y
O 00
11
Fig. 6.2 Polar orthotropic surface.

Spiral Orthotropy on a Cylindrical Surface. Structure of the BC

and B Tensors in Cylindrical Coordinates
As a more complex case, we consider a circular cylinder and the surface or-
thotropy resulting from spiral coordinate lines on the cylinder. This example
can be practically interesting to model e.g. screw connections. First, we define
a rigid cylinder with a surface described by the following parameterization
ρ(α, z), see Fig. 6.3 and similar Fig. 14.3:
⎧ ⎫
⎨ R cos α ⎬
ρ = R sin α , (6.77)
⎩ ⎭
z
where α is an angular parameter in plane XOY. The necessary surface char-

acteristics are the tangent vectors and the normal vector
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨ −R sin α ⎬ ⎨0⎬ ⎨ cos α ⎬
ρ1 = R cos α , ρ2 = 0 , n = sin α , (6.78)
⎩ ⎭ ⎩ ⎭ ⎩ ⎭
0 1 0
resulting in a covariant surface metrics tensor

2
(ρ1 · ρ1 ) (ρ1 · ρ2 ) R 0
[aij ] = = . (6.79)
(ρ2 · ρ1 ) (ρ2 · ρ2 ) 0 1
Orthotropic properties of the surface are obtained as follows. The equation

for a family of cylindrical spiral lines on the cylinder (6.77) is, see also the
geometry given in Fig. 6.3 and similar Fig. 14.3:
⎧ ⎫
⎪
⎪ R cos α ⎪
⎨ R sin α ⎪ ⎬
r= . (6.80)
⎪
⎪ ⎪
⎩ H α + const ⎪ ⎭
2π
The first tangent vector r1 along the spiral line necessary for the tensor
representation (6.62) becomes:
⎧ ⎫
⎪
⎪ −R sin α ⎪
∂r ⎨ R cos α ⎪⎬
r1 = = . (6.81)
∂α ⎪ ⎪ H ⎪
⎪
⎩ ⎭
2π
The second tangent vector r2 is defined to be orthogonal to the first vector r1
and to the normal on the cylinder surface, see the third equation in eqn. (6.78)
and Fig. 6.3:
⎧ ⎫
⎪ H ⎪
⎪
⎪ sin α ⎪
⎪
⎪
⎪ 2π ⎪
⎪
⎪
⎨ ⎪
⎬
r2 = [n × r1 ] = H . (6.82)
⎪
⎪ − cos α ⎪
⎪
⎪
⎪ 2π ⎪
⎪
⎪
⎪ ⎪
⎪
⎩ ⎭
R
With this expression an equation for the line orthogonal to the main spiral
line (6.80), see line AC in Fig. 6.3, can be found from the condition that the
integrated and scaled tangent vector r2 must belong to the cylinder surface
#
2πR
A r2 dα ∈ cylinder =⇒ A = , (6.83)
H
which leads to the following definition of a vector r̂, orthogonal to r:
⎧ ⎫
⎪
⎪
R cos α ⎪
⎪
⎨ R sin α ⎬
r̂ = . (6.84)
⎪
⎪ ⎪
⎩ 2πR α + const ⎪
2
⎭
H
Thus, orthotropic properties are inherited from the orthogonal spiral net on
the cylinder via eqns. (6.80) and (6.84).
Remark 6.9. For further analyses, one can define from eqns. (6.80) and (6.84)
the distances between two adjacent threads as H for the main spiral and
(2πR)2
Ĥ = for the orthogonal spiral AC, see Fig. 6.3.
H
The covariant components of the metric tensor gij are defined as
⎡
2 ⎤
2 H
⎢R + 2π
0 ⎥
[gij ] = ⎢
⎣
2 ⎥
⎦. (6.85)
H
0 R2 +
2π
The tensor of orthotropy Bα in the Cartesian basis, see eqn. (6.62), becomes
then:
r1 ⊗ r1 r2 ⊗ r2
BC = −1 − 2 = (6.86)
g11 g22
⎡ ⎤
RH
gε sin2 α −gε sin α cos α −(ε1 − ε2 )
sin α
⎢ 2π ⎥
⎢ ⎥
⎢ ⎥
⎢ RH ⎥
1 ⎢ −gε sin α cos α 2
− ⎥
=− 2 ⎢ g ε cos α (ε 1 ε 2 ) cos α ⎥,
H ⎢ 2π ⎥
2
R + ⎢ ⎥
2π ⎢ 2 ⎥
⎣ RH RH H ⎦
−(ε1 − ε2 ) sin α (ε1 − ε2 ) cos α ε1 + ε2 R 2
2π 2π 2π
with
2
H
g ε = ε1 R 2 + ε2 . (6.87)
2π
The backward tensor transformation (6.68) with the matrix H defined via
the cylindrical coordinates
⎡ ⎤
−R sin α 0
∂xi ∂ρ
H= = = ⎣ R cos α 0 ⎦ (6.88)
∂ξ j ∂ξ
0 1
gives us the covariant components bij of the tensor B in the contravariant

surface basis ρ1 , ρ2 :
⎡ ⎤
R2 H
gε R 2 (ε1 − ε2 )
⎢ 2π ⎥
1 ⎢ ⎥
B = [bij ] = HT BC H = −
2 ⎢
2 ⎥.
H ⎣ R2 H H ⎦
R2 + (ε1 − ε2 ) ε1 + ε2 R 2
2π 2π 2π
(6.89)
Matrix B in eqn. (6.89) represents the constant spiral orthotropy for the
cylindrical surface and, therefore, is a generalization of the constant plane
orthotropy in eqn. (6.55) for the case of a cylindrical geometry.
Remark 6.10. With the assumption of isotropy ε1 = ε2 = ε the unit matrix

is recovered only in mixed components. In covariant components we obtain
2
R 0
Bε1 =ε2 =ε = [bij ]ε1 =ε2 =ε = −ε . (6.90)
0 1
6.3 Derivation of the Anisotropic Adhesion-Friction

Contact Problem via the Principle of Maximum
Dissipation
The principle of maximum dissipation is known in plasticity for the formula-
tion of the necessary characteristics, such as plastic strains etc.. The applica-
tion of this principle for the construction of computational algorithms in lin-
ear and nonlinear isotropic plasticity was developed in Simo and Hughes [161].
He and Curnier [61] formulated the dissipation function for the anisotropic
function and investigated its extremal properties apply convex analysis. The
correspondence between the dissipation function and the sliding rule was
shown. Here we will also formulate the frictional problem as an extremal
problem in a continuous form, and then applying the finite difference scheme
in an incremental form. Afterwards, the return-mapping algorithm is applied
to obtain all characteristics for sliding, such as a sliding force Tsl and a
6.3 Statement via the Principle of Maximum Dissipation 157
Fig. 6.3 Spiral orthotropy on the cylindrical surface.
sliding displacement vector Δξsl . These variables can be viewed as a pair of

conjugate variables for the energy dissipation function, which allows to define
them separately.
6.3.1 Continuous Formulation

According to the elastic-plastic analogy, a frictional contact problem via the
energy dissipation function can be formulated as follows:
a) The relative velocity vector of the contact point is decomposed additively
into an elastic part vel and a sliding part vsl
vr = vel + vsl . (6.91)

b) The elastic part vel is responsible for reversible deformations (adhesion)
and satisfies the evolution equations (6.41):
dT
= Bvel . (6.92)
dt
c) The tangential force T must satisfy the following inequalities defined via
the yield function eqn. (6.46), which in tensor form can be written as:
√
Φ := f ij Ti Tj − |N | = T · FT − |N | : (6.93)
•
if Φ < 0 then the contact point is inside the elastic domain and T =
Tel is an elastic force,
• if Φ = 0 then the contact point is sliding and T = Tsl is a sliding
force.
d) The power of the sliding forces , described by the energy dissipation
function D achieves its maximum:
D := ξ˙sl
i
Tisl = vsl · Tsl , D −→ max . (6.94)
For the convenient application of standard methods of convex analysis [119],
[19], [45] we transform the problem (6.94) into a minimization problem
Dmin := −ξ˙sl
i
Tisl = −vsl · Tsl , Dmin −→ min . (6.95)
The principle of maximum dissipation requires that the plastic dissipation

function D subjected to the inequality conditions (6.93) achieves a maximum.
A system of ordinary equations (6.91) - (6.92) is defined in convective surface
coordinates ξ 1 , ξ 2 , identifying a contact point on the surface.
6.3.2 Incremental Formulation

The application of the backward Euler scheme to the continuous problem
a)-d) described above, namely to a system of ordinary differential equa-
tions (6.91)-(6.92) with an additional extremal condition (6.95) leads to an
incremental formulation. Here we investigate only quasi-static contact prob-
lems, therefore we can take Δt = 1. The return-mapping scheme – for plas-
ticity see [161] and among the first applications for contact problems see [44]
– is applied to obtain the real sliding force and sliding displacements: the
trial tangential force Ttr is assumed to be elastic and can be computed from
the incremental solution. Thus, the following incremental formulation for the
trial tangential force Ttr is found:
i) The full displacement vector Δξ = ξ(n+1) −ξ(n) is decomposed additively
into an elastic increment Δξ el and into a sliding increment Δξsl :
Δξ = Δξ el + Δξ sl , (6.96)
where both vectors are defined in the surface metrics, namely,
Δξ := Δξ i ρi = (ξ(n+1)
i
− ξ(n)
i
)ρi . (6.97)
ii) The trial force Ttr

(n+1) is computed via the incremental evolution equa-
tions:
(n+1) − T(n) = B (ξ (n+1) − ξ el

(n+1) el
Ttr (n) ). (6.98)
iii) In order to decide whether the trial force Ttr is a sliding force Tsl or a
sticking force Tst the yield condition is checked in each load step:

Φtr : = Ttr (n+1) · F (n+1) − |N(n+1) |
(n+1) Ttr

= f ij Titr(n+1) Tjtr(n+1) − |N(n+1) |, (6.99)
• If Φtr < 0 then the trial force is a real sticking force T = Ttr .
• If Φtr ≥ 0 then the sliding force must be obtained via the maximum
of the energy dissipation function given in the incremental form.
iv) The incremental analog of the continuous formulation eqn. (6.95) is then:
(n+1) (n+1)
Dmin := −Δξ sl · Tsl
(n+1) = −Δξsl Ti (n+1) ,
i sl
Dmin −→ min . (6.100)
For large sliding problems especially with reversible loading, it is necessary to

define both, the sliding force and the sliding displacements. Taking the sliding
distance Δξ sl as an independent variable we can determine the sliding force
(n+1)
Tsl via the minimum of the function Dmin in eqn. (6.100). The expression
for the sliding force will be used for numerical computation within each load
step during an iterative Newton solution scheme. The sliding distance, in
due course, is defined after convergence is achieved in the load step and is
computed via the consistency condition.
A Lagrange function for the constraint minimization problem is given as:

L(n+1) := −Δξ sl · Tsl (n+1) + λ (n+1) · F(n+1) T(n+1) − |N(n+1) | ,
Ttr tr
L −→ min, (6.101)
where the complementary Kuhn-Tucker conditions (see [119], [19]) are given
as:

Φ := Ttr (n+1) · F(n+1) T(n+1) −|N(n+1) | ≤ 0, λ ≥ 0, λΦ(n+1) = 0. (6.102)
tr
For the next transformations a gradient of the yield function is necessary:
∂Φ(n+1) F(n+1) Ttr

(n+1)
= , (6.103)
∂Ttr
(n+1) T(n+1) · F(n+1) Ttr
tr
(n+1)
as well as a derivative of the trial force at load step (n + 1), computed via
the chain rule, see eqns. (6.98) and (6.96):
∂Ttr
(n+1)
= −B(n+1) . (6.104)
∂Δξ sl
In the following sections the subscript (n + 1) will be omitted everywhere for

simplicity reasons.
Derivation of the Sliding Force Tsl

In order to obtain the sliding force Tsl , the sliding incremental displacement
Δξ sl is taken as an independent variable. A formal application of convex anal-
ysis, see [119], [19], to the Lagrange function (6.101) gives us the necessary
condition of the minimum:
∂L
= 0, (6.105)
∂Δξsl
leading to the definition of the sliding force as:
∂Φ
Tsl = λ . (6.106)
∂Δξ sl
Exploiting the chain rule and using eqns. (6.103), (6.104), we obtain:
∂Φ ∂Ttr FTtr
Tsl = λ = −λB √ . (6.107)
tr
∂T ∂Δξ sl
Ttr · FTtr
For the current sliding case, i.e. when λ > 0, we have to satisfy the Kuhn-
sl
Tucker condition
√ Φ = 0, in order to define λ, i.e. substitute T in the yield
function Φ = T · FT − |N |. This leads to the following equation for λ:
sl sl
,
Ttr · FTtr
λ = |N | , (6.108)
BFTtr · FBFTtr
where the positive λ is taken due to the second Kuhn-Tacker condition in
eqn. (6.102). Thus, the sliding force Tsl in eqn. (6.107) is defined as
BFTtr
Tsl = − √ |N |. (6.109)
BFTtr · FBFTtr
We now introduce an auxiliary vector
T̂ = BFTtr (6.110)
to compute the sliding force Tsl . The covariant components of the auxiliary
vector are defined via various components of the tensor B and F as:
T̂i = bij f jk Tktr = bij fln ajl akn Tktr = bji fjk Tktr . (6.111)
Then the covariant components of the sliding vector (6.109) can be computed
as
T̂ T̂i
Tsl = − =⇒ Tisl = − (6.112)
T̂ · FT̂ T̂k T̂l f kl
The isotropic case is recovered from eqn. (6.109) by taking B = −T E and
F = E/μ2 to
Ttr
Tsl = μ|N | √ = μ|N |. (6.113)
Ttr · Ttr
6.3.3 Specification of Initial Conditions for the

Return-Mapping Scheme
Initial conditions are necessary for the incremental solution of eqns.(6.96)
and (6.98) formulated as Cauchy problem for a system of ordinary differential
equations. These initial conditions can be defined assuming that the initial
configuration is unstressed, thus with zero external loading:
T = 0, N = 0 =⇒ ξ = ξ(0) , ξ sl
(0) = 0 for ξ ≤ 0.
3
(6.114)
The conditions are formulated for all points which are in contact at the
initial configuration, i.e. for all points satisfying ξ 3 ≤ 0. The vector ξ(0) is
obtained via a projection procedure, and defines the center of the ellipse for
the adhesion domain, see Fig. 6.4. The additional initialization of the sliding
displacement and the update procedure will be discussed after the geometrical
interpretation of the solution process.
6.3.4 Derivation of the Sliding Incremental

Displacement Δξ sl and Update Scheme for the
History Variables
We consider here a ”step-by-step” scheme for the case with nonlinear tensors
B and F concentrating on computational aspects for the numerical imple-
mentation. For the nonlinear case let us assume that the first converged load
step is elastic, i.e. Φ < 0 in eqn. (6.102), while for the second load step the
sliding condition is achieved, i.e. Φ = 0 and the load step was computed with
the sliding force Tsl . Let ξ (1) and ξ (2) are convective coordinates of a con-
tact point after the corresponding first and the second load steps. The trial
force for the second load step Ttr (2) is then computed as in an Euler backward
scheme :
Ttr
(2) = T(1) + B
(2)
(ξ (2) − ξ (1) ), (6.115)
where for the first load step a force T(1) is computed taking into account the
initial conditions in eqn. (6.114)
T(1) = B(1) (ξ (1) − ξ(0) ). (6.116)

The value of the sliding displacement ξ sl

(2) and resp. the elastic part ξ (2) are
el
defined via the strict fulfillment of the Kuhn-Tucker condition Φ(2) = 0 for
the elastic part ξ el
(2) :
Φ(2) = Φ(2) (ξ, ξ sl ) =

(2) − ξ (1) )) · F
(2) (ξ el − ξ
(T(1) + B(2) (ξ el (1) )) − |N |, (6.117)
(2) (T
(1) + B (2)
where the full displacement vector ξ (2) after the converged second load step
can be decomposed as:
ξ (2) = ξ el
(2) + ξ (2) .
sl
(6.118)
The consistency condition (see [119], [19]) for the constraint function Φ(2)
leads to an additional equation allowing to determine the direction of the
sliding displacement:
sl
∂Φ dξ(2) ∂Φ dξ (2)
Φ̇ = · + sl · = 0. (6.119)
∂ξ(2) dt ∂ξ(2) dt
Continuing with the chain rule we obtain:

el el sl
- .
∂Φ ∂ξ(2) dξ (2) ∂Φ ∂ξ (2) dξ (2) ∂Φ dξ (2) dξ sl
(2)
el ∂ξ
· + el sl
· = · − = 0.
∂ξ(2) (2) dt ∂ξ(2) ∂ξ (2) dt ∂ξ el
(2)
dt dt
(6.120)
From the last equation, we can obtain the following condition
dξ sl
(2) dξ (2)
= . (6.121)
dt dt
Then the sliding displacement update vector ξ sl

(2) can be defined in the direc-
tion of the last converged full displacement vector ξ(2) . For the computation
of the nonlinear case this direction can be approximately taken as
ξ(2) − ξ (1)
e= (6.122)
|ξ (2) − ξ (1) |
leading to the sliding displacement
ξ sl
(2) = λe. (6.123)
The parameter λ defining the length of the vector is obtained from the Kuhn-
Tucker condition for the function Φ(2) in eqn. (6.117). This leads to the
following algebraic equation:

λ2 B(2) e · F(2) B(2) e − 2λ B(2) e · F(2) T(2) +
+T(2) · F(2) T(2) − N 2 = 0, (6.124)
where the positive root λ > 0, minimizing globally the function in eqn. (6.100),
should be taken.
6.3.5 Computational Aspects for Further

Implementation Considering Nonlinear and
Constant Tensors
For the further implementations we consider the following cases: a) with
nonlinear tensors for large displacement problems; b) with constant tensors,
i.e. the case of constant orthotropy, for both, large and small displacement
problems.
A Case with Nonlinear Tensors for Large Displacement Problems

Within the adjustment of the sliding force the strict execution of the back-
ward scheme in eqn. (6.98) leads to the necessity to store as history variables
(n) also the updated sliding variables ξ (n) at load step (n),
in addition to ξ el sl
which is computationally rather expensive. However, the numerical experi-

ence from some cases even with non-constant B, e.g. for polar orthotropy,
shows that it may be mostly sufficient to use a simplified scheme, which
is identical with the backward scheme computed with the updated matrix
B(n+1) at load step (n+1), namely, with the following finite difference scheme
:
Ttr
(n+1) = T(n) + B
(n+1)
(ξ (n+1) − ξ (n) ), (6.125)
where ξ (n) is a displacement vector from the converged load step (n). For large
displacement problems the elastic part ξ el can be neglected leading to the
result that the update sliding vector ξsl
(n) is equal to the displacement vector
ξ (n) in the last converged load step. Computations show that the scheme in
eqn. (6.125) is robust and requires only T(n) and ξ (n) as history variables.
A Case with Constant Orthotropy

For the case with a constant tensor B, we can proceed recursively transform-
ing eqn. (6.98) similar to the isotropic case, see eqn. (6.27), and obtain
(n+1) = T(n) + BΔξ = ... = B(Δξ (n+1) − Δξ ),

el sl
Ttr (6.126)
Δξ (n+1) = ξ(n+1) − ξ (0) ,
where ξ(0) is a center of the elliptical adhesion domain, for which a geomet-
rical interpretation is given later.
For the case with constant tensors the update algorithm in Sect. 6.3.4 leads
to the exact definition of the sliding displacement as:
ξ − ξ(0)
Δξ sl = ξ − ξ(0) − |N | (6.127)
(ξ − ξ (0) ) · BFB(ξ − ξ (0) ),
where ξ is the full displacement vector after the converged load step. A geo-
metrical interpretation of this scheme is discussed in the next section.
6.3.6 Geometrical Interpretation of the Solution

Process
A simple geometrical interpretation for the solution can be given for a plane
surface with constant orthotropy in the case of both, elastic sticking and
sliding behavior. Interpretations can be formulated in the trial force space
T1tr , T2tr and on the tangent plane ξ 1 , ξ 2 where we assume Cartesian met-
rics, i.e. aij = δij . In the trial force space the yield function in eqn. (6.99)
represents an ellipse due to the positivity of the friction tensor F:
Ttr · FTtr = N 2 . (6.128)
In order to obtain the interpretation on the tangent plane, the incremental

evolution equation (6.126) is used. Then, we also obtain an ellipse on the
tangent plane
BΔξ · FBΔξ = N 2 . (6.129)
In order to derive further characteristics, the symmetrical tensors F and B
are expressed via the spectral representation (6.54):
B = Qα DB QTα , F = Qβ DF QTβ , (6.130)
where Qα and Qβ are rotational matrices with corresponding angles α and

β, see eqn. (6.53), and DB , DF are diagonal matrices:
⎡ 1 ⎤
0
ε1 0 ⎢ μ2 ⎥
DB = − , DF = ⎣ 1 1 ⎦ . (6.131)
0 ε2 0 2
μ2
Spectral representations leads to the canonical form of ellipses. Thus,

eqn. (6.128) becomes then
⎡ 1 ⎤
0
⎢ μ21 N 2 ⎥ T tr
QTβ Ttr · ⎣ 1 ⎦ Qβ T = 1, (6.132)
0
μ22 N 2
describing a central ellipse inclined with angle β in the force plane T1tr , T2tr
with the main axes μi N .
Eqn. (6.129) describes a domain with orthotropic elastic properties, where

the contact point S is attracted by the center (adhesion domain). Its transfor-
mation according to the representation (6.130) and (6.131) gives the following
canonical equation:
⎡ 1 ⎤
0
⎢ 2 ⎥
(Qα Qα QTβ )T Δξ · ⎣ (μ1 N/ε1 ) 1
T T
⎦ (Qα Qα Qβ ) Δξ = 1, (6.133)
0
(μ2 N/ε2 )2
Δξ = ξ(n+1) − ξ (0)
describing an ellipse inclined by the matrix Qα Qα QTβ . The ellipse center is
shifted by the distance ξ(0) on the tangent plane ξ 1 , ξ 2 , see Fig. 6.4. The
lengths of the main axes of the ellipse are a resp. b = μi N/εi . The inclination
angle becomes φ = 2α − β, which is verified from the matrix:

T cos(2α − β) − sin(2α − β)
Qα Qα Qβ = (6.134)
sin(2α − β) cos(2α − β)
In the numerical examples, we will also investigate a case with nonlinear or-
thotropy with ε1 = 0. In this limit case with lim a = ∞ the ellipse degenerates
μ2 N
into an infinite strip of width 2b = 2 , see Fig. 6.4. The properties inside
ε2
the strip defined by eqn. (6.126) are elastic, but the motion along the strip
causes the corresponding elastic force to be zero Ta = 0.
2
ξ
in
ma
μ 2N
1111
0000 it d
o 0
b= ε 2
il m ε 1=
μ1 N
sticking a= 000
111
ε1
O φ
11
00
00
11
00
11
00S
11
1
0
0S
1
ξ (0)
sliding
ξ1
1
0
0
1
Fig. 6.4 Allowable elastic region (adhesion domain ).

ξ2
μ 2 Ν2
ε2
0
1 0
1
1
0
1
0
0
1 μ1 Ν1
0
1 1
O(0) 0
1
0
1 A ε1 O(k)
0
1
0
1 B ξ
1
0 1
0 0
10
1 00
11 01
10
01
1 0
1
0
1 0
1 00
1
0
1
00
11
c 0
1
0
1
0
1
ξ (0) 0
1
0
1 ξ (k) 0
1
0
1
0
1 0
1
0
1
0
1
0
1 0
1
0
1 0
1
0
1
ξ (0) ξ sl(k)1
0
0
1 0
1
1111111111111
0000000000000 0
1
11111111111
00000000000
0
1
0
1
Δξ 0
1
0
1
0
1
0
1 0
1
0
1
0
1
Fig. 6.5 Update scheme. Particular case.
Thus, a geometrical interpretation of the solution is as follows. The ellipse

describes an elastic domain with orthotropic properties obtained by the in-
cremental evolution equation (6.126). The sticking condition is fulfilled when
a contact point S remains inside the ellipse. If a contact point S appears out-
side of the ellipse, then this point is sliding, i.e. the acting force is computed
via eqn. (6.112), see Fig. 6.4. Due to the condition Φ = 0 this point is on the
boundary of the elliptic region, which leads to a shifting of the ellipse in order
to define the forces in the next load step correctly. This shifting is originated
by the sliding vector Δξ sl which leads to an update of the ellipse center and
which is computed via the update scheme in eqn. (6.127). The ellipse center
is an attraction point for the domain with anisotropic elastic central forces
defined by the evolution equation (6.126). As long as the contact point S is
inside of the adhesion domain, the sliding displacements as well as the sliding
forces have not to be computed. This is, the so-called, ”sticking zone” . Let k
be the number of the load step, when sliding is detected the first time, i.e. the
contact point has been moved outside the adhesion domain. Then, the sliding
displacement ξ sl(k) is computed via eqn. (6.127). Now, the trial procedure in
eqn. (6.126) is considered in the next load step (k + 1). Since the trial pro-
cedure is a computation in the elastic region, the sliding displacement ξ sl (k+1)
is assumed to be zero.
(k+1) = B(Δξ (k+1) − Δξ (k+1) ) =

sl
Ttr
= B(ξ (k+1) − ξ (0) − ξsl
(k) ) =
= B(ξ (k+1) − (ξ (0) + ξ sl(k) )). (6.135)

!
ξ c(k)
Vector ξ c(k) defines the update scheme for the sliding displacements and al-
lows to describe the shift of the ellipse center. The incremental evolution
equation (6.126) is corrected then in accordance to this update scheme as
(n+1) = B(Δξ (n+1) − ξ n ),

c
Ttr
with ξ cn = ξ cn−1 + ξ sl
n = ... = ξ + ξ k + ... + ξ n .
0 c sl
(6.136)
Now a particular case is considered, when the orthotropy axes coincide with
the Cartesian coordinate axes (i.e. Qα = Qβ = E) and a contact point being
at the position A during the load step (k) moved to the position B during
the load step (k + 1) along the ξ 1 axis with the distance Δξ, see Fig. 6.5.
In this case, the trial force is obtained as T1tr = −ε1 Δξ, and the following
sliding displacement Δξ sl in eqn. (6.127) becomes then
Δξμ1 T1tr
Δξ sl = Δξ − |N | = (−|T1tr | + μ1 |N |). (6.137)
(Δξε1 )2 ε1 |T1tr |
From Fig. 6.5 it becomes obvious that the sliding displacement Δξ sl is updat-
ing the position of the ellipse center O. The last result (6.137) can already be
found for the isotropic case in Wriggers and Krstulovic-Opara [192] and [102].
In addition, the analogy between the geometrical interpretations of friction
and plasticity with kinematical hardening, as described in Simo and Hughes
[161] becomes obvious.
6.3.7 Rheological Model of the Orthotropic

Adhesion-Friction Problem
As is well known [161], the return-mapping scheme used for the model of
elasto-plasticity can be interpreted via a one-dimensional spring-sliding sys-
tem. A generalization of this model into 2D anisotropy is a point on the
plane with a two spring-two slider system with different properties: ε1 , ε2 as
stiffnesses of the springs and μ1 , μ2 as coefficients of friction for the sliding
devices, see Fig. 6.6. A constant force F is applied to the point at an angle
γ. It becomes obvious, that after transformation of the coordinate system in
such a way that e.g. the X-axis coincides with the direction a force F, the
problem exactly corresponds to the constant orthotropy on the plane. The
latter is given by a tensor with the spectral decomposition in the case of
coinciding orthotropy angles for adhesion α and for friction β, see Sect. 6.2.4
eqns. (6.55), (6.56). The trajectory of the point is then a straight line inclined
with an angle ϕ, the value of which depends on the ratio of the eigenvalues
defining the adhesion ellipse, see the geometrical interpretation Fig. 6.4:
λ1 ε 1 μ2
rλ = = · (6.138)
λ2 μ1 ε 2
In forthcoming computations, see Chapter 11, we will show, that it is pos-
sible to represent geometrically isotropic behavior, in such a way that the
trajectory of the point is coinciding with the direction of the force F, though
the properties of this contact surface remain orthotropic.
λ2
μ2
>
2
λ
1
=
λ
1
λ
<λ
2
λ1
ε2
Δ xel
ε1 μ1
X
F γ Δ x sl
Δ x el Δ x sl
Δx
Fig. 6.6 Mechanical interpretation of orthotropic friction – orthotropic adhesion

model. A material point on a plane with two spring-slider systems driven by a
tangential force.
6.4 Analysis of Various Models for Anisotropic

Friction and Adhesion
In this section, the range of applicability of a classical model of orthotropic
friction, based only on the orthotropic friction tensor and its generalization
including orthotropy for both friction (inelastic region) and adhesion (elastic
region) is discussed. The necessity to assume in addition elastic properties
for the surface will be shown. Also recovering of circular motion in the case
of polar orthotropy is proved. This section is necessary for the validation of
adhesion-friction law for some structural surfaces described in Chapter 12.
As a first simple model which can be investigated analytically a material
point on a plane is considered. According to the experimental tests, provided
later in Chapter 12, we assume a quasi-statical motion of the material point
A with weight P loaded by the force F acting along the X 1 -axis, see Fig. 6.7.
The orthotropic properties of the surface are defined in the coordinate sys-
tem ξ 1 , ξ 2 inclined with an angle α to the original coordinate system. During
quasi-statical loading, point A is moving along a line with velocity vector v
inclined with an angle ϕ. The reaction force T with Cartesian components
6.4 Analysis of Various Anisotropic Friction Models 169
T1 , T2 is acting on the point. The values of components depend on the hy-

pothesis concerning the orthotropic friction law. Here two variants of the
orthotropic law are considered: the well known orthotropic friction Coulomb
law and a contact interface model including orthotropy for both friction and
adhesion, as discussed in previous sections.
2 v
X A ϕ
2
ξ −T1 F
−T2
1
ξ
α X
1
O
Fig. 6.7 Motion of material point A on an orthotropic plane loaded by force F.
The equilibrium equations for the system in Fig. 6.7 are given as:
⎧ 1
⎨ X : F + T1 = 0;
X 2 : T2 = 0; , (6.139)
⎩ 3
X : −P + N = 0.
where N is the reaction force along the X 3 axis.

The principle of maximum dissipation is applied to obtain the relation
between the sliding force T and sliding displacements Δrsl . This principle
requires that the dissipation function Ψ reaches its maximum
Ψ := Δrsl · T = Δxisl Ti −→ max, (6.140)
where Δrsl is an increment of the sliding vector. The dissipation function Ψ

must also satisfy the sliding condition, formulated via inequalities, reflecting
the assumed friction law, e.g. Coulomb’s law.
6.4.1 Orthotropic Coulomb Friction Law

First, we recall the standard case known in literature, see e.g. [128], [31], [130],
[214], where orthotropy is defined only for the sliding forces. The model is
formulated according to the generalization of the sliding criteria. The yield
function for the Coulomb friction law is then written as
√
Φ := T · FT − |N | = Ti Tj f ij − |N |. (6.141)
The sticking and the sliding conditions are defined by the rule:
Φ < 0 → sticking; Φ ≥ 0 → sliding. (6.142)
According to equations (6.141-6.142) the material point is not moving during

sticking (no adhesion) and the motion starts when Φ = 0. The components
of the friction tensor f ij are defined for the orthotropy on the plane via e.g.
the spectral representation plane as follows:
F = Qα ΛF QTα = (6.143)
⎡ 1 ⎤
0 T
cos α − sin α ⎢ μ1 2 ⎥ cos α − sin α
= ·⎣ ·
1 ⎦ sin α cos α
sin α cos α 0 2
μ2
⎡ 1 1 1 1 ⎤
cos2 α + 2 sin2 α ( 2 − 2 ) sin α cos α
⎢ μ12 μ2 μ1 μ2 ⎥
⎢ ⎥
=⎢ ⎥,
⎣ 1 1 1 1 ⎦
( 2 − 2 ) sin α cos α 2 sin α + 2 cos α
2 2
μ1 μ2 μ1 μ2
where μi > 0 are friction coefficients along the axis ξ i inclined at angle α.
The standard method of the convex analysis is applied to obtain the sliding
forces with regard to the principle of maximum dissipation (6.140). Thus, the
Lagrange function with the multiplier λ is specified as

L := −Ψ + λΦ = −Δxisl Ti + λ Ti Tj f ij − |N | (6.144)
together with the complementary Kuhn-Tucker conditions :
λ ≥ 0, λΦ = 0. (6.145)
∂L
The optimality conditions = 0 lead to the following sliding displacement
∂T i
components:
Tj f ij
Δxisl = λ . (6.146)
Tk Tl f kl
These equations recover the trajectory of a block as a straight line declined
by angle ϕ, which is confirmed by experiments.
Now, taking into account the second equilibrium equation (6.139) tan ϕ
can be determined:
⎫
T1 f 11
Δx1 = λ ,⎪
⎪
⎬
Tk Tl f kl Δx2 f 12
=⇒ tan ϕ = = 11 , (6.147)
T1 f 12
⎪
⎪ Δx 1 f
2
Δx = λ ,⎭
Tk Tl f kl
6.4 Analysis of Various Anisotropic Friction Models 171
and, after transformations taking into account the values determined in

eqn. (6.143), we finally obtain:
(μ22 − μ21 )
tan ϕ = tan α. (6.148)
μ22 + μ21 tan2 α
6.4.2 Model for Orthotropic Contact Interfaces

Including Both Adhesion and Friction
An alternative model involving coupling orthotropy for both adhesion and
friction is Sect. 6.3 including the elastic-plastic analogy.
For the corresponding sliding force Tsl derived in Sect. 6.3, eqn. (6.109),
we define the inclination angle ϕ. The problem is considered as a displace-
ment driven one, therefore the incremental displacement Δr = {Δx1 , Δx2 }
is applied. Thus, in each load step the sliding force Tsl in eqn. (6.109) is
computed taking into account that Ttr = BΔr as:
BFBΔr
Tsl = − √ |N | = AΔr. (6.149)
BFTtr · FBFTtr
Now, if sliding is assumed, the second component of the sliding force T2 in the
formulation depicted in Fig. 6.7 becomes zero, see equilibrium eqn. (6.139).
Thus, the displacement vector components Δx1 , Δx2 are coupled via the
equation:
T2 = 0 =⇒ a21 Δx1 + a22 Δx2 = 0, (6.150)
leading to the equation for the angle ϕ:
Δx2 a21
tan ϕ = 1
=− . (6.151)
Δx a22
Analysis of the Model by General Spectral Representation

In order to study possible trajectories of a block (a theoretical curve ϕ(α)),
we consider a spectral decomposition of the matrix A given in eqn. (6.149)
as
⎡ 2 ⎤
λ1 cos2 α + λ22 sin2 α (λ21 − λ22 ) sin α cos α
A = [aij ] = ⎣ ⎦, (6.152)
(λ1 − λ2 ) sin α cos α λ1 sin α + λ2 cos α
2 2 2 2 2 2
leading together with the condition (6.151) to the observed sliding angle ϕ
defined as
a21 (λ2 − λ2 ) sin α cos α (λ21 − λ22 ) tan α
tan ϕ = − = − 21 2 2 = − . (6.153)
a22 λ1 sin α + λ22 cos2 α λ21 tan2 α + λ22
An analysis for extremal values gives us

d tan ϕ
= 0, =⇒ (λ21 − λ22 )(λ22 − λ21 tan2 α) = 0. (6.154)
d tan α
The first bracket leads to the isotropic case, whereas from the second one the
following critical value is obtained:
λ2
tan αext = , (6.155)
λ1
leading to the extremum of the observed inclination angle ϕext for the motion
of the point
λ2 − λ21
tan ϕext = 2 . (6.156)
2λ1 λ2
Considering the last equation (6.156) we can obtain a critical ratio of the
eigenvalues
λ1
ratioext = = − tan ϕext ± tan ϕ2ext + 1. (6.157)
λ2
This value will be used during the validation procedure.
For further considerations we adopt the spectral decomposition also for
the adhesion tensor
B = Qα ΛB QTα = (6.158)
T
cos α − sin α −ε1 0 cos α − sin α
= · · =
sin α cos α 0 −ε2 sin α cos α

= 1 ,
where εi > 0 are stiffnesses along the axis ξ i inclined at angle α.
Remark 6.11. The trajectory of a block is a straight line inclined by the angle
ϕ. Moreover, the observed inclination angle ϕ does not depend on particular
values of λ1 , λ2 , but only on its ratio λ1 /λ2 .
6.4.3 Recovering Circular Motion for Polar

Orthotropy
A family of curves in the case of polar orthotropy with the parameter εϕ = 0
for the problem considered numerically in Sect. 11.3.2 is a family of circles.
In order to prove this statement we consider the equilibrium equations for
the quasi-statical sliding of a block, see Fig. 11.17, Sect. 11.3.2.
6.5 2D Contact Problems – Evolution Equations for Contact Tractions 173
T1 = 0
T2 + F = 0. (6.159)
Here a force F is associated with applied displacements along OY axis and

T1 , T2 are components of a sliding force Tsl defined in eqn. (7.48). Following
the incremental displacement loading process and the definition of the sliding
force via the trial force, see also eqn. (7.50) we can write:

BFB Δu
Tsl = −|N | , (6.160)
Ψ Δv
where Δu is a prescribed displacement component and Δv is a computed

component from the equilibrium equation. Then from the first equilibrium
equation (6.159) we obtain:
BFB11 Δu + BFB12 Δv = 0. (6.161)
The limit of the ratio of the displacement increments when Δu goes to zero
leads to the exact definition of the derivative:
Δv BFB11
lim = y = − . (6.162)
Δu→0 Δu BFB12
Now, we specify the particular case with an isotropic friction tensor F = E/μ2
and εϕ = 0 in the adhesion tensor B in eqn. (11.16) leading to
2
εr x xy
B=− 2 . (6.163)
x + y 2 xy y 2
Inserting the corresponding components from eqn. (6.163) into eqn. (6.162)
we obtain an ordinary differential equation (ODE) describing a family of
curves (trajectory of the ODE):
x
y = − . (6.164)
y
The integration of this differential equation leads to
y 2 + x2 = const, (6.165)
which describes a family of circles.
6.5 2D Contact Problems – Evolution Equations for

Contact Tractions
The evolution equations for normal N and tangential T tractions are for-
mulated in a 2D curvilinear coordinate system, see the kinematics of 2D
interaction in Section 4.2. The evolution equations can be regarded as rate

equations for the contact tractions. The contact traction vector Rs is defined
for the slave point s in the local coordinate system on the master curve in
the covariant form as:
ρξ
Rs = T ρ1 + N ρ2 = T + N ρ2 = T + N ν. (6.166)
(ρξ · ρξ )
For the normal traction N , the following regularized equation is similar to
3D case in eqn. (6.6) applied
N = N ζ, H(−ζ)ζ (6.167)
where N is a penalty parameter for the normal interaction and H(x) is a

Heaviside function eqn. (6.7). According to eqn. (4.118), the rate of a normal
traction can be computed as
Ṅ = N ζ̇H(−ζ) = N H(−ζ)(vs − v) · ν. (6.168)
As a reasonable equation for the regularization of the tangent traction vector

dT
T we choose a proportional relation between the full time derivative and
dt
the relative velocity vector expressed on the tangent line ζ = 0:
D1 T 1 ˙ ,
ρ = −T ξρ ξ (6.169)
dt
in component form written as
D1 T ˙
= −T ξ(ρ ξ · ρξ ), (6.170)
dt
where T is a penalty parameter for the tangential interaction. Applying
the results from eqn. (4.106) leads to the evolution equations in the form of
covariant derivatives:
dT ρξξ · ρξ h11
= −T (ρξ · ρξ )ξ˙ + ξ˙ − ζ̇. (6.171)
dt (ρξ · ρξ ) a11
which is used to compute a trial tangent traction.
6.6 Constitutive Equations for Point-To-Curve Contact

Interaction
The constitutive equations for the normal N and tangential force T are for-
mulated in the coordinate system according to the kinematics of the Point-To-
Curve Interaction presented in Section 4.4. This can be regarded, of course,
dually, either as the regularization of contact conditions via the penalty
method, or specification of the new constitutive equations. Thus, the normal
6.7 Curve-To-Curve Contact 175
component N is formulated as follows (constitutive equation or regulariza-

tion):
N = N ζH(−ζ), (6.172)
where N is an elasticity parameter – a penalty parameter. For definition of
ζ see Section 4.4. The tangent component T is formulated in the rate form
as:
dT
= −T ṡ, (6.173)
dt
where T is an elasticity parameter (a penalty parameter) for the tangential
interaction according to the formulation of the weak form in Section 5.3.
A computational algorithm for the tangent traction can be constructed
according to the standard return-mapping scheme for the Coulomb friction
law. First, the trial force Ttr is computed assuming to be an elastic predictor:
(n+1)
Ttr = T (n) − T (s(n+1) − s(n) ), (6.174)
then the real force is computed as:

⎧
⎪ (n+1)
⎨ Ttr
(n+1)
if |Ttr | < μ|N | sticking
T (n+1) = . (6.175)
⎪
⎩ sign(T (n+1))μ|N | if |T (n+1) | ≥ μ|N | sliding
tr tr
Remark 6.12. This formulation is used as simplified model of PTC interac-

tion, weak form for which is given in Sect. 5.3. More precise form includ-
ing also the rotational (torsional) interaction must be taken as a part of
Curve-To-Curve (CTC) interaction for arbitrary geometry considered in the
corresponding Section.
6.7 Curve-To-Curve Contact – Constitutive Equations

for Contact Tractions
The components of a contact force vector for the CTC contact is obtained
as energy conjugated pairs for the rate of defined contact measures as a con-
travariant vector in the corresponding basis, according to the kinematics of
Curve-To-Curve (CTC) contact, see Sect. 4.5. A rotational moment between
curves is appearing in addition to normal and tangential forces. Let us con-
sider the full contact force R1 in the curvilinear coordinate system τ 1 , e1 ,
g1 see kinematics in Sect. 4.5 attached to the first curve, see eqn. (5.27),
obtained in Sect. 5.4:

(1 − rk1 cos ϕ1 ) τ 1 + κ1 r g1
R1 = T1 √ + N e1 + Rϕ1 g1 =
g11

(1 − rk1 cos ϕ1 ) κ1 r
= T1 √ τ 1 + N e1 + Rϕ1 + T1 √ g1 . (6.176)
g11 g11
Fig. 6.8 Normal force N is acting along the vector of the shortest distance between
the curves e1 .
The force R2 acting on the second curve is formulated symmetrically. Now,

we have to define the constitutive equations following components: for the
mutual normal force N , tangential forces Ti and rotational (torsional) forces
Rϕi . As we discussed already in Sect. 5.4 an equivalent moment M1 obtained
M1
by the expression Rϕ = can be defined instead of formulating the ro-
r
tational (torsional) forces Rϕi , see eqn. (5.28). Thus, the full force at the
projection point of the first curve with r = 0 is expressed as follows
M1
R 1 = T 1 τ 1 + N e1 + g1 1 ↔ 2. (6.177)
r
6.7.1 Normal Contact. Specification of Constitutive

Laws for the Traction N Coupled with Contact
Constraints for the Variable r
The following situations with the curve-to-curve contact kinematics may arise
for different mechanical models, namely,
• beam-to-beam contact,
• a special case of solid-to-solid contact when contact is appearing along
sharp edges.
These two cases are influencing the specification of normal contact constraints
for the variable r and, in due course, the specification of constitutive laws for
a normal traction N acting along the vector of the shortest distance between
the curves e, see Fig. 6.8. They are now considered separately.
Case of Beam-to-Beam Contact

Beam models in mechanics are characterized as a line with corresponding
cross sections. The line represents a reduction of a 3D solid to a 1D line,
though in 3D space, according to the chosen kinematical hypothesis (e.g.
Bernoulli, Timoshenko, Reissner). Thus we can always define for each I-th
curve a function RI (ϕI ) defining a cross section of the beam in our coordinate
system, i.e. in the direction normal to a curve in the plane spanned by the
vectors ν I , βI , see Fig. 1.10.
First, we define a mutual penetration of two beams as a part of intersection
of two cross sections R1 (ϕ1 ) and R2 (ϕ2 ) along coordinate r as
p = r − (R1 (ϕ1 ) + R2 (ϕ2 )). (6.178)

A constitutive relation coupled with the corresponding contact constraints
for the normal force N mutual for both curves can be written as:
"
0 if p > 0 no contact
N= (6.179)
εN (r − (R1 (ϕ1 ) + R2 (ϕ2 )) if p ≤ 0 contact.
Here, a parameter εN is defining a penalty parameter, or a stiffness for the

normal interaction. If a penetration p is positive then the curves are out of
contact. For the negative penetration the contact force N is defined as a
linear elastic force.
However, even for the beam-to-beam contact some difficulties may arise
for complex shapes of the cross-section used in practice (e.g. for I, ,
-
shaped domains). In those cases taking the edge-to-edge contact instead of
specifying R1 (ϕ1 ) and R2 (ϕ2 ) may lead to a simpler procedure.
Case of Edge-to-Edge Contact

The case is representing a situation when 3D solid bodies are contacting along
their sharp edges. Assuming that the parametrizations of both edge curves
are separately given we can define the following constitutive equation for the
normal traction:
"
0 if ϕ ∈
/ [ϕ0 , ϕ1 ] no contact
N= (6.180)
−εN r if ϕ0 ≤ ϕ ≤ ϕ1 and p = r ≤ 0 contact.
Here, the normal force is acting along the common normal e1 , see Fig. 6.9,
and is defined in proportion to the distance r (representing the penetration
into the edge) if the corresponding vector e1 is laying inside the sector defined
in plane ν 1 , β1 via the angles ϕ0 and ϕ1 . As in the previous case εN is defining
a penalty parameter, or a stiffness for the normal interaction.
rg1
−Rϕ g 1 r e1 ϕi
ϕf ν1 g1
τ1
s (ξ 1) β1
C1
−M 1
−T 1 τ 1
−N e 1
Z ρ1
O
Y
X
Fig. 6.9 Contact tractions in the spatial coordinate system. Edge-to-Edge contact
case.
6.7.2 Tangential Contact. Specification of

Constitutive Laws for Tractions TI Coupled
with Contact Constraints for the Variables sI
The specification of constitutive laws for tangential variables sI is unique for
the curve-to-curve kinematics, i.e. is common for both beam-to-beam and
edge-to-edge cases. One of the advantages of the current theory is the pos-
sibility to specify the contact law separately for each curve. In each case
the corresponding tangential force TI is acting along the corresponding tan-
gent vector τ I , see Fig. 6.10. For tangential contact the sticking case and the
sliding case should be differentiated. Thus we give a definition of tangential
sticking as follows:
• if ṡ1 = 0 then the first curve is tangentially sticking along the second
curve;
• if ṡ1 = 0 then the first curve is tangentially sliding along the second curve.
This definition is symmetric with regard to the choice of the curves, therefore,
can be repeated in the sense the first curve ↔ the second curve. Using
the elasto-plastic analogy for the Coulomb friction law we can specify the
Fig. 6.10 Tangential force TI is acting along the corresponding tangent vector τ I
for the I th -curve.
constitutive law for tangential friction between curves as elasto-plastic one

(allowing elastic deformations for a sticking case) in the following rate forms:
⎧
⎪
⎨ −ε1 ṡ1 if |T1 | < μ1 |N | tangential sticking along 2nd curve.
Ṫ1 = ṡ1
⎪
⎩ −μ1 |N | if |T1 | ≥ μ1 |N | tangential sliding along 2nd curve.
|ṡ1 |
(6.181)
Here,
ε1 – tangential stiffness along the first curve;
μ1 – coefficient of friction along the first curve.
Applying then the return-mapping scheme to eqn. (6.181) we can get a trial
tangential force as
(trial) (trial,n+1) (n) (n+1) (n)
ΔT1 = −ε1 Δs1 =⇒ T1 = T1 − ε1 (s1 − s1 ), (6.182)
(n) (n)
T 1 , s1 – tangential force (resp. arc-length coordinate)
for the previous converged load step;
(trial,n+1) (n+1)
T1 , s1 – tangential force (resp. arc-length coordinate)
in the current iteration.
Fig. 6.11 Moment M represents the rotation along the τ axis and represents the
forces acting in the circumferential direction.
After the return-mapping step we get the real tangential force as

(trial,n+1) (trial,n+1)
T1 if |T1 | < μ1 |N1 | sticking.
T1 = (sl,n+1) (trial,n+1) (trial,n+1)
T1 = μ1 |N1 |sign(T1 ) if |T1 | ≥ μ1 |N1 | sliding.
(6.183)
All equations (6.181), (6.182), (6.183) are, of course, symmetric concerning
the choice of the curves 1 ↔ 2.
6.7.3 Rotational Contact. Specification of a

Constitutive Law for the Rotational Moment
MI Coupled with Contact Constraints for the
Variables ϕI
The moment M1 appearing in the current theory is energetically coupled
with the variable ϕ1 , and therefore, is responsible for the rotational motion
along the τ 1 axis (plane e1 , g1 ) representing then the forces acting in the
circumferential direction to the corresponding cross-section of the curvilinear
beam and projected onto g-axis, see Fig. 6.11. A specification of this moment
allows e.g. to describe a full sticking case specifying the rotational sticking
in addition to the tangential sticking to prevent the motion of the first curve
relative to the second curve. The rotational sticking can be defined as follows:
• if ϕ̇1 = 0 then the first curve is rotationally sticking along the second
curve, i.e. rotation along the τ 1 -axis is not possible.
• if ϕ̇1 = 0 then the first curve is rotationally sliding along the second curve,
i.e. can be rotated along the τ 1 -axis.
6.8 Curve-To-Curve Contact: Rate of Contact Forces in a Covariant Form 181
This definition is again symmetric with regard to the choice of the curves 1 ↔
2. One of the simplest way to define the constitutive relation M1 is by using
an analogy to the Tresca friction law, but now defined for the angular variable
ϕ1 . We can specify the following relation for rotational friction between curves
in rate form:
⎧
⎪
⎨ −ε1 ϕ̇1
M
if |M1 | < M cr rotational sticking along 2nd curve.
Ṁ1 = M ϕ̇1
⎪
⎩ −μ1 if |M1 | ≥ μM1 M
cr
rotational sliding along 2nd curve.
|ϕ̇1 |
(6.184)
Here,
εM
1 – is a rotational stiffness for the first curve;
μM1 – is a coefficient of rotational friction for the first curve;
M cr – is a critical given value for the rotational moment after which the curve
begins to rotate. The simplicity of this law is emphasized by the independence
of the critical moment M cr on any other parameters, e.g. a normal contact
force. Applying then the return-mapping scheme to eqn. (6.184) we can get
a trial rotational moment as
(trial) (trial,n+1) (n) (n+1) (n)
ΔM1 = −εM
1 Δϕ1 =⇒ M1 = M1 − εM
1 (ϕ1 − ϕ1 ),
(6.185)
similar to the scheme (6.182). The return-mapping step is constructed fully
similar to (6.183)
"
(trial,n+1) (trial,n+1)
M1 if |M1 | < M cr sticking;
M1= (sl,n+1) (trial,n+1) (trial,n+1)
M1 M
= μ1 sign(M1 ) if |M1 | ≥ M cr sliding.
(6.186)
Remark 6.13. Full sticking will require the enforcement of both tangential
sticking and rotational sticking.
6.8 Curve-To-Curve Contact: Rate of Contact Forces

in a Covariant Form
For the further linearization it is necessary to obtain the frame-indifferent
rate of the force vector R1 . As is known, the rate in the form of covariant
derivatives is a frame indifferent one. Following its formal definition the total
time derivative of a vector can be written in a covariant form
dR dRi ri
= = ∇j Ri ξ˙j ri , (6.187)
dt dt
where the components in the coordinate system r1 , r2 , r3 are defined as fol-
lows:
∂Ri
∇j Ri = + Rk Γjki
, (6.188)
∂ξ j
where again the Christoffel symbols representing the derivation of coordi-

nate vectors are necessary. All necessary derivations have been established in
Section 4.5.4 via the direct derivation of coordinate vectors. Thus, the main
step now is to express all constitutive relations presented in Section 6.7 in
the form of covariant derivatives with respect to a metric given by gij in
eqn. (4.184).
6.8.1 Covariant Form for Sticking

All sticking cases, discussed in Section 6.7 are characterized by the propor-
tionality of the force rate components Ṫ , Ṅ , Ṁ to the corresponding velocity
components ṡ, ṙ, ϕ̇. Summarizing, we can write in a vector form as
dR
= −D(v2 − v1t ) =⇒ ∇j Ri ξ˙j ri = −di ξ̇ i ri . (6.189)
dt
Here, the tensor expressed in the mixed metrics D = di ri ⊗ ri is given by the
diagonal matrix containing the stiffness parameters ε1 , εN , εM
1
⎡ ⎤
ε1 0 0
D = ⎣ 0 εN 0 ⎦ . (6.190)
0 0 εM1
Remark 6.14. Since the rate forms given in the definition of constitutive laws,
see eqns. (6.181) and (6.184) are coinciding with derivatives of the correspond-
ing trial forces in eqns. (6.182) and (6.185) for the return-mapping scheme in
eqns. (6.183) and (6.186) we can directly use the vector form in eqn. (6.189)
for the forthcoming linearization.
6.8.2 Covariant Form for Sliding

Similar to the sticking cases the rate for the sliding cases can be written in
a proportional form, however, the form suitable for the linearization should
contain the linearization of the real forces given in eqns. (6.183) and (6.186)
of the return-mapping scheme. Thus, we will linearize the weak form directly
for the sliding cases in Sect. 7.5.
6.9 Curve-To-Rigid Surface Contact – Transformation

of the Contact Forces
Taking into account the kinematics for Curve-To-Rigid Surface (CTRS) con-
tact, see Sect. 4.6, the contact traction vector RC defined first in the local
surface coordinate system at the point C, see Fig. 4.13, as
RC = NS n + T i ρi , (6.191)
6.9 Curve-To-Rigid Surface Contact – Transformation of the Contact Forces 183
Fig. 6.12 Coordinate systems used for the Curve-To-Rigid Surface (CTRS) Con-
tact: the Gaussian surface and the Serret-Frenet curve coordinate systems.
where contravariant components of the trial tangent vector T i can be defined

numerically in the surface coordinate system similar to STS algorithm in
eqn. (6.22) as
(n+1), i (n), i
Ttr = Treal − εT (ξ (n+1), i − ξ (n), i ). (6.192)
The return-mapping scheme in eqn. (6.26) should be involved in order to com-

(n), i
pute the real components of the tangent vector Treal at the corresponding
load (time) step. The curve is deformable, however, all parameters are initially
computed on the rigid master surface. Thus, taking into account equilibrium
conditions at the contact point, we can compute the contact traction RS act-
ing on the deformable curve and expanded by coordinate vectors τ , e, g, see
kinematics of the Curve-To-Curve interaction in Sect. 4.5 as well as Fig. 6.12
and 4.13 for CTRS kinematics:
RC + RS = 0
RC = NS n + T i ρi
RS = N e + T τ + Bg (6.193)
During the contact we have n = −e (see Kinematics in Sect. 4.6), therefore,

we can construct the following relations to compute the contact force in the
curve coordinate system:
• normal force
N = NS (6.194)
• pulling force
T = −T i (ρi · τ ) (6.195)
• dragging force
B = −T i (ρi · g) (6.196)
Thus, components of the contact force RS in the curve coordinate system N ,

T , B are just obtained by the projection of the contact force RC computed
in the surface coordinate system.
7
Linearization of the Weak Forms –
Tangent Matrices in a Covariant Form
The full Newton iterative method will be applied to solve the global equilib-
rium equations on the stage of the numerical solution. This requires the full
linearization of the corresponding weak forms representing the equilibrium
conditions on the contact boundaries described in Chapter 5 for all contact
cases. Linearization is obtained using the covariant derivatives in the corre-
sponding local surface coordinate system, where derivatives of contact trac-
tions are taken in covariant forms as described in Chapter 6 and derivatives of
corresponding convective coordinates are described in Chapter 4. Linearized
weak forms are the basis to create tangent matrices in the form independent
of any approximations of the object (surfaces, curves, beams etc.). All tan-
gent matrices are formulated then via the abstract approximation operator
A. The matrices are split into several parts and possessing a clear geometri-
cal structure for all studied geometrical contact cases – Surface-To-Surface,
Point-To-Curve, Curve-To-Curve including also various constitutive relations
for contact tractions.
7.1 Linearization of the Weak Form for the

Surface-To-Surface Contact
The weak form representing the virtual work of normal traction N and tan-
gential tractions Ti obtained in Section 5.1 eqn. (5.1), is split into a weak
form for normal traction δWcN and a weak form for tangential traction δWcT
# #
δWc = N δξ 3 ds + Ti δξ i ds (7.1)

s
! s
!
δWcN δWcT
The idea behind the consistent linearization for a Newton type solution pro-
cess within the geometrically exact theory is to exploit the full material time
derivative in the form of the covariant derivative in the spatial coordinate
186 7 Linearization of the Weak Forms – Tangent Matrices in a Covariant Form
system, together with the evolution equations for the contact normal and
tangential frictional forces. We introduce the linearization operator L acting
on the weak form as follows:
# # #
L[δWc ] = L ai δbi ds = L[ai δbi ]ds + ai δbi L[ds] (7.2)
s s s
where ai δbi is the scalar product describing the virtual work of corresponding
contact tractions.
Remark 7.1. The integral in eqn. (7.2) describing the global virtual work of
contact tractions is computed over the “slave” surface ds, which is defined by
a set of “slave” points after fulfillment of the Closest Point Projection (CPP)
procedure. Each parameter ai , bi in the contact integral is considered in the
spatial local coordinate system of the “master” surface, (i.e. as a function of
the convective coordinates ξ i ), therefore, linearization of the “slave” surface
element ds will not be included in the process. Thus ds is assumed to remain
constant within the linearization and L[ds] = 0. In other words, we are
studying how the projection of the slave point is moving on the slave surface
and this process is reflected by the linearization. Further it must be noted that
the use of different quadrature schemes for the computation of the contact
integral may lead to different contact elements.
For the linearization either a covariant ai or a contravariant component bi ,

two operators, based on the covariant derivatives are necessary. The operator
for the linearization of the contravariant component bi has the form

i
∂ ∂bi ∂b
L(b ) ≡
i
+ ξ̇ ∇j (b ) =
j i
+ + Γkj b ξ˙j
i k
(7.3)
∂t ∂t ∂ξ j
and the linearization operator for the covariant component ai has the form

∂ ∂ai ∂ai
L(ai ) ≡ + ξ˙j ∇j (ai ) = + − Γ k ˙j
ij i ξ .
a (7.4)
∂t ∂t ∂ξ j
It is obvious that the Christoffel symbols disappear in the final result after
the linearization of the scalar, i.e. the full time derivative of the scalar is the
covariant derivative of the scalar

i
i
∂ai ∂ai ∂b ∂b
i
L(ai b ) = + − Γij ai ξ̇ b +
k j i
+ + Γkj b ξ̇ j ai
i k
∂t ∂ξ j ∂t ∂ξ j
i
∂ai ∂ai ˙j i ∂b ∂bi ˙j dbi dai
= + jξ b + + j ξ ai = ai + bi (7.5)
∂t ∂ξ ∂t ∂ξ dt dt
Combining the rule obtained in eqn. (7.5) and Remark 7.1, the following
linearization rule of the weak form is achieved from eqn. (7.2)
7.1 Linearization of the Weak Form for the Surface-To-Surface Contact 187
#
dai dδbi
L[δWc ] = δbi + ai ds. (7.6)
s dt dt
The general rule in eqn. (7.6) gives then a rule for the linearization of the
tangential part δWcT in eqn. (7.1) as
# #
dTi i dδξ i
L[δWcT ] = L Ti δξ i ds = δξ + Ti ds. (7.7)
s s dt dt
7.1.1 Linearization of the Normal Contact Part δWcN

According to the linearization rule in eqn. (7.6) the linearized expression for
the normal contact integral δWcN in eqn. (7.1) is written as follows:
# # #
dδξ 3
L(δWcN ) = L(N δξ 3 )ds = Ṅ δξ 3 ds + N ds . (7.8)
s s s dt
Equation (6.8) for the rate of the normal traction Ṅ in the case of the penalty
regularization and equation (4.70) for the linearization of δξ 3 are used di-
rectly:
L(δWcN ) =
#
= N H(−ξ 3 )(δrs − δρ) · (n ⊗ n)(vs − v)ds (7.9a)
s
!
Lm
N
#

− N δρ,j · aij (n ⊗ ρi )(vs − v) + (δrs − δρ) · aij (ρj ⊗ n)v,i ds (7.9b)
s !
LrN
#
− N (δrs − δρ) · hij (ρi ⊗ ρj )(vs − v)ds . (7.9c)
s
!
LcN
The full contact tangent matrix is subdivided into the main part eqn. (7.9a),
the “rotational” part (7.9b) and the “curvature” part (7.9c). The last two
terms are highly depending on the value of the penetration ξ 3 , because of
the relation N = N ξ 3 . The “rotational” part contains derivatives of δρ and
v with respect to the convective coordinates ξ j and, therefore, represents
the rotation of a contact surface during the incremental solution procedure.
The “curvature” part contains components of the curvature tensor hij and,
therefore, represents the change of the curvature of the master surface.
7.1.2 Tangent Matrices for the Normal Contact Part

δWcN
Tangent matrices are obtained as a result of the approximation and the dis-
cretization of the corresponding weak forms. We introduce the approximation
operator A as follows:
• Let us denote that the master surface is described by the vector ρ(ξ 1 , ξ 2 )
and the slave surface is described by the vector rs (η 1 , η 2 ).
• Both master and slave surfaces can be either represented exactly (by
e.g. NURBS), or approximated with regards to a discrete vector x. This
vector can be a vector of nodal points in the case of classical FE, a vector
of knot points in the case of NURBS etc. This representation is linear and
is written via the operator A as follows:
ρ(ξ 1 , ξ 2 ) − rs (η 1 , η 2 ) = A(ξ 1 , ξ 2 , η 1 , η 2 )δx (7.10)

Remark 7.2. The exact structure of the operator A depends on a type of ap-
proximations involved in the finite element implementations of master and
slave surfaces. This also influences the dimension of a discrete vector x. Var-
ious discretizations will be considered in the next Chapter.
A relative virtual displacement and a velocity vector is given then as
δρ(ξ 1 , ξ 2 ) − δrs (η 1 , η 2 ) = Aδx for a virtual displacement vector;

v(ξ 1 , ξ 2 ) − vs (η 1 , η 2 ) = Av for a velocity vector. (7.11)
In addition only derivatives with respect to the master convective coordinates

ξ i should be computed for further representation of the tangent matrices:
∂A
δρ,i = δx = A,i δx. (7.12)
∂ξ i
Representation of the Tangent Matrix for Normal Interaction

All parts of the linearized weak form in eqns. (7.9a), (7.9b) and (7.9c) contain
either a vector of virtual relative displacements δρ − δrs or a relative velocity
v − vs . Thus we employ directly the approximation operator A in eqn. (7.11)
and for the derivatives eqn. (7.12).
The main, or constitutive, part is transformed using the approximation
operator A in eqn. (7.11) as follows:
#
Lm N = N (δrs − δρ) · (n ⊗ n)(vs − v)ds
s
#
= δx · N AT (n ⊗ n)Adsv = δx · [Km N ]v (7.13)
s
where [Km
N ] is the corresponding main part of the tangent matrix.
The rotational part is transformed using the operator A,i in eqn. (7.12).
Here the part LrN from eqn. (7.9b) is taken
#
$ %
LN = − N δρ,j · aij (n ⊗ ρi )(vs − v) + (δrs − δρ) · aij (ρj ⊗ n)v,i ds
r
s
#
$ %
= δx · N AT,j aij (n ⊗ ρi )A + AT aij (ρj ⊗ n)A,i dsv (7.14)
s
= δx · [KrN ]v
The curvature part from eqn. (7.9c) is transformed using the operator A as
follows:
#
LcN = − N (δrs − δρ) · hij (ρi ⊗ ρj )(vs − v)ds
# s
$ %
= N AT hij (ρi ⊗ ρj )A dsv (7.15)
s
= δx · [KcN ]v
with a contact force defined in eqn. (6.6) as N = H(−ξ 3 )N ξ 3

Thus, the full normal tangent matrix for normal interaction is set up as
KN = Km + KrN + KcN (7.16)

#N
= N AT (n ⊗ n)Ads (7.17)
s
⎡ ⎤
#
⎢ ⎥
+ N ⎣AT,j aij (n ⊗ ρi )A + AT aij (ρj ⊗ n)A,i ⎦ ds (7.18)
s ! !
K1 K2
#
$ %
+ N AT hij (ρi ⊗ ρj )A ds (7.19)
s
The matrix is obviously symmetric. The symmetry of the main part (7.17)
and the curvature part (7.19) are straightforwardly observed. The symmetry
of the rotational part (7.18) follows as a sum of transposed matrices K2 =
KT1 .
Remark 7.3. Non-frictional contact problems are solved with the help of the
tangent matrix defined in eqn. (7.16) only.
Case of the Non-Linear Approximation Operator A

For complex situations with an exact representation of boundaries via the
blended function method, see Düster [34] in general nonlinear approximations
with respect to a discrete vector x can be involved. Thus the approxima-

tion operator A is nonlinear with respect to variables (Degree of Freedoms)
for the finite element approximation (or knot variables for the NURBS, iso-
geometrical FE etc.)
ρ(ξ 1 , ξ 2 ) − rs (η 1 , η 2 ) = A(x, ξ i , η i ). (7.20)
In such a case, the approximation operator itself should be linearized with

respect to the global variable vector x in order to represent the globally
consistent tangent matrices. We introduce DA in the sense of the Frechet
derivative with respect to a global variable vector x. Considering the Tay-
lor expansion for the relative virtual displacement vector in eqn. (7.10) and
taking into account the approximation property we obtain A(0, ξ i , η i ) = 0
δρ(ξ 1 , ξ 2 ) − δrs (η 1 , η 2 ) = A(δx, ξ i , η i )

= A(0, ξ i , η i ) + DA(0, ξ i , η i )δx + ...
= DA(0, ξ i , η i )δx + ... (7.21)
Taking a partial derivative with respect to the convective variables ξ i the

operator DA,i is introduced as:
∂DA(0, ξ i , η i )
δρ,i = δx + ... = DA,i δx + ... (7.22)
∂ξ i
Thus, in the case of nonlinear approximations for contact boundaries all
parts of tangent matrices are computed following the pattern presented
in eqns. (7.13), (7.14) taking into account the linearized operator DA in
eqn. (7.21) and its derivative DA,i in eqn. (7.22). It leads formally to the
substitution of A with DA and respectively of A,i with DA,i .
7.1.3 Linearization of the Tangential Contact Part

δWcT – Case of Sticking
In the sticking case, the trial tangential traction terms Ti are identical with
the tangential traction defined by the evolution equation, therefore, the lin-
earized traction terms are obtained from the evolution equation in (6.19)
directly. Starting with eqn. (7.7) together with the evolution equation (6.19),
and equations (4.81a), (4.81b), (4.81c) for the linearization of the term
d
(δξ i ) we finally obtain
dt
#

dTi dδξ i
L(δWcT ) = δξ i + Ti (7.23)
s dt dt
# - .
= ˙j k ˙
(−T aij + Γij Tk )ξ − hi Tk ξ δξ i ds
k 3
s !
#
$
− Ti (δrs − δρ ail ajk ρk ⊗ ρl vj + δρ,j aik ajl ρk ⊗ ρl (vs − v)
s

+hij (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v)+
/
+ hin ξ̇ 3 δξ n − Γkj
i j
ξ̇ δξ k ds.
!
The underlined terms are vanishing. Using the tensor notation and the equa-
tion for convective velocities (4.36) for the main part aij ξ˙j δξ i , we obtain the
following form for the tangential tangent matrix in the case of sticking.
L(δWcT ) =
#
−εT (δrs − δρ)aij ρi ⊗ ρj (vs − v)ds (7.24a)
s
#

− Ti (δrs − δρ) ail ajk ρk ⊗ ρl vj + δρ,j aik ajl ρk ⊗ ρl (vs − v) ds
s
(7.24b)
#

+ Ti hij (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v)ds. (7.24c)
s
As we have a conservative problem for the sticking case it is obvious that the
symmetric form is correct.
Similar to the normal tangent matrix, the tangential tangent matrix can
be subdivided into a main (7.24a), a rotational (7.24b) and a pure curvature
part (7.24c).
Remark 7.4. The artificial non-symmetry of the tangent matrix for the stick
condition, based on the evolution equation (6.13) was mentioned by Laursen
and Simo in [109], [106] and [104]. As an appropriate alternative within a
solution scheme, a symmetrization based on a split technique with the Aug-
mented Lagrangian method was proposed by Laursen in [162]. Wriggers [190],
suggested to use the consistent linearization of the sticking conditions in the
form rs − ρ2 directly, which then leads to the correct symmetric matrix.
Here, it becomes obvious, that it is particularly important to use the evolution
equation in the form of the covariant derivatives (see eqn. (6.19)) together
with the linearization of the metric components aij as 3D metric compo-
nents gij (see eqn. 4.11). This allows to avoid the artificial non-symmetry
and obtain the correct symmetric tangent matrix for sticking.
Remark 7.5. The non-symmetry of the tangent matrix even for conservative
problems due to the usage of the directional derivative formula (Frechet
derivative) in Cartesian coordinate system and recovery of the symmetry
by the usage of the covariant derivative was also discussed by Simo in [160].
Representation of the Tangent Matrix for the Tangential Sticking

Interaction
Applying the approximation operator A in eqn. (7.11) and the operator A,i
for derivatives eqn. (7.12) as we have shown in Sect. 7.1.2 to eqns. (7.24a),
(7.24b), (7.24c) the following tangent matrix for tangential sticking is
obtained:
st, m
Kst
T = KT + Kst, r
+ KcT (7.25)
# T
− εT AT aij (ρi ⊗ ρj )Ads (7.26)

# s

− Ti AT ail ajk ρk ⊗ ρl A,j + AT,j aik ajl (ρk ⊗ ρl )A ds (7.27)
#s

+ Ti hij AT ρj ⊗ n + n ⊗ ρj Ads. (7.28)
s
7.1.4 Linearization of the Tangential Contact Part

δWcT – Case of Sliding
The expressions for the linearized variation of the convective velocity eqn.
(4.81a), (4.81b) and (4.81c) are also used in this case. In addition, the tan-
gential force in the case of sliding, see eqn. (6.26) of the return-mapping
algorithm, has to be linearized.
- .
(n+1) tr (n+1)
dTi d (n+1) (T )i
= μN = (7.29)
dt dt Ttr(n+1)
- .
(n+1) tr (n+1)
dN (n+1) (T tr )i d (T )
=μ + μN (n+1) i
dt Ttr(n+1) dt Ttr (n+1)
For the derivative of the unit vector on the tangent plane

(n+1)
(T trial )i
e= aij ρj
Ttrial
(n+1)
we will use the following formula, see e.g. Simo and Hughes [161]
de 1
= [I − e ⊗ e] (7.30)
dT T
and the chain rule

de de dT 1 dT
= = [I − e ⊗ e] . (7.31)
dt dT dt T dt
dT
Here the full time derivative of the tangential traction is given by the evo-
dt
lution equation (6.19). The tensor operations are considered on the tangent
plane:
dT
[I − e ⊗ e] =
dt

Tk Tl aik ajl
= aij ρi ⊗ ρj − ρ ⊗ ρ (−T amn + Γmn r
Tr )ξ˙n − hrm Tr ξ˙3 ρm =
T 2 i j

= −T ξ˙i + aik Tl Γkj ξ − hkj aij Tk ξ˙3 ρi +
l ˙j
(7.32a)
Tk Tl aik
˙3 ρ .
+ T ξ̇ l − ajl Tm Γjn
m n
ξ̇ + ajl Tm hm
j ξ (7.32b)
T 2 i
The time derivative of the normal force N (n+1) gives:
dN (n+1) d
= (N |ξ 3 |) = −N ξ˙3 , (7.33)
dt dt
where the minus sign is a result from the conditions that the contact integral
is computed only if ξ 3 < 0. Summarizing, we get:
L(δWcT ) =
# -
N μξ˙3 Ti δξ i T μ|N |aij ξ˙i δξ j
− − (7.34a)
s T T
μ|N |Tk Γijk ξ̇ j δξ i μ|N |Ti hij ξ˙3 δξ j

+ − (7.34b)
T T
μ|N |Ts Tl δξ s ˙l m ˙n m ˙3

+ T ξ − a jl
T m Γ ξ + a jl
T m h ξ (7.34c)
T3 jn j
μ|N |Ti $
− (δrs − δρ) ail ajk ρk ⊗ ρl vj + δρ,j aik ajl ρk ⊗ ρl (vs − v)
T
(7.34d)

+hij (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v)+ (7.34e)
0
+hin ξ̇ 3 δξ n − Γkj
i j
ξ̇ δξ k ds (7.34f)
The sum of the parts (7.34b) and (7.34f) is zero. After some tensor algebra
the other parts can be grouped into the following form:
L(δWcT ) =
#

N μTi aij
− (δrs − δρ) ρj ⊗ n(vs − v) ds (7.35a)
s T
#

T μ|N |aij
− (δrs − δρ) ρi ⊗ ρj (vs − v) ds (7.35b)
s T
#

T μ|N |Ti Tj aik ajl
+ (δrs − δρ) ρ ⊗ ρ (vs − v) ds (7.35c)
s T3 k l
#
μ|N |Ti
− (δrs − δρ) ail ajk ρk ⊗ ρl vj + δρ,j aik ajl ρk ⊗ ρl (vs − v) ds
s T
(7.35d)
#

μ|N |Ti ij
+ h (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v) ds (7.35e)
s T
#
μ|N |Ts Tl δξ s jl m ˙n

+ −a T m Γ ξ + a jl
T m h m 3
ξ̇ ds. (7.35f)
s T3 jn j
The matrix consists then of a constitutive non-symmetric part (7.35a), a

constitutive symmetric part (7.35b) and (7.35c), a symmetric rotational part
(7.35d), a symmetric curvature part (7.35e) and a non-symmetric part cur-
vature part (7.35f) which is preserved for curved surfaces. All geometrical
parameters are computed for the master surface.
Remark 7.6. One can find from comparison with Peric and Owen [140], that
they have considered a tangent matrix which is represented by the the main
parts of the full tangent matrix.
7.2 Linearization of the Weak Form for the 2D Case

Here we derive the tangent matrices for 2D plane contact case, see kinematics
in Sect. 4.2 and variational formulation in Sect. 5.2, in the case of a natural
parametrization. In order to avoid the complexity for the sticking – sliding
cases for the tangential part, the derivation is given as a reduction of the 3D
case.
7.2 Linearization of the Weak Form for the 2D Case 195
7.2.1 Tangent Matrix for the Normal Part

The normal part of the weak form is defined by the following integral:
#
δWcN = N δζdl. (7.36)
l
The integral is computed over the slave surface l, while all functions are
defined on the master surface. Thus, a linearization of dl is not necessary
within the process:
D(δWcN ) =
#

dN dδζ
δζ + N dl =
l dt dt
then the application of eqn. (6.168) and eqn. (4.132) leads to
#
= H(−ζ)N (δrs − δρ) · (ν ⊗ ν)(vs − v)dl− (7.37a)
l
#

∂τ
− H(−ζ)N ζ δτ · (ν ⊗ τ )(vs − v) + (δrs − δρ) · (τ ⊗ ν) dl−
l ∂t
(7.37b)
#
− H(−ζ)N ζκ(δrs − δρ) · (τ ⊗ τ )(vs − v)dl. (7.37c)
l
Remark 7.7. The contact matrix obtained via eqn. (7.37) is computed only
for the case ζ < 0. This fact is emphasized by the presence of the Heaviside
function H(−ζ).
The form in natural coordinates allows a simple geometrical interpretation

of each part in eqn. (7.37) and even allows to determine situations where
some of them are zero. The first part, eqn. (7.37a) is called the main part
and defines the constitutive relation for normal contact conditions. The sec-
ond part, eqn. (7.37b) is called rotational part and defines the geometrical
stiffness due to the rotation of the tangent vector of the master curve. It
disappears when a master segment is moving in parallel, because only in this
case the derivative of a unit vector τ becomes zero, see Fig. 4.4. The third
part eqn. (7.37c) is called curvature part. This part disappears when the cur-
vature κ of a master segment is zero, i.e. in the case of linear approximations
of the master segment.
7.2.2 Tangent Matrix for the Tangential Part

The part of the contact integral which includes the effect of the tangential
interaction is given as:
#
δWcT = T δξdl (7.38)
l
The linearized equation has to be subdivided into a part for sticking and
another part for sliding, which differ concerning the return-mapping scheme.
Sticking Case
In this case, the tangential force T has to be computed from the solution of
the evolution equation eqn. (6.171), e.g. via the backward Euler scheme. The
simplest case with linear approximations will be presented in the following
section concerning the finite element implementation.
Sticking is fulfilled according to Coulomb’s friction law, i.e. the inequality
T ≤ μ|N | has to be valid in each load step. The linearized contact integral
has then the following form:
#

dT dδξ
Dv (δWcT ) = δξ + T dl =
l dt dt
#
εT
− (δrs − δρ) · ρξ ⊗ ρξ (vs − v)dl (7.39a)
l (ρ ξ · ρξ )
#
T $ %
− (δrs − δρ) · ρξ ⊗ ρξ vξ + δρξ · ρξ ⊗ ρξ (vs − v) dl (7.39b)
l (ρ ξ · ρ ξ )2
#
T h11
+ (δrs − δρ) · ρξ ⊗ ν + ν ⊗ ρξ (vs − v)dl. (7.39c)
l (ρ ξ · ρ ξ )2
Sliding Case
If sliding is detected, i.e. if T > μ|N |, then the sliding force is computed
according to Coulomb’s friction law. We also keep a covariant form:
Ttr
T sl = μ|N | = μ|N | ρξ sgn(Ttr ), (7.40)
Ttr
with
√
ρξ = (ρξ · ρξ )1/2 = a11 .
7.3 Linearization – Case of Coupled Anisotropic Adhesion-Friction 197
The linearized contact integral gets the following form:
Dv (δWcT ) =
#
N μ sgn(Ttr )
− (δrs − δρ) · (ρξ ⊗ ν)(vs − v)dl (7.41a)
l (ρξ · ρξ )1/2
#
μ|N | sgn(Ttr )
− (δrs − δρ) · ρξ ⊗ ρξ vξ + δρξ · ρξ ⊗ ρξ (vs − v) dl
l (ρξ · ρξ )3/2
(7.41b)
#
μh11 |N | sgn(Ttr )
+ (δrs − δρ) · 2ρξ ⊗ ν + ν ⊗ ρξ (vs − v)dl. (7.41c)
l (ρξ · ρξ )3/2
The non-symmetric part eqn. (7.41c) now is resulting from the last term of
the evolution equation (6.171).
Remark 7.8. The derivations for the two dimensional case allow to describe
all parts of the tangent matrix and to find all cases, when some of them
become zero. The main parts, eqns. (7.39a) and (7.41a), the so-called con-
stitutive parts, contain a penalty parameter and describe the stiffness of the
contact interaction due to the chosen interface model. This is based on an
allowable elastic deformation due to the regularization in the case of sticking
resp. due to the applied sliding force μ|N | in the tangential direction in the
case of sliding. The rotational parts eqns. (7.39b) and (7.41b), contain a met-
ric coefficient a11 = (ρξ ·ρξ ) and a vector ρξ . A metric coefficient is a measure
of the tensile deformation of the contact master line, e.g. a component of the
Cauchy-Green tensor for the contact line can be written as ε11 = (a11 − 1)/2.
The vector ρξ is a measure of the rotation of the master segment, which
becomes obvious if the length is chosen as a coordinate s = ξ. In this case
we find δρξ = δτ and vξ = ∂∂tτ . The vector τ is a unit vector, therefore the
vectors δτ and ∂∂tτ are describing the rotation of the unit vector τ (see Fig.
4.4). It is identical to zero only in the case of parallel motions. Thus, the rota-
tional part is obviously negligible in the case of small deformations and small
rotations of the master line. The curvature parts eqns. (7.39c) and (7.41c)
describe the stiffness of the contact interaction due to the curvature of the
contact surface. If the surface has zero curvature or is approximated by linear
elements, then the curvature part becomes zero.
7.3 Linearization of the Weak Form for the

Surface-To-Surface Contact – Case of Coupled
Anisotropic Adhesion-Friction Interfaces
The anisotropic adhesion-friction model, characterized by the new relations
for the tangent traction including the adhesion B and friction F tensors, is
developed in Sections 6.2 and 6.3. Due to decoupling of constitutive relations

into the normal and tangential parts, the linearized normal part in the case
of the penalty regularization remains the same as described for non-frictional
problems in Section 7.1.1. Thus, we present the linearization of the tangential
part δWcT for the sticking and sliding cases.
7.3.1 Linearization of the Tangential Part δWcT

The tangential part of the contact integral δWcT is considered taking into ac-
count the anisotropic evolution equations and the return mapping algorithm.
The cases of sticking and sliding have to be treated separately.
For the linearization of the tangent traction we recall the evolution equa-
tions for the elastic reversible part, obtained in Sect. 6.2.2 in the covariant
vector form
dT
= B(vs − v), (7.42)
dt
or in the component form in the surface metrics as
∂Ti
+ ∇j Ti ξ˙j = bij ξ˙j , j = 1, 2, (7.43)
∂t
where the adhesion tensor B = bij ρi ⊗ ρj is defined in the surface metrics.
The Sticking Case

Sticking is identified when the trial contact tractions Titr computed at load
step (n + 1) satisfy the conditions imposed by Coulomb’s law :

Φ := f ij Tjtr Titr − N < 0. (7.44)
In this case, the real tractions are identical to the trial ones Ti = Titr ,
therefore, the linearized traction terms are obtained from the evolution equa-
tion in (7.43) directly. For the convective velocities δξ i , the linearized equa-
tions (4.81a, 4.81b, 4.81c) can be used.
#

dTi i dδξ i
L(δWc ) =
T
δξ + Ti ds = (7.45)
s dt dt
#
= (δrs − δρ) · B(vs − v)ds− (7.46a)
s
#

− Ti (δrs − δρ) · ail ajk ρk ⊗ ρl vj + δρ,j · aik ajl ρk ⊗ ρl (vs − v) ds,
s
(7.46b)
#

s
or component-wise
#
= (δrs − δρ) · bij ρi ⊗ ρj (vs − v)ds− (7.47a)
s
#

− Ti (δrs − δρ) · ail ajk ρk ⊗ ρl vj + δρ,j · aik ajl ρk ⊗ ρl (vs − v) ds.
s
(7.47b)
#

s
The matrices included in this integral obviously preserve symmetry.
The Sliding Case

The sliding case is identified if the inequality in eqn. (7.44) is not satisfied.
Then the sliding force Tsl is derived from the principle of the maximum
dissipation (see Section 6.3) as
BFTtr T̂|N |
Tsl = − √ |N | = − , (7.48)
BFT · FBFT
tr tr Ψ
or component-wise as
bij f jk Tktr
Tisl = − (7.49)
bij blm f jk f il f mn Tktr Tntr
In eqn. (7.48) an auxiliary force T̂ – allowing some reductions in the following

expressions – is introduced as
T̂ = BFTtr , (7.50)
or component-wise as
T̂i = bij f jk Tktr = bij fln ajl akn Tktr = bji fjk Tktr . (7.51)
and a function Ψ as
√
Ψ := BFTtr · FBFTtr = T̂tr · FT̂tr . (7.52)
The derivative of the sliding force Tsl is computed according to the chain
rule as:
- . - .
dTsl d T̂|N | d|N | T̂ |N | dT̂ |N |T̂ ∂Ψ dT̂
= − =− + − ·
dt dt Ψ dt Ψ Ψ dt Ψ 2 ∂ T̂ dt
- 1 /.
d|N | T̂ |N | dT̂ FT̂ dT̂
=− + − T̂ 2 · (7.53)
dt Ψ Ψ dt Ψ dt
The evolution equation (6.8) for the normal traction N is used for the lin-
earization of |N |. The auxiliary force T̂ defined in eqn. (7.50) is linearized
according to the chain rule
dT̂ ∂ T̂ dTtr
= = BFB(vs − v), (7.54)
dt ∂Ttr dt
where for the linearization of the trial traction Ttr the evolution equation
(7.42) is used directly. It is an interesting fact that due to the tensor rep-
resentation the linearization is valid even in the case of arbitrarily varying
surface tensors B and F. Eqn. (7.53) is then transformed into
1 /
dTsl T̂ BFB T̂ ⊗ (BFB) T
FT̂
= N ξ̇ 3 − |N | − (vs − v). (7.55)
dt Ψ Ψ Ψ3
Eqn. (7.55) is used for the further linearization. After some transformations
the following expression in components is obtained for the tangential part of
the contact integral
#

dTi i dδξ i
Dv (δWcT ) = δξ + Ti ds = (7.56)
s dt dt
in tensor form denoted by (... t)

# - .
N T̂ ⊗ n
= (δrs − δρ) · (vs − v) ds (7.57a)
s Ψ
#

|N | BFB
− (δrs − δρ) · (vs − v) ds (7.57b)
s Ψ
# - .
|N | T̂ ⊗ (BFB)T FT̂
+ (δrs − δρ) · (vs − v) ds (7.57c)
s Ψ3
#
$ %
− Tisl (δrs − δρ) · ail ajk ρk ⊗ ρl vj + δρ,j · aik ajl ρk ⊗ ρl (vs − v) ds.,
s
(7.57d)
or component-wise denoted by (... c)
# - .
N T̂i aij
= (δrs − δρ) · ρj ⊗ n(vs − v) ds (7.58a)
s Ψ
# - .
|N | bki fji bjl
− (δrs − δρ) · ρk ⊗ ρl (vs − v) ds (7.58b)
s Ψ
# - j ml q
.
|N | bij fm b fi T̂q T̂n ank
+ (δrs − δρ) · ρk ⊗ ρl (vs − v) ds (7.58c)
s Ψ3
#
$ %
− Tisl (δrs − δρ) · ail ajk ρk ⊗ ρl vj + δρ,j · aik ajl ρk ⊗ ρl (vs − v) ds.
s
(7.58d)
Here, the components of the sliding force Tisl are computed via eqn. (7.49),
and the components of the auxiliary vector T̂i via eqn. (7.51). It becomes obvi-
ous, that anisotropy leads to the loss of symmetry of part (7.57c) and (7.58c),
by looking at the non-diagonal components of T̂ ⊗ AT̂, where
A = (BFB)T F (7.59)
then we have

c12 = T̂1 A12 T̂1 + T̂1 A22 T̂2
C = T̂ ⊗ AT̂ =⇒ . (7.60)
c21 = T̂2 A11 T̂1 + T̂2 A21 T̂2
Symmetry in eqn. (7.60) is recovered only in the isotropic case, i.e. if Aij = δji .
δji T
In the case of isotropy we have fji = 2 and bij = −T δji leading to Ψ = 3
μ μ
T tr
and T̂ = − 2 T .
μ
Summarizing we obtain, that the tangent matrix in eqn. (7.56) consists
of the standard constitutive non-symmetric part (7.57a), a constitutive sym-
metric part (7.57b), and a constitutive non-symmetric part (7.57c), which is
symmetric only in the isotropic case, and, finally, the standard symmetric
rotational part (7.57d).
Remark 7.9. For a consistent linearization of the sliding part we adopt the
assumption that all terms describing the curvature properties of the mas-
ter surface, i.e. including the second derivatives with respect to convective
coordinates, can be neglected based on the numerical investigations in [87]
and [88]. This allows to reduce the size of various expressions considerably.
However, one can easily recover all terms based on the full linearization of
δξ i in Section 4.1.8.
Remark 7.10. The component-wise formulas in eqns. (7.58a, 7.58b, 7.58c,

7.58d) show the possible representations of the corresponding tensor formu-
las in eqns. (7.57a, 7.57b, 7.57c, 7.57d). It obvious that variations within the
sequence of covariant and contravariant components are possible.
7.4 Linearization of the Weak Form for

Point-To-Curve Interaction
We present only the linearization of the weak form in the most simplest
form as discussed in Section 5.3. The linearization is provided in the form of
covariant derivatives in the corresponding curvilinear coordinate system at
the tangent line with r = 0.
Linearized part for the normal interaction:
L(δWN ) = L(N δζ) = L(N ζδζ) = (δrs − δρ)N e ⊗ e(vs − v). (7.61)
Linearized part for the tangential interaction:

– sticking
L(δWTstick ) = L(Ttr δs) = (δrs − δρ)T τ ⊗ τ (vs − v). (7.62)
– sliding
(n+1)
L(δWTslide ) = L(T δs) = (δrs − δρ)sign(Ttr )N τ ⊗ e(vs − v). (7.63)
Remark 7.11. The presented formulation is rather limited either to linear

geometry with a small penetration r, or to small load (time) steps. The
formulation allowing arbitrary geometry will be discussed in the Section 7.5
for arbitrary Curve-To-Curve contact interaction.

Curve-To-Curve Contact
Numerical solution schemes for contact between bodies are realized via a
Newton type iterative solver, thus a knowledge about derivatives of the func-
tional δW is required. This procedure known as linearization will be taken
in a covariant form, i.e. provided in a coordinate system exploiting covariant
derivatives. We start with the full form in eqn. (5.29) without any simplifi-
cation on small distance r and curvature. For the linearization the total time
derivative in a form of the covariant derivative operator in eqn. (4.202) is
used:
Ls,r,ϕ [δW ] = Ls,r,ϕ [Sym{R1 · (δρ1 − δρ2 )}] =
= Sym{Ls,r,ϕ [R1 ] · (δρ1 − δρ2 )} + Sym{R1 · Ls,r,ϕ [(δρ1 − δρ2 )]}. (7.64)
7.5 Linearization of the Weak Form for Curve-To-Curve Contact 203
The second term contains derivatives of the variations and remains in the
same form for the sticking and sliding cases for the real force vector R1 .
This part represents a tangent matrix due to the geometrical nonlinearity of
contact interaction, while the first term in eqn. (7.64) represents a constitutive
relation for the contact force.
7.5.1 First Part, Representing Geometrical

Nonlinearity
The derivatives in eqn. (4.204a) and the full expression for the force in
eqn. (5.27) are used to obtain the part representing the geometrical non-
linearity.
Sym{R1 · Ls,r,ϕ [(δρ1 − δρ2 )]} = (7.65a)

(1 − rk1 cos ϕ1 ) τ 1 + κ1 r g1
Sym T1 √ + N e1 + Rϕ1 g1 · L s,r,ϕ [(δρ1 − δρ2 )]
g11
= −Sym{ T1 [rk1 κ1 sin ϕṡ1 δs1 − k1 cos ϕ1 (ṡ1 δr + ṙδs1 ) (7.65b)
(1 − rk1 cos ϕ1 )
+ rk1 sin ϕ1 (ṡ1 δϕ1 + ϕ̇1 δs1 )]
(1 − rk1 cos ϕ1 )2 + (rκ1 )2
+N [k1 cos ϕ1 (1 − rk1 cos ϕ1 )ṡ1 δs1 − r(κ1 ṡ1 + ϕ̇1 )(κ1 δs1 + δϕ1 )] (7.65c)
- .
rκ1
+ Rϕ1 +T [−k1 sin ϕ1 (1 − rk1 cos ϕ1 )ṡ1 δs1
(1 − rk1 cos ϕ1 )2 + (rκ1 )2
(7.65d)
+ (ṙδϕ1 + ϕ̇1 δr) + κ1 (ṡ1 δr + ṙδs1 )] }.

The long expression can be tremendously simplified if we consider the lin-
earized term at r = 0:
Sym{R1 · Ls,r,ϕ [(δρ1 − δρ2 )]} = (7.66a)
= −Sym{− T1 k1 cos ϕ1 (ṡ1 δr + ṙδs1 ) (7.66b)
+N [k1 cos ϕ1 ṡ1 δs1 ] (7.66c)

M1
+ [−k1 sin ϕ1 ṡ1 δs1 + (ṙδϕ1 + ϕ̇1 δr) + κ1 (ṡ1 δr + ṙδs1 )] }. (7.66d)
r
For finite element implementations it is rather convenient to give a tensor
form directly leading to the corresponding tangent matrices. In order to
derive this form all pairs containing derivatives and variations of coordi-
nates in eqns. (7.66b), (7.66c) and (7.66d) are transformed using eqns. (4.193)
and (4.197). The following example for ṙδs1 is illustrating the simple trans-
formation procedure for two scalar products into a tensor product:
ṙδs1 = (v2 − v1t ) · e1 (δρ2 − δρ1 ) · τ 1 = (δρ2 − δρ1 ) · τ 1 ⊗ e1 (v2 − v1t ). (7.67)
Transformation of the linearized equation (7.66a) according to this pattern

leads to
Sym{R1 · Ls,r,ϕ [(δρ1 − δρ2 )]} = (7.68a)
= −Sym{(δρ2 − δρ1 ) · [ (− T1 k1 cos ϕ1 (τ 1 ⊗ e1 + e1 ⊗ τ 1 )+ (7.68b)
+N k1 cos ϕ1 (τ 1 ⊗ τ 1 )+ (7.68c)

M1 e1 ⊗ g1 + g1 ⊗ e1
+ −k1 sin ϕ1 τ 1 ⊗ τ 1 + ] (v2 − v1t ) }. (7.68d)
r r
The term with κ1 is disappearing because of the following transformation
(ṙδϕ1 + ϕ̇1 δr) + κ1 (ṡ1 δr + ṙδs1 ) −→

e1 ⊗ g1 + g1 ⊗ e1
− κ1 (τ 1 ⊗ e1 + e1 ⊗ τ 1 ) + κ1 (τ 1 ⊗ e1 + e1 ⊗ τ 1 ) =
r
e1 ⊗ g1 + g1 ⊗ e1
= .
r
7.5.2 Constitutive Part for Sticking

Due to the Remark 6.14 in Section 6.8.1 we can use the rate equation given
d
in eqn. (6.189) keeping in mind the equivalence of operators ∼ Ls,r,ϕ . We
dt
are starting with a tensor form as
Sym{Ls,r,ϕ [R1 ] · (δρ1 − δρ2 )} = −D(v2 − v1t ) · (δρ1 − δρ2 ) = (7.69)
= Sym{di ξ˙i ri · δξ j rj } = Sym{di gij ξ˙i δξ j }

and then obtaining

= Sym{ε1 (1 − rk1 cos ϕ1 )2 + (rκ1 )2 ṡ1 δs1 + εr ṙδr+ (7.70a)

1 r ϕ̇1 δϕ1 + r κ1 ε1 ṡ1 δϕ1 + ε1 ϕ̇1 δs1 }.
+εM 2 2 M
(7.70b)
Remark 7.12. Obviously in a case of not equal stiffnesses for the spatial curve
ε1 = εM
1 the symmetry is destroyed. Due to the coupled kinematics for the
7.5 Linearization of the Weak Form for Curve-To-Curve Contact 205
spatial curve, see the velocity vector in eqn. (4.191), we are restricted to take
ε1 = εM1 , otherwise the sum of force rates along r1 and r2 will not be parallel
to the relative velocity vector along the curve. For the latter we can take then
εM
1 = ε1 . (7.71)
Taking into account this remark after grouping eqns. (7.70a) and (7.70b) the
following form is obtained:
Sym{Ls,r,ϕ [R1 ] · (δρ1 − δρ2 )} =

2 3
Sym ε1 (1 − rk1 cos ϕ1 )2 ṡ1 δs1 + εr ṙδr + ε1 r2 (ϕ̇1 + κ1 ṡ1 )(δϕ1 + κ1 δs1 ) .
(7.72)
Now for the tensor form we will use the non-reduced equations for the con-
vective velocities in eqn. (4.194) and their variational analog. The final result
has a simple form as
Sym{Ls,r,ϕ [R1 ] · (δρ1 − δρ2 )} =

2 3
= Sym (δρ2 − δρ1 ) · (ε1 τ 1 ⊗ τ 1 + εr e1 ⊗ e1 + ε1 g1 ⊗ g1 ) (v2 − v1t ) ,
(7.73)
leading to a constitutive part of the tangent matrix in the sticking case.
7.5.3 Constitutive Part for Tangential Sliding

The tangential sliding force obtained in the computation via the return-
mapping scheme in eqn. (6.183) can be written then in vector form as:
r1
R = −μ|N | √ . (7.74)
g11
Inserting this into the first term of the linearized weak form in eqn. (7.64)
we obtain:
r1
Sym Ls,r,ϕ −μ|N | √ · (δρ1 − δρ2 ) = (7.75)
g11

μ|N |
= Sym Ls,r,ϕ √ r1 · (δρ2 − δρ1 ) + (7.76a)
g11

μ|N |
+Sym √ Ls,r,ϕ [r1 ] · (δρ2 − δρ1 ) . (7.76b)
g11
The first term (eqn. (7.76a)) is transformed as

−εr ṙ (−k1 cos ϕ1 (1 − rk1 cos ϕ1 )ṙ

(1st) = Sym μ1 √ − |N | √ +
g11 ( g11 )3

rk1 sin ϕ1 (1 − rk1 cos ϕ1 )ϕ̇1 − κ12 rṙ
+ |N | √ ·
( g11 )3
[(1 − rk1 cos ϕ1 )2 δs1 + r2 κ1 (κ1 δs1 + δϕ1 )] } (7.77a)

and after taking at r = 0
= Sym {μ1 (−εr + |N |k1 cos ϕ1 )ṙδs1 } . (7.77b)
One can prove that the second term (eqn. (7.76b)) taking its value at r = 0
is transformed then as
(2nd) = Sym {−μ1 |N |k1 cos ϕ1 ṙδs1 } . (7.78)
Summing eqns. (7.77b) and (7.78) we obtain the simple value for the linearized
constitutive part in the case of tangential sliding

r1
Sym Ls,r,ϕ −μ|N | √ · (δρ1 − δρ2 ) = −Sym{μ1εr ṙδs1 } = (7.79a)
g11
or in tensor form
= −Sym{(δρ2 − δρ1 ) · εr μ1 τ 1 ⊗ e1 (v2 − v1t )} (7.79b)
7.5.4 Linearized Part for Rotational Sliding

The rotational sliding force representing the moment obtained in the compu-
tation via the return-mapping scheme in eqn. (6.186) can be written then in
vector form as:
M1 1
R = Rϕ1 g1 = g1 = −μM
1 sign(ϕ̇1 ) g1 . (7.80)
r r
Inserting this into the linearized weak form we obtain:
Ls,r,ϕ [δWM ] =

1
= Sym Ls,r,ϕ −μM 1 sign( ϕ̇1 ) g 1 · (δρ1 − δρ2 ) (7.81a)
r

1
−μM
1 sign(ϕ̇1 ) g 1 · L s,r,ϕ [(δρ 1 − δρ 2 )] (7.81b)
r
7.6 Linearization of the Weak Form for Curve-To-Rigid Surface Contact 207
The first term is transformed as

μM
1 sign(ϕ̇1 )
(1st) = Sym{ [k1 sin ϕ1 (1 − rk1 cos ϕ1 )ṡ1 δs1
r
−(κ1 ṡ1 + ϕ̇1 )δr − ṙ(κ1 δs1 + δϕ1 )]}. (7.82)
The second term is representing the geometrically nonlinear part and can be
taken directly from eqn. (7.65d) for the sliding value of M1
μM
1 sign(ϕ̇1 )
(2nd) = Sym{ [−k1 sin ϕ1 (1 − rk1 cos ϕ1 ) ṡ1 δs1
r
+(ṙδϕ1 + ϕ̇1 δr) + κ1 (ṡ1 δr + ṙδs1 )] }. (7.83)
The sum of the two terms is zero. Thus, the tangent matrix for a rotational
sliding due to the Tresca model is zero.

Curve-To-Rigid Surface Contact
The weak form for the Curve-To-Rigid Surface given in Sect. 5.5 is linearized
in the curve Serret-Frenet coordinate system according to the results obtained
for the Curve-To-Curve algorithm in Sect. 7.5.
For the sticking case identified, however, on the surface, we are taking
the corresponding tangent matrix for the curve similar to eqn. (7.73)
{Ls,r,ϕ [R] · (δρrs − δρ)} =

#
= { (δρs − δρ) · (ετ τ ⊗ τ + εr e ⊗ e + εg g ⊗ g) (vs − v)} dl. (7.84)
l
The part representing geometrical nonlinearity is obtained fully following the

CTC algorithm in Sect. 7.5.1, eqn. (7.68a), but now in the form of an integral
along the curve: #
{R · Ls,r,ϕ [(δrs − δρ)]} = (7.85a)
l
= {(δrs − δρ) · [ ( T k cos ϕ (τ ⊗ e + e ⊗ τ ) + N k cos ϕ (τ ⊗ τ )+ (7.85b)
+B (k sin ϕ τ ⊗ τ ) ] (vs − v) }. (7.85c)

The last term in eqn. (7.68d) is omitted assuming then during the contact
the rolling of a rope is infinitesimally small δϕ = 0.
The parts representing geometrical nonlinearity in eqn. (7.85a) are used for
both sticking and sliding cases. The constitutive parts in the case of sliding
are depending on the model of friction and can be taken similar to CTC
contact in Sect. 7.5.3 and 7.5.4.
8
Surface-To-Surface Contact – Various
Aspects for Implementations within
the Finite Element Method
The current Chapter is devoted to the finite implementation of various al-

gorithms for Surface-To-Surface contact pair. Newton iterative scheme is the
main solution method for the most algorithms. Thus, the result of lineariza-
tion developed in Sect. 7.1 is directly employed to construct the tangent
matrices. Diversity of contact approaches leading to the corresponding con-
tact elements are depending on the type of approximation involved into the
discretization within the finite element method. First, the standard Node-
To-Surface (NTS) approach for non-frictional and frictional problems are
considered. Implementation of the Mortar method within the Segment-To-
Segment (STS) type of contact element together with integration by subdivi-
sion is shown to be effective to satisfy the Contact Patch Test . The simplest
smoothing technique for contact surfaces is shown as the implementation of
the Segment-To-Segment type of contact element – this technique is the basis
for the isogeometric implementation. A special Segment-To-Analytical Sur-
face (STAS) contact approach is illustrated for the contact with rigid surfaces
described analytically. The Large Penetration algorithm is presented as addi-
tional technique to accelerate the global solution. It is shown to be effective
for some problems where the large penetration and as a result large load
steps are applicable. Finally, two versions of implementation of the Nitsche
method are shown.
8.1 Finite Element Discretization for Various Contact

Approaches – NTS, STS and STAS Contact
Elements
The discretization techniques are further considered in the current section.
First, the well known “Node-To-Segment”, also known as “Node-To-Surface”,
(NTS) contact element is constructed using the covariant approach. Then the
discretization for the “Segment-To-Segment” (STS) approach (also called as
Surface-To-Surface in many engineering finite element codes) for bi-linear and
210 8 Surface-To-Surface Contact – Various Aspects for Implementations
bi-quadratic STS elements, combined with various integration techniques (see

Sect. 8.6) is considered. A smoothing technique based on a NURB spline ap-
proximation, see [38], is used to obtain C 1 continuous contact elements within
the STS approach. Further a general discretization strategy is presented for
the “Segment-To-Analytical-Surface” (STAS) approach.
8.1.1 Node-To-Segment (NTS) Contact Approach

The master “surface” is arbitrarily approximated and a “slave” node is taken
directly from the finite element mesh in the case of the “Node-To-Segment”
approach. A pair approximation for the master segment and a separate node
leads to the Node-To-Segment (NTS) contact element. If the approximation
of the ’master’ surface is defined by n nodes and the “slave” node is the
(n + 1)’th node, then the nodal vector for the contact element can be written
in the following form:
xT ={x(1) , x(2) , . . . , x(n) , x(n+1) }T (8.1)

(1) (1) (1) (2) (2) (2) (n) (n) (n) (n+1) (n+1) (n+1) T
={x1 , x2 , x3 , x1 , x2 , x3 , . . . , x1 , x2 , x3 , x1 , x2 , x3 } ,
where x(i) , i = 1, 2, . . . n+1 are vectors for nodal points. Let N(k) (ξ 1 , ξ 2 ), k =
1, 2, ..., n be the shape functions for the master surface parametrization. We
then define the approximation matrix A(ξ 1 , ξ 2 ) with 3 × (n + 1) dimension
as follows
⎡ ⎤
N1 0 0 N2 0 0 ... N(n) 0 0 −1 0 0
A = ⎣ 0 N1 0 0 N2 0 ... 0 N(n) 0 0 −1 0 ⎦ . (8.2)
0 0 N1 0 0 N2 ... 0 0 N(n) 0 0 −1
On the basis of this notation, the relative displacement vector ρ(ξ 1 , ξ 2 ) − rs

has the following form
ρ(ξ 1 , ξ 2 ) − rs = Ax. (8.3)
Both the relative variation, and the relative velocity vector have identical
structure:
δρ − δrs = Aδx, v − vs = A{v}. (8.4)
For further developments only the matrix of the shape function derivatives
A,j , i = 1, 2 has to be specified
⎡ ⎤
N1,j 0 0 N2,j 0 0 ... N(n),j 0 0 000
∂A
= A,j = ⎣ 0 N1,j 0 0 N2,j 0 ... 0 N(n),j 0 0 0 0⎦,
∂ξ j
0 0 N1,j 0 0 N2,j ... 0 0 N(n),j 0 0 0
(8.5)
in order to describe the vectors δρ,j and v,j
δρ,j = A,j x , v,j = A,j {v}. (8.6)

8.1 FE Discretization – NTS, STS and STAS Contact Elements 211
8.1.2 Segment-To-Segment (STS) Contact Approach

Both “master” and “slave” segments should be approximated separately for
a contact pair within the Segment-To-Segment (STS) approach, see Fig. 8.1.
The STS contact element is constructed as follows. Since, the integration of
all parameters is performed over the slave segment, however, being defined
via the “master” convective variables ξ 1 , ξ 2 , we have to specify the location
of the integration points ηI1 , ηJ2 on the slave element via the chosen inte-
gration rule. The slave points are then projected onto the master segment.
Thus, the contact area is approximated as a set of integration points which
have penetrated into the master segment. All necessary geometrical contact
parameters, such as the penetration ξ 3 , the normal n, the tangent vectors
ρ1 (ξ 1 , ξ 2 ), etc. are computed from the master segment.
1
0
0
1
0
1
11
00
00
11
N11
00
00
11 M
η2 slave
1
0
0
1
S η1
11
00 11
00
00
11 00
11
00
11 L
K ξ3 n
1
0
0
1
ρ ξ2 11
00
E 2 ρ1 00
11
00
11
D
1
0 ξ1
0
1
C
11
00 11
00
00
11 master 00
11
00
11
A B
Fig. 8.1 Segment-To-Segment (STS) contact approach
Let ABDE be a master segment (see Fig. 8.1) defined by n nodes
xTmaster ={x(1) , x(2) , . . . , x(n) }T (8.7)

(1) (1) (1) (2) (2) (2) (n) (n) (n)
={x1 , x2 , x3 , x1 , x2 , x3 , ..., x1 , x2 , x3 }T ,
and KLMN be a slave segment defined by m nodes

T
yslave ={y(1) , y(2) , . . . , y(m) }T (8.8)
(1) (1) (1) (2) (2) (2) (m) (m) (m)
={y1 , y2 , y3 , y1 , y2 , y3 , ..., y1 , y2 , y3 }T .
The shape functions Nk (ξ 1 , ξ 2 ), k = 1, 2, ...n and Mk (η 1 , η 2 ), k = 1, 2, ...m

are defined for the “master” and the “slave” segment respectively. A dis-
placement vector for the contact element is defined by n + m nodes as
(1) (1) (1) (2) (2) (2) (n) (n) (n)
uT ={u1 , u2 , u3 , u1 , u2 , u3 , ..., u1 , u2 , u3 , (8.9)
(n+1) (n+1) (n+1) (n+2) (n+2) (n+2) (n+m) (n+m) (n+m) T
u1 , u2 , u3 , u1 , u2 , u3 , ..., u1 , u2 , u3 } .
The approximation matrix A now contains both Nk (ξ 1 , ξ 2 ), and Mk (η 1 , η 2 )

shape functions and has the dimension 3 × (n + m):
⎡
N1 0 0 N2 0 0 ... N(n) 0 0
A(ξ 1 , ξ 2 , η 1 , η 2 ) = ⎣ 0 N1 0 0 N2 0 ... 0 N(n) 0 (8.10)
0 0 N1 0 0 N2 ... 0 0 N(n)
⎤
−M1 0 0 −M2 0 0 ... −M(m) 0 0
0 −M1 0 0 −M2 0 ... 0 −M(m) 0 ⎦.
0 0 −M1 0 0 −M2 ... 0 0 −M(m)
The slave points are found on the slave segment after the selection of the
integration rule, and therefore, are computed as interpolation over the slave
nodal points y(k) , see eqn. (8.8):

m
rs (ηI1 , ηJ2 ) |(η1 =ηI1 , 2)
η 2 =ηJ = Mk (ηI1 , ηJ2 )y(k) , (8.11)
k=1
where ηI1 , ηJ2 are integration points. Their projections onto the master seg-
ment are computed then as

n
ρ(ξc1 , ξc2 ) = Nk (ξc1 , ξc2 )x(k) , (8.12)
k=1
where the coordinates ξc1 , ξc2 are found as the solution of the closest point
projection (CPP) procedure on the surface, see Sect. 3.1.1.
The relative displacement vector is approximated via the matrix A for the
full coordinate vector x = {xTmaster , yslave
T
}T
ρ(ξc1 , ξc2 ) − rs (ηI1 , ηJ2 ) = A(ξc1 , ξc2 , ηI1 , ηJ2 )x, (8.13)
Since the differential operations for all parts of the tangent matrix, see e.g.
for non-frictional contact eqns. (7.9a), (7.9b), (7.9c) are defined only on the
master surface, the matrix of the shape function derivatives is computed only
for the ξ i derivatives as
8.1 FE Discretization – NTS, STS and STAS Contact Elements 213
⎡
N1,j 0 0 N2,j 0 0 ... N(n),j 0 0
∂A ⎣ 0 N1,j 0
= A,j = 0 N 2,j 0 ... 0 N (n),j 0
∂ξ j
0 0 N1,j 0 0 N2,j ... 0 0 N(n),j
⎤
0 0 0 ... 0 0 0
0 0 0 ... 0 0 0 ⎦ . (8.14)
0 0 0 ... 0 0 0
Based on the matrices defined in eqn. (8.10) and (8.14), the relative vectors
for the variation and for the velocity will have the same form as in eqns. (8.4)
and (8.6).
Remark 8.1. An integration rule should be specified by the user for the
“Segment-To-Segment” contact approach. It should be noted that the Lo-
batto quadrature formula (see e.g. [32] for the further discussion about nu-
merical integration) with only 2 × 2 integration points leads exactly to the
result of the “Node-To-Surface” element.
8.1.3 Segment-To-Analytical Surface (STAS) Contact

Approach
The STAS approach is used in combination with one of the rules to compute
the penetration for the analytically given surfaces, described in Sect. 4.3.
Thus, an approximation is given only for the contact segment covering the
surface of the meshed deformed body. Consider then the nodal displacement
vector with n nodes
(1) (1) (1) (2) (2) (2) (n) (n) (n)
uT = {u1 , u2 , u3 , u1 , u2 , u3 , ..., u1 , u2 , u3 }T , (8.15)
which is taken from the finite element discretization of the deformable body.
The approximation matrix A contains then only shape functions N(k) , k =
1, 2, . . . n:
⎡ ⎤
N1 0 0 N2 0 0 ... ... ... N(n) 0 0
A = ⎣ 0 N1 0 0 N2 0 ... ... ... 0 N(n) 0 ⎦ ; (8.16)
0 0 N1 0 0 N2 ... ... ... 0 0 N(n)
the matrix of the shape function derivatives A,j , j = 1, 2 has the following
form
⎡ ⎤
N1,j 0 0 N2,j 0 0 ... N(n),j 0 0
∂A
= A,j = ⎣ 0 N1,j 0 0 N2,j 0 ... 0 N(n),j 0 ⎦ .
∂ξ j
0 0 N1,j 0 0 N2,j ... 0 0 N(n),j
(8.17)
All geometrical parameters are computed, of course, based on the segment
geometry, see also the discussion in Sect. 4.3.2.
8.2 Various Approximations of Contact Surfaces

Defined by Finite Elements
As shown in Sect. 8.1, the approximation matrix A and its derivatives should
be supplied for all contact approaches. A huge variety of approximations
is known in finite element practice, starting from the classical results in
Zienkiewicz and Taylor [212] extended then to high-order finite element tech-
niques as in Solin et.al. [165] or the isogeometric finite element approxima-
tions via NURBS in Hughes et.al. [70]. We consider here only quadrilateral
elements with linear and quadratic approximation and smoothing techniques
based on simplified NURB splines for a quadrilateral area.
8.2.1 Quadrilateral Segment with Linear

Approximation
The geometry for the quadrilateral segment with linear approximation, see
Fig. 8.2, is given by 4 nodes, and therefore, 4 shape functions are necessary
to define the segment. Shape functions and their first derivatives for −1 ≤
ξ 1 , ξ 2 < +1 are given in Table 8.1.
3 ξ2 2
1
−1 0 1 ξ1
4 1
−1
Fig. 8.2 Quadrilateral segment with linear approximation
8.2.2 Closed Form Solution for CPP Procedure for

Linear Segment
In the case of the linear segment it is possible to obtain the closed form
solution of Closest Point Projection (CPP) procedure under assumption that
the segment is a parallelogram. Namely, taking the approximation in the form
given in Table 8.1 as

4
ρ(ξ 1 , ξ 2 ) = N k x(k) (8.18)
k=1
8.2 Various Approximations of Contact Surfaces 215
Table 8.1 Shape functions for a quadrilateral segment with linear approximation
∂N k ∂N k
N k (ξ 1 , ξ 2 )
∂ξ 1 ∂ξ 2
1 1 1
N1 (1 + ξ 1 )(1 − ξ 2 ) (1 − ξ 2 ) − (1 + ξ 1 )
4 4 4
1 1 1
N2 (1 + ξ 1 )(1 + ξ 2 ) (1 + ξ 2 ) (1 + ξ 1 )
4 4 4
1 1 1
N3 (1 − ξ 1 )(1 + ξ 2 ) − (1 + ξ 2 ) (1 − ξ 1 )
4 4 4
1 1 1
N4 (1 − ξ 1 )(1 − ξ 2 ) − (1 − ξ 2 ) − (1 − ξ 1 )
4 4 4
we can obtain for the tangent vectors the following expression:
x(1) + x(2) − x(3) − x(4) −x(1) + x(2) − x(3) + x(4)

ρ1 = + ξ2
4 4
−x(1) + x(2) + x(3) − x(4) −x (1)
+ x(2)
− x(3) + x(4)
ρ2 = + ξ1 (8.19)
4 4
If the segment is a parallelogram then
−x(1) + x(2) − x(3) + x(4) = 0. (8.20)
Thus, for the linear segment is spanned by two constant tangent vectors:
x(1) + x(2) − x(3) − x(4)

ρ∗1 =
4
−x (1)
+ x(2)
+ x(3) − x(4)
ρ∗2 = (8.21)
4
and approximated as follows
ρ(ξ 1 , ξ 2 ) = ρ∗1 ξ 1 + ρ∗2 ξ 2 + x(0) (8.22)
where x(0) is a central point of the segment:
x(1) + x(2) + x(3) + x(4)

x(0) = . (8.23)
4
In due course, a master node S can be described in the segment coordinate
system as follows:
rs = ρ∗1 ξ 1 + ρ∗2 ξ 2 + x(0) + n∗ ξ 3 , (8.24)
ρ∗1 × ρ∗2
where n∗ = is a normal vector to the segment.
ρ∗1 × ρ∗2
All convective coordinates are obtained then from eqn. (8.24) using the
scalar product:
ξ i = (rs − x(0) ) · ρ∗j aij

∗ for convective variables;
ξ 3 = (rs − x(0) ) · n for the penetration, (8.25)
Here only contravariant components aij ∗ of the metric tensor for the par-
allelogram should be computed as an inverse of the matrix with covariant
components a∗ij :
a∗ij = ρ∗i · ρ∗j . (8.26)
Remark 8.2. Numerical tests with arbitrary shaped quadrilateral elements
show gut results for non-frictional cases.
Remark 8.3. Computational of convective coordinates according to the rule
in eqn. (8.25) is exact for the triangular segment, because the triangular
segment can be exactly represented in the form of eqn. (8.22).
8.2.3 Quadrilateral Segment with Quadratic

Lagrangian Approximation
The geometry for the quadrilateral segment with quadratic Lagrangian ap-
proximation, see Fig. 8.3, is given by 9 nodes. The corresponding 9 shape
functions and their first derivatives for −1 ≤ ξ 1 , ξ 2 < +1 are given in
Table 8.2
4 ξ2 3
7 1
−1 0 1 ξ1
8 9 6
1 5 2
−1
Fig. 8.3 Quadrilateral segment with quadratic Lagrangian approximation

Table 8.2 Shape functions for a quadrilateral segment with quadratic Lagrangian
approximation
∂N k ∂N k
N k (ξ 1 , ξ 2 )
∂ξ 1 ∂ξ 2
N 1 ξ 1 (ξ 1 − 1)ξ 2 (ξ 2 − 1)/4 (2ξ 1 − 1)ξ 2 (ξ 2 − 1)/4 ξ 1 (ξ 1 − 1)(2ξ 2 − 1)/4
N 2 ξ 1 (ξ 1 + 1)ξ 2 (ξ 2 − 1)/4 (2ξ 1 + 1)ξ 2 (ξ 2 − 1)/4 ξ 1 (ξ 1 + 1)(2ξ 2 − 1)/4
N 3 ξ 1 (ξ 1 + 1)ξ 2 (ξ 2 + 1)/4 (2ξ 1 + 1)ξ 2 (ξ 2 + 1)/4 ξ 1 (ξ 1 + 1)(2ξ 2 + 1)/4
N 4 ξ 1 (ξ 1 − 1)ξ 2 (ξ 2 + 1)/4 (2ξ 1 − 1)ξ 2 (ξ 2 + 1)/4 ξ 1 (ξ 1 − 1)(2ξ 2 + 1)/4
N 5 (1 − ξ 1 ξ 1 )ξ 2 (ξ 2 − 1)/2 −ξ 1 ξ 2 (ξ 2 − 1) (1 − ξ 1 ξ 1 )(2ξ 2 − 1)/2
N 6 ξ 1 (ξ 1 + 1)(1 − ξ 2 ξ 2 )/2 (2ξ 1 + 1)(1 − ξ 2 ξ 2 )/2 −ξ 1 (ξ 1 + 1)ξ 2
N 7 (1 − ξ 1 ξ 1 )ξ 2 (ξ 2 + 1)/2 −ξ 1 ξ 2 (ξ 2 + 1) (1 − ξ 1 ξ 1 )(2ξ 2 + 1)/2
N 8 ξ 1 (ξ 1 − 1)(1 − ξ 2 ξ 2 )/2 (2ξ 1 − 1)(1 − ξ 2 ξ 2 )/2 −ξ 1 (ξ 1 − 1)ξ 2
N 9 (1 − ξ 1 ξ 1 )(1 − ξ 2 ξ 2 ) −2ξ 1 (1 − ξ 2 ξ 2 ) −2(1 − ξ 1 ξ 1 )ξ 2
8.2.4 Surface Smoothing Techniques in a Covariant

Approach
Since linear finite elements have been mostly used for FE models, it was soon
recognized that faceted geometries are causing problems within the contact
algorithm due to “jumping” normals on contact boundaries. Thus, the idea
of smoothing for linear contact surfaces appeared. The first publications have
been focusing on smoothing for rigid obstacles with splines, see Schweizer-
hof and Hallquist [156] (1992), Shimizu and Sano [159] (1995)), Heege and
Alart [64] (1996) for forming problems. Pietrzak and Curnier [145] (1998) dis-
cussed a problem resulting from the non-smoothness of the contact surfaces
with the example of two coaxial cylinders and used different splines for cov-
ering the linear meshes. The major difficulties using a smooth approximation
are the algorithm for the computation of sliding forces and the lineariza-
tion procedure, especially for the 3D case. Thus, a special approach based
on a so-called moving friction cone description for the computation of slid-

ing forces in combination with various spline smoothing techniques have been
proposed for 2D cases in Wriggers at.el. [192] (2001) and in Krstulovic-Opara
and Wriggers [101] (2002), then generalized for 3D cases in Krstulovic-Opara
et.al. [102] (2002). The difficulties in the linearization have then been resolved
for numerical examples by using an automatic differentiation software.
The effect of C 2 and C 1 smoothing for frictionless 2D problems was inves-
tigated in Stadler et.al. [170] (2003) and a subdivision scheme for smooth con-
tact surfaces, describing a procedure to sew different meshes (various quadri-
lateral patches, triangular + quadrilateral) has been considered in Holzapfel
and Korelc [169] (2004). Again an automatic differentiation software has been
used in both publications.
An approach with convective coordinates together with different approxi-
mations including splines and Hermite interpolations for the contact surface
were used to enforce C 1 continuity in the 2D case in Padmanabhan and
Laursen [135] (2001). A smoothing procedure for the 3D case, based on the
usage of Gregory patches was suggested in Puso and Laursen [149] (2002).
The problem with the linearization for curvilinear surfaces has been overcome
by taking the Mortar approach for the linear geometry. The application of
a control spline polygon for the searching procedure has been discussed in
El-Abbasi et.al. [37] (2001) and in Al-Dojayli and Meguid [2] (2002).
Since, in this book the linearization problems are resolved by the covariant
derivation, we are free now to chose any kind of approximations. Here, a Her-
mitian approximation as a simple subclass of the NURB spline approximation
is taken. Recently NURB splines are shown for the contact implementation,
see [175] and [176].
However, a very short information about spline and NURBS surfaces is
necessary to describe even the simple structure of the smooth element. Ad-
vanced approximation techniques are described in special monographs for
CAD surfaces, see e.g. Farin [38], Piegl and Tiller [144]; they can be applied
within the covariant description without any limitation.
Spline Interpolation of Curves

We start with the construction of splines for curves and consider the inter-
polation problem with different splines.
Bernstein Polynomials
are forming a basis function space to construct polynomial splines and can be
generalized into N on-U niform Rational B -S pline (NURBS ). The Bernstein
Polynomials are defined as follows:
Bin (t) = Cin ti (1 − t)n−i , i = 0, 1, ..., n, (8.27)

n(n − 1)...(n − i + 1)
where Cin = are binomial coefficients. The first three
i!
polynomial sets up to the third order contain the following polynomials:
B00 (t) = 1, B01 (t) = 1 − t, B11 (t) = t;
B02 (t) = (1 − t)2 , B12 (t) = 2t(1 − t), B22 (t) = t2 ; (8.28)
B03 (t) = (1 − t)3 , B13 (t) = 3t(1 − t)2 , B23 (t) = 3t2 (1 − t), B33 (t) = t3 .
The First Problem of Spline Interpolation
is specified as a construction of a spline which is passing through selected

points. Let x0 , x1 , ..., xn ∈ R3 be given points with parameters u0 , u1 , ..., un .
Construct a curve r(u) which satisfies the following conditions:
r(uk ) = xk , k ∈ {0, 1, ...n}. (8.29)
The Second Problem of Spline Interpolation
is specified as a construction of a spline line which is passing through selected

points and is possessing certain tangent vectors. Let x0 , x1 , ..., xn ∈ R3 be
given points with parameters u0 , u1 , ..., un and m0 , m1 , ..., mn ∈ R3 be a
given tangential vector with the same parameters.
Construct a curve r(u) which satisfies the following conditions:
dr(u)
r(uk ) = xk , = mk , k ∈ {0, 1, ...n}. (8.30)
du u=uk
The solution of the second problem allows to obtain then the C 1 -continuous
spline for the first problem.
Parametrization of the Spline Curve
A global spline should be parametrized based on the coordinates of the inter-

polation points. One of the simplest parameterizations is to use the distance
between points as a parameter (a chord length parametrization):
u0 = 0, u1 = ||x1 − x0 ||, ... , ui = ui−1 + ||xi − xi−1 ||, i = 2, ..., n. (8.31)
The Second Interpolation Problem – Solution in the Form of a Piecewise

Cubic C 1 Interpolation
We consider only piecewise interpolations, because this case is more suitable

for further finite element constructions. In the first problem, interpolation
points together with their tangential vectors are known. Both the continuity
of tangential vectors (C 1 continuity), and the local support are satisfied. The
solution can be constructed in the form of a Bezier curve, see Farin [38]. The
global Bezier curve consists of local Bezier curves of third order; it passes
through the points x0 , x1 , ..., xn ∈ R3 and satisfies C 1 continuity being a
Bezier curve of third order on each interval (xi , xi+1 ), i ∈ {0, 1, ..., n − 1}. It
should be mentioned that this problem does not have a unique solution. As
an additional constraint equation, the normalization of the tangential vector
can be taken
||mi || = 1, i ∈ {0, 1, ..., n}. (8.32)
Boundary Bezier points for each interval (xi , xi+1 ) are defined from the in-
terpolation condition:
b3i = xi , i = 0, 1, ..., n; (8.33)
the inner Bezier points are defined as
b3i+1 = b3i + αi mi , b3i−1 = b3i − βi mi . (8.34)
Several choices for αi and βi values are discussed in Farin [38], recommended
ones are
αi = βi = 0.4||Δxi ||. (8.35)
The global Bezier curve is then defined on the interval between points xi and
xi+1 by local Bezier curves:

3
x(u) = b3i B03 (t) + b3i+1 B13 (t) + b3i+2 B23 (t) + b3i+3 B33 (t) = b3i+j Bj3 (t);
j=0
(8.36)
where the local variable t ∈ [0, 1] is given via the global parameter u, see in
eqn. (8.31):
t = (u − ui )/Δi , Δi = ui+1 − ui , i = 0, 1, ..., n − 1. (8.37)
The First Interpolation Problem – Solution in the Form of Piecewise Cubic

C 1 Interpolation
The solution of the first problem as piecewise cubic C 1 interpolation in Bezier

form is, in general, not unique – all tangential vectors mi have to be defined
to achieve uniqueness. One of the possible approximations of the tangential
vectors is a vector describing a chord through the points xi−1 and xi+1 , i.e.
mi = xi−1 − xi+1 , (8.38)
thus leading to the inner Bezier points in the form:

Δi−1
b3i−1 =b3i − mi
3(Δi−1 + Δi )
Δi
b3i+1 =b3i + mi . (8.39)
3(Δi−1 + Δi )
This method is known as the Catmull-Rom spline.

This construction of the spline fails at boundary points x0 and xn . Several
methods are known to overcome this problem. The idea behind the Bessel
method is to find an interpolating parabola q(u) through the starting points
x0 , x1 , x2 (or through the ending points x0 , x1 , x2 ) and to take the spline
tangential direction as the tangential direction for this parabola:
dq
m0 = (8.40)
du
Another method (Overhauser spline) is to construct a special cubic inter-
polant in the form:
ui+1 − u u − ui
x(u) = qi (u) + qi+1 (u), i = 1, ..., n − 2 (8.41)
Δi Δi
where qi (u) is the parabola passing through three points, then at the begin-
ning points one sets x(u0 ) = q0 (0).
Piecewise Cubic C 1 Interpolation in Hermite Form
The Hermite form of the spline is one of the most straightforward imple-
mentations into the finite element method. One can show that the Hermite
interpolation is simply another form of the Bezier interpolation. Consider,
as example, the first spline problem. The boundary Bezier points are the in-
terpolation points. In order to obtain the inner Bezier points, the derivative
formula for Bezier curves has to be used:
d 3 3
x(ui ) = (b3i − b3i−1 ) = (b3i+1 − b3i ) = mi (8.42)
du Δi−1 Δi
Thus, the inner Bezier points are given as

Δi
b3i+1 =b3i + mi , i = 0, 1, ..., n − 1
3
Δi−1
b3i−1 =b3i − mi , i = 0, 1, ..., n (8.43)
3
The spline in Bezier form can be written after some algebra in the form:
x(u) = xi H03 (u) + mi H13 (u) + mi+1 H23 (u) + xi+1 H33 (u), (8.44)
where Hi3 (u) are Hermite polynomials. They can be expressed in terms of
Bernstein polynomials as follows:
H03 (u) =B03 (t) + B13 (t)

Δi
H13 (u) = B13 (t)
3
Δi 3
H23 (u) = − B (t)
3 2
H33 (u) =B23 (t) + B33 (t) (8.45)
where t = (u − ui )/Δi ∈ [0, 1] is the local parameter of the interval [ui , ui+1 ].
Equations (8.44) defines the relationship between the Hermite and the
Bezier form of the spline. If the Hermite form is known, then the inner Bezier
points can be calculated and vice versa.
Piecewise Cubic C 2 Interpolation in Bezier Form
The requirement of C 2 -continuity leads to a linear system of equations with

a unique solution, however, the local support property is lost then which
makes this scheme inconvenient in FEM. Another strategy of keeping the
C 2 -continuity is to use full NURBS definitions.
Rational B-Spline – NURBS
Rational B-splines are one of the standard curve and surface descriptions
in CAD and computer graphics systems. A generalization of the polynomial
curve is a rational curve with order n of Bernstein’s polynomials. The Bezier
form of it can be written as
w0 b0 B0n (t) + w1 b1 B1n (t) + ...wn bn Bnn (t)
x(u) = . (8.46)
w0 B0n (t) + w1 B1n (t) + ...wn Bnn (t)
This description of a spline has many advantages due to the additional degrees
of freedom because of additional weights wi . It leads to multiple possibilities
to control the geometry. Rational B-splines constructed in this way are called
N on-U niform Rational B -S pline or NURBS. Special literature, however, is
available containing all NURBS properties in detail, see e.g. Farin [38], Piegl
and Tiller [144].
Spline Surfaces Interpolation

The theory outlined for the spline curve interpolation is directly applied for
the surface interpolation problem. However, the situation with surfaces is far
more complicated: e.g. it is necessary to distinguish a regular or an irregular
net of interpolation points, rules for mixed derivatives should be additionally
developed etc. All these problems are forming a huge area of research, and
we will take only the simplest structures out of it.
Our main assumption is
• Interpolating points are defining a regular quadrilateral surface mesh in
a Cartesian coordinate system.
Thus, we can directly use the tensor product of Bezier splines, considered in
Sect. 8.2.4.
Composite Surfaces
consist of patches. A patch is a part of the global surface and has its own
parametrization. This property is called a local support property. The first
interpolation problem is formulated then for surfaces as follows:
Let the matrix ⎡ ⎤
x00 x01 ... x0m
⎢ x10 x11 ... x1m ⎥
⎢ ⎥ (8.47)
⎣ ... ... ... ... ⎦
xn0 xn1 ... xnm
be a given as a regular net of points xij ∈ R3 with parameters u0 , u1 , ..., un
and v0 , v1 , ..., vm . These points are vertices’s of the quadrilateral mesh. Con-
struct the smooth C 1 surface parametrization x(u, v) that satisfies the fol-
lowing condition:
x(ui , vj ) = xij , i = 0, 1, ..., n; j = 0, 1, ..., m. (8.48)
The interpolation problem can then be solved with patches constructed as

tensor products of 1D bicubic Bezier curves. However, additional equations
for the twist estimation xuv on the vertices’s are necessary. This leads to
an additional degree of freedom, and therefore to different models of the
composite surface.
A Bicubic Hermite Patch
is constructed as a tensor product of C 1 -splines given in Hermite form. The

surface interpolation vector x(u, v) gets then the following form:

3
3
x(u, v) = hij Hi3 (u)Hj3 (v); 0 ≤ u, v ≤ 1, (8.49)
i=0 j=0
where Hi3 (u) are the cubic Hermite functions from eqn. (8.45) and the terms
hi,j are defining the following matrix
⎡ ⎤
x(0, 0) xv (0, 0) xv (0, 1) x(0, 1)
⎢ xu (0, 0) xuv (0, 0) xuv (0, 1) xu (0, 1) ⎥
⎢
[hij ] = ⎣ ⎥ (8.50)
xu (1, 0) xuv (1, 0) xuv (1, 1) xu (1, 1) ⎦
x(1, 0) xv (1, 0) xv (1, 1) x(1, 1)
Additional information for the mixed derivative xuv at vertices’s is required

leading to the so-called twist estimation problem. Various methods include
then the Coons patch, the Adini twist, the Bessel twist, the Gregory patch
and others, see Farin [38]. In our numerical examples we further use the
simplest Zero twist just setting all mixed derivatives to zero at the vertices’s
xuv (ui , vj ) = 0. (8.51)
Finally, after a lengthy introduction into spline interpolation theory we are

ready to construct a simple smooth 3D contact element.
A Quadrilateral Smooth Contact Surface

The idea of surface smoothing discussed in computational contact mechanics
is to cover the given linear surface mesh describing a possible contact surface
with a smooth spline surface. This is a mesh dependent task for a certain
type of spline. Thus, for simplicity we consider here only a quadrilateral area
with a regular quadrilateral mesh, see Fig. 8.4. This area can be covered
by a composite spline surface constructed via regular patches as discussed in
Sect. 8.2.4. Three different types of patches are necessary for the quadrilateral
surface: a corner patch for four corners – No. 1-4 in Fig. 8.4, a boundary patch
for boundaries – No. 5-8, and a central patch for the inner region – No. 9.
Following the technique introduced in Sect. 8.2.4, the structure of the
matrix [hij ] in eqn. (8.50) should be defined depending on the position of the
nodal vector. Assuming that nodal points from the regular net in Fig. 8.4
are given in the form xij ∈ R3 i = 0, 1, ..., n; j = 0, 1, ..., m we have to
introduce parameters u0 , u1 , ..., un and v0 , v1 , ..., vm in order to obtain the
parametrization in the form of eqn. (8.48). This parametrization with ui , vj is
constructed as a tensor product using the chord length for lines in eqn. (8.31).
Starting with the line xi0 , i = 0, 1, ..., n and choosing parameters u0 , u1 , ..., un
u0 = 0, u1 = ||x10 − x00 ||, ... , ui = ui−1 + ||xi0 − xi−10 ||, i = 2, ..., n;

(8.52)
then, parameters v0 , v1 , ..., vn are chosen with the line x0j , j = 0, 1, ..., m in
the form
v0 = 0, v1 = ||x01 − x02 ||, ... , vj = vj−1 + ||x0j − x0j−1 ||, i = 2, ..., m.

(8.53)
4 7 3
corner patch
central patch
9
8 6
1 5 boundary patch 2
Fig. 8.4 A quadrilateral regularly meshed surface. 3 types of patches are necessary
for a smooth contact surface: central, boundary and corner patches.
01 11 00
10 11
10 00
1010 9
1100000000
00 01 00
11111111
11
11
4 1010 0
3
8
1
1100000000
0011111111
01 00 11 1010 01
10
1 2
1010 11
00
12 7
5
00
11 6
10
Fig. 8.5 Central patch – pattern of nodal points.
If the cells of the mesh have approximately the same size, it is even better to
choose the simplest parametrization:

ui = i, i = 0, ..., n;
(8.54)
vj = j, j = 1, ..., m.
The Central Patch
is defined by 12 nodal points and has the pattern of nodal points presented
in Fig. 8.5. For simplicity, zero twist xuv = 0 is taken further for the finite
elements. Taking into account the simplest parametrization in eqn. (8.54) (in
this case the computation rule for the first partial derivatives is coinciding
with the finite difference scheme), the matrix [hij ] has the following form:
⎡ ⎤
1 1
⎢ x1 (x4 − x5 ) (x10 − x1 ) x4 ⎥
⎢ 2 2 ⎥
⎢1 1 ⎥
⎢ (x2 − x12 ) 0 0 (x3 − x11 ) ⎥
⎢2 2 ⎥
[hij ] = ⎢
⎢ 1
⎥.
⎥ (8.55)
⎢ (x − x ) 1
⎢ 2 7 1 0 0 (x8 − x4 ) ⎥
⎥
⎢ 2 ⎥
⎣ 1 1 ⎦
x2 (x3 − x6 ) (x9 − x2 ) x3
2 2
Using the formulas for the Hermite patch, the tensor form can be rewritten
in the vector form as follows:

3
3
12
x(u, v) = hi,j Hi3 (u)Hj3 (v) = N i (u, v)xi , 0 ≤ u, v ≤ 1. (8.56)
i=0 j=0 i=1
This vector form is used with the shape functions N i (u, v) and corresponds
to a nodal vector defined by nodes xi . This form is standard for the finite
element method. Thus, the approximation of the vector is given as

12
x(u, v) = N i (u, v)xi = (8.57)
i=1
1 1
= x1 [H1 (u)H1 (v) − H1 (u)H3 (v) − H3 (u)H1 (v)]
2 2
1 1
+ x2 [ H2 (u)H1 (v) + H4 (u)H1 (v) − H4 (u)H3 (v)]
2 2
1 1
+ x3 [H4 (u)H4 (v) + H4 (u)H2 (v) + H2 (u)H4 (v)]
2 2
1 1
+ x4 [ H1 (u)H2 (v) + H1 (u)H4 (v) − H3 (u)H4 (v)]
2 2
1 1
+ x5 (−H1 (u)H2 (v)) + x6 (−H4 (u)H2 (v))
2 2
1 1
+ x7 H3 (u)H1 (v) + x8 H3 (u)H4 (v)
2 2
1 1
+ x9 H4 (u)H3 (v)) + x10 H1 (u)H3 (v))
2 2
1 1
+ x11 (−H2 (u)H4 (v)) + x12 (−H2 (u)H1 (v)).
2 2
8 7
4 3 6
1 2 5
Fig. 8.6 Corner patch – pattern of nodal points.
The Corner Patch
is defined by 8 nodes and has the pattern of nodal points presented in Fig. 8.6.
Now, the rules discussed in Sect. 8.2.4 eqn. (8.40) for boundary points for
curves are used to compute the derivatives for the boundary points for a
surface spline. Derivatives for the inner points are computed similar to the
central patch. The following approximations coinciding with the finite differ-
ence scheme of order O(h2 ) is obtained:
3 1 3 1
xu (0, 0) = − x1 + 2x2 − x5 xv (0, 0) = − x1 + 2x4 − x8
2 2 2 2
3 1 1 1
xu (0, 1) = − x4 + 2x3 − x6 xv (0, 1) = − x1 + x8
2 2 2 2
(8.58)
1 1 3 1
xu (1, 0) = x5 − x1 xv (1, 0) = − x2 + 2x3 − x7
2 2 2 2
1 1 1 1
xu (1, 1) = x6 − x4 xv (1, 1) = − x2 − x7 .
2 2 2 2
Then the matrix [hij ] gets the following form:

⎡ ⎤
3 1 1 1
⎢ x1 − x1 + 2x4 − x8 − x1 + x8 x4 ⎥
⎢ 3 2 2 2 2 ⎥
⎢ − x + 2x − 1 x 3 1
− x4 + 2x3 − x6 ⎥
⎢ 1 2 5 0 0 ⎥
⎢ 2 2 2 2 ⎥
[hij ] = ⎢ ⎥.
⎢ 1 1 1 ⎥
⎢ 1
x − x1 0 0 x6 − x4 ⎥
⎢ 2 5 2 2 2 ⎥
⎣ 3 1 1 1 ⎦
x2 − x2 + 2x3 − x7 − x2 − x7 x3
2 2 2 2
(8.59)
The vector form of the approximation of the surface vector x(u, v) is given
as

8
x(u, v) = N i (u, v)xi = (8.60)
i=1

3 1
= x1 H1 (u)H1 (v) − H1 (u)H2 (v) − H1 (u)H3 (v)
2 2

3 1
− H2 (u)H1 (v) − H3 (u)H1 (v)
2 2
3 1
+ x2 [2H2 (u)H1 (v) + H4 (u)H1 (v) − H4 (u)H2 (v) − H4 (u)H3 (v)]
2 2
+ x3 [2H2 (u)H4 (v) + 2H4 (u)H2 (v) + H4 (u)H4 (v)]+
3 1
+ x4 [2H1 (u)H2 (v) + H1 (u)H4 (v) − H2 (u)H4 (v) − H3 (u)H4 (v)]
2 2
1
+ x5 (−H2 (u)H1 (v) + H3 (u)H1 (v))
2
1
+ x6 (−H2 (u)H4 (v) + H3 (u)H4 (v))
2
1
+ x7 (−H4 (u)H2 (v) + H4 (u)H3 (v))
2
1
+ x8 (−H1 (u)H2 (v) + H1 (u)H3 (v))
2
The Boundary Patch
is defined by 10 nodes and has the pattern of nodal points presented in

Fig. 8.7. The partial derivatives for boundary points are computed as
1 1 3 1
xu (0, 0) = x2 − x10 xv (0, 0) = − x1 + 2x4 − x8
2 2 2 2
1 1 1 1
xu (0, 1) = x3 − x9 xv (0, 1) = x8 − x1
2 2 2 2 (8.61)
1 1 3 1
xu (1, 0) = x5 − x1 xv (1, 0) = − x2 + 2x3 − x7
2 2 2 2
1 1 1 1
xu (1, 1) = x6 − x4 xv (1, 1) = x7 − x2 .
2 2 2 2
8 7
9 4 3 6
10 1 2 5
Fig. 8.7 Boundary patch – pattern of nodal points.

The matrix [hij ] has the following form:

⎡ ⎤
3 1 1 1
x1 − x1 + 2x4 − x8 x8 − x1 x4
⎢ 2 2 2 2 ⎥
⎢1 1 1 1 ⎥
⎢ x2 − x10 0 0 x3 − x9 ⎥
⎢ 2
[hij ] = ⎢ 1 2 2 2 ⎥ (8.62)
1 1 1 ⎥
⎢ x5 − x1 0 0 x6 − x4 ⎥
⎣ 2 2 2 2 ⎦
3 1 1 1
x2 − x2 + 2x3 − x7 x7 − x2 x3
2 2 2 2
The vector form approximation of the boundary patch vector x(u, v) is given
as

10
x(u, v) = N i (u, v)xi = (8.63)
i=1
3 1 1
= x1 [H1 (u)H1 (v) − H1 (u)H2 (v) − H1 (u)H3 (v) − H3 (u)H1 (v)]
2 2 2
1 3 1
+ x2 [ H2 (u)H1 (v) + H4 (u)H1 (v) − H4 (u)H2 (v) − H4 (u)H3 (v)]
2 2 2
1
+ x3 [ H2 (u)H4 (v) + 2H4 (u)H2 (v) + H4 (u)H4 (v)]
2
1
+ x4 [2H1 (u)H2 (v) − H3 (u)H4 (v) + H1 (u)H4 (v)]
2
1 1
+ x5 H3 (u)H1 (v) + x6 H3 (u)H4 (v)
2 2
1
+ x7 (−H4 (u)H2 (v) + H4 (u)H3 (v))
2
1
+ x8 (−H1 (u)H2 (v) + H1 (u)H3 (v))
2
1 1
− x9 H2 (u)H4 (v) − x10 H2 (u)H1 (v).
2 2
Writing all necessary shape functions in vector form together with
their derivatives is an absolutely tedious task. During implementa-
tion it is organized via subroutines containing the Hermite functions
H1 (t), H2 (t), H3 (t), H4 (t) used for all shape functions. These functions are
just consequently renumbered after their definition in eqn. (8.45). The func-
tions and their derivatives are presented in Table 8.3.
Table 8.3 Hermite functions Hi (t), t ∈ [0, 1] and their derivatives are forming a
shape function space for smooth contact elements.
∂Hk ∂ 2 Hk
Hk (t)
∂t ∂t2
H1 (t) 1 − 3t2 + 2t3 −6t + 6t2 −6 + 12t
H2 (t) t − 2t2 + t3 1 − 4t + 3t2 −4 + 6t
H3 (t) t3 − t2 3t2 − 2t 6t − 2
H4 (t) 3t2 − 2t3 6t − 6t2 6 − 12t
8.3 Non Frictional Contact Analysis:

Node-To-Segment Approach
In this section a series of numerical examples of non-frictional contact prob-
lems between a flexible structure and rigid surfaces of second order (cylinder
and sphere) are investigated. Bilinear and biquadratic contact elements are
used for parameterization of both master and slave contact surfaces. The
Node-To-Segment (NTS) contact approach shown in Sect. 8.2.1 and in
Sect. 8.2.3 is realized. The aim is to estimate the influence of rotational and
curvature parts of the contact tangent matrix on the convergence of the
iterative algorithm. In order to investigate the corresponding contribution
of each part of the tangent matrix, the following three alternatives are
considered
a. Use of the full tangent matrix K = K(m) + K(curv) .
b. Use the main part K(m) and the rotational part Kr ,

(1) (2)
i.e. the contact matrix K = K(m) + Kr + Kr .
c. Use the main part K(m) and the pure curvature part Kh ,
i.e. the contact matrix K = K(m) + K(h) .
d. Use only the main part K(m) as a contact matrix, i.e. K = K(m)
8.3 Non Frictional Contact Analysis:Node-To-Segment Approach 231
The main contribution into the full contact tangent matrix comes usually
from the main matrix K(m) , therefore a relative measure ε is chosen as an
estimate
||K − K(m) ||
ε= · 100% (8.64)
||K(m) ||
with the following matrix norm:

n
||K|| = max |kij | (8.65)
i
j=1
The influence on the convergence rate is given by the number of equilibrium

iterations at each load step, while the value ε, computed at each load step
after the equilibrium iteration, is given in order to estimate the contribution
of the “curvature” part.
8.3.1 Bending of a Beam over a Rigid Cylinder

Linear Approximation of the Contact Surfaces
A clamped elastic beam is loaded by prescribed displacements at the free
end in vertical direction, that leads to bending over a rigid cylinder. The
parameters of the beam are chosen as: length l=24 cm, height h=0.25 cm,
width b=1 cm; the material model is St. Venant elastic material with an
elasticity modulus of 1.0 · 104 ; the Poisson ratio is 0.3; for contact a penalty
factor of 1.0 · 104 MPa/cm is chosen. The beam is modeled with 24 “solid-
shell” elements [60] with linear shape functions plus some added shell specific
enhancements. The rigid cylinder with radius R=2 cm is modelled with 49
contact elements in the circumferential direction. The central axis of the
cylinder is positioned at 12 cm from the clamping. For the contact elements
a linear approximation of the contact surfaces is used; the beam surface was
treated as a “master” surface. The prescribed final displacement uext = 9 cm
is subdivided into 90 identical load steps. Fig. 2 shows a sequence of the
deformed beam during loading at the 0, 20, 40, 60, 80 and 90’th load step,
respectively. The value of penetration does not exceed 0.18 % of the beam
thickness.
The results of the computation are presented in Table 8.4. The convergence
rate in each load step is estimated by the number of equilibrium iterations
(column No. it. / l.s. in Table 8.4), which has been changing over load steps,
e.g. the computation shows 3 equilibrium iterations per load step within the
first 15 load steps. The contribution of various parts of the tangent matrix is
estimated by the norm in eqn. (8.64).
We obtain 319 total iterations for the cases a and b and 345 iterations for
the cases c and d. As expected, there is no difference between the results if
Fig. 8.8 Bending of a clamped beam over a rigid cylinder. Sequence of

deformations.
Table 8.4 Bending over a rigid cylinder. Bilinear elements for the beam. Node-to-
surface contact elements. Influence of various contact stiffness parts on convergence.
Case a: full matrix, case b: without curvature part; case c: without rotational part;
case d: only main matrix. Comparison of no. of iterations in each load step (l.s.)
and accumulated over several load steps.
Case a/b Case c/d

No. l.s. it./l.s. Cum. it. ε · 10−2 % eqn. (8.64) No. l.s. it./l.s. Cum. it.
1-15 3 45 0.18 1-15 3 45
16 5 50 0.19 16 5 50
17-45 3 137 0.55 17-26 3 80
46-73 4 249 0.45 27-73 4 268
74 5 254 0.62 74 5 273
75 4 258 1.04 75 4 277
76 5 263 0.62 76 5 282
77-90 4 319 1.26 77-83 4 310
84-90 5 345
the “pure curvature” part is taken into account or not, because the curva-
ture tensor is zero. It is also obvious that the rotational part is much more
important.
Quadratic Approximation of the Contact Surfaces

Now the beam from the previous example is modeled with 12 quadratic “solid-
shell” elements [58]. The penalty factor is chosen as 5.0 · 104 MPa/cm, which
leads to a maximum penetration of 0.048 % of the beam thickness.
8.3 Non Frictional Contact Analysis:Node-To-Segment Approach 233
Table 8.5 Bending over a rigid cylinder. Biquadratic elements for the beam. Node-
to-surface contact elements. Influence of various contact stiffness parts on conver-
gence. Case a: full matrix; case b: without curvature part; case c: without rotational
part; case d: only main matrix. Comparison of no. of iterations in each load step
(l.s.) and accumulated over several load steps.
Case a/b Case c

−2
No. l.s. No. Cum. ε · 10 % No. l.s. No. Cum. ε · 10−4 %
it./l.s. No. it. it./l.s. No. it.
1-16 3 48 0.19 1-16 3 48 0.09
17-18 4 56 0.19 17-18 4 56 0.12
19-46 3 140 0.55 19-30 3 92 0.35
47 4 144 0.45 31-47 4 160 0.49
48-50 3 153 0.62 48 3 163 1.03
51-69 4 229 1.04 49-86 4 315 4.81
70 5 234 0.63 87-90 5 335 9.76
71-86 4 298 1.27
87 5 303 2.00
88-90 4 315 2.37
Case d
No. l.s. No. Cum.
it./l.s. No. it.
1-16 3 48
17-18 4 56
19-30 3 92
31-47 4 160
48 3 163
49-69 4 247
70 5 252
71-86 4 316
87 6 322
88-90 5 337
The total number of iterations for case a with full matrix and case b, when
the rotational matrix is taken into account, is identical, therefore, Table 8.5
presents the influence of the “curvature” matrix on convergence in case of
a, c and d. Though, elements with quadratic shape functions are used, the
influence of the “pure curvature” part is still negligible. It is even possible to
obtain the result with the main matrix only with a loss of 7 % of the total
number of iterations.
8.3.2 Bending of a Beam over a Rigid Sphere

A more general case to examine the influence of all parts of the contact matrix
is to consider the contact with a body with a curvature in both directions.
The clamped elastic beam of the previous example, but with a width of
b = 4 cm, is now bending over a rigid sphere. The radius of the sphere is
4 cm. The center of the sphere is positioned at 0.5 cm from the edge of
the beam and at 12 cm along the beam measured from the clamping. The
rigid sphere was modeled with 512 bilinear contact elements. The contact
elements of the beam inherit the geometry of the beam and, therefore, have
biquadratic approximations; the beam surface is represented as a “master”
surface. Two asymmetric forces F1 = 17.5 N and F2 = 70 N are applied in
the nodes at the free end as presented in Fig. 8.9. In the computation they
were applied incrementally with 100 identical load steps. Fig. 8.9 shows the
evolution diagram of the deformation for the 0, 20, 40, 60, 80, 100’th load
step respectively. In this example we have bending and twisting of the beam
as well.
Again the composition of the tangent matrix is varied. The value of the
contact penalty is taken as 0.5·104 MPa/cm for the cases c and d, but 0.5·103
MPa/cm for the cases a and b due to convergence problems. The last value
leads to a maximum penetration of 0.75 % of the beam thickness. Table 8.6
shows the result of the analysis with biquadratic “solid-shell” elements and
biquadratic contact elements.
In this case the curvature is changing in both directions and, as it can be
seen from the result for case b, the influence of the “rotational” and the “pure
curvature” part is larger. The higher the curvature, the more equilibrium
iterations are necessary. One can see that the influence of the “rotational”
part in this example is crucial, because the exclusion of the “rotational” part
leads to an increase of the total number of iterations of more than 40 % ,
while the influence of the “pure curvature” part is still small. For the case
c, which is not presented in the Table 8.6, we obtain 609 iterations. For the
full matrix (case a), however, the computation of the curvature part is rather
costly in comparison with the “main” and with the “rotational” part, due to
the necessity to compute the second derivative and the double summation
for the contravariant components of the curvature tensor hij , see eqn. (2.16).
Thus it appears to be the most efficient choice to consider the analysis without
“pure curvature” matrix.
8.3.3 Discussion
The numerical examples show that in the case of linear approximations and
aligned contact elements keeping of the ”pure curvature” part is meaningless.
Then, it even appears sufficient to keep only the main part as a contact
tangent matrix.
8.4 Non Frictional Contact Analysis: Large Penetration Algorithm 235
Fig. 8.9 Bending of a beam over a rigid sphere. Sequence of deformations.
Table 8.6 Bending over a rigid sphere. Biquadratic elements for the beam. Node-to-
surface contact elements. Influence of various contact stiffness parts on convergence.
Case a: full matrix; case b: without curvature part; case d: only main matrix.
Comparison of no. of iterations in all load steps (l.s.)
Case a Case b Case d

No. No. Cum. ε · 10−2 % No. l.s. No. Cum. No. l.s. No. Cum.
l.s.] it./l.s. No. it. it./l.s. No. it. it./l.s. No. it.
1 27 27 0.731 1 27 27 1 20 20
2-23 4 115 7.382 2-22 4 111 2-8 4 48
24 6 121 18.08 23 6 117 9-23 5 123
25-33 4 157 16.26 24-33 4 157 24 6 129
34 5 162 12.20 34 6 163 25-33 5 174
35-100 4 426 17.32 35-100 4 427 34-48 6 264
49-58 5 314
59-68 6 374
69-85 7 493
86-100 8 613
If elements with higher order approximations are used, the influence of

the ”rotational” part is larger, but the influence of the ”pure curvature”
part remains still small. Therefore, the last part, which is computationally
more expensive then the others, can be eliminated from the complete tangent
matrix without loss of efficiency.
8.4 Non Frictional Contact Analysis: Large

Penetration Algorithm
The large penetration algorithm is proposed originally by Zavarise, De Loren-
zis and Taylor in [200] and [201] for 2D geometries discretized with linear
elements. The method is based on separating the consistent tangent matrix

in a ”basic” and a ”geometrical” part and starting the computation only with
the ”basic” part combined with a special algorithm switching after some iter-
ations to the computation with consistently linearized matrices. The current
Section is dealing with a 3D extension of the large penetration algorithm
developed by Izi, Konyukhov and Schweizerhof in [75]. The split of the full
consistent tangent matrix, studied numerically in Sect. 8.3 for non-frictional
problems solved by NTS approach, is employed to implement the large pen-
etration algorithm in a covariant fashion for linear and quadratic finite ele-
ments. The algorithm is tested first on numerical 2D like examples similarly
proposed in [201] and then on real 3D-numerical examples. It is found to be
very effective, especially, for applications with curvilinear geometries showing
convergence even within one load step.
8.4.1 Large Penetration Algorithm – General

Considerations
In recent publications by Zavarise et al., see [200] and [201], dealing with 2D
linear geometry contact problems a specific approach of separating the full
tangent matrix into a ”basic” and a ”geometrical” part is proposed where
the ”basic” part in 2D is comparable to the main part gained within this
contribution in eqn. (7.17) for 3D. Both are consisting of the dyadic product
of their contact normal vectors n ⊗ n. Based on the separation of the tangent
matrix a scheme for large load steps in the presence of large penetrations is
developed. This scheme relies on the observation that consistent linearization
is only convenient for a field of unknowns being close to the solution point. A
further assumption is that a priori estimate of the finally resulting maximum
contact pressure for a given contact setup is possible. The a priori estimated
contact pressure denoted as pe is used then as an indicator to switch between
the two phases of the large penetration scheme. As mentioned above this
proposed procedure is for a 2D linear geometry essentially not enabling a
geometrical exact approach and is specifically not directly applicable to 3D
contact problems, due to the linearization concentrating only on a linear ge-
ometry approximation for the 2D case. However, the derived formulation in
the current contribution regarding the geometrical exact description accord-
ing to the covariant approach is further on used enabling a clear separation
of the consistent tangent matrix into main, rotational and curvature part.
Therefore, the covariant formulation provides in the following the basis for
the large penetration scheme in 3D. The setup of the large penetration scheme
nevertheless remains the same according to Zavarise et al., see [200] and [201]
and is presented for completeness reasons in the following in covariant form.
The first phase takes place for the first iterations of each load step
characterized by computed contact pressures orders of magnitudes higher
than the final pressure. Within this first phase the Newton strategy with
consistent linearization correlating to the full tangent matrix should not be

used as the rotational and curvature part in eqns. (7.18) and (7.19) are de-
pending on the penetration ξ 3 via the contact traction as N = N ξ 3 and are
for large penetrations often leading to negative entries on the diagonals of the
full tangent matrix. In order to prevent this problem arising for large pen-
etrations two modifications within the computation were proposed in [201].
The first modification concerning the tangent matrix is now applied to the
aforementioned covariant form and has the consequence to disregard tangent
matrix parts such as the rotational and curvature parts. The second modi-
fication is concerning the residual of a nonlinear computation where instead
of the linear relationship between contact pressure and normal penetration
an upper bound for the maximum contact pressure represented by the a pri-
ori estimated contact pressure pe is used. Coupling the upper bound with
pe
the use of a modified contact stiffness contribution gained by 3 instead of
ξ
the constant penalty parameter N , permits to enforce gradually the violated
impenetrability condition within this modified Newton iterations in the first
phase. This is subsequently enabling the contact solution procedure to shift
smoothly from phase one to the phase two, where the standard consistently
linearized terms - the full tangent matrix - are then taken resulting in a
quadratic rate of convergence in the last iterations.
Enhanced Large Penetration Algorithm

As outlined already in [200] and [201] there is also a limitation to this strategy
concerning the a priori estimated contact pressure. It has to be ensured that
the a priori estimated contact pressure is set within a range of the finally
resulting maximum contact pressure at convergence. Especially, an under-
estimation often results in converging to an incorrect solution. In order to
prevent this within the current contribution the identical strategy as pro-
posed in [200] and [201] is followed updating the a priori estimated contact
pressure of consecutive iteration steps n and n + 1 by some multiplicative of
pe as
pe n+1 = pe n + αc pe , (8.66)
where αc represents a chosen multiplier. In order to judge if such an update
is necessary an indicator R based on the penetration is introduced checking
if the penetration of consecutive iteration steps is changing rapidly. Desig-
nating penetrations within iteration step n and n + 1 with ξ 3 n and ξ 3 n+1 ,
respectively, the value of R is computed as
⎧
⎪ ξ 3 n+1
⎪
⎪ 1 − ξ 3 n+1 < ξ 3 n
⎪
⎨ ξ3n
R= for . (8.67)
⎪
⎪
⎪
⎪ ξ 3
⎩1 − 3 n ξ 3 n+1 > ξ 3 n
ξ n+1
A value of R close to zero indicates that the penetration is undergoing very

little change between subsequent iterations which is due to forcing the contact
pressures below an underestimated contact pressure. Therefore, an update of
the a priori estimated contact pressure is conducted for values of R dropping
below a chosen Rc . The used large penetration algorithm in addition to the
contact algorithm following the covariant approach is outlined in Fig. 8.10.
Fig. 8.10 Outline of applied enhanced LP scheme for 3D covariant form

8.4.2 Numerical Examples

A series of numerical examples regarding the above presented formulations
are carried out within this section. They are aligned to examples already
presented in Sect. 8.3 and in [200] in order to show the major benefit of
a covariant velocity description in combination with the large penetration
scheme for 3D contact problems. The first two examples are representing
rather simple setups where the rotational and curvature parts are playing no
major role while the following examples are highly depending on these parts of
the full tangent matrix. All examples represent different setups of frictionless
contact problems between two bodies and demonstrate the effectiveness of
the described large penetration algorithm in combination with the covariant
contact approach.
Within all examples a dimensionless setup using unit lengths and unit
forces is considered. The loading is in general displacement controlled except
in the last example. The contact formulation is carried out using NTS contact
elements consisting of quadrilateral master surface elements with bilinear and
biquadratic approximation.
In order to investigate the efficiency of the proposed algorithm the influ-
ence on the convergence rate is given by the number of equilibrium iterations
of a Newton algorithm accumulated within all load steps for each example.
For a reference all given examples are computed according to the consistent
linearization using the full tangent matrix presented in eqns. (7.17), (7.18)
and (7.19). As discussed in Sect. 8.4.1 using a full tangent matrix the con-
sistent linearization is limited regarding the initial penetration of a loading
step and, therefore, the loading increment enabling quadratic convergence is
restricted for the consistent linearization (CL). Thus, for each example the
upper limit of the loading increment for the full tangent matrix based on con-
sistent linearization is figured out and the mentioned reference computation
is carried out.
The resulting amount of total iterations is used then compared with the
large penetration (LP) scheme for different loading increments and their re-
sulting total number of iterations. Within this contribution only some load
step size variations are presented for each example. The investigations are in-
cluding ranges of the loading increment starting at values computed for the
limits using the full tangent matrix and going up to very large load increments
up to the limits of the large penetration scheme.
8.4.3 Wedge Plate Indented into Cantilever Plate

A wedge plate - height 5.0, bottom edge length 14.0 and top edge length 3.0
(E = 25000.0, ν = 0.0) - is indenting into a cantilever plate as shown in the
system setup in Fig. 8.11(a). The body of the cantilever has the dimensions
height 10.0, length of the bottom edge 8.0, length of the top edge 2.0 and
material properties of E = 2500.0 and ν = 0.0. The thickness of both bodies
is chosen to be 0.3. The wedge is discretized with 450 (15 × 30 × 1) and

the cantilever with 900 (30 × 30 × 1) 8-node structural solid elements with
neo-Hookean material model according to the decoupled free energy with
volumetric and isochoric parts, as described in [170]. The indenting wedge,
thereby, is driven up to a maximum prescribed displacement of 8.0 applied to
its vertical edge while the bottom side is restrained allowing only horizontal
motion. The cantilever is restrained on its top and vertical edge.
cantilever plate
wedge plate
'$
&%
(a) system setup (b) defining cutout area for Fig. 8.12
Fig. 8.11 General setup of wedge plate indenting cantilever plate with linear
approximation
The inclined surface of the wedge is used as the master contact surface
and the surface nodes of the bottom edge and the inclined surface on the
cantilever are taken as slave nodes within the NTS contact approach. A global
searching algorithm is applied where the master surfaces are used with a
linear approximation. The contact parameters are chosen: N = 25000.0 for
the penalty parameter; for the large penetration scheme cut-off pressure pe =
10.0, αc = 1.0 for the update multiplier and Rc = 0.1 for the update indicator.
The investigated load step sizes are 1.0, 2.0, 4.0 and 8.0 - the latter being
the final load - for which the obtained results are summarized in Table 8.4.3.
While the LP scheme is working for load step sizes of up to 2.0, the CL
is achieving convergence only with load step size up to 1.0. Comparing the
complete number of iterations necessary for the load step size 1.0 (8 load
steps) where the CL requires 64 and the LP 206 iterations it is obvious that
the LP has definitely a lower convergence rate within a load step compared
to the CL in cases when the CL converges at all. This is in accordance with
the observations with numerical examples in Sect. 8.3 that quadratic con-
vergence is gained only for the CL and a usage of only part of the tangent
matrix as in the case of the LP scheme (phase 1 - just main part of tangent
matrix) does not enable this rate. However, an observation becoming more
apparent for the following examples is that with increasing the load step size
the amount of accumulative iterations gets reduced within the LP scheme
and often convergence is only achieved with the LP scheme.
Finally, a cutout of the shown setup in Fig. 8.11(b) is presented for a
deformation sequence in Fig. 8.12 with the contact area developing during
the iterations for the first load step - load step size of 2.0 - for the LP scheme.
In the first load step convergence is achieved within 26 iterations. It should
be mentioned that within the example neither the rotational eqn. (7.18) nor
the curvature eqn. (7.19) parts of the tangent matrix were activated, as the
master surface is being placed on the stiffer body with almost no rotation of
the contact element and the used linear approximation enables no curvature
part, respectively.
Table 8.7 Wedge plate indenting into cantilever plate. Influence of CL and LP
scheme by varying the load step size for reaching convergence and accumulative
number of iterations necessary for the prescribed final displacement of 8.0
scheme convergence load step cum. no. it.

CL not achieved 8.0 ∞
not achieved 4.0 ∞
achieved 1.0 64
LP not achieved 8.0 ∞
achieved 2.0 126
achieved 1.0 206
8.4.4 Semicircular Plate Indented into Rectangular

Plate
The second example is a semicircular plate indentor pressed into a rectangular
plate, see Fig. 8.13(a). The geometrical and material data for the rectangular
plate are: width 20.0, length 12.0, thickness 0.6, E = 2500.0 and ν = 0.0.
The stiffer semicircular plate has the properties: radius r = 5.0, thickness 0.6,
E = 25000.0 and ν = 0.0. The meshing is performed with trilinear solid
elements as in the example before where the amount of elements is chosen
512 (32 × 16 × 1) and 343 (consisting of 7 subdivisions of 7 × 7 × 1) for the
rectangular and the semicircular plate, respectively. The semicircular plate is
driven up to a maximum displacement of 2.0 applied on the straight edge in
vertical upward direction while the rectangular plate is completely restrained
along its long edge.
Contact is defined using the bottom surface of the rectangular plate as
master surface and the nodes on the curved edge of the semicircular plate
(a) 0 (b) 1 (c) 2 (d) 3
(e) 5 (f) 10 (g) 14 (h) 26
Fig. 8.12 Development of the contact area for several iteration states of the first
load step for the LP scheme with load step size 2.0 and 26 iterations necessary for
convergence within this first load step
rectangular
plate
'$
semicircular
plate
&%
(a) system setup (b) defining cutout area for Fig. 8.14
Fig. 8.13 General setup of semicircular plate indenting rectangular plate with
linear approximation
as slave points. The penalty parameter is given as N = 25000.0, whereas,

pe = 100.0, αc = 1.0 and Rc = 0.1 are chosen as the parameters for the large
penetration scheme.
Table 8.4.4 contains the results with the analyzed identical load step sizes
in the range of 0.005 (400 load steps overall) up to 2 (final load in 1 load step).
Obviously the LP scheme allows on the one hand to increase the maximum
load step size by a factor of 50 and on the other hand shows a reduced number
of iterations by a factor of about 15. This major advantage appears to be due
to the fact that the rotational part of the tangent matrix becomes active
within this example as the surface of the indented part is chosen as master
surface. For large penetrations this leads to negative entries on the diagonal
of the full tangent matrix (CL) and, therefore, just smaller load steps are
possible. However, the LP scheme using only the main part of the tangent
matrix for large penetrations in phase 1 is not restricted concerning this.
The cutout given in Fig. 8.13(b) is presented in a deformation sequence in
Fig. 8.14 and gives an idea regarding the ”behaviour” of the scheme in the
case of large penetrations. It is obvious that rotations of the master surfaces
of the contact elements are involved, especially, for the first iterations of the
presented load step. Fig. 8.14 represents hereby the computation with the
LP scheme for a load step size of 0.5 (4 load steps overall). Out of the 36
accumulative iterations the first load step requires 11 iteration steps.
Table 8.8 Semicircular plate indenting into a rectangular plate. Influence of CL and
LP scheme on maximum load step for reaching convergence and on accumulative
number of iterations necessary for the prescribed final displacement of 2.0

achieved 0.01 523
achieved 0.005 813
LP not achieved 2.0 ∞
achieved 0.5 36
achieved 0.1 84
achieved 0.01 541
achieved 0.005 830
8.4.5 Bending of a Beam over a Rigid Cylinder

Linear Approximation for Structure and Contact Elements
The following example is following the scheme in Sect. 8.3.1 and is aimed to
extend the convergence study including LP scheme. A beam is bended over
a rigid cylinder as shown in the system setup of Fig. 8.15. The rigid cylinder
possesses an outer radius of 2.0 while the clamped elastic beam has a length
of 24.0 and height of 0.25. The beam undergoing large displacements but
small strains consists of a St. Venant material with an elasticity modulus
of E = 1.0 · 104 and a Poisson ratio of ν = 0.3. The FE discretization is
carried out using 24 ”solid-shell” elements according to [59] with linear shape
(a) 0 (b) 1 (c) 2 (d) 3
(e) 4 (f) 5 (g) 8 (h) 11
Fig. 8.14 Development of the contact area after several iterations of the first load
step for the LP scheme with load step size 0.5 (4 load steps overall) and 11 iteration
steps necessary within first load step for convergence
functions plus some added shell specific enhancements to avoid locking. The
thickness of both bodies is chosen to 1.0. The loading consists of a prescribed
displacement at the free end in vertical direction up to a final value of 9.0
that leads to considerable bending over the rigid cylinder.
Fig. 8.15 General setup of bending a clamped beam over a rigid cylinder with
linear approximation
For contact the bottom surface of the elastic beam is chosen as master
surface, whereas, the outer surface nodes of the rigid cylinder are then the
slave nodes within the NTS contact approach. A global contact searching
algorithm is applied where the master surface is discretized with linear shape
functions. The contact parameters are chosen as N = 1000.0 for the penalty
parameter and the parameters specific for the large penetration scheme are
taken as 0.5 for the cut-off pressure pe , αc = 1.0 for the update multiplier
and Rc = 0.1 for the update indicator.
Table 8.9 Bending of a beam over a rigid cylinder with linear approximation of
the contact surfaces. Influence of CL and LP scheme on convergence varying the
load step size and accumulative number of iterations necessary for the prescribed
final displacement of 9.0

achieved 2.25 36
achieved 1.8 28
achieved 1.0 41
LP achieved 9.0 15
achieved 4.5 23
achieved 3.0 25
achieved 2.25 32
achieved 1.8 43
achieved 1.0 54
As can be seen in Table 8.4.5 the setup of this example is enabling a

maximum load step converging for the CL with 2.25 (4 load steps overall)
while for the LP scheme a value of 9.0, the final prescribed displacement value
within a single load step, is possible. As before, the LP scheme is showing
for smaller load steps high accumulated numbers of iterations compared to
the CL scheme. Fig. 8.16 contains the deformed figure at eight significant
iteration states for the load step size 9.0 including the unloaded (0) and the
last iteration step (15).
Quadratic Approximation for Structure and Contact Elements

The same system setup (see Fig. 8.17) as in the previous example is now used
with quadratic shape functions for the incorporated structural and contact
elements. While the dimensions of the analyzed problem remain as before,
the amount of used structural elements is reduced from 24 linear ”solid-shell”
elements to 12 quadratic elements. The beam not showing large deformations
consists of St. Venant material with an elasticity modulus of E = 1.0 · 105
and a Poisson ratio of ν = 0.0. The loading again consists of a prescribed
displacement at the free end in vertical direction with a value of 9.0.
A global searching algorithm for contact is applied using a quadratic ap-
proximation of the master surface within the NTS contact approach. The
quadratic approximation leads for both the CL and for phase 2 of the LP
scheme, too, (compared to the linear example before in Sect. 8.4.5) to a curva-
ture part of the full tangent matrix. The contact parameters including those
specific for the LP scheme are chosen as in Sect. 8.4.5.
(a) 0 (b) 1 (c) 2 (d) 3
(e) 4 (f) 5 (g) 8 (h) 15
Fig. 8.16 Development of the contact area for several iteration states of the LP
scheme with load step size 9.0 (only one load step) and 15 iterations necessary for
convergence
Fig. 8.17 General setup of bending a clamped beam over a rigid cylinder with
quadratic approximation
As is obvious from Table 8.4.5 a much larger load step size is possible
for the LP scheme (9.0) compared to the CL (2.25). In general, again an
advantage is concerning the accumulated number of iterations with increasing
load step size for the LP. The advantage, hereby, is due to the example not
relying much on the curvature part. Furthermore, in Fig. 8.18 the deformation
during a series of iteration states for the load step size of 9.0 (one load step)
with 20 iterations necessary to reach convergence is shown. As in the previous
example the penetrations are really large for the first 5 iterations.
8.4.6 Bending of a Beam over a Rigid Sphere

A more general case to examine the influence of all parts of the contact
tangent matrix is to consider contact with a doubly curved body. Influence of
various parts of the tangent matrix on the convergence for the NTS studied
Table 8.10 Bending of a beam over a rigid cylinder with quadratic approximation
of the contact surfaces. Influence of CL and LP scheme on convergence for varying
load step size and accumulative number of iterations necessary for the prescribed
final displacement of 9.0

achieved 2.25 35
achieved 1.8 40
achieved 1.0 39
LP achieved 9.0 20
achieved 4.5 18
achieved 3.0 29
achieved 2.25 31
achieved 1.8 34
achieved 1.0 40
(a) 0 (b) 1 (c) 2 (d) 3
(e) 4 (f) 5 (g) 7 (h) 20
Fig. 8.18 Development of the contact area for several iteration states of the first
load step for the LP scheme with load step size 9.0 (only one load step) and 20
iterations necessary for convergence
for this example in Sect. 8.3.2. The clamped elastic beam of the previous
example, but with a width of 4.0, is now bending over a rigid sphere. The
general setup is shown in Fig. 8.19. The rigid sphere possesses a radius of 4.0
while the clamped elastic beam has a length of 24.0 and a height of 0.25. The
center of the sphere is positioned at 0.5 from the edge of the beam and at 12.0
along the beam measured from the clamping. The rigid sphere is modeled with
512 bilinear elements. The beam not showing large deformations consists of
St. Venant material with an elasticity modulus of 1.0 · 105 and a Poisson ratio
of 0.0. The FE discretization is carried out using 12 ”solid-shell” elements

according to [59] with quadratic shape functions. The loading consists of two
asymmetric forces F1 = 0.8 and F2 = F1 /4 applied at the nodes of the free
end, see Fig. 8.19.
F1
F2
Fig. 8.19 General setup of bending a beam over a rigid sphere with quadratic
approximation
Concerning contact the bottom surface of the elastic beam is taken as

master surface, whereas, the surface nodes of the rigid sphere are considered
as slave nodes within the NTS contact approach. Again a global searching
algorithm was involved where the master surfaces are used with a quadratic
approximation. The contact parameters are chosen as N = 500.0 for the
penalty parameter, and, the parameters specific for the large penetration
scheme are given as 1.0 for the cut-off pressure pe , αc = 1.0 for the update
multiplier and Rc = 0.1 for the update indicator.
In this example bending and twisting of the beam is observed and, there-
fore, the rotational and curvature (hij ) parts of the tangent matrix are fully
active for the consistent linearization. In Table 8.4.6 the result of the analysis
with biquadratic ”solid-shell” elements and biquadratic contact elements for
the beam are given. A maximum load step of 0.00625 is applicable for the
CL scheme. The LP scheme is showing a much better performance allowing
to increase the maximum load step size by a factor beyond 100. The amount
of accumulative iterations is reduced by a factor of about 10 compared to the
CL scheme. The extreme penetrations allowed during the first iterations of
the LP scheme are visualized in Fig. 8.20 where the deformations are shown
for various iteration states.
8.5 Frictional Contact Analysis: Node-To-Segment Approach 249
Table 8.11 Bending of a beam over a rigid sphere with quadratic approximation
of the contact surfaces. Influence of CL and LP scheme on convergence for varying
load step size and accumulative number of iterations necessary for the final load
F1 = 0.8

achieved 0.00625 394
achieved 0.003125 704
LP achieved 0.8 36
achieved 0.4 40
achieved 0.2 67
achieved 0.1 65
achieved 0.05 91
achieved 0.025 144
achieved 0.0125 231
achieved 0.00625 397
achieved 0.003125 706
8.4.7 Discussion
The covariant contact description proved to enable a clear setup based on
the consistent contact tangent matrix where ”main”, ”rotational” and ”cur-
vature” parts are easily distinguishable. The large penetration scheme highly
relies on this specific setup. The numerical examples show that in both con-
sidered cases, linear and quadratic shape functions for the master surface
within the NTS approach, the investigated algorithm is very efficient. The
efficiency even increases with the complexity of the analyzed examples re-
garding the influence of rotational and curvature parts of the tangent matrix
on the convergence.
8.5 Frictional Contact Analysis: Node-To-Segment

Approach
In this section a series of numerical examples of frictional contact prob-
lems between a flexible structure and rigid surfaces of second order (cylinder
and sphere) are investigated. Bilinear and biquadratic contact elements are
(a) 0 (b) 1 (c) 12 (d) 14
(e) 17 (f) 20 (g) 23 (h) 36
Fig. 8.20 Development of the contact area for several iteration states of the LP
scheme with load step size 0.8 (only one load step) and 36 iteration steps necessary
for convergence
used for parameterization of both contact surfaces and the Node-To-Segment

(NTS) contact approach shown in Sect. 8.2.1 and in Sect. 8.2.3 are realized.
8.5.1 Global Solution Scheme. Summary of the

Results
Summarizing the theoretical discussion about the covariant description, we
present the global solution scheme for the numerical implementation in
Table 8.12 and 8.13. All parts of tangent matrices contain either a term
(δrs − δρ), or a term δρ,j , resp. terms (vs − v) and v,j , and, therefore, can
be algorithmically computed. For discretization of any surface only two po-
sition matrices A and Aξ are necessary. The proposed approach has been
implemented in the FEAP code see [173], ”solid-shell” elements are used for
modelling of elastic structures, see [59] and [58].

Contact Surfaces. Two Types of a Frictional
Contact Problem
During the solution of the frictional problem, it is necessary to solve the
evolution equation (6.19) with a return-mapping algorithm, as described in
(6.24), (6.25) and (6.26). As was mentioned in Remark 6.4, Sect. 6.1.2, it is
important to know the value of the coordinate increments Δξ i and, therefore,
the displacement increments Δui in order to capture the “sticking-sliding”
zone correctly. As a representative example for a-priori estimation of the
Table 8.12 Global solution scheme. Summary of the results for numerical
implementation.
1. Initialization of convective coordinates ξ i .

The Closest Point Projection procedure on a surface
i
with no external loads gives ξ(0) .
2. Loop over load increments and Newton iterations
for the contact integral
+ +
δWc = s N δgds + s Tj δξ j ds where δξ j = aij (δrs − δρ) · ρi
3. Loop over all contact elements and all contact points
• compute projection points ξ(n) i
(CPP for a surface)
• Check penetration g = (rs − ρ) · n. If g > 0 then exit loop 3.
• Compute contact tractions and corresponding tangent matrices
Normal traction: N = N g
Tangent matrix KN for normal traction is defined via
+
s N (δrs − δρ) · (n ⊗ n)(vs − v)ds
+
− s N aij δρ,j · (n ⊗ ρi )(vs − v) + (δrs − δρ) · (ρj ⊗ n)v,i ds

−
s N
g (δrs − δρ) · hij (ρi ⊗ ρj )(vs − v)ds
Tangential traction Ti is defined via the return-mapping algo-

rithm.
(n+1) (n) (n) (n+1)
Trial step: Ti = Tk akj(n) (ρj · ρi )−
⎧
⎪
⎪ (Δρ · ρi ) for node-to-surface (NTS) and
⎨ surface-to-surface (STS) approaches, where
−T · Δρ = ρC (n+1) |ξ(n+1) − (ρC (n) + uC (n) ) |ξ(n)
⎪
⎪
1 2
, ξ(n+1) 1 , ξ2
⎩ (n)
−(u · ρi ) for segment-to-analytical surface (STAS) approach
Coulomb friction law:

(n+1) (n+1) ij
Φ(n+1) = Ti Tj a(n+1) − μN (n+1)
Return-mapping step see Table 8.13.
• Compute residual R from the contact integral in 2

• Compute the full contact tangent matrix K = KN + KT
Table 8.13 Return-mapping scheme and tangent matrices for tangential traction.
(n+1) ≤ 0
if Φtr
sticking condition
(n+1) (n+1)
Ti stick = Ti
Tangent matrix KT

−εT s
(δrs − δρ)aij ρi ⊗ ρj (vs − v)ds

− s Ti (δrs − δρ) ail ajk ρk ⊗ ρl vj + δρj aik ajl ρk ⊗ ρl (vs − v) ds

+ s
Ti hij (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v)ds.
if Φtr
(n+1) > 0
sliding condition
(n+1)
(n+1) (T tr )i
Ti = μN (n+1)
slide
Ttr
(n+1)

Tangent matrix KT

N μTi aij
− (δrs − δρ) ρj ⊗ n(vs − v) ds
s T

T μ|N |aij
− (δrs − δρ) ρi ⊗ ρj (vs − v) ds
s T

T μ|N |Ti Tj aik ajl
+ (δrs − δρ) 3
ρk ⊗ ρl (vs − v) ds
s T

μ|N |Ti
− (δrs − δρ) ail ajk ρk ⊗ ρl vj + δρ,j aik ajl ρk ⊗ ρl (vs − v) ds
s T

μ|N |Ti ij
+ h (δrs − δρ) · ρj ⊗ n + n ⊗ ρj (vs − v) ds
s T

μ|N |Ts Tl δξ s jl 3

+ −a Tm Γjn
m ˙n
ξ + ajl Tm hmj ξ̇ ds.
s T 3
value of incremental displacements, the stresses in an infinite layer have to

be considered, see Fig. (8.21). Both a vertical displacement h and a horizontal
displacement u are applied at the upper boundary. During the deformation
the rectangle ABCD is changing into a parallelogram AB1 C1 D. Under the
assumption of linear elasticity and a plane strain deformation, the stresses in
the layer are obtained via superposition of the normal compressive stress σ
and the pure shear stress τ :
Remark 8.4. Curvature parts in boxes (Table 8.12 and 8.13) can be omitted
with very little loss of efficiency
E h E u E
σ=ε = ; τ = γG = . (8.68)
1−ν 2 b 1 − ν2 b 2(1 + ν)
Now we assume Coloumb friction with μ as a friction coefficient at the lower
boundary. Sliding starts if the condition τ = μσ is fulfilled. Thus, the con-
dition of sticking of the thin layer can be estimated by the following ratio:
γ(1 − ν)
≤ μ, (8.69)
2ε
from which we obtain the threshold value of the horizontal displacement u:
2μh
ucr = . (8.70)
1−ν
u1
0 0
1
0
1 0
1
0
1
11100
000 110
1
111
000
0
1
0 1
1 0
0
1 0
1
0
1 0
1
0
1
0
1
01
1 0
10
1
01
010
10
1
010
1
0
100
11
0 1
1
00
11 0
01
10
01
10
01
10
01
10
01
10
01
10
01
10
01
1 0
01
10
011
100
0
1
0 1
00
11
1 0
01
10
01
1 0
0
1
0 0
1
0000000000000
1111111111111
0
1
B 1111111111111
0000000000000
0
1 C
111111
0000000000000000000
1111111111111
0
1
0000000000000
1111111111111
0
1
0000000000000
1111111111111
0000000000000
1111111111111
B1
0000000000000
1111111111111 C1 b
h 0000000000000
1111111111111
0000000000000
1111111111111
A 1111111111111
0000000000000
0000000000000
1111111111111 D
0
1 0
10
1 0
10000000000000
1111111111111
0
10
100
11
0
1 0
10
10
10
10
10
10
10
1
01
0
01
0
011
00
0
1
0 1
0
01
0
01
0
01
1 01
0101
01
011
00
0 1
1 01
01
01
01
01
01
01 1100
11
1 11 0
1
Fig. 8.21 Plane deformation of a layer.
One can see from the infinite layer, that sliding starts immediately at the
complete lower boundary. However, though this is not a case for a finite
dimensional block, or an arbitrarily thin layer, where a developing zone of
sticking and sliding exists, eqns. (8.69) and (8.70) can be used as a rough
estimation of the presence of the sticking condition, and, therefore, for the
estimation of the displacement increments.
As an example for the computation, we consider a rectangular block (Fig.
8.22) with the following parameters: elasticity modulus E = 2.1 · 104, Poisson
Y u
v flexible
X
rigid
Fig. 8.22 Sliding block on the base. Meshed surfaces. STS contact approach.
0.008
0.0030
0.0060
0.0075
0.007 0.0090
0.0100
0.0110
0.0120
0.006
0.005
Displacement, U
0.004
0.003
0.002
0.001
0
0 2 4 6 8 10 12 14 16 18 20
Coordinate, X
Fig. 8.23 Horizontal displacements of the contact surface for various states of the
displacement loading.
ratio ν = 0.3, length a = 20, height b = 5, thickness c = 0.5. The dimension

system is assumed to be consistent. The lower supplementary block is added
to model a rigid base. Coloumb friction with a coefficient μ = 0.3 is specified
between two bodies. The contact surface of the upper block is assumed to be a
“master”, while the upper surface of the lower block is a “slave” surface within
the “segment-to-segment” approach. The penalty parameters are chosen as
εN = εT = 2.1 · 106 .
Since the problem is path-dependent, we will investigate a case when dis-
placements at the upper edge are applied in two steps: at the first step, a
vertical displacement v = −7.0 · 10−3 is applied, then, a horizontal displace-
ment is applied incrementally. Here, we should mention that initial conditions
for the history variables are defined at the zero load step with zero external
loads, i.e. the initial projection points are the sticking points, see step 1 in Ta-
ble 8.12. An estimation of the critical horizontal displacement in eqn. (8.70)
0.3
0.0030
0.0060
0.0075
0.0090
0.2 0.0100
0.0110
0.0120
0.1
0
Ratio Fx/Fy
-0.1
-0.2
-0.3
-0.4
0 5 10 15 20
Coordinate, X
Fig. 8.24 Reaction forces ratio Fx /Fy on the contact surface for various states of
the displacement loading.
gives ucr = 6.0 · 10−3 , so in order to capture the sticking-sliding zone we

choose a displacement increment Δu = 2.5 · 10−4 and apply it in 100 load
steps. Our aim in the first computation is to show the development of the
sticking-sliding zone. In order to verify this zone carefully we will consider a
plot of the horizontal displacements and a plot of the reaction forces ratio on
the boundary Fx /Fy . Of course, this zone is precisely specified by the return-
mapping algorithm, but we are interested in various parameters. Fig. 8.23
contains the spatial distribution of the horizontal displacements at the lower
boundary, if the following displacements u = 3.0 · 10−3 ; 6.0 · 10−3 ; 7.5 · 10−3 ;
9.0 · 10−3; 10.0 · 10−3; 11.0 · 10−3; 12.0 · 10−3 are applied at the upper bound-
ary. As shown in the corresponding reaction forces ratio diagram in Fig. 8.24,
sliding starts from u = 7.5 · 10−3 , when the ratio Fx /Fy = −0.3 is reached.
The block is considered to be sliding at the full lower boundary, when the
applied displacement reaches the value u = 11.0 · 10−3. We can also conclude
that the estimation of the threshold displacement given by eqn. (8.70) is a
good approximation.
The spreading of the zone of sliding is found to be within a relatively short
interval of loading. In some practical problems, as e.g. metal forming, the
energy loss due to large sliding can be more important. Assume for the next
discussion that the stresses are approximated by eqn. (8.68) for the finite-
dimensional block with size AD = a, e.g. for a very long block. Then the
elastic energy accumulated at the critical state in the block has the following
form:
σεab τ γab Eah2 2μ2
Eel = + = 1+ . (8.71)
2 2 2b(1 − ν 2 ) 1−ν
If the sliding process is developing, when the block is dragged along the
distance l, then the work of the critical sliding stresses τsl , is evaluated as:
τsl al Eahlμ
Esl = = . (8.72)
2 b(1 − ν 2 )
It is obvious, that during large sliding a thin layer along a relatively large dis-
tance l, and the dissipation of energy due to sliding, eqn. (8.72) can be rather
more important then the initial threshold value. Thus, frictional problems
can be subdivided into two problems:
a) compute the global threshold value for sliding and the development of
the distribution of the sticking-sliding zone;
b) compute forces which are necessary to drag the structure under the
assumption of full sliding.
Obviously, for the first problem the evolution equation (6.19) must be com-
puted with small steps within the return-mapping algorithm, but for the
second problem the sticking zone is out of interest and for the analysis rela-
tively large steps can be taken. Such problems are certainly present in forming
processes with large plastic deformations. In order to show an example for
the problem type b), another analysis is performed with a displacement in-
crement Δu = 12.0 · 10−3 , which is even larger than the critical one and
corresponds to the developed sliding zone, see Fig. 8.23. In order to compare
the influence of the various parts of the tangent matrix we compute two cases
1) with the full tangent matrix;
2) only with the main part of the tangent matrix.
The penalty parameter is chosen as εN = εT = 2.1 · 105 . Table 8.14 shows
the comparison of the numerical results between both cases by the number of
iterations per load step. As we can see in the developed sliding region the full
matrix in comparison with only the main matrix leads to a reduction of the
number of equilibrium iterations per load step from 4 to 3. We should men-
tion that for the previous example during the incremental horizontal loading
there is no difference between the number of equilibrium iterations for both
cases. Obviously this is due to the fairly small load steps. Thus, as expected,
keeping all parts of the matrix appears to be only necessary in the case of
large load increments.
Fig. 8.25 shows the spatial distribution of the relative horizontal displace-
ments u − uapplied at the lower boundary if the displacement at the upper
boundary is taken exemplarily as u = 0.012, then u = 0.048 and finally
u = 0.120. It is obvious, that the relative horizontal displacements hardly
change during the fairly large sliding process.
-0.005
0.0120
0.0480
0.1200
-0.006
Relative displacement -0.007
-0.008
-0.009
-0.01
-0.011
-0.012
0 2 4 6 8 10 12 14 16 18 20
Coordinate, X
Fig. 8.25 Relative horizontal displacements of the contact surface for various states
of the displacement loading.
Table 8.14 Sliding of a block. Bilinear elements. Segment-to-segment (STS) con-

tact approach. Influence of various contact stiffness parts on convergence. Case 1:
full matrix; case 2: only main matrix. Comparison of no. of iterations in all load
steps (l.s.)
Case 1 Case 2
No. l.s. No. Cum. No. l.s. No. Cum.
1 4 4 1 5 5
2 6 10 2 6 11
3 5 15 3 5 16
4-20 3 66 4-20 4 84
8.5.3 Sliding of a Block. Quadratical Approximation

of the Contact Surfaces
Since general smoothing techniques for contact surfaces are out of the scope
of this article, in this example we will use contact elements with quadratical
approximation of the master surface together with a specially chosen geom-
etry of both contact bodies in order to preserve C 1 -continuity of the contact
surfaces. Namely, we consider contact between a parabolic block sliding on a
parabolic cylindrical base, see Fig. 8.26. The block is meshed with 18-node
solid-shell elements with density 20 × 1 × 5. Both master and slave contact
surface geometries are satisfying the equation
z = c · x2 , with c = 0.03 . (8.73)

The contact is modeled by the node-to-surface approach with the master

surface from the parabolic block. The parabolic slave surface of the fixed
base is represented by slave nodes with the same mesh density as the master,
which are not shown in Fig. 8.26. The geometrical parameters are H = 5,
L = 10; the material is linear elastic with Young’s modulus E = 2.1 · 104 ,
Poisson ratio ν = 0.3, Coulomb friction coefficient μ = 0.3.
w u
Z
H
−L
L X
Fig. 8.26 Sliding of a parabolic block on a parabolic base. Meshed block. NTS
contact approach.
In the case of contact with a curvilinear surface, even with homogeneous

loading, zones with sticking and sliding can be present. One can expect from
the rigid body mechanics that the sliding zone during vertical loading w
in the current example is satisfying the following condition |x| > 5. From
the friction cone for the parabolic cylinder follows that: tan α|x=5 = z =
2 · 0.03x|x=5 = 0.3. In order to inspect this effect in the deformable body, we
apply at the upper edge the vertical displacement w = 0.007 in 7 load steps. In
Fig. 8.27 the distribution of the tangent displacement uτ = ux cos α+uy sin α
over x on the contact surface is depicted, where now and for the next exam-
ple the angle α is computed in the reference configuration. For the further
discussion we will distinguish based on the OZ axis, the left sliding zone
with negative displacements and the right sliding zone with positive dis-
placements. The tangential displacements from both zones are directed to-
wards the OZ axis, therefore the distribution looks mirror-symmetric. One
can see that the sticking zone is approximately satisfying the condition
|x| < 3.
0.00125
1.0E-03
2.0E-03
0.001 3.0E-03
4.0E-03
5.0E-03
0.00075 6.0E-03
7.0E-03
0.0005
Displacement, U
0.00025
-0.00025
-0.0005
-0.00075
-0.001
-0.00125
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Coordinate, X
Fig. 8.27 Distribution of the tangential displacements in OX direction. Parabolic

cylinder. NTS contact approach. Incremental vertical loading.
Equidistant Motion on a Cylinder
As continuation of the numerical example, we choose an equidistant motion of

the upper edge of the parabolic block at the distance h from the generatrix of
the parabolic base. The curve r of this motion satisfies the following equation:
r = ρ + hn, (8.74)
which for the parabola (8.73) can be written as:

⎧ ⎫ ⎧ ⎫
⎨ x ⎬ ⎨ −2cx ⎬
h
r= 0 +√ 0 , (8.75)
⎩ 2⎭ 1 + 4c2 x2 ⎩ 1 ⎭
cx
where n is a normal on the initial curve, and h is an initial vertical displace-

ment. From eqn. (8.75) it is clear that the trajectory of the body is no longer
a parabola. If the curvature of the cylinder is small, i.e. c << 1, then we
consider a Taylor expansion with a linear term for the first coordinate and
with a quadratic term for the second coordinate. Thus, we obtain as a first
approximation of the trajectory in eqn. (8.75) a parabolic motion in the form:
⎧ ⎫
⎨ x(1 − 2ch) ⎬
r= 0 . (8.76)
⎩ ⎭
h + cx2 (1 − 2ch)
In this displacement driven problem the parabolic block is moving in the X-Z
plane, providing an approximately constant compression.
Next, the loading is applied in two steps also: the first step is a vertical
loading with w = −h = 0.007, then both a horizontal and a vertical loading
are incrementally applied at the upper edge with Δu = Δx = 2.5 · 10−5
according to eqn. (8.76), providing the equidistant motion of the parabolic
block. Now, two phases of the development of the sticking-sliding zone can
be observed. The first phase corresponds to the situation when the right
sliding zone disappears during horizontal loading, as presented in Fig. 8.28
for the following load steps: u = 1.0 · 10−3 , 2.0 · 10−3 , 3.0 · 10−3 , 5.0 · 10−3 .
Fig. 8.29 a) shows scaled deformed and undeformed states when only vertical
displacements are applied and, therefore, the two sliding zones are symmetric.
The configuration with the vanishing resp. vanished sliding zone on the right
side is presented in Fig. 8.29 b). This moment can be detected also from
the reaction forces ratio diagram Fτ /Fn in Fig. 8.31, where the right part of
the sliding zone is also disappearing with Fτ /Fn = 0.3. The second phase is
the spreading of the left sliding zone through the contact surface shown in
Fig. 8.30 exemplarily for u = 5.0 · 10−3 , 10.0 · 10−3 , 15.0 · 10−3 , 20.0 · 10−3 ,
25.0·10−3, 30.0·10−3. Here the zone without contact is detected as a zone with
zero normal nodal forces fn = 0 starting at a loading with u ≥ 1.52 · 10−2 .
These sub-zones are marked with thicker lines in Fig. 8.30.
0.005
0.000
1.0e-3
2.0e-3
3.0e-3
0.004 5.0e-3
0.003
Displacement, U
0.002
0.001
-0.001
-0.002
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Coordinate, X
Fig. 8.28 Parabolic cylinder. Horizontal loading. Distribution of the tangential

displacements in X direction. Phase 1 — vanishing right sliding zone.
Again we now compare the influence of the various parts of the tangent
matrix on the convergence rate when the applied displacements correspond
to the developed sliding. Namely, the load is applied in 20 load steps with the
(a) ux = 0.0
(b) ux = 5.0 · 10−3
Fig. 8.29 Parabolic cylinder. Initial vertical loading. Undeformed and deformed
states with additionally applied horizontal displacement a) u = 0.0 – two sliding
zones; b) u = 5.0 · 10−3 – only left sliding zone. (Displacements scaled: 250 times
in x-direction, 40 times in z-direction.)
displacement increment Δu = 4.0 · 10−2 , providing a fully developed sliding

motion from the first step on. The following cases are shown in table 8.15:
1) full tangent matrix;
2) without curvature parts;
3) only with main part of the tangent matrix.
We see that excluding the curvature matrix leads to a minor reduction of the
convergence, while excluding the rotational part too causes a considerable
increase of the number of equilibrium iterations per load step. We should
also mention that during the analysis of the threshold value before full sliding
the number of equilibrium iterations remains the same for each case due to
a small load step. Thus, the computation with the rotational part is more
important for the developed sliding problem of type b.
0.025
5.0e-3
10.0e-3
15.0e-3
20.0e-3
0.02 25.0e-3
30.0e-3
0.015
Displacement, U
0.01
0.005
-0.005
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Coordinate, X
Fig. 8.30 Parabolic cylinder. Horizontal loading. Distribution of the tangent dis-
placement over X-coordinate. Phase 2 — spreading of the left sliding zone.
0.6
0.000
1.0e-3
0.5 3.0e-3
5.0e-3
10.0e-3
0.4 15.0e-3
30.0e-3
0.3
0.2
ratio, Fx/Fy
0.1
-0.1
-0.2
-0.3
-0.4
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Coordinate, X
Fig. 8.31 Parabolic cylinder. Horizontal loading. Reaction forces ratio Fτ /Fn on
the contact surface for various states of loading.
8.5.4 Discussion
In this Section devoted to the frictional examples, it was shown in the
numerical examples that frictional contact problems can be subdivided into
two types. The first type contains the development of a sticking-sliding zone.
In this case, small loads steps, which can be estimated by considering an
elastic layer under friction conditions, are necessary. In due course, it appears
that then the differences in the convergence rate between computations with
various tangent matrix are meaningless. For the second type a fully developed
8.6 Computation of Contact Integrals – Mortar Type Contact 263
Table 8.15 Full sliding of a parabolic block. Biquadratic elements. Node-to-

segment contact approach. Influence of various contact stiffness parts in tangent
matrix on convergence. Case 1: full matrix; case 2: without curvature parts; case 3:
only main matrix. Comparison of no. of iterations in all load steps (l.s.)
Case 1 Case 2 Case 3

No. l.s. No. Cum. No. l.s. No. Cum. No. l.s. No. Cum.
it./l.s. No. it. it./l.s. No. it. it./l.s. No. it.
1 6 6 1 6 6 1-9 6 54
2 5 11 2 5 11 10-12 7 75
3 5 16 3 5 16 13-15 8 99
4-18 4 76 4-16 4 68 15-17 9 126
19-20 5 86 17-20 3 88 18-20 10 156
sliding is assumed and, therefore, fairly large steps beyond the threshold
value can be applied. In this case it is important to compute with the matrix
containing the rotational part. Keeping the curvature matrix leads only to a
small improvement.
8.6 Computation of Contact Integrals – Mortar Type

Contact
Within the finite element method the contact integral in eqn. (5.9) leading to
the residual, as well as the integrals leading to the consistent tangent matrix,
see eqns. (7.9a), (7.9b), (7.9c) for non-frictional contact, and see enqs. (7.24a),
(7.24b), (7.24c) for sticking part and eqns. (7.35a), (7.35b), (7.35c), (7.35d),
(7.35e), (7.35f) for sliding part as additional integrals for frictional contact,
have to be computed using one or another quadrature formula. In the most
common approach known as “node-to-segment” NTS technique the value at
the nodes from the finite element discretization of the “slave” part is taken
directly. As is well known, this technique can be only directly applied in the
case of linear approximations for both “slave” and “master” parts, see [195],
and it does not satisfy the patch test, see [30]. This fact can be explained
as under-integration of the contact integral, because the Lobatto quadrature
formula with only two integration points gives exactly a nodal collocation
formula in this case. As an improvement different quadrature formulae of
higher order can be used. In this situation the question arrises: How many
integration points have to be taken in order to achieve a certain error bound?
The usual formula to estimate the integration error does not give a correct
answer, because it requires differentiability of the integrand up to a certain
order. This is not the case for the computation of the contact integral, which
is discussed in the following: the function in the integral is defined on the
master element, but the computation of the integral has to be done over the
unknown slave surface. In practice, the penetration of “slave” points, e.g.

integration points, from different “slave” segments into the master segment
is checked, see Fig. 8.1. This can be considered as integration of auxiliary
functions over the known master surface which again define a function which
is discontinuous on the master surface.
Remark 8.5. The “Segment-To-Segment” approach based on the projection of

integration points onto the master segment considered in the current section is
known also as the Mortar method with a penalty regularization of contact
tractions, see Fischer and Wriggers [40] (2005).
8.6.1 Convergence Test for the Integration

Algorithm: Computation of the Energy
Associated with the Penalty Functional
As a representative example to show that the problem of integration of dis-
continuous functions arises during the ”master-slave” approach we consider
the classical Hertz problem. Assume that the contact problem of a cylin-
der and a semi-infinite elastic plane, see Fig. 8.32, is solved by the standard
penalty approach for the finite element method. Let CD be a 2D contact
element. Controlling the process with an applied vertical displacement h,
the cylinder penetrates into the plane within the first iteration as shown in
Fig. 8.32. A characteristic quantity for the satisfaction of contact is the value
of the energy associated with the penalty form in e.g. the first iteration. It
has the following form:
#
1 2 εN
Eg = εN g N dx = J . (8.77)
2 AB 2
The value of the integral J in eqn. (8.32) can be evaluated in closed form,
because both the penetration gN and the contact zone AB are defined from
the specific geometry as:

gN = R − h − R2 − x2 , AB = 2 h(2R − h) . (8.78)
The integral J after evaluation and some transformations has the following
form:
# +√h(2R−h)
J= √ (R − h − R2 − x2 )2 dx = (8.79)
− h(2R−h)

2 h(2R − h)
= (3R2 − 2Rh + h2 ) h(2R − h) − 2R2 (R − h) arcsin .
3 R
O X
11
00 1
0
0
1 1
0
0
1 1
0
0
1 1
0
00
11 0
1
C A B D
Fig. 8.32 Cylinder penetrating into plane during the first iteration
Within the finite element solution the value of the integral J becomes:
#
J= gN 2 dx, (8.80)
CD
where denotes the Macauley brackets in the form

0, if gN > 0
gN = (8.81)
gN , if gN ≤ 0
The value of the integral in eqn. (8.80) is computed over the contact element
CD, while the contact region AB is detected via the integration points, which
then leads in general to a discontinuous function defined over the contact ele-
ment CD. Before comparing the results, we describe in the following section
one of the techniques to integrate discontinuous functions [32].
8.6.2 Integration Schemes with Subdomains

The a-priori error estimation in the case of the application of Gauss quadra-
ture rules for discontinuous functions is a rather complicated question, be-
cause it is necessary to know the behavior of the integrand, see e.g. [32].
However, this is in general not known in the considered cases of rather
general contact surfaces. One can only expect, that increasing the number
of integration points leads to a reduction of the integration error. As an
improved and efficient technique to decrease the integration error a subdivi-

sion of the integration area into subdomains together with lower order inte-
gration in each subdomain can be used (see e.g. [32]). With the same number
of integration points as for a standard Gauss integration this technique leads
to a smaller integration error, as is shown in the following.
Let A be an area of element. In the case of a quadrilateral contact element
with area A, this can be subdivided into non-overlapping subdomains Aij :
4
A= Aij . (8.82)
ij
η
n
ηj
j+1
ξi
ξ
2i
1
m
2 2
m
1
0 1 2 i i+1 m
Fig. 8.33 Subdivision of a contact segment into subdomains for integration
Now we consider the subdivision of A with the local coordinate system

ξ, η into rectangular subdomains, see Fig. 8.33, e.g. into m parts along the ξ
axis and into n parts along the η axis. In each subdomain a separate local
coordinate system ξi , ηj , which has to satisfy the following conditions, is
introduced:
2i 2(i + 1)
ξi = −1 if ξ = −1, ξi = 1 if ξ = −1, i = 0, 1, 2, ...m−1 (8.83)
m m
and
2j 2(j + 1)
ηj = −1 if η = − 1, ηj = 1 if η = − 1, j = 0, 1, 2, ...n − 1
n n
(8.84)
Then, the transformation of the coordinates ξi , ηj into ξ, η can be written as:
ξi 2i + 1 ηj 2j + 1
+
ξ= − 1, η= + − 1. (8.85)
m m n n
Finally, the integration of a function f (ξ, η) over the area A in the local
coordinate system ξ, η leads to a sum of integrals over each subdomain Aξi ,ηj :
# +1 # +1 #
n−1
m−1
f (ξ, η)dξdη = f (ξ, η)dξdη = (8.86)
−1 −1 i=0 j=0 Aξi ,ηj
m−1 n−1 # #

1 +1 +1 ξi 2i + 1 ηj 2j + 1
= f + − 1, + − 1 dξi dηj .
m · n i=0 j=0 −1 −1 m m n n
Each integral in formula (8.86) is computed via standard quadrature formu-

lae, e.g. Gauss integration.
In order to consider general features of the proposed approach, first the
exact value of the integral J in eqn. (8.79) is compared with the computed
value in (8.80) for a line contact. The following parameters are taken: radius of
cylinder R = 2.0, vertical displacement h = 0.1, length of element CD = 2.0.
In this case as exact value of the integral J = 6.7099 · 10−3 and as length of
the contact zone AB = 1.249 are obtained. The following relative error e is
used for comparison of the computed value Jcom with the exact value J:
Jexact − Jcom
e= · 100% . (8.87)
Jexact
Table 8.16 shows the relative error e in the case of various numbers of Gauss
points and subdivisions.
As expected, the formula with subdivisions leads to a smaller error than the
standard single domain Gauss formula. It is obvious that among the formulae
with a fixed total number of integration points the smallest error is obtained
by the formula that combines both the maximum number of subdivisions and
the maximum number of Gauss points which can be independently chosen.
After the description of the contact elements in the following section we
will show that the proposed approach allows first to diminish the error for
the patch test and second to improve the quality of the results, e.g. the load-
displacement curve.
Table 8.16 Relative error in energy of the contact integral for the penalty formu-
lation; comparing standard Gauss integration with subdivisional Gauss integration
in subdomains
No. of Gauss points No. of subdivisions e %

2 1 93.4232
3 1 -32.4737
5 1 5.4618
6 1 -2.4670
3 2 4.7420
2 3 -0.1113
7 1 -1.0916
10 1 -0.6844
5 2 0.2399
2 5 0.2018
20 1 -0.0669
10 2 0.1147
5 4 0.0978
4 5 0.0153
2 10 0.0271
40 1 0.0114
20 2 0.0196
10 4 -0.0137
8 5 -0.0021
5 8 0.0001
4 10 0.0029
2 20 0.0142
8.7 Segment-To-Segment (Mortar) Approach: Analysis

of the Patch Test
In the current section the Segment-To-Segment STS (Mortar) approach with
various orders of approximations the contact surfaces is shown for the Patch
Test problem and is compared with the NTS contact approach. Integration
schemes using subdivision into subdomains developed in Sect. 8.6.2 are ap-
plied together with the STS contact approach. Thus, the STS approach with
various numerical schemes is applied for the standard patch test with lin-
ear approximations, then the STS approach with smooth contact surfaces is
studied to satisfy the Contact Patch Test.
8.7 Segment-To-Segment (Mortar) Approach: Analysis of the Patch Test 269
8.7.1 Classical Contact Patch Test – Linear

Approximations
One of the important problems for contact algorithms is the ability to transfer
the stresses correctly through the contact surface. The contact patch test
serves to check the ability to transfer an uniform stress state through the
contact surface. Different techniques were proposed to pass the contact patch
test. Taylor and Papadopoulos [174] proposed the two-pass algorithm based
on interchanging the master and slave parts to pass the patch test in the
case of a linear approximation for 2D problems. Zavarise and Wriggers [202]
proposed the integration over overlapping regions in the 2D case in order to
more accurately treat the contact conditions. Crisfield [30] considered contact
elements with higher order approximation to satisfy a patch test in the 2D
case. Also in the 2D case the integration of the contact integral over the
overlapping zone, which is constructed by projection was investigated in El-
Abbasi and Bathe [36] to satisfy the patch test. Jones and Papadopoulos [80]
considered a special pressure interpolation in the overlapping region of the
two contact elements to pass the patch test in 3D. Heinstein and Laursen [65]
developed an algorithm based on the construction of a special 3D element
in the overlapping region for mesh-matching problems as well as to pass the
patch test.
Here we consider the application of the Segment-To-Segment contact ap-
proach with various integration schemes to the modified patch test problem,
originally proposed in Crisfield [30] for the 2D patch test. The upper block
with the dimensions 1 × 1 × 0.5 is meshed with a regular rectangular 3 × 3 × 2
mesh. The lower block has the same geometry as the upper block, and a finer,
but distorted 6 × 6 × 2 mesh is used, see Fig. 8.34. Both blocks are made of
elastic material with the following parameters: Young’s modulus E = 1.0·105,
Poisson ratio ν = 0.3. The value of the penalty is chosen as ε = 1.0 · 107 .
During contact the upper block is considered as a slave. An uniform vertical
displacement of Δ = 0.05 is applied on the top surface.
The integration algorithm based on integration of subdomains is an ap-
proximate approach to integrate discontinuous functions. Here we show, that
with this technique it is possible to construct a sequence of results with di-
minishing error to finally satisfy the patch test. In order to investigate in the
case of uniform stresses the normal contact traction N , the normal stress σz
and the vertical nodal displacement uz of the contact surface of the lower
block are checked. Their values are controlled by parameters used in statis-
tics: by the mean value x̄ and by the standard deviation σ and the coefficient
of variation Cv = 100% · σ/x̄ in order to estimate the variation. In Table 8.17
the results concerning the mean value and the coefficient of variation for the
following quantities are given: sum of contact tractions N = εp gN over the
surface, computed at Gauss points of the contact surface; normal stresses
σz , computed for the upper and lower surface of each element of the lower
block; nodal vertical displacements uz for the contact surface. For compari-
upper block
contact surface of lower block
lower block
Fig. 8.34 Blocks for the patch test; Upper block – regular mesh; Lower block –
distorted mesh
Fig. 8.35 Node-To-Segment (NTS) approach fails to pass the patch test; “roof-like”
contact surface
son, the first computation was made for the node-to-segment approach with
a of penalty value of 105 which was reduced due to the convergence problems.
As is known, this approach fails the patch test. Fig. 8.35 shows the “roof-like”
contact surface of the lower block for this case. As expected this approach
leads to the maximum of the coefficient of variation.
From Table 8.17 it becomes clear that also the integration with subdo-
mains leads to – though only slightly – smaller variations than an algorithm
with standard Gauss integration. It appears rather remarkable that the vari-
ations of the tractions remain constant while the variation of the stresses and
displacements falls below one percent. Visually, the application of the STS
8.7 Segment-To-Segment (Mortar) Approach: Analysis of the Patch Test 271
Table 8.17 Patch test; linear surfaces, influence of different integration schemes;
mean value and coefficient of variation for the following quantities: contact tractions
N on the contact surface, normal stresses σz in the lower block and vertical nodal
displacements uz on the contact surface; N T S ∗ - node-to-segment approach
No. Gpt No. sbd. tractions N stress σz displ. uz

x̄ · 104 v % x̄ · 103 v % x̄ · 10−2 v %
N T S∗ 1 -1.3084 -68.087 -1.5248 -120.039 -0.9150 -157.93
2 1 -1.7767 -21.474 -4.6787 -2.5772 -2.4286 -7.7430
6 1 -1.6738 -19.851 -4.7077 -1.1663 -2.4143 -1.5571
3 2 -1.6210 -16.760 -4.7137 -0.8570 -2.4180 -1.2770
2 3 -1.6354 -17.331 -4.7115 -0.8789 -2.4170 -1.4131
10 1 -1.6614 -19.696 -4.7109 -0.8595 -2.4150 -1.3121
2 5 -1.6477 -15.793 -4.7113 -0.7798 -2.4166 -1.2097
5 2 -1.6408 -17.572 -4.7139 -0.7942 -2.4160 -1.2383
20 1 -1.6537 -19.226 -4.7124 -0.7791 -2.4164 -1.1597
10 2 -1.6408 -16.667 -4.7128 -0.7585 -2.4159 -1.1304
4 5 -1.6299 -16.790 -4.7141 -0.7136 -2.4158 -1.0822
5 4 -1.6447 -16.395 -4.7125 -0.7239 -2.4154 -1.1190
2 10 -1.6337 -16.341 -4.7126 -0.7578 -2.4170 -1.1335
approach together with higher order of integration rules, and integration with
subdomains leads then to a flat surface, and therefore, disappearance of the
“roof-like” structure.
8.7.2 Contact Patch Test with Smooth Surfaces

As we have seen the STS approach either with increasing the number of inte-
gration points, or with combining the integration with subdomains technique
allows to improve the contact patch test result. This visually leads to the
flattening of the roof-like surfaces in the case of linear approximations. This
example is illustrating the situation for the smooth surface. For the patch test
two elastic blocks are taken, see Fig. 8.36. The lower block consists of nine
regular elements with linear approximation. The upper surface of this block
is covered with a smooth contact surface – four corner patches, four boundary
patches and one central patch, as discussed in Sect. 8.2.4. The upper block is
only one element with linear approximation. The size of the upper element is
chosen such that the nodes are projected to the middle of the corner elements
of the lower block, see Fig. 8.36. The upper surface of the lower block is a
master surface, and the lower surface of the upper block is considered as a
slave one. As in the previous example, the vertical displacements are applied
to the four upper nodes of the upper block.
From the meshed geometry and loading it is obvious that the NTS contact
approach with linear approximation will completely fail the patch test leading
to “the roof-like” surface, see Fig. 8.37 a). For the second test, the surface has
been covered with a smooth patch, but the NTS algorithm has been applied
– the result is shown in Fig. 8.37 b) – the contact surface is smooth, but it is
erroneously pointing upward. Only a combination of the STS approach (here
with 3 × 3 Gauss integration points) with a smooth contact surface leads to
the correct deformed surface positioned downward, see Fig. 8.37 c).
Fig. 8.36 Geometry for the patch test with smooth surfaces. The upper surface of
the lower block is covered with a smooth spline surface.
8.8 Segment-To-Analytical Surface (STAS) Approach:

Various Applications
In the current section various computational aspects and arising problems
are shown discussing some numerical examples. As an example of a problem
with a 3D spatial large sliding, a motion of a semi-circular cylinder on the
rigid surface of a parabolical cylinder is considered with the STAS approach
described in Sect. 4.3.3. Integration schemes using subdivision into subdo-
mains developed in Sect. 8.6.2 are applied together with the STS and STAS
contact approaches with various order of approximations for finite elements.
The influence of the integration scheme on the force-displacement curve is
presented for a deep drawing example – various approximations of solid shell
elements are discussed. An example with deep drawing of a plate into a pot
is chosen to illustrate the STAS algorithm with various combinations of rigid
surfaces. A special example with a deep drawing of a strip is devoted to the
selection of the approximation order for shell elements and contact approach
in order to obtain the correct thickness strain vs. loading displacement curve.
8.8 Segment-To-Analytical Surface (STAS) Approach: Various Applications 273
Case a) Failure of the patch test for the NTS contact approach with linear
approximation. Vertical displacements are scaled with the factor 5.
STS, smooth
1.01
1
0.99
NTS, smooth
0.98
0.97
0.96
0.95
0.94
NTS, linear
0.93
3.5
3
2.5
2
-0.5 1.5
0
0.5 1
1
1.5 0.5
2 0
2.5
3
3.5 -0.5
Fig. 8.37 Upper surface for three computations: Case a) NTS with linear approx-
imation (see above the whole block); Case b) NTS with smooth spline approxima-
tion; Case c) STS with smooth spline approximation of the master contact surface.
Vertical displacements are scaled.
8.8.1 Large Sliding on a Rigid Parabolic Cylinder

As an example of a problem with a 3D spatial large sliding, we consider here
a motion of a semi-circular cylinder on a rigid surface profiled as a parabolic
cylinder. Consider a parabolic cylinder in the canonical form, see details in
Sect. 4.3.3:
xs = α
(8.88)
ys = cα2
The convective coordinate α of the contact point is computed via the iterative
procedure given in eqn. (4.171) according to the rigid surface as a “master”
strategy discussed in Sect. 4.3.2. Penetration ξ 3 is computed afterward fol-
lowing eqn. (4.173).
Spiral Equidistant Motion to a Cylinder

Now we consider a 3D motion on the surface of the rigid parabolic cylinder.
The implementation is performed according to the STAS algorithm described
in Sect. 4.3.3, the approximation matrix A and other implementation details
are given in Sect. 8.1.3.
l
R
Z
w
v
Y
u
X
Δz
L
Fig. 8.38 Spiral motion of a circular semi-cylinder on the parabolic cylinder.

Segment-To-Analytical Surface (STAS) approach.
In order to generalize the equidistant motion in eqn. (8.76) into a spiral

one we consider the parametrization in the form:
x = vt, y = Ht/T, (8.89)
where v is a loading rate, T is the final load step. Thus, the spiral motion
with the trajectory shown in Fig. 8.38 is defined as
⎧ ⎫
⎨ vt(1 − 2ch) ⎬
r= Ht/T , t = 0, 1, 2, ..., T. (8.90)
⎩ ⎭
h + c(vt)2 (1 − 2ch)
For the numerical example, we chose the spiral motion of a short deformable
circular semi-cylinder with radius R = 1 and l = 0.4 on the surface of the
parabolic cylinder with parameters c = 5 · 10−2 , H = 20, see Fig. 8.38.
The semi-cylinder is made of linear elastic material: E = 2.10 · 104 ; ν = 0.3
and meshed with linear “solid-shell” elements 16 × 3 as shown in Fig. 8.38.
Coulomb friction with μ = 0.3 is specified between the bodies.
The loading process consists of two stages:
1) the circular semi-cylinder positioned at the initial point is pressed into
the parabolic cylinder with Δz = h = −0.01
2) then the circular semi-cylinder is moving on the parabolic surface, pro-
viding an equidistant motion of the central axis with the distance R + Δz =
0.99 according to eqn. (8.90) together with the upper surface parallel to the
X-Z plane. Thus, the central axis of the semi-cylinder is moving according to
the following equation:
⎧ ⎫
⎨ 1.001vt ⎬
r= 20t/T . (8.91)
⎩ ⎭
h + 5.005 · 10−2 (vt)2
The displacements are applied in 1000 load steps with v = 10−2 , providing an
increment Δy = 0.02 which is larger than the critical threshold value in eqn.
(8.70). The Gauss integration formula with 4 × 4 integration points is used to
check the value of penetration. Here we concentrate again on the investigation
of the influence of the various parts of the matrix on the convergence. Since
the contact geometry is linear only due to the discretization of the semi-
cylinder, the cases with the main matrix and the tangent matrix without
curvature part are compared, see Table 8.18 for the first 20 loads steps.
Obviously, it is definitely advantageous to use the tangent matrix without
curvature parts, but keeping the rotational parts in this problem.
8.8.2 Free Bending of a Metal Sheet on Two

Cylinders
Another example with STAS contact approach is the free bending problem
of a metal sheet (thickness t). Geometry and deformed configuration are
shown in Fig. 8.39 for the bilinear and biquadratic elements. The material is
taken to be elasto-plastic of Hencky’s type with isotropic plastic hardening
with the following data: Bulk modulus κ = 1.75 · 104 cm kN
2 ; Shear modulus
μ = 8.077 · 10 cm2 ; σi 16.0 cm2 ; σ∞ = 40.0 cm2 ; δ = 20.0; H = 20.0.

3 kN kN kN
At the beginning the metal sheet with thickness t = 0.25 cm is positioned

on two cylindrical rigid bodies. As loading a displacement u is prescribed in
the center of the sheet. Due to symmetry only one half of the system has to be
modeled and discretized using 12 bilinear resp. 6 biquadratic elements and a
rather fine mesh with 100 bilinear resp. 50 biquadratic elements. Concerning
FE−model
1111
0000
0000
1111 1111
0000
0000
1111
0000
1111 u 0000
1111
0000
1111 0000
1111
b=12cm
a=24cm
Fig. 8.39 Geometry and loading process for a metall sheet on two rigid cylinders.
Case with biquadratic elements is shown.
Table 8.18 Sliding of a semi-cylinder on a parabolic block. Bilinear contact ele-

ments. Segment-To-Analytical Surface (STAS) contact approach. Influence of vari-
ous contact stiffness parts on convergence. Case 1: excluding only curvature matrix;
case 2: only main matrix. Comparison of no. of iterations for the fist 20 load steps
(l.s.)
Case 1 Case 2
No. l.s. No. Cum. No. l.s. No. Cum.
1 9 9 1 10 10
2 8 17 2-18 13 231
3 5 22 19 12 243
4-20 4 90 20 11 254
... ... ... ... ... ...
the application of different contact approaches the following variations were

investigated:
1) The rigid cylinder is modeled by linear finite elements with 48 elements
in the circumferential direction. The metal sheet is treated as “master”
part, the cylinder as “slave” part. Contact is modeled by the “Node-To-
Segment” (NTS) approach.
2) The model is the same as in 1, but contact is modeled by the “Segment-
To-Segment” (STS) approach.
3) The surface of the rigid cylinder is described analytically (STAS ap-
proach), the metal sheet is then the “master” part. The contact integral
is computed by a nodal collocation formula with additional gap interpo-
lation over the element surface.
4) The model is the same as in 3. The quadrature formula is chosen either
with different numbers of integration Gauss points, or with subdomains
on the “master” part.
Case with Bilinear Solid-Shell Elements

In Fig. 8.40 the results for the global central reaction force for case 1, case
2 with 2 × 2 Gauss points and case 3 each with a mesh of 12 bilinear solid-
shell elements are plotted, see the structure of solid-shell elements in [56].
The result with the fine mesh of 100 bilinear solid-shell elements of elements
with one integration point for contact evaluation is taken for comparison as
“exact” solution. The nodal collocation formula of case 3 shows the largest
oscillations, because the value of the penetration is checked only at the nodes
of the sheet and the mesh is relatively coarse in comparison with the geo-
metrical size of the cylinder despite the analytical description of the cylinder.
If contact is checked at the nodes of the mesh of the rigid cylinder, in case
25
case 1; 12 el.
case 2; 12 el.
case 3; 12 el.
case 4; 100 el.; 1 Gp
20
15
Force
10
0 2 4 6 8 10 12
Displacement
Fig. 8.40 Force-deflection curves for free bending problem; bilinear solid-shell ele-
ments; comparing quadrature formulae of low order element and mesh refinement
25
2 Gpt
3 Gpt
7 Gpt
10 Gpt
case 4; 100 el.; 1 Gp
20
15
Force
10
0 2 4 6 8 10 12
Displacement
Fig. 8.41 Force-deflection curves for free bending problem; bilinear solid-shell ele-
ments; contact against analytically defined contact surface; influence of the number
of Gauss points
1 of the “Node-To-Segment” approach, jumps appear with smaller ampli-

tudes. This is an obvious consequence of the finer mesh for the cylinder in
comparison with the sheet mesh. As is well known, this would improve with
a finer mesh on both sides. Using a Gauss quadrature for the “Segment-
To-Segment” strategy leads to only slightly reduced jumps, because in both
“Node-To-Segment” and “Segment-To-Segment” approaches the rigid cylin-

der is modeled still with a rather coarse finite element mesh. In order to
investigate the influence of the order of the Gauss integration for the STAS
approach when the rigid cylinder is given exactly by an analytical function,
case 4 was extended with 2 × 2, 3 × 3, 7 × 7 and 10 × 10 integration points
and, finally, compared to the results with a refined mesh of 100 elements, but
with 1 Gauss point only, see Fig. 8.41. Obviously, the quadrature formulae
with 2 × 2 integration points leads to rather large oscillations. This is due to
25
3 Gpt, 2 subdm, 6 el.
' $ 2 Gpt, 3 subdm, 6 el.
6 Gpt, 6 el.
3 Gpt, 12 el.
20
' 2 Gpt, 50 el. $
15
& %
Force
10
& %
0
0 2 4 6 8 10 12
Displacement
Fig. 8.42 Force-deflection curves for free bending problem; biquadratic solid-shell
elements; Influence of the number of Gauss points and of refined mesh
the fact that rather non-smooth contact checking is performed, which can be
improved using more contact points. As mentioned earlier, checking contact
at Gauss points can be interpreted as integration of a discontinuous function
for which no a-priori error estimation is avaliable. One can see that a con-
vergent sequence of curves is achieved if the number of integration points is
increased even in the case of a coarse mesh for the sheet. The influence of the
mesh refinement with a softer response of the sheet is also obvious.
Case with Biquadratic Solid-Shell Elements

The next step is to consider the influence of the number of Gauss points
for the sheet meshed with biquadratic elements. Fig. 8.42 shows the result
of the computation for the beam meshed with 6 elements, if the integration
formula with 6 × 6 integration points is used and if as an alternative the
integration formula with 2 subdomains and 3 × 3 integration points in each
and with 3 subdomains and 2 × 2 integration points is used. The result is
25
10 Gpt, 6 el.
' $ 2 Gpt, 5 subdm, 6 el.
2 Gpt, 50 el.
20
' $
15
& %
Force
10
& %
0
0 2 4 6 8 10 12
Displacement
Fig. 8.43 Force-deflection curves for free bending problem; biquadratic solid-shell
elements; subdivision of the integration domain into subdomains results in smaller
oscillations
compared with a refined mesh of 12 elements for the beam meshed with 12
elements with 3 × 3 integration points. For comparison a 50 element mesh
and 2×2 integration points per element is chosen. The density of the Gauss
points to check the penetration is identical in the first three cases, but as a
consequence of the smallest a-priori integration error for the algorithm with
subdomains, the latter leads to a reduction of the oscillations. In Fig. 8.43
it is shown that even a relatively large number of 10×10 Gauss points per
element still preserves oscillations. The integration with 5 subdomains and 2
Gauss points for the 6 element mesh leads again to a solution with a smaller
deviation.
Frictional Case with Bilinear Solid-Shell Elements

A comparison with the frictional case for the following friction coefficients:
μ = 0.1, μ = 0.2, μ = 0.3 is presented in Fig. 8.44, where the computation
was performed with the “best” 10 × 10 integration formula. All contact pairs
with rigid surfaces are modeled with the STAS approach. As an obvious
result, the reaction force is increasing following the modification of the friction
coefficient.
Remark 8.6. Here the STAS approach has been applied successfully for the
C 1 -continuous rigid surfaces. If the rigid boundary is not smooth, or another
approach is favorable (e.g. boundaries are not rigid surfaces and STS is nec-
essary), then the full history transfer algorithm is absolutely necessary to
30
no friction
0.1
0.2
0.3
25
20
15
Force
10
0 2 4 6 8 10 12
Displacement
Fig. 8.44 Force-deflection curves for free bending problem; influence of the choice
for the friction coefficient; 10 × 10 Gauss points
resolve the frictional contact problem, see the study of the example “drawing
of an elastic strip into a channel with sharp corners” in Sect. 9.2.2.
8.8.3 Deep Drawing of a Cylindrical Pot –

Combination of STAS Contact Elements
In this numerical example the deep drawing process of a circular plate into a
cylindrical pot with counter die is simulated, see the geometry in Fig. 8.45.
The material of a circular plate is elasto-plastic with the following parameters:
κ = 1.75 · 104 kN/cm2 ; μ = 8077 kN/cm2 ; σi = 16 kN/cm2 . The circular
sheet has a uniform thickness t = 1 mm and a diameter D = 16.0 cm. The
geometry of the tools is shown in Fig. 8.45. Due to symmetry only a quarter
of the structure is discretized using a mesh with 175 bilinear elements. As
loading the displacement u is applied incrementally with Δu = 0.0025 for the
punch, as well as for the counter die, however, for the latter after the punch
is contacting the counter die.
The contact is defined with rigid surfaces using the STAS approach. The
example is illustrating the ability of the STAS approach to model a rigid tool
via a combination of planes, cylinders and toruses. Only the non-frictional
case is simulated here.
It can be seen from the deformation in Fig. 8.46 that the blank is domi-
nantly drawn along the upper part of the die and the punch.
16.0 cm
8.0 cm
u
punch
RS =0.5 cm
11111111111111111111111111111111111111111111111
00000000000000000000000000000000000000000000000
00000000000000000000000000000000000000000000000
11111111111111111111111111111111111111111111111
RG=1.0 cm
u + Δu
counter die
12 cm
10cm
Fig. 8.45 Geometry for deep drawing process.
8.8.4 Deep Drawing – Test for the Quality of Shell

Elements as Well as for the Quality of the
Contact Algorithm
One of the advantages of the “solid-shell” being 3D continuum formulation
is the ability to obtain the thickness strain. However, obtaining such charac-
teristic curves as thickness strain vs. loading displacement can be a crucial
test for the quality of both finite elements for shell problems and of contact
elements. Thus, the necessity of a correct algorithm to transport the history
variables is studied for special geometrical cases such as drawing of an elas-
tic strip into a channel with sharp corners, see Sect. 9.2.2 originated to the
publication in [92].
In order to study this case in a purified situation a deep drawing with a
counter die is modeled for one strip with thickness h = 0.125, see the geometry
in Fig. 8.47. Only half of the object is modeled due to symmetry: a = 2.750×2,
b = 2.875 × 2, c = 3.000 × 2, a = 2.750 × 2, l = 10.000 × 2. The punch and the
counter die are modeled via the STAS approach as a combination of planes
and cylinders with r = 0.500, R = 1.000, see the geometry in Fig. 8.47. The
geometry is chosen such that during drawing the distance between the punch
and the counter die remains equal to the thickness of the blank h = c − b. As
a) u = 0.00 cm
b) u = 1.25 cm
c) u = 2.50 cm
d) u = 3.75 cm
d) u = 5.00 cm
Fig. 8.46 Deep drawing process. Blank at various deformation states.

in the example from Sect. 8.8.3, the blank (the strip) is drawn by an applied
vertical displacement u to both the punch and the counter die – though, small
distances δ1 = 0.02 and δ2 = 0.01 are supplied at the beginning, see Fig. 8.47.
Both, bilinear and biquadratic solid-shell elements are used for comparison
of the thickness strain – loading displacement curve. Thus, first 40 bilinear
elements are applied, then 20 biquadratic elements are applied, see mesh and
deformed configuration in Fig. 8.48.
r
B A x = 3.75
δ2
h
δ1
O X
a
b
c
l
Fig. 8.47 Geometry for the deep drawing process.
The problem finally has been resolved using only the STAS approach with
10 × 10 integration points, however, only a perfect smooth geometry of the
rigid body allows to overcome the problem with the correct transfer of his-
tory variables, see example in Sect. 9.2.2. Thus, the combination of the STS
approach to model rigid bodies leads to difficulties with convergence. The
example with bilinear solid shell elements requires smaller load steps, and
therefore, more incremental load steps than the example with biquadratic
solid shell elements: 1000 incremental load steps for bilinear and only 100 in-
cremental load steps until the strip is reaching the configuration at u = 5.000
presented in Figure 8.48. The advantage of the biquadratic solid shell ele-
ments becomes more pronounced if we consider the thickness strain com-
puted at points during the deformation process. In Fig. 8.49 the strain vs.
loading displacement curve is given for the non-frictional case at point A
Model with bilinear solid shell elements.
Model with biquadratic solid shell elements.
Fig. 8.48 Deformed configuration for both meshes at applied vertical displacements
u = 5.000
with x = 3.750 (see geometry in Fig. 8.47). The highly oscillatory behavior
of the curve for the bilinear mesh is due to enforcing a circular geometry
by linear meshes. Each peak in the figure is reflecting the rolling of a linear
segment from the strip on a circular part with radius R, see Fig. 8.47. The
frictional case with bilinear meshes is causing even more severe convergence
problem, therefore, only a case with biquadratic meshes is shown in Fig. 8.50
for different points A (x = 3.750) and B (x = 3.000) for a small coefficient
of friction μ = 0.01.
0.01
bilinear solid shell
biquadratic solid shell
0.008
0.006
thickness strain
0.004
0.002
-0.002
-0.004
-0.006
0 1 2 3 4 5 6 7
displacement
Fig. 8.49 Thickness strain vs. loading displacement. Comparison of bilinear and
biquadratic solid shell elements.
8.8.5 Discussion
In this Section STAS contact approach together with continuum finite “Solid-
Shell” elements have been investigated for applications in large deformation
contact analysis including specific deep drawing situations. Thus, a large
number of examples is focusing on the comparison of the contact algorithms
used especially for problems in sheet metal forming. The discussed contact
elements with different degree of geometrical approximation of the contact
surfaces were tested in the example of a free bending of metal sheet with large
sliding in contact. A force-deflection curve was chosen to represent the main
characteristics of the results. Large oscillations appeared if relatively coarse
meshes together with a low order integration are used. It was found that
increasing the number of integration points leads to an improved reduction of
these oscillations, but an integration procedure with an additional subdivision
into subdomains leads to a sufficiently further reduction of oscillations with
the same total number of Gauss points over the contact area. It was shown
that it is possible within the STS approach together with an integration over
subdomains to efficiently improve the quality of different characteristics such
as force-displacement and strain displacement curve. It was shown that the
higher than linear order finite elements (here biquadratic ones) are superior
concerning the results and using the same number of degrees of freedom,
especially for modeling of deep drawing processes.
8.9 Implementation of the Nitsche Approache 287
0.0005
no friction - node A
no friction - node B
0 friction, mu=0.01, - node A
friction, mu=0.01, - node B
-0.0005
-0.001
thickness strain
-0.0015
-0.002
-0.0025
-0.003
-0.0035
-0.004
0 1 2 3 4 5 6 7
displacement
Fig. 8.50 Thickness strain vs. loading displacement at different points. Frictional
case. Biquadratic solid shell elements.
Summarizing the discussion over various techniques one may conclude that
the higherst quality can be achieved only within the smooth contact finite
element with higher then linear approximation of finite elements together
with the full transfer of history variables algorithm, however, the skilled user
of contact algorithms can achieve the good result depending on the type of
modeling with lower efforts.
8.9 Implementation of the Nitsche Approache

In this Section varios types of the Nitsche approach described in Sect. 5.6.3
Surface-To-Analytical-Surface (STAS) contact approach in the sense of rigid
surface as a “master” surface, see Sect. 4.3.2 is implemented, where one of
the contacting bodies is a rigid body. The necessary for the Nitsche method
discretization of the continuum is carried out for a geometrically nonlinear
solid element with Neo-Hookean material law allowing the decoupled repre-
sentation of the strain-energy function for compressible isotropic hyperelastic
materials, see [170].
8.9.1 Gauss Point-Wise Substituted Formulation

Starting with eqn. (5.57) and using the approximation matrix A for the dis-
cretization of the penetration ξ 3 within the contact kinematics the expression
ξ 3 = (rS − ρM ) · nM = nTM A uB (8.92)
is gained where uB is the nodal displacement vector of the contacting sur-

face of the solid element. nM denotes the normal vector on the contacting
master surface of the nonlinear solid element. Furtheron, for implementation
purposes the vector n2M is defined by entries of the normal vector nM as
$ %
n2M = n1 n1 , n2 n2 , n3 n3 , 2n1 n2 , 2n2 n3 , 2n1 n3 (8.93)
enabling to compute the normal projection of the stress vector out of the
T
stress tensor as σnM = n2M DBu. This leads to the discretization
#
T
δu (BT DB) dV u +
# V
+εN δuB T
(A n nT A) dSC uB −
T
SC
#
T
− δuTB (AT n n2M DB) dSC u −
SC
#
−δuT (BT DT n2M nT A) dSC uB = 0, (8.94)
SC
finally, ending up with matrices of the same order as

3 3 3 3
δuT (Ku + εN Kξu ξ
− Kξu σn − Kσun ξ ) u = 0. (8.95)
The assembly directly results in the element stiffness matrix and its additions
in the case of contact.
8.9.2 Bubnov-Galerkin-Wise Partial Substituted

Formulation
Discretization of eqn. (5.59) leads to the following two equation sets
#
δuT (BT DB) dV u
V
#
+δuB T
(AT nC) dSC λ = 0
SC
#
δλT (N CT nT A) dSC uB
SC
#
T
− δλ T
(CT n2M DB) dSC u
# SC
−δλT (CT C) dSC λ = 0. (8.96)
SC
C indicates, hereby, the matrix consisting of shape functions for the Lagrange
multiplier λ defined on the contacting surface of the solid element. The as-
sembly of eqn. (8.96) leads to
3
δuT Kuu u + δuT Kξuλ λ = 0
3
δλT (N Kξλu − Kσλun ) u − δλT Kλλ λ = 0 (8.97)
!
Kλu
where the Lagrange multipliers can be preeliminated as they are only element-
wisely defined by applying the Schur complement to the equation system of
1 3
/
Kuu Kξuλ u
=
0
. (8.98)
Kλu −Kλλ λ 0
Regarding the check of contact an active set strategy is applied. Within the
Gauss point-wise substituted formulation an active set is defined as the set
of Gauss points with negative penetration. For the Bubnov-Galerkin-wise
partial substituted formulation a special searching algorithm with a set of
Gauss points enabling the inversion of the matrix Kλu is used.
8.9.3 Numerical Example

For verification purposes both Nitsche formulations are used for the simula-
tion of the frictionless free bending of a metal sheet on two rigid qclyinders
which involves large sliding. The geometrical setup is presented in Fig. 8.51(a)
and is following the setup given in the numerical example from Sect. 8.8.2.
R=2
12
24
(a) (b)
Fig. 8.51 Setup of metal sheet on two rigid cylinders: (a) geometrical properties,
(b) deformed situation with predefined u
The Neo-Hookean material properties chosen are equaling E = 105 and

ν = 0.3, the element size used is 1 × 0.25 × 0.5. For a displacement con-
trolled loading the deformed state is given in Fig. 8.51(b). As a reference
4
12 Penalty with εN= 10
Nitsche (Gauss Point − wise) with εN=3*105
Nitsche (Bubnov−Galerkin − wise) with εN= 105
10
E = 105, ν=0.3
8 Gauss Points
8
Reaction Force
4
(b)
−2
0 2 4 6 8 10
Displacement
(a)
Fig. 8.52 Displacement-reaction force diagram with 8 Gauss points and best per-
forming N for each formulation: (a) overall result, (b) cutout at initiation of sliding
12 Penalty
Nitsche (Gauss Point − wise)
Nitsche (Bubnov−Galerkin − wise)
10
5
E = 10 , ν=0.3
5
εN=10
8
Reaction Force
2 (b)
0
−2
0 2 4 6 8 10
Displacement
(a)
Fig. 8.53 Displacement-reaction force diagram with 10 Gauss points and N = E

for each formulation: (a) overall result, (b) cutout.
formulation for the Nitsche approaches the standard Penalty scheme is used.
Within the given setup both Nitsche formulations perform similar to the
Penalty scheme, but show higher sensitivity regarding the amount of Gauss
points (see Fig. 8.52(a)) which is more obvious at the initiation of sliding (see
Fig. 8.52(b)). These oscillations are due to the fact that rather non-smooth
contact checking is performed, which can be improved using more contact
points – either more Gauss points or higher mesh refinement, see also ex-
amples in Sect. 8.8.2. The Nitsche approach with the Bubnov-Galerkin-wise
partial substituted Lagrange multipliers is performing more robust, thus, al-
lowing a larger range of values for Gauss points and Penalty parameters.
On the other hand the Gauss point-wise substituted Lagrange multipliers
formulation is highly sensitive, even for N = E, see Fig. 8.53(a). While

for this specific choice of the Penalty paremeter the approach with Bubnov-
Galerkin-wise partial substituted Lagrange multipliers is performing again
similar to the Penalty scheme, the cutout in Fig. 8.53(b) shows the interrup-
tion of the computation for the Gauss point-wise substituted formulation at
an early stage.
9
Special Case of Implementation –
Reduction into 2D Case
2D implementation case is constructed as a sub-part of the Surface-To-Surface

contact similar to NTS, STS and other strategies. It is more simple algo-
rithm for understanding and implementation and can be recommended for
the beginner to start with. Also some special algorithms such as continuous
numerical integration algorithm introduced for Surface-To-Surface contact in
Sect. 4.1.6 and update of history variables derived for the coupled anisotropic
adhesion-friction model in Sections 6.3.5 and 6.3.6 are much easier to study
for 2D case considered in this Section.
9.1 Finite Element Implementation of 2D Contact

Interaction
As the simplest technique we consider here Node-To-Segment (NTS) ap-
proach. The structure of all parts of the tangent matrix is algorithmic. It
is sufficient for the discretization to define only approximations of a relative
displacement vector (rs − ρ) and its derivative with respect to ξ. Following
the standard iso-parametric technique as for finite elements, we consider a
contact line element with the same order of approximation for the geometry
as for the displacement field. The boundary curve can be given with any curve
description (spline, NURB, etc.). For simplicity, we consider only the node-
to-segment contact approach. Let a boundary curve or a master segment of
it be defined by n nodes with x(1) , x(2) , ..., x(n) ; and x(n+1) for a slave node.
A standard grouping of a displacement vector can then be written as
(1) (1) (2) (2) (n) (n) (n+1) (n+1) T
uT = {u1 , u2 , u1 , u2 , ..., u1 , u2 , u1 , u2 } , (9.1)
where the first n nodes resp. 2n displacements belong to the master surface,
while the (n + 1)’th term is belonging to the “slave” node resp. is describing
the “slave” displacements.
294 9 Special Case of Implementation – Reduction into 2D Case
We introduce a matrix of shape functions A

−N1 0 0 −N2 0 0 ... −Nn 0 1 0
A= , (9.2)
0 −N1 0 0 −N2 0 ... 0 −Nn 0 1
where Ni , i = 1, 2, ..., n are shape functions. The matrix of the derivatives of

the shape functions A is defined then as

N1 0 0 N2 0 0 ... Nn 0 0 0
A = − , (9.3)
0 N1 0 0 N2 0 ... 0 Nn 0 0
The relative vector of variations (δrs − δρ) and the relative velocity vector
(vs − v) are then written as
δrs − δρ = Aδu , vs − v = Au̇ , (9.4)
where u̇ is the nodal velocity vector. With the matrix of the derivatives A
e.g. a derivative δρξ of a vector δρ can be defined as
δρξ = −A δu. (9.5)
We note that the matrices A and A are sufficient to build the tangent
matrices as well as the residual vector for any arbitrary contact surface.
9.1.1 Linear NTS Contact Element

The simplest approximation is a linear contact element within the Node-To-
Segment (NTS) approach. This linear contact element has 3 nodes: the first
two nodes x1 , x2 are approximating a contact boundary, while the third node
xS is the slave node. The approximation on the master element is defined as

1−ξ 1 1+ξ 2 1−ξ x1 1+ξ x2
ρ(ξ) := x + x = + . (9.6)
2 2 2 y1 2 y2
The tangent vector ρξ is then given as:

x2 − x1 1 x2 − x1
ρξ = = . (9.7)
2 2 y2 − y1
A single metric coefficient becomes:
a11 = ρξ · ρξ = 0.25 · ((x1 − x2 )2 + (y1 − y2 )2 ), (9.8)
which is the square of the length of the vector ρξ . The unit normal vec-
tor ν to the contact segment is defined in a Cartesian coordinate system via a
9.1 Finite Element Implementation of 2D Contact Interaction 295
cross product of the tangent vector ρξ and the third unit vector k which is
normal to the plane:

[k × ρξ ] 1 y2 − y1
ν := = . (9.9)
(ρξ · ρξ ) 2a11 x1 − x2
In this definition according to eqn. (4.88) and to Fig. 4.3, we assume that the
solid body occupies the lower part relative to the contact element in Fig. 9.1.
The approximation matrix containing the shape functions A
⎡ 1−ξ 1+ξ ⎤
− 0 − 0 10
⎢ 2 2 ⎥
A=⎣ ⎦, (9.10)
1−ξ 1+ξ
0 − 0 − 01
2 2
and the matrix of the derivatives of the shape functions A

1 1 0 −1 0 0 0
A = , (9.11)
2 0 1 0 −1 0 0
are used to approximate the displacement field as well as the derivatives. The
displacement vector u is defined as
uT = {u(1)
x , uy , ux , uy , ux , uy } .
(1) (2) (2) (S) (S) T
(9.12)
1
0S
0
1
0
1
Y
ν 2
00
11
11
00
O 00
11
ξc 1
0
C 1
0
0
1 ξ= 1
1
11
00
00
11
0
1
0 ρ
1
00
11 ξ
ξ =−1
X
1
0
1
0
Fig. 9.1 Linear contact element. Node-to-segment approach.
9.1.2 Closed Form Solution for CPP Procedure

Consider now the increment Δξ in load step (m) for the Closest Point Projec-
tion (CPP) procedure in eqn. (4.85) in order to define the internal coordinate
ξC of the projection point C. Due to the linear approximation in the contact
element we get ρξξ = 0.
(rs − ρ) · ρξ
Δξ = =
(ρξ · ρξ )

2
4 x1 + x2 x2 − x1 (m) x − x1
= xS
− − ξ · =
x2 − x1 2 2 2 2
2xS · (x2 − x1 ) x2 2 − x1 2
= − − ξ (m) . (9.13)
x2 − x1 2 x2 − x1 2
It is obvious that the Newton update scheme ξ (m+1) = ξ (m) +Δξ in eqn. (4.85)
is independent of the initial guess ξ (m) , therefore, the last expression in
eqn. (9.13) together with this scheme leads to the exact value for the in-
ternal coordinate ξC , so called, the closed form solution of CPP procedure:
2xS · (x2 − x1 ) x2 2 − x1 2

ξC = − . (9.14)
x2 − x1 2 x2 − x1 2
The result can be also interpreted as convergence in one iteration. Since the
geometry of this element is quite simple, the same result can be also obtained
directly via the triangle in Fig. 9.1.
A simple searching algorithm leads to computations of all further compo-
nents for a contact element at the projection point ξC only if the slave point
S is projected onto this element, i.e. if |ξC | ≤ 1.
9.1.3 Return-Mapping Scheme – 2D Case

In order to define the tangent traction vector T, we apply the standard
return-mapping scheme based on the elasto-plastic analogy as discussed for
the Surface-To-Suface contact in Sect. 6.1.3. The trial tangential traction is
computed via the evolution equation (6.171), which in the case of a linear
approximation is reduced to:
˙
Ṫ = −T (ρξ · ρξ )ξ. (9.15)
We consider here the simplest case for frictional problems: a quasi-statical

motion leading to the development from sticking to sliding. In addition, we
assume that during motion a slave point is not crossing an element boundary.
Thus, for our case we need only one additional variable ξ 0 per contact element
in order to define the initial position of the slave point on the master segment.
The trial tangent traction is obtained via the application of the backward
Euler scheme within the evolution equation (9.15). Since we have a linear
approximation for the geometry, we have
T (m+1) = T (m) − T a11 (ξ (m+1) − ξ (m) ) = (9.16)

and continuing recursively, we obtain
= T (m−1) − T a11 (ξ (m+1) − ξ (m−1) ) = ... = T (0) − T a11 (ξ (m+1) − ξ 0 ).
Assuming in addition, that at the initial position the tangential traction T (0)
was zero, we obtain
(m+1)
Ttr = −T a11 (ξ (m+1) − ξ 0 ). (9.17)
The last eqn. (9.17) serves now to compute the trial tangential reaction.
The return-mapping following the Coulomb friction condition leads then
with:
N (m+1) = N ζ (m+1) ,
to
⎧ (m+1) √
⎪
⎪ Ttr
(m+1)
if |Ttr | < μ|N (m+1) | a11
⎪
⎪
⎨ (sticking)
T (m+1) = .
⎪
⎪ (m+1) √ (m+1) (m+1) (m+1) √
⎪
⎪ μ|N | a 11 sgn(T ) if |T | ≥ μ|N | a 11
⎩ tr tr
(sliding)
(9.18)
The inequality condition in eqn. (9.18) is obtained from the following:
(m+1)
(m+1) Ttr
Ttr < μ|N | =⇒ √ < μ|N |, (9.19)
a11
where eqn. (4.104) for the absolute value of the covariant vector is taken into
account.
The global solution scheme for the simplest case discussed here is presented
in Table 9.1.
9.1.4 Treatment of Special Cases

In this section, we consider, how to treat some particularities, which were
mentioned and excluded in section 9.1.3. The first problem is arising when
the applied load is not simply modified proportionally. In this situation a trial
load can not be computed only via eqn. (9.17), because the attraction point
ξ 0 (initial point of sticking) must be updated. Thus we have to extend the
algorithm as is shown in the following. The second problem is arising when
the projection point is crossing an element boundary during the incremental
loading. In this case, the computation according to eqn. (9.17) will produce a
jump, because the convective coordinate ξ belongs to different elements, see
Wriggers [190] and Laursen [106].
9.1.5 Update of the Sliding Displacements in the

Case of Reversible Loading
We consider a geometrical interpretation of the return-mapping scheme in
eqn. (9.18) together with the evolution equation (9.17), see Fig. 9.2, repre-
senting 2D simplification of the more general case developed in Sections 6.3.5
and 6.3.6
(m) √
|Ttr | < μ|N (m) | a11 =⇒ T |ξ (m) − ξ 0 | < μ|N (m) | (9.20a)
(m) (m) μ|N (m) |

|ξ (m) − ξ 0 | < Rξ , Rξ = (9.20b)
T
Eqn. (9.20b) describes an allowable elastic region A(m) B(m) with a center
of attraction O(m) . All points inside this domain are in “sticking condition”.
If now a point ξ (m+1) appears to be outside of the domain at load step
(m + 1), then its only admissible position is on the boundary of the domain,
i.e. must coincide with B(m+1) . A sliding force is applied then at the contact
point, see eqn. (9.18). As long as we have a motion of the contact point
(m+1)
only in one direction the sign function for the sliding force sgn(Ttr )=
sgn(Δξ (m+1) ) does not change and the computation will be correct. However,
when a reversible load is applied which forces the contact point to move
forward or backward, the attraction point O(m) must be updated, in order
to define the sign function for the sliding force correctly. This update can be
defined geometrically from Fig. 9.2:
(m+1) μ|N (m) |

|Δξ (m+1) | = |Δξsl | + Rξ = |Δξ (m+1) | − . (9.21)
T
The absolute value of the sliding displacement is then computed at load step
(m + 1) as:
μ|N (m+1) |
|Δξsl | = |ξ (m+1) − ξ (0) | − , (9.22)
T
and the updated center of the elastic domain becomes:
ξc(up) = ξ (0) + sgn(ξ (m+1) − ξ (0) )|Δξsl |. (9.23)
For the next step, the evolution equation (9.17) is corrected as
T (m+2) = −T a11 (ξ (m+2) − ξcup ). (9.24)
Remark 9.1. As an alternative procedure the back-substitution of the evolu-

tion equation (9.17) into eqn. (9.22) gives the updated scheme via the trial
force:
1 (m+1)
|Δξsl | = |Ttr | − μ|N (m+1)| , (9.25)
T
which can be found e.g. in Wriggers [190], see also eqn. (6.137).
(m)
1
0 Rξ 1 0
0
1
11111
0000
0
1 0
0
1
(m)1
0
A
(m)
O 1 0 B 0
1
0
(m)1
0 A(m+1) O (m+1) B(m+1)
1
0
1
1
0 0
10
1 00
1100
11 0
1
0
100
11 0
1
0
1 00
11 00
11 0
1
0
1
0
1 0
10
1
0
1 00
1100
11 011
00 00
11 00
11
0 (m)1
1
(0) 1
0
0
01
1 0 1
1
0
1
0 (up) 1
0
1
0
1 0
1
0(m+1)
1
ξ 1 0 ξ 0
1
0
1 ξ 0
0
1 0
1
ξ
0
1
c
0
1 0
1 0
1 0
1
0
1
0
1 0
1 0
1 0
1
0
1 0
1 0
1 0
1
Δ 1
(m)0 0
1 0
1
0ξ 1
0
1
1
0
1
0
0
1
0
1
0
1 0
1
0
1
0
1 0
1
0
1 (m+1) 1
0
0
1 0
1
0
1
0
1 0
1 Δξ 0
1 0
1
0
1
0
1 0
1 0
1
0
1 0
1
0
1
0
1
0
1 Δ ξ
sl 0
1
0
1R
(m+1) 0
1
0
1
ξ
0
1
0
1 0
1
0
1 0
1
0
1
Fig. 9.2 Coulomb friction. Updating of sliding displacements in convective coordi-

nates. Motion of friction cone and center of attraction.
9.1.6 Crossing an Element Boundary – Continuous

Integration Scheme
Consider two adjacent elements I(m) J(m) and J(m) L(m) at load step (m), see
Fig. 9.3. Let the contact point S(m) be projected onto the element I(m) J(m)
and its projection is a point K(m) with the convective coordinate ξ (m) . At
load step (m + 1) these two elements move into the position I(m+1) J(m+1)
and J(m+1) L(m+1) , but the contact point is moved into position S(m+1) and
is projected now onto element J(m+1) L(m+1) . We assume that the angular de-
formation of these elements is small in comparison with a rigid body motion,
thus the elements are drawn as straight lines. It is obvious, that the direct
computation according to the evolution equation (9.17) results in a jump. Let
e.g. the point K(m) have the convective coordinate ξ (m) = 0.9 close to the
right element boundary, (see parameterization in Fig. 9.1), and then point
K(m+1) has ξ (m+1) = −0.9 close to the left element boundary. The distance
between them is only Δξ = 0.2, but the evolution equation gives:
(m+1)
Ttr = −T a11 (ξ (m+1) − ξ (m) ) = 1.8T . (9.26)
The maximum possible jump following this straightforward action is easily

determined from the limit values of the convective coordinates:

Tjump = −T a11 lim (ξ)ξ∈IJ − lim (ξ)ξ∈JL = 2T a11 . (9.27)
ξ→−1+0 ξ→+1−0
This jump appears only due to the different approximation of the adjacent
elements. In order to overcome this, we can compute the force in geometrical
form. The incremental tangential displacements Δρ can be expressed in the

metrics of the second element J(m+1) L(m+1) at time step (m + 1):
(m+1)
Δρ = Δξρξ , (9.28)
and alternatively, it can be geometrically defined via the incremental dis-

placement vector Δu:

Δρ = ρ(ξ (m+1) )ξ∈JL − ρ(ξ (m) ) + Δu(ξ (m) ) . (9.29)
ξ∈IJ
Then Δξ is defined as

(m+1)
ρ(ξ (m+1) )ξ∈JL − ρ(ξ (m) ) + Δu(ξ (m) ) ξ∈IJ · ρξ
Δξ = (m+1)
(9.30)
a11
and the evolution equation becomes

(m+1) (m+1)
Ttr = T (m) − T a11 Δξ. (9.31)
Modifications of the global solution scheme given in Table 9.1 are represented
in Table 9.2 according to the special cases.
9.1.7 Remarks on Additional Developments

One can see, that the continuous integration scheme as presented in Sec-
tion 9.1.6 leads to an increasing number of history variables, in fact, in ad-
dition to ξ (m) the vector ρ(ξ (m) ) must be stored. Moreover, other history
variables such as the updated sliding displacements must be transfered in a
similar fashion. The continuous integration scheme, of course, is particularly
important for contact problems with singularities, e.g. sliding of an edge along
a curve as also shown in Fig. 9.3. For other cases with nonsingular geome-
try it is more efficient to exclude contact points, once they appear outside
the master element, but then, as a compensation, introduce additional con-
tact points in the slave segment, e.g. integration points, within the so-called
segment-to-segment approach, for details see Zavarise and Wriggers [202] es-
pecially for the 2D case and Harnau, Konyukhov and Schweizerhof [56] for
the 3D case. Another approach to increase the number of contact points is
the mortar method, see e.g. in McDevitt and Laursen [124] and in Puso and
Laursen [151].
Another important problem arises due to the non-smoothness of contact
boundaries, if for the boundary a low order finite element mesh is used. This
leads to jumps in both the normal and the tangential characteristics when
crossing the element boundary see Fig. 9.4, because neither a normal vec-
tor n, nor a tangent vector ρξ are defined at the edge point B. If the real
(m)
Y L1
0 0
1
(m)
J
0
1
0
1 S
(m) 11
00
00
11 00
11(m)
I 00
11
K 11
00
00
11 00
11
11
00 00
11
00S(m)
11
00
11
Δ u 11
00
00
11
00
11 ρξ
(m+1)
(m+1) (m+1) (m+1)
I J L
11
00 111
000 1
0 00
11
00
11 011
100 0
1 00
11
0
1
0
1
K up Δ ρ 0 K(m+1)
1
(m)
0
1
0
1
0
1
0
1
0
1
01
10 (m+1)
0S
1
X
Fig. 9.3 Crossing an element boundary within a load increment. Typical case for
the continuous integration scheme.
boundary is an edge then e.g. adaptive methods can improve the result for
a straight geometry, see Wriggers and Scherf [193] and Wriggers [190]. If the
real boundary is smooth, then various smoothing techniques based on the
approximation of the boundary with e.g. splines can be used. There are nu-
merous publications on this subject, see Wriggers et. al. [192], Padmanabhan
and Laursen [135] and Stadler et.al. [170] especially for 2D problems, and
then in Puso and Laursen [149], Krstulovic-Opara et. al. [102], Stadler and
Holzapfel [169] for 3D problems. In this case, the geometrical singularity is
removed, i.e. the normal n and the tangent vector ρξ are uniquely defined
at point B. However, the continuous integration scheme in eqn. (9.30-9.31)
is still necessary, as the smooth patches have in all above publications local
support, i.e. their convective coordinate is defined separately.
n
n ΑΒ ρξ n BD
1
0
B
A1
0 ρξΑΒ ρξBD 11 D
00
0
1
Fig. 9.4 Low-order finite element approximation. Approximation of a real bound-

ary with a smooth curve.
Table 9.1 Global solution scheme for a linear contact element.
1. Initialization of the convective coordinate ξ 0 .

Initial condition for the evolution equation for all contact elements:
• compute projection point ξC in eqn. (9.14), with no external loads
−→ ξ 0 = ξC .
2. Loop over load increments m, m = 1, ..., mend .
3. Loop over global Newton iterations i, i = 1, ..., iend .
4. Loop over all contact elements
(i) (i)
• compute projection points ξC eqn. (9.14). If | ξC | ≥ 1 then exit loop 4.
(i) (i)
• check penetration ζ = (rs − ρ) · ν. If ζ > 0 then exit loop 4.
• compute contact tractions and corresponding tangent matrices at contact
point ξC .
Normal traction: N (i) = N ζ (i)
Tangent matrix KN for normal traction
N ζ (i) T
KN = N AT (ν ⊗ ν)A + A (ν ⊗ ρξ )A + AT (ρξ ⊗ ν)A
(ρξ · ρξ )
(i) (i)
Trial tangential traction: Ttr = −T (ρξ · ρξ )(ξC − ξ 0 )
Real tangential traction T (i) and corresponding matrices
are defined via the return-mapping algorithm
(i) (i)
if |Ttr | ≤ μ|N (i) |ρξ if |Ttr | > μ|N (i) |ρξ
sticking sliding
T (i) = Ttr (i)

T (i) = μ|N (i) |ρξ sgn(Ttr )
Tangent matrix KT Tangent matrix KT
(i)
εT εN μ sgn(Ttr ) T
KT = − AT (ρξ ⊗ ρξ )A KT = − A (ρξ ⊗ ν)A
(ρξ · ρξ ) (ρξ · ρξ )1/2
μ|N |sgn(Ttr ) T
(i)
T (i)
− AT (ρξ ⊗ ρξ )A + A (ρξ ⊗ ρξ )A
(ρξ · ρξ )2 (ρξ · ρξ )3/2

+AT (ρξ ⊗ ρξ )AT +AT (ρξ ⊗ ρξ )AT
• Compute the full contact tangent matrix K = KN + KT

• Compute residual R
T (i)
RN = N (i) AT ν; RT = AT ρξ ,
(ρξ · ρξ )
R = RN + RT .
end loop over contact elements

end loop over global Newton iterations
end loop over load increments
Table 9.2 Modifications of the global solution scheme according to special cases.
1. Initialization of the convective coordinate ξ 0 .

Initial condition for the evolution equation for all contact elements:
• compute projection point ξC in eqn. (9.14), with no external loads
−→ ξ 0 = ξC .
2. Loop over load increments m, m = 1, ..., mend .
3. Loop over global Newton iterations i, i = 1, ..., iend .
4. Loop over all contact elements
(i) (i)
• compute projection points ξC eqn. (9.14). If | ξC | ≥ 1 then exit loop 4.
(i) (i)
• check penetration ζ = (rs − ρ) · ν. If ζ > 0 then exit loop 4.
• compute contact tractions and corresponding tangent matrices at contact point
ξC .
Compute normal traction: N (i) and corr. matrix, see Table 9.1
(i)
Trial tangential traction Ttr according to the specific algorithm:
a) reversible loading b) continuous integration
Δξ see eqn. (9.30)

T (i) = −T a11 (ξ (i) − ξcup ).
(i) (i)
Ttr = T (m−1) − T a11 Δξ.
• Compute real tangential traction T (i) and corr. tangent matrices KT ac-
cording to the return mapping scheme, see Table 9.1
• Compute the full contact tangent matrix K = KN + KT , see Table 9.1
• Compute residual R, see Table 9.1
end loop over contact elements

end loop over global Newton iterations
Update and store necessary history variables
a) reversible loading b) continuous integration
Compute and store the update cen- Store history variables for Δξ accord-
(up)
ter ξc according to eqn. (9.23) ing to eqn. (9.30) and T (m) .
end loop over load increments

9.2 Frictional Contact Analysis: Requirements of

Special Algorithms
In this section a series of numerical examples of frictional contact problems
showing the implementation of 2D algorithms as well as the treatment of spe-
cial cases such as the sliding displacements update and continuous integration
scheme discussed in Section 9.1 are shown.

Contact Surfaces. Reversible Loading Process
We consider the sliding of an elastic block similar to the numerical example in
Sect. 8.5.2, however, here the block will be loaded with horizontally prescribed
reversible displacements. The main point is to show the update procedure for
sliding displacements and investigate the development of the sticking-sliding
zone.
As an example for the computation, we consider a rectangular block (Fig.
9.5) resting on a rigid foundation with the following parameters: elasticity
modulus E = 2.1·104 , Poisson ratio ν = 0.3, length a = 20, height b = 5. The
dimensions are assumed to be consistent. The block is uniformly meshed by
linear finite elements: 40 elements in horizontal direction and 10 elements in
vertical direction. The lower surface represents a rigid base. Coulomb friction
with a coefficient μ = 0.3 is specified between the surfaces. The contact
surface of the deformable block is assumed to be a “master”, while the surface
of the rigid base is a “slave” surface within the “node-to-segment” approach.
The penalty parameters are chosen as εN = εT = 2.1 · 106 . The loading is
applied as prescribed displacements at the top side of the deformable block.
This example is chosen to show the robustness of the contact algorithm and
the update scheme within a covariant description, though the current results
can be achieved certainly with other known techniques.
The main question arising before the computation is, which size of the load
step is allowed, as it is important to capture the spreading of the sticking-
sliding zone correctly. We use the concept of the critical displacement ucr , in-
troduced for the numerical example in Sect. 8.5.2 (originated to [88]), namely,
a value of the applied displacements after which the block fully slides. As
shown for the infinite layer this value is given as
2μh
ucr = . (9.32)
1−ν
The estimation of the critical horizontal displacement in eqn. (9.32) gives
ucr = 6.0 ·10−3, thus in order to investigate the sticking-sliding zone properly
we choose a displacement increment Δu = 1.0 · 10−4 . The displacements are
applied according to the loading process given in Table 9.3.
9.2 Frictional Contact Analysis: Requirements of Special Algorithms 305
u2 1
0 v
0
1 1
0
0
1
0
1 0
1
u1
0
1 0
1
0
1
0
1
0
1 0
1
0
1 0
1
1111111111111
0000000000000
0
1
0
1
0000000000000
1111111111111
B 0
1
0
1
C
0000000000000
1111111111111
0
1
0000000000000
1111111111111
B1 C1 b
A 1111111111111
0000000000000
h
0
1
01
1 0
10
1
01
010
10000000000000
1111111111111
0
1
010
1
010
1
01
010
1
010
10
1
010
1
010
1
010
1
010
1
010
1
010
1
01
D
010
01
10
011
100
0
1
0 1
00
11
1 0
01
10
01
1 0
0
1
Fig. 9.5 Plane deformation of a block. Applied displacement loading at top of the
block.
Table 9.3 Plane deformation of a block. Loading procedure with prescribed dis-
placements on the top side of the deformable block.
No. l.s. Δu u · 10−3 Δv · 10−3 v Loading

0 0.0 0.0 0.0 0 initialization of conv. coord.
1 0.0 0.0 7.0 · 10−3 0.0 − 7.0 vertical displ. v
2–80 10−4 0.0 − 8.0 0.0 7.0 forward horizontal displ. u1
81–84 −2.5 · 10−5 8.0 − 7.9 0.0 7.0 backward horizontal displ. u2
85–163 −1.0 · 10−4 7.9 − 0.0 0.0 7.0 backward horizontal displ. u2
164–280 −1.0 · 10−4 0.0 − (−1.17) 0.0 7.0 backward horizontal displ. u2
As a consequence of the reversible loading a hysteresis curve as shown

in Fig. 9.6 is developed. The applied displacement at point C, see Fig. 9.5,
is depicted along the x-axis, and the computed horizontal displacement at
point D is depicted along the y-axis. We obtain a spreading of the sliding
zone during the forward loading process (curve OF) as well as during the
backward loading process (curve FG). The horizontal displacements u along
the contact line AD, see Fig. 9.5, together with the distribution of the stress
ratio T /N allow us to define the sticking-sliding zone during the loading
process.
Forward Loading. Path OF on the Hysteresis Curve
The horizontal displacement distribution as well as the stress ratio T /N dis-

tribution on the contact boundary are shown in Fig. 9.7 and Fig. 9.8 for the
following loading points:
a) Load step 1, see Table 9.3, i.e. only vertical displacements are applied
v = 0.007. This is the starting point O on the hysteresis curve Fig. 9.6;
b) Intermediate points with applied horizontal displacements u = 0.0040
and u = 0.0060, resp. points O1 and O2 on the hysteresis curve Fig. 9.6;
c) The load step No. 80 with u = 0.0080, see Table 9.3, is chosen as a final
point of the forward loading, see also the point F on the hysteresis curve
Fig. 9.6.
The development of the sliding zone with increasing displacement loading u
is given in Fig. 9.7 and the development of the stress ratio T /N in Fig. 9.8.
At the end of the forward loading the sliding zone is increased to about
8 ≤ x ≤ 20.
Backward Loading. Part FGH on the Hysteresis Curve
The horizontal displacement distribution and the contact stress ratio T /N

distribution on the contact boundary are shown in Fig. 9.9 and in Fig. 9.10
for the following loading points:
a) Last load step of the forward loading with u = 0.0080, resp. point F on
the hysteresis curve Fig. 9.6 as a starting point;
b) Intermediate points on the unloading part FG with applied horizontal
displacements u = 0.0040, u = 0.000, u = −0.0040, u = −0.0060, u =
−0.0080, resp. points F2 , F3 , F4 , F5 , F6 on the hysteresis curve;
c) Selected point with u = 0.010 on the full backward sliding part GH of
the hysteresis curve.
When the load is reversed, starting from load step No. 81, see Table 9.3,
all points on the contact boundary are sticking. On the contact boundary
we have so-called residual horizontal displacements (in analogy to plastic-
ity). During the following unloading back to u = 0.000 the whole bound-
ary is still sticking: points F1 , F2 , F3 . The latter we can only detect from
0.0025
G F
F6 F5 F4 F3 F2 F1
0.002
Observed displacement (pt. D)
0.0015
0.001 O2
0.0005
O1
H O
0
-0.0005
-0.012 -0.01 -0.008 -0.006 -0.004 -0.002 0 0.002 0.004 0.006 0.008
Applied displacement (pt. C)
Fig. 9.6 Plane deformation of a block. Hysteresis curve. Observed horizontal dis-
placement at point D vs. applied horizontal displacement at point C.
0.003
Pt. O, u=0.0000
Pt. O1, u=0.0040
Pt. O2, u=0.0060
0.0025 Pt. F, u=0.0080
0.002
0.0015
Displacement, U
0.001
0.0005
-0.0005
-0.001
0 2 4 6 8 10 12 14 16 18 20
Coordinate, X
Fig. 9.7 Plane deformation of a block. Horizontal displacement of the contact

boundary. Forward loading.
the stress ratio diagram in Fig. 9.10, where the curves vary inside the layer
−0.3 ≤ T /N ≤ 0.3. The residual displacements are not changing, see dia-
gram 9.9, until sliding is beginning. Starting from the applied displacements
u = −0.0040 (point F4 ) we can detect the beginning of sliding at the left cor-
ner of the block, as the stress ratio curve is approaching its limit ratio 0.3. The
final part of the hysteresis diagram, from point F4 to point G, is responsible
for the development of the sliding zone in the backward direction, which can
be observed either by the development of the horizontal displacements in the
left part of the displacement diagram 9.9, or by the development of the zone
with a stress ratio T /N = 0.3 in diagram 9.10. Full sliding of the block starts
at point G with u ≈ −0.0092. Beyond that, horizontal displacements on the
contact are changing proportionally to the applied displacement loading, i.e.
linearly, as we see from the linear part GH in the hysteresis curve.
Remark 9.2. In the presented example the displacements are small and slaves
nodes never cross the element boundaries, therefore the continuous integra-
tion scheme is not necessary. In the next example we show a particular
case with large sliding in which the application of the continuous integra-
tion scheme is necessary.
Pt. O, u=0.0000
0.4 Pt. O1, u=0.0040
Pt O2, u=0.0060
Pt. F, u=0.0080
0.3
0.2
0.1
Ratio T / N
-0.1
-0.2
-0.3
-0.4
0 2 4 6 8 10 12 14 16 18 20
Coordinate, X
Fig. 9.8 Plane deformation of a block. Stress ratio T /N on the master contact
surface. Forward loading.
9.2.2 Drawing of an Elastic Strip into a Channel

with Sharp Corners
In this section, we consider a special contact case, for which the application
of the continuous integration scheme described in sect. 9.1.6 is absolutely
necessary. An elastic strip AD, see Fig. 9.11 (E = 2.1 · 104, ν = 0.3, thickness
h = 0.5, length L = 24) is positioned at the beginning of a channel with width
a = 13. The corners of the channel are rather sharp. The channel itself is
modeled by two rigid blocks B1 and B2 . The strip is loaded incrementally by
a prescribed displacement v at the center until it is inserted into the channel,
see Fig. 9.12. The strip here is modeled with 24 linear solid-shell elements,
see Hauptmann and Schweizerhof [59] and Hauptmann et.al. [58], and due to
symmetry only one half of the system is modeled. The crucial point during
the analysis is the sliding of a sharp corner C over the element boundaries
1, 2, 3, see Fig. 9.12. A load-displacement curve computed for the loading
point is chosen as the representative parameter to compare various contact
approaches. The following variations were investigated:
1. Non-frictional case with the “node-to-surface” approach without the con-
tinuous integration scheme proposed in eqn. (9.30) and (9.31).
2. Non-frictional case with the “segment-to-segment” approach. Here the
number of integration points in the contact segment is varied.
3. Frictional case with the “node-to-surface” approach without the contin-
uous integration scheme.
-0.005
0.0120
0.0480
0.1200
-0.006
-0.007
Relative displacement
-0.008
-0.009
-0.01
-0.011
-0.012
0 2 4 6 8 10 12 14 16 18 20
Coordinate, X
Fig. 9.9 Plane deformation of a block. Backward loading. Horizontal displacement

of the contact boundary.
4. Frictional case with the “node-to-surface” approach with the continuous

integration scheme.
5. Frictional case with the “segment-to-segment” approach. Here the num-
ber of integration points in the contact segment are varied.
We start the investigation with the non-frictional problem (case 1, 2) applying
the load increment v = 0.005 with the penalty parameter εN = 2.1 · 105 .
The elements from the strip are chosen to be a master, while the sharp
corner is a singular slave node. The load-displacement curve for the “node-
to-surface” approach contains a jump when the sharp corner is crossing the
boundary nodes 1 resp. 2, see Fig. 9.12. The solution process is no longer
converging after the sharp corner is crossing the boundary node 3. As an
alternative for improvements, we chose the “segment-to-segment” approach,
described in Harnau, Konyukhov and Schweizerhof [56]. The sharp corner C
is modeled with two slave segments which are orthogonal to each other and
take contact points as the Lobatto integration points. Taking 2 integration
points with 2 sub-domains, or e.g. 5 integration points allows to compute the
full load-displacement curve without the jumps obtained with NTS scheme,
see Fig. 9.13. The smoothing effect in the last cases happens because the
contact is checked not only against the single edge node, but also against the
set of contact points which covers fairly densely the sharp edge.
The next study is devoted to the frictional problem (case 3, 4) with the load
increment v = 0.0025, the penalty parameters εN = 2.1·105 , εT = 2.1·105 and
a friction coefficient μ = 0.2. The straightforward analysis without the con-
0.5
Pt. F, u=0.0080
Pt. F2, u=0.0040
Pt. F3, u=0.0000
0.4 Pt. F4, u=-0.0040
Pt. F5, u=-0.0060
Pt. F6, u=-0.0080
0.3
0.2
Ratio T / N
0.1
-0.1
-0.2
-0.3
-0.4
0 2 4 6 8 10 12 14 16 18 20
Coordinate, X
Fig. 9.10 Plane deformation of a block. Backward loading. Stress ratio T /N on

the master contact surface.
h
A D
X
11111111
00000000
00000000
11111111 C 11111111
00000000
00000000
11111111
00000000
11111111 v 00000000
11111111
B1 B2
a
L
Fig. 9.11 Drawing of an elastic strip into a channel with sharp corners. Geometrical
parameters.
tinuous integration scheme (case 3) leads to a jump in the force-displacement

curve when the sharp corner is crossing the first boundary node 1. The so-
lution is no longer converging when the sharp corner is crossing the second
boundary node 2, see Fig. 9.14. The application of the continuous integration
scheme, however, allows to obtain the full force-displacement curve, even in
the part when the strip is fully inserted into the channel, see the straight part
Fig. 9.12 Sequence of deformations for the elastic strip. Nodes are sliding over the
sharp corner C.
NTS
0 STS 2 * 2
STS 5 pts.
NTS
-5
NTS
-10 NTS
STS
force
-15
-20
-25
-30
0 2 4 6 8 10
Displacement
Fig. 9.13 Drawing of an elastic strip over a sharp corner. Load-displacement curve.
Non-frictional case. NTS and STS approaches with different Lobatto integration –
a) 2 × 2 int. points with 2 sub-domains and b) 5 × 5 int. points.
of the curves in Fig. 9.14. For this case a side part of the channel is modeled
as a rigid surface described by an analytical function. For comparison, the
analysis is carried out with various friction coefficients μ = 0.1 and μ = 0.3,
see Fig. 9.14.
It would be favorable also for the frictional case to perform the analysis
without the continuous integration scheme avoiding to store a lot of informa-
tion about slave nodes and apply the “segment-to-segment” approach just
increasing the number of integration points (case 2). The results of such an
analysis using the same loading parameters as for the analysis in Fig. 9.14 are
shown in Fig. 9.15. 5 and 10 Lobatto integration points are taken. Despite
the fact that the solution is converging, the load-displacement curve shows
large oscillations in the sliding regions after passing node 2. This result is
due to the fact, that from one load step to the other the history variables are
not transported correctly. Only the upper envelope of the oscillatory curve
NTS; mu = 0.2
0 NTS + cont. int.; mu = 0.1
NTS + cont. int.; mu = 0.2
NTS + cont. int.; mu = 0.3
-5
-10
NTS
-15
force
-20
NTS + cont. int.

-25
-30
-35
0 2 4 6 8 10
Displacement
Frictional case – various friction coefficients. Straightforward NTS approach and
continuous integration schemes.
5 int. pts.
0 10 int. pts.
-5
-10
-15
Force
-20
-25
-30
-35
-40
0 2 4 6 8 10 12 14
Displacement
Frictional case. Straightforward application of the “segment-to-segment” scheme
without the continuous integration scheme leads to high oscillations also for a larger
number of integration points.
9.3 Discussion 313
could be used as representation of the correct load-displacement curve. This

confirms the necessity of the continuous integration scheme in particular for
frictional contact.
Remark 9.3. We have to note that reversing the master surface, or the so-
called symmetric treatment of the contact, in the current example would also
not resolve the problem in a sufficient manner.
9.3 Discussion
Different situations requiring the application of more advanced techniques,
such as an update of sliding displacements and a continuous integration
scheme for the frictional case are illustrated with numerical examples. The
update technique is absolutely necessary for the simulation of reversible load-
ing processes, as the residual deformations have to be described correctly.
This is also conforming to the elasto-plastic analogy involved for formulation
of the return-mapping scheme for frictional problems. The continuous inte-
gration technique allows to transport all history variables correctly over the
contact segment boundaries, however, additional storage (history variables)
is required for the implementation. In the particular example of a sliding
edge on a surface, it is shown that for the non-frictional contact problem
the ”segment-to-segment” approach with different integration schemes can
improve the result, but for the frictional contact problem the continuous in-
tegration scheme is absolutely necessary independent of the approaches NTS
and STS.
10
Implementation of Contact Algorithms
with High Order FE
The Chapter is showing the implementation of high-order finite elements

including the possibility of contact interactions within a special strategy such
as contact layer finite element. Some numerical difficulties within the classical
Hertz problem are studied (originated to [96]). Later, this example became
the benchmark test for iso-geometric finite element, see [175] and [176]. The
contact layer finite element is illustrating good correlation with the analytical
solution of the famous Hertz contact problem even within a single element
covering the contact zone.
10.1 Introduction
One of the big advantages of high order finite elements along with the high
accuracy is the possibility to describe the given geometry of surfaces exactly.
The high quality for shell structure analysis has been shown e.g. in Rank
et.al [153] (2005). Also high order FE has became a basis for a, so-called,
iso-geometrical analysis for arbitrary 3D structures, see Hughes et.al. [70]
(2005). The exact geometry naturally leads also to improved results for con-
tact problems, however, some difficulties occur due to the nonlinear nature
of contact problems.
For small displacement problems an hp-version of the finite element
method has been discussed in Paczelt et.al. [134] (1999). A gap function
defined as a difference between approximated normal displacements from
two contacting bodies and corresponding to a penetration in the direction
of an initial normal vector nc is used as a measure of contact interaction.
The penalty method and the augmented Lagrangian method have been used
to enforce contact conditions within the simple iteration method possessing
only a linear rate of convergence. For large displacement problems simulated
with linear finite elements the gap has been introduced via the closest point
projection procedure taking into account an updated normal vector earlier
in Wriggers and Simo [194] (1985). The problem was solved via an iterative
316 10 Implementation of Contact Algorithms with High Order FE
solver of Newton’s type possessing quadratic rate of convergence for which a

consistent linearization was necessary. A linearization procedure via convec-
tive coordinates has been used for 3D problems in Laursen and Simo [109]
(1993), though the linearization has been carried out in the global coordi-
nate system. A fully covariant description in the local surface coordinate
system corresponding to the closest point procedure has been developed in
Konyukhov and Schweizerhof [87] (2004), [88] (2005). Since all operations
such as formulation of a weak form in accordance with the applied method
(penalty, augmented Lagrangian), the return-mapping algorithm and the lin-
earization are carried out in the local surface coordinate system via convective
coordinates i.e. in the fully covariant form, the description is applicable to
any parametric approximations of the surfaces. A straightforward geometri-
cal interpretation of various contact stiffness parts, see [91] and the possi-
bility of generalization into coupled anisotropy including tangential adhesion
and friction, see [89], [90] are among other advantages of the fully covariant
description.
Non-smoothness of linear approximations for contact surfaces (namely
faceted surfaces) is causing artificial oscillations in contact tractions and also
can lead to dis-convergence of the iterative solution. This has been recog-
nized in numerical contact mechanics and several approaches have been pro-
posed: earlier proposals are dealing with a smoothing of a rigid surface, see
Schweizerhof and Hallquist [156] (1992) and in Heege and Alart [64] (1996).
Then various aspects of smoothing techniques for master contact surfaces
for both non-frictional and frictional cases have been extensively studied in
Pietrzak and Curnier [145] (1999), Padmanabhan and Laursen [135] (2001),
Puso and Laursen [149] (2002), Krstulovic-Opara and Wriggers [102] (2002),
Stadler, Holzapfel and Korelc [170] (2003) and others. Despite the number of
publications all developments have been concerned predominantly with the
geometrical smoothing of the given linear mesh. The discussion about the
unknown influence of the void between the smoothed surfaces and the linear
mesh surface remains open and a high-order finite element technique seems
to be a key to fill this void. Thus, the current contribution is aimed at consid-
erations of the covariant contact description together with high-order finite
elements with exact representation of the contact boundaries as a smooth-
ing technique for piecewise linear meshes. A special anisotropically refined
p-finite element adopting refinement only in a single layer is constructed to
take advantage both from the covariant contact description, and from the
p-finite element technique. Since the contact description itself is considered
independently from the p-finite element techniques it can be easily applied
for other general cases with high-order FE including iso-geometric FE.
10.2 Finite Element Implementation of High Order Contact FE 317
10.2 Finite Element Implementation of High Order

Contact FE
In order to implement the possibility of contact for high order finite elements
all elements on the potential contact boundary are modified for the Contact
Layer finite element shown in Fig. 10.1 according to the following scheme:
1. Each element on the boundary is constructed as an anisotropically refined
element, see the terminology in Solin et.al. [165]. Thus, the shape function
space is hierarchically constructed via the Lobatto shape functions with
the enrichment possibility for the edge degree of freedom (DOF) pedge ,
the side DOF pside and the bubble DOF pbubble possessing necessary
conformity requirements to the internal elements (for further numerical
examples the layer element is satisfying conformity conditions with the
linear mesh i.e. pint. mesh = 1).
2. The shape function space is modified according to the blending function
method in order to represent the boundary surface exactly (e.g. linear
edge shape functions which have support at vertices’s V5 , V6 , V7 , V8 are
modified).
3. The truncated shape function space organized via all shape functions with
a support on a curvilinear boundary V5 V6 V7 V8 becomes a shape function
space for a contact element. Thus, all derived tangent matrices and resid-
uals are assembled with regard to the DOF’s of the corresponding high
order element.
According to the described scheme it is possible then to subtract the ap-
proximation of the master surface given by the vector ρ(ξ 1 , ξ 2 ) together with
the approximation of the slave surface rs (η 1 , η 2 ). Since all possible contact-
ing pairs “master-slave” should be tested it leads to the necessity of a global
searching routine which is out of scope of the current contribution. One of the
most robust scheme for linear “master-slave” segments allowing self-contact
is described in Benson and Hallquist [11]. The approximation operator A
for a relative displacement vector ρ(ξ 1 , ξ 2 ) − rs (η 1 , η 2 ) in the standard finite
element method also for the p-version of finite elements is defined as a linear
operator with regard to a DOF vector x
(1) (1) (1) (n) (n) (n)
x = {x(1)
ms , yms , zms , ..., xms , yms , zms , xsl , ysl , zsl , ..., xsl , ysl , zsl } ,
(1) (1) (m) (m) (m) T
(10.1)
where the index ms indicates correspondence to the master surface and the
index sl indicates correspondence to the slave surface. The number of the
degrees of freedom depending on the involved method can be related either
to the nodal DOF for the classical nodal version of finite element, or to the
generalized DOF including a hierarchical structure of the vertex, edge, side
and bubble degree of freedom for a p-element. The linear operator is defined
by the approximation matrix A as follows
S
ξ3
n ρ2
V8
V7
C
ρ1
V3
V6
V5
p
side
pedge p bubble
V2
p int. mesh
V1
Fig. 10.1 Structure of contact layer finite element
⎡
M 1 0 0 M 2 0 0 ... ... ... M m 0 0
A= ⎣ 0 M 1 0 0 M 2 0 ... ... ... 0 M m 0
0 0 M 1 0 0 M 2 ... ... ... 0 0 Mm
⎤
−N 1 0 0 −N 2 0 0 ... ... ... −N n 0 0
0 −N 1 0 0 −N 2 0 ... ... ... 0 −N n 0 ⎦ . (10.2)
0 0 −N 1
0 0 −N 2 ... ... ... 0 0 −N n
with M k = M k (ξ 1 , ξ 2 ), k = 1, m as shape functions for a master surface;

N k = N k (η 1 , η 2 ), k = 1, n are shape functions for a slave surface. A relative
virtual displacement or velocity vector is given then as
δρ(ξ 1 , ξ 2 ) − δrs (η 1 , η 2 ) = Aδx for a virtual displacement vector;

v(ξ 1 , ξ 2 ) − vs (η 1 , η 2 ) = Av for a velocity vector. (10.3)
In addition only derivatives with respect to the master convective coordinates

ξ i should be computed for further representation of the tangent matrices:
∂A
δρ,i = δx = A,i δx. (10.4)
∂ξ i
10.2.1 Computation of Contact Integrals

Two situations should be distinguished for computations of all integrals for
equilibrium conditions as well as for corresponding tangent matrices: contact
between deformable and rigid bodies (so-called Signorini problems); and con-
tact between two deformable bodies. The first problem leads to the Contact
Layer - Rigid Surface (CLRS) finite element and the second problems leads
to Contact Layer - Contact Layer (CLCL) finite element. The structure of
these elements is considered here for the 2D case as layer-wise enrichment
of the initially linear mesh. However, the general contact algorithm is not
limited to this case.
10.2.2 Contact Layer - Rigid Surface (CLRS) Finite

Element
An initially linear element, see Fig. 10.2, is enriched hierarchically with edge
pel − 1 shape functions for the left side, per − 1 edge shape functions for the
right side, peb −1 edge shape functions for the bottom side and (pb −1)×(pb −1)
bubble shape functions, where p∗ is the highest polynomial degree of the
corresponding shape function. An exemplary element with quadratic degree
p = 2 of freedom in 2D leads to 2 × (4 vertex + 3 edge + 1 bubble) = 16
DOF. The linear boundary is represented then exactly by the blending func-
tion method (e.g. an exact circular arch in the following numerical exam-
ples). The contact with a rigid surface can be resolved assuming that the
parametrization of the rigid surface (becoming then a “slave” surface) is
given as rS (η 1 , η 2 ). All contact integrals are computed then over the master
surface in the following sequence:
1. A master segment is covered by a set of integration points C (e.g. by
Gauss points) ξi1 , ξj2 i, j = 1, Np according to the chosen order Np .
2. The CPP procedure is interpreted now vice versa: both the penetration
ξ 3 and the convective coordinate η 1 , η 2 of the point S on the slave surface
where a normal n (computed at the integration point C of the master
surface) is “penetrating” the slave surface, or in other words, according
to “the Rigid Surface is a “Master” Surface“ approach, see Sect. 4.3.2
F(η 1 , η 2 , ξ 3 ) ≡ rS (η 1 , η 2 ) − ρ(ξC
1 2
, ξC ) − ξ 3 n(ξC
1 2
, ξC ) = 0. (10.5)
3. All contact integrals (for residuals as well as for tangent matrices) are for-
mally computed via the quadrature formula, where a component with an
integration point ξi1 , ξj2 is included only if the corresponding penetration
is less then zero ξ 3 |ξi1 ,ξj2 ≤ 0.
#
f (ξ 1 , ξ 2 )ds = H(−ξ 3 )f (ξi1 , ξj2 )detJ(ξi1 , ξj2 )Ai Aj , (10.6)
i,j
where Ai , Aj are weights of the quadrature formula, J is the Jacobian

of the transformation ds → dξ 1 dξ 2 for a master segment, H(−ξ 3 ) is a
Heaviside function.
Remark 10.1. The CLRS finite element follows the STAS strategy to define
the penetration, thus all results obtained in Sect. 4.3 including the closed form
solutions for some surfaces are fully applicable for CLRS finite elements.
linear mesh
pb
pel
per
p eb
A C ξI B
ρ (ξ)
ζ
n ρξ
rs S
rigid body
Fig. 10.2 Contact Layer - Rigid Surface contact element. 2D-representation.
All necessary parameters for further numerical examples are reduced here for
the 2D case as shown in Fig. 10.2.
10.2.3 Contact Layer - Contact Layer (CLCL) Finite

Element
Both master and slave elements are enriched hierarchically in a similar fash-
ion as described in the previous Section 10.2.2. An exemplary element with
quadratic degree p = 2 of freedom in 2D leads to 2 × 2 × (4 vertex + 3 edge +
1 bubble) = 32 DOF. All contact integrals are computed over the slave surface
in the following sequence:
1. A slave segment is covered by a set of integration points C (e.g. by Gauss
points) ηi1 , ηj2 i, j = 1, Np according to the chosen order Np and the
corresponding vectors rs (ηi1 , ηj2 ) are computed.
2. The CPP procedure is executed to find out the convective coordinate

ξi1 , ξj2 on the master segment corresponding to projections of rs (ηi1 , ηj2 ).
Points with coordinates ξi1 , ξj2 not laying inside the master segment are
excluded (a, so-called, local searching procedure).
3. All contact integrals are formally computed via the quadrature formula,
where a component with an integration point ξi1 , ξj2 is included only if
the corresponding penetration is less then zero ξ 3 |ξi1 ,ξj2 ≤ 0,
#
f (ξ 1 , ξ 2 )ds = H(−ξ 3 )f (ξi1 , ξj2 )detJ(ηi1 , ηj2 )Ai Aj , (10.7)
i,j
where Ai , Aj are weights of the quadrature formula, J is the Jacobian

of the transformation dsS → dη 1 dη 2 for a slave element, H(−ξ 3 ) is a
Heaviside function.
The presented approach to compute integrals follows STS approach and the
Mortar method discussed in Sections 8.1.2 and 8.7.
linear mesh
master (mortar) part

S
A B
ζ
C C
ρξ D
n
slave (non−mortar) part
linear mesh
Fig. 10.3 Contact Layer - Contact Layer finite element. 2D-representation.

10.2.4 Lagrange Multiplier Method for Normal

Traction
As discussed before a robust scheme based on the full Lagrange multiplier
method for arbitrary large sliding problems would include the enforcement
of only normal non-penetration conditions combined with the augmented
Lagrangian method for the tangential condition. In a case of contact with a
rigid surface (CLRS) the finite element is modified according to the following
scheme:
1. Both, displacement vector and normal traction are approximated (mixed
formulation). Let A be an approximation matrix for a relative displace-
ment vector corresponding to the DOF vector x, see eqn. (10.3), and B
[1 × Nλ ] be an approximation matrix for the normal traction N corre-
sponding to a vector with Nλ DOF’s for a normal traction λ. An element
vector for the mixed formulation X contains then both, displacement
DOF’s x and variables for normal traction λ:

δρ − δrs = Aδx x
→ {X} = (10.8)
N = Bλ λ
2. The consistent linearization shown in eqn. (7.8) is derived according to
both independent variables x and λ. Thus, eqn. (7.8) leads to the follow-
ing matrix

Kxx Kxλ
[K] = , (10.9)
Kλx O
where the matrix Kxx is derived from the rotational part LrN and the
curvature LcN part in eqns. (7.9b)-(7.9c).
Kxx = [KrN ] + [KcN ]. (10.10)

Since the constitutive equation for normal traction is not involved into
the linearization the main part LrN in eqn. (7.9a) is representing then a
coupled part Kxλ
#
Kx λ = AT nBdsS . (10.11)
sS
Discussion About BB-Stability of the Proposed Lagrange

Multiplier Approach
As is known, a mixed formulation within the finite element method should
satisfy the discrete Babuska-Brezzi (BB) condition in order to ensure conver-
gence and stability with respect to the mesh size and discretized constraints.
The Lagrange method described in this section in combination with both the
approximation of the stress field and the integration rule of corresponding
10.3 Numerical Examples for High Order Contact Layer Finite Elements 323
contact integrals via a set of projected slave integration points over the mas-
ter segment (not via the set of slave nodes!) is falling into the case “one-pass
pressure interpolation method” which has been intensively studied in Solberg
and Papadopoulos [164] and proved to be BB-stable upon sufficiently accurate
integration of contact boundary terms P.2755 [164]. The BB-stability in due
course leads to the satisfaction of the contact patch test. Various integration
techniques including combinations of Gauss or Lobatto quadrature formulae
of higher order together with subdivisions of the integrand into sub-domains
(integration with subdomains) were studied in Sect. 8.7 in order to satisfy
the contact patch-test.
10.2.5 Global Solution Scheme

The global solution scheme in the case of a Contact Layer -Contact Layer
finite element is presented in Table 10.1. Some remarks are necessary for
better understanding.
• Since tangential tractions are defined in the rate form, see eqn. (6.19),
initial conditions should be supplied for further numerical solutions. This
is done in Step 0 of the global scheme, in which initial coordinates ξ0i
of the sticking point are computed and initial tangential tractions are
assigned Ti, 0 = 0.
• For a Contact Layer-Rigid Surface finite element the CPP procedure in
a) is changed as discussed in Section 10.2.2.
• For the mixed formulation discussed in Section 10.2.4 condition b) of
No contact in loop 4 is changed into N = Bλ > 0, see definition in
eqn. (10.8).
• The simplest backward Euler scheme is employed in step d) to solve in
general the evolution equation (6.19) (here eqn. (6.171) for 2D case) nu-
merically. The improved numerical scheme based on a geometrical inter-
pretation of the covariant derivative allows to overcome jumps occurring
in the contact segments due to different approximations, see Sect. 4.1.6.
10.3 Numerical Examples for High Order Contact

Layer Finite Elements
Contact between a deformable cylinder and a rigid plane, the classical Hertz
problem possessing an analytical solution, see e.g. in Johnson [77], is chosen
to test the ability of the anisotropically p-refined layered element to approxi-
mate the exact solution. The problem is well known also as a benchmark-test
for h-adaptive schemes in contact mechanics, see Wriggers and Scherf [193].
For the computation only a quarter of the cylinder in Fig. 10.4 is meshed
with 1120 bilinear finite elements. All parameters are chosen in order to sat-
isfy the assumptions for the Hertz contact problem . The material of the
Table 10.1 Global solution scheme for Contact Layer – Contact Layer (CLCL)
finite element
0. Initialization of convective coordinates ξ0i and

tangential tractions Ti, 0 = 0.
The CPP procedure is computed without any external applied load.
1. Loop over load increments I
2. Newton iteration loop k:
Iterative solution of global equilibrium equations.
3. Loop over contact pairs “master-slave”

4. Loop over integration points η 1 , η 2 of the slave segment
a) compute projection ξ 1 , ξ 2 onto the master segment (CPP procedure).
if ξ 1 , ξ 2 do not belong to the master segment (Local searching proce-
dure) then exit loop 4.
b) compute penetration ξ 3 at the projection point
if ξ 3 > 0 then exit loop 4. (No contact).
c) compute the normal traction N = N ξ 3 and the corresponding tan-
gent matrix [KN ] in eqns. (7.9a), (7.9b), (7.9c)
d) compute the trial tangential traction Tiel
Tiel = TiI−1 − T (ξ j − ξ0j )aij ,
where TiI−1 are real tangential tractions from the previous converged
load step I − 1.
e) return-mapping algorithm
compute the trial yield function

Φel = amn Tm el T el − μ|N |
n
if Φel ≤ 0 if Φel > 0
sticking condition sliding condition

= Tiel Tiel
Ti stick Ti slide = μ|N | √ el T el amn
Tm n
Compute the corresponding Compute the corresponding

tangent matrix [KT, st ] in tangent matrix [KT, sl ] in
enqs. (7.24a), (7.24b), (7.24c) eqns. (7.35a), (7.35b), (7.35c),
(7.35d).
e) Compute residual arising from eqn. (5.9)
R = N A T n + T i A T ρi
e) Assemble residual and tangent matrices

Check global convergence
cylinder is linear elastic with Young’s modulus E = 1.0 · 105 N/mm2 and
Poisson ratio ν = 0.3; radius of the cylinder R = 100mm. The quarter of
cylinder is loaded by a distributed pressure at the upper side equivalent to
the global load P/2. The level of the load P is intentionally chosen to keep
the contact radius a small in order to stay in the frame of the Hertz hypoth-
esis , namely a << R. Thus, only six bounding elements are modified into
Contact Layer-Rigid Surface (CLRS) finite elements (the sixth element sat-
isfies additionally the conformity condition , with a linear mesh on the right
4P R(1 − ν 2 )
side), see Fig. 10.4. The contact radius a = and the contact
πE
2P
pressure p = a2 − x2 (see derivation e.g. in [77]) are main parameters
πa2
for the verification. Several incremental loading states corresponding to the
extension of the contact zone through several elements are studied.
1 2 3 4 5 6
x0 x1 x2 x3 x4 x5
a) b)
Fig. 10.4 a) Initial linear mesh. b) Anisotropic p-refinement in contact layer. Only
six elements are modified into Contact Layer-Rigid Surface (CLRS) elements.
10.3.1 Loading Case 1. Contact Zone within One

Element
The first load level is taken as P = 0.657 kN corresponding to the contact
radius a = 0.857 mm laying completely inside the first contacting element, see
Fig. (10.4) b). The computation is performed first with an initially linear mesh
employing a Node-To-Segment (NTS) contact approach and then with CLRS
elements with increasing polynomial degree p = 2, 3, 4. The penalty method
has been used to enforce the contact conditions. As discussed already in
Sect. 8.6 a problem of integration of non-smooth functions appears during the
integration of contact integrals. This leads to the impossibility to set up the
order of the quadrature formula, a-priori, based only on the knowledge of the
polynomials involved into the shape function space. The current computation
is performed only with 10 Gauss points. A comparison with the analytical
solution is given in Fig. 10.5.
600
Hertz solution
NTS, initial mesh
500 p=2
Contact pressure
p=3 undeformed geometry

p=4 deformed geometry
400
300
Radius
200
100
0 0 0.5 1 1.5 2 2.5

0 0.2 0.4 0.6 0.8 1 1.2 Contact radius
Contact radius
a) b)
Fig. 10.5 Contact zone is within a single element. a) Contact pressure vs. radius.
Variation of the polynomial degree for the CLRS element. b) Deformed shape.
Obviously a good approximation is achieved even with the second order

polynomial p = 2; also the deformed shape (Fig. 10.5 b) is represented exactly.
It is obvious that this result is impossible to achieve with a linear mesh and
a node-to-segment approach.
10.3.2 Loading Case 2. Contact Zone within Several

Elements
If the contact zone is spreading through several elements a specific problem
is arising: if the the contact zone is ending inside a contact element then a
highly oscillatory solution occurs for the contact stresses. This phenomenon
is shown in Fig. 10.6. In this case the load P = 7.5 kN is applied resulting
in the contact zone ending inside the fourth element. The amplitude of the
oscillations is even higher than the maximum of stresses which makes the
computed results inadmissible for applications. Moreover, oscillations occur
not only inside the bounding element (fourth element in the example), but
also in neighboring elements. Neither increasing the polynomial degree, nor
application of the mentioned above composed integration rules of higher order
could reduce the oscillations. Oscillations are present for both the penalty
and the Lagrange multiplier method. Only if the contact zone is ending in
the vicinity of the element boundary (e.g. represented by a node for 2D) then
the contact stresses are well approximated. The problem has been recognized
for high-order FE already in Paczelt et.al. [134] and as remedy the re-meshing
with the goal to move the node to the end of the contact zone was suggested.
2500
Exact Hertz
p=2
2000 p=3
p=4
Contact pressure
1500
1000
500
0
0 0.5 1 1.5 2 2.5 3
Contact radius
Fig. 10.6 Oscillations for contact pressure occur at P = 7.5kN when the con-
tact zone is ending inside an element. Neither increasing the polynomial degree p,
nor application of the composed integration rules of higher order could reduce the
oscillations.
In the case of arbitrary loading contact detection is performed via the CPP
procedure, therefore, it would be advantageous to find out techniques with-
out moving a node. It was found in computations that the under-integration
of the CLRS elements together with the lowering of the order of polynomial
degree only for bounding elements, e.g. where the contact zone is ending,
leads to a reduction of the oscillations. This effect is shown in Fig. 10.7,
where the polynomial degree is taken to be p = 2 and the contact integrals
for the bounding element only have been computed with only 2 Gauss points
(neither the exact function representing the circular arch, nor the quadratic
shape functions are fully integrated). Algorithmically this leads to the con-
tact detection in only two points, therefore, the size of the contact zone is
underestimated.
This technique is tested then for the development of the contact zone
through the element for p = 2. The applied force P is then rising in the range
5.90 kN ≤ P ≤ 10.5 kN while the contact zone is spreading through the
fourth element, see Fig. 10.8. Obviously, the oscillations are not eliminated
completely, but their amplitudes are sufficiently reduced.
2500
Exact Hertz
Full integration
2000 Under-integration, 2 Gauss pts.
Contact pressure
1500
1000
500
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Contact radius
Fig. 10.7 Contact pressure at P = 7.5kN . Under-integration of the bounding

element with p = 2 leads to a reduction of oscillations.
2000
P=5.90 kN
P=6.50 kN
P=7.50 kN
P=8.50 kN
1500 P=9.50 kN
P=10.5 kN
Contact pressure
1000
500
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Contact radius
Fig. 10.8 Contact pressure. Under-integration of the bounding element with p = 2.

The contact zone is spreading through the bounding element while the load P is
increasing.
Remark 10.2. A fairly good correlation with the analytical Hertz solution
even for the cases where only the contact layer is hierarchically enriched with
polynomial order p allows to consider the proposed contact layer elements
as a generalization of smoothing techniques known in contact mechanics,
where only contact surfaces are smoothed. However, the full p-refinement
combined with contact can be applied without loss of generality including
the refinement of the internal zone as well, which can lead even to a better
correlation.
10.3.3 Discussion
In this chapter the special contact layer finite elements are constructed as
anisotropically refined p-finite elements allowing refinement only in the con-
tact layers. Contact with rigid bodies leads then to a Contact Layer - Rigid
Surface finite element, and contact between deformable bodies leads to a
Contact Layer - Contact Layer finite element. The elements are verified with
the classical Hertz contact problem. It is shown, that good correlations can
be achieved even within a few elements. However, oscillations may occur if
the contact zone appears inside the master contact segment. In such cases,
the reduction of the polynomial order together with under-integration of only
these bounding elements leads to improved acceptable results.
From a general point of view the proposed remedy for the oscillations
has limitations. However, from the mathematical point of view the source of
oscillations lays in the inability to approximate non-smooth functions (contact
pressure) via the set of functions given on the fixed domain represented by the
master segment, known as the Gibbs phenomenon. This explains the effect
of under-integration as a simple elimination of the singular point (end of the
contact zone as a point of non-smoothness) from the computation. A more
advanced technique would fall into the local enrichment of the shape function
space, i.e. XFEM method.
11
Anisotropic Adhesion-Friction Models
– Some Particular Details of
Implementation and Numerical
Examples
11.1 Introduction
This chapter is focusing on the finite element implementation of coupled
anisotropic adhesion-friction model which has been developed and analyzed
in Sections 6.2, 6.3 and 6.4. The model, first, is implemented for the full
Newton method and then the Augmented Lagrangian method. The tangent
matrices is then based on the linearization derived in Sect. 7.3.
11.1.1 Various Models for Friction

The majority of contact problems is solved under the assumption of smooth-
ness of contact surfaces. However, some cases appear in practice when it is
impossible to neglect the roughness of the contact surfaces. Essentially two
types can be distinguished: a) when a surface has randomly distributed as-
perities, and b) when asperities have algorithmic structure, e.g. the surface
shows different macro properties in different directions.
Mechanical characteristics for the contact problem of the first type a)
are obtained via statistically distributed asperities. Statistical analysis of a
real rough surface and experimental aspects of its measurements have been
developed in series of publications: Longuet-Higgins [116], Greenwood and
Williamson [49], Whitehouse and Archard [182] and more recently White-
house and Phillips [184] and Greenwood [47]. A comparative analysis of these
surface models is presented in McCool [123].
A statistical concept for the contact in the context of finite ele-
ment methods was developed then in Zavarise et.al. [204], Wriggers and
Zavarise [197], [196] for the non-frictional contact with normally distributed
asperities. Buczkowski and Kleiber [23] considered first non-frictional contact
with an isotropical statistical distribution of asperities, and then in [24] non-
frictional contact with an anisotropic statistical distribution of asperities. The
modeling of a contact surface with Bezier splines according to the statistical
332 11 Anisotropic Adhesion-Friction Models – Implementation
distribution of asperities was considered in Bandeira, Wriggers and Pimenta

[9]. Various nonlinear friction models are considered in the monography of
Wriggers [190].
A generalization of the isotropic macro characteristics is used to describe
frictional contact problems of the second type b). Michalowski and Mroz [130]
considered the sliding of a rigid block on an inclined surface and formulated
various sliding rules for sliding displacements which depend on directions.
Thus, an anisotropic friction model for sliding was introduced. In a purely
theoretical description Zmitrowicz [213] developed the structure of the fric-
tion tensor for sliding forces based on the motion of an elementary block
on an anisotropic surface and described its properties based on symmetry
groups for the tensor. Curnier [31] presented a rate independent theory of
anisotropic friction for contact interaction mentioning adhesion as a possible
elastic part without any further development. Zmitrowicz [214] developed the
structure of the sliding friction tensor according to a relative sliding velocity
and introduced a classification of anisotropic surfaces based on the number
of eigenvalues of the friction tensor. These cases were numerically illustrated
for a material point on the anisotropic plane. He and Curnier [61] used the
theory of tensor function representations to obtain the structure of the fric-
tion tensor for an arbitrary nonlinear case according to the relative sliding
velocity and derived also thermodynamical restrictions for the friction ten-
sor components. Mroz and Stupkiewicz [130] considered the structure of the
friction tensor based on a statical model of interaction of springs located in
a plane which has periodically inclined asperities.
Despite the extensive literature on finite element solutions for contact prob-
lems, there are only few publications on finite element models for anisotropic
friction. Buczkowski and Kleiber [22], [24] created an interface element con-
taining the orthotropic sliding law. The return-mapping scheme in a Carte-
sian coordinate system was then used to obtain the sliding displacements.
The effect of orthotropy was interpreted considering small displacements for
a flat punch on an elastic foundation. Hjiaj et. al. [67] formulated the prob-
lem via the bi-potential and applied Lagrangian multiplier methods to solve
a problem with orthotropic friction considering also small displacements in
Cartesian coordinates. Parametric quadratic programming was used in Zhang
et. al. [206] to solve the almost identical problem. Recently, Jones and Pa-
padopoulos [82] developed a finite element model for anisotropic friction,
where the stick-slip condition is enforced via Lagrange multipliers.
The aforementioned publications include anisotropy only for the friction
model and do not assume any anisotropy for the elastic region, the
adhesion. In the current contribution we propose a general approach for
the finite element solution of quasi-statical frictional contact problems in-
cluding anisotropy for both adhesion and sliding assuming that contact sur-
faces in general possess an anisotropic structure for both, elastic and friction
domains. This approach is based on the covariant description for contact
problems which is applicable for an arbitrary geometry of contact surfaces
11.1 Introduction 333
and large displacements. Within the geometrically exact description, contact

conditions are described on the tangent plane of the contact surface exploit-
ing tensor analysis. Using a penalty regularization of Coulomb’s friction law
and the return-mapping algorithm leads to the evolution equations for con-
tact friction. It becomes obvious, that the evolution equations are not only
a regularization technique, but act also as the constitutive relation to model
friction behavior for an adhesion domain. Keeping this idea in mind, the
evolution equations are generalized for a more complex mechanical behavior
exploiting tensor algebra on the tangent plane of the contact surface. The co-
variant description allows to formulate the main characteristics for surfaces
with arbitrary geometry, e.g. the yield function is formulated via the friction
tensor defined in surface metrics. Then both, anisotropy for adhesion resp.
sticking and anisotropy for sliding are treated. Anisotropic resp. orthotropic
tensors inherit their properties either from the spectral decomposition or, in
the more general case, from arbitrary curvilinear coordinate systems defined
on the surface. The last case has advantages in practical applications as e.g.
for a homogenized average model of machined surfaces. This structure of
tensors automatically satisfies all necessary theoretical restrictions developed
earlier in [61] and [214]. Thus, a consistent model for anisotropic surfaces can
be developed.
In order to define sliding forces as well as sliding displacements the prin-
ciple of maximum dissipation is used. All models allow representative geo-
metrical interpretations on the tangent plane. Via this principle, the sliding
forces and sliding displacements are derived in the covariant form. The formu-
lation in the covariant form easily allows to derive the consistently linearized
equations, which are necessary for the iterative solution within a Newton
type scheme, even for the case with nonlinear anisotropy in the reference
Cartesian coordinate system.
11.1.2 Available Finite Element Models

The finite implementation of anisotropy only for friction law has been briefly
discussed in Montmitonnet and Hasquin [129] with an application to hot
rolling processes, and presented in details in Alart and Heege [4]. A sym-
bolic computation software has been exploited to derive the corresponding
tangent matrices in [4]. In the current chapter, the covariant form of tangent
matrices allows the straightforward implementation into a finite element code
even for arbitrary curved contact surfaces possessing anisotropic properties
for both friction and adhesion. The set of numerical examples is chosen to
show the influence of the orthotropy type on the development of the sticking-
sliding zone as well as on the kinematic behavior of the contact bodies. Thus,
constant orthotropy is classified by the eigenvalue ratio of the correspond-
ing tensor defining the adhesion region. These cases are thoroughly investi-
gated by the development of the sticking-sliding zone. It will be shown that a
specific combination of both, orthotropy for adhesion and orthotropy for fric-
tion, can lead to the so-called geometrical isotropy, when the contact bodies
show kinematically isotropic behavior. It is also demonstrated that various
kinematical structural properties of arbitrarily curved contact surfaces such
as a bolt connection can be modeled by means of the adhesion tensor.
The full Newton method together with the penalty method for frictional
contact problems such as non-penetration and sticking conditions, only within
a certain tolerance. This tolerance is defined by the penalty parameters for
both normal and tangential direction. The classical method of Lagrangian
multipliers leads to an exact satisfaction of contact constraints, however, ad-
ditional degrees of freedoms for the contact tractions are often mentioned
among the disadvantages of this method. As a good alternative, available in
many finite element software, the Augmented Lagrangian approach is em-
ployed. It allows to satisfy the contact constraints for non-penetration and
sticking within the nested iterative algorithm. The method is described theo-
retically in Bertsekas [12] and Fortin and Glowinski [41]. Pietrzak and Curnier
[145] developed the Augmented Lagrangian approach for frictional contact
including the corresponding saddle point functional.
In the second part of this chapter, the nested Uzawa algorithm is ap-
plied for symmetrization within the Augmented Lagrangian approach for an
anisotropic friction model including adhesion and friction. The geometrical
structure of the corresponding tangent matrices allows us to construct very
simple symmetric tangent matrices for the anisotropic case. The isotropic
case can then be defined as a reduction of the anisotropic case. The numeri-
cal examples show the good convergence behavior for various problems such
as small and large sliding problems.
11.2 Examples of the Finite Element Implementation

of the Coupled Anisotropic Adhesion-Friction
Model
In this section we will discuss details of the finite element implementation and
necessary definitions for the coupled anisotropic adhesion-friction model pro-
posed in Sect. 6.3 for the full Newton method. In particular, the anisotropic
structure has to be defined on the whole contact surface. This leads to addi-
tional difficulties concerning a unique description for the whole surface and
not only an approximation for the corresponding contact elements. There-
fore, we start with the simplest contact element defining contact with an
anisotropic rigid surface a so-called point-to-analytical surface contact ele-
ment. For this a node of a FE-mesh as well as an integration point of an
element can be taken as the contact point. In this case, the contact point
itself can be seen as containing history variables. If the anisotropic surface is
deformable then the node-to-segment strategy has to be applied. In this case
the history variables of passing nodes are stored with the associated contact
11.2 FE Implementation of Anisotropic Adhesion-Friction Model 335
segments. As a more general case, the re-parametrization of the complete

contact surface in the case of contact of two deformable bodies is discussed.

Linear Surface Approximation of a
Deformable Body
The contact of a body meshed with bilinear finite elements – and thus bilin-
ear contact surface elements – with a rigid anisotropic surface is one of the
simplest cases to define a contact element. In this case, a node of the FE
meshed surface is taken as a contact point, while all necessary anisotropic
tensors are defined on the rigid surface which can geometrically be described
by analytical functions. The corresponding geometrical characteristics as nor-
mal vector n and tangent vector ρi are then directly given by the description
of the analytical surface. All integrals for the tangent matrix as well as for
the residuum are defined for one nodal point. Since the anisotropic surface
is rigid, all rotational parts can be set to zero. Thus, we obtain the following
matrices for the contact contributions.
Matrix for the Normal Part
KN = −N n ⊗ n. (11.1)
Matrix for the Tangential Part. Sticking Case
Kstick
T = bij ρi ⊗ ρj . (11.2)
Matrix for the Tangential Part. Sliding Case
N T̂i aij bji fjn bln aik |N |

Kslide
T = ρj ⊗ n − ρk ⊗ ρl (11.3)
Ψ Ψ
fiq bij fm b T̂q T̂n ank |N |
j ml
+ ρk ⊗ ρl .
Ψ3
All matrices contain only constitutive parts and belong to the correspond-
ing nodes. The contact node then owns also the necessary history variables
1 2
from the previous converged step (n): convective coordinates ξ(n) , ξ(n) and
(n) (n)
tangential contact forces T1 , T2 .
The function Ψ is given in eqn. (7.52). The components T̂i of the auxiliary
force T̂ are computed in eqn. (7.51) (in vector form in eqn. (7.50)) via the
trial force Tktr , see other details in the corresponding Sect. 7.3.
Remark 11.1. During computation of eqn. (7.49) an expression of order

O(max ε2i / min μ2i )
appears, which can cause a large numerical error due to
O(max ε2i / min μ3i )
the loss of numerical tolerance either in the case of large εi , or small μi . Thus,
such cases must be treated separately, e.g. the computation of the auxiliary
force in eqn. (7.50) as well as of the sliding force in eqn. (7.49) can be ob-
tained with a scaled matrix B/ max εi etc. Of cause, a limit non-frictional
case leading to Tsl = 0, must be separated leading numerically to expression
1/0 in friction tensor components.

Arbitrary Surface Approximation of the
Deformable Body
If a body is meshed with finite elements of higher order of approximation
for the surface – leading to a contact element of the same high order – then
an integration point of the FE surface elements has to be taken as a contact
point. In this case, the vector rs describes the analytical surface and the vector
ρ is computed from the finite element mesh. Algorithmic aspects of contact
problems with a surface described by analytical functions are discussed in
Sect. 4.3. The algorithmic discretization of the tangent matrices, presented
in Sect. 7.3, is obtained as follows. Let ue be the nodal displacement vector
taken from the finite element discretization as
(1) (1) (1) (2) (2) (2) (n) (n) (n)
uTe = {u1 , u2 , u3 , u1 , u2 , u3 , ..., u1 , u2 , u3 }T , (11.4)
where (n) is a number of nodal points of the contact surface element. As-
suming that the approximation is performed with (n) shape functions, only
a single position matrix A is necessary for discretization of all contact con-
tributions.
⎡ ⎤
N1 0 0 N2 0 0 ... ... ... N(n) 0 0
A = ⎣ 0 N1 0 0 N2 0 ... ... ... 0 N(n) 0 ⎦ . (11.5)
0 0 N1 0 0 N2 ... ... ... 0 0 N(n)
The contact matrix for the normal part in eqns. (7.9a), (7.9b), (7.9c) is then
obtained as (only the main part left):
#
KN = − N AT n ⊗ nAds = (11.6)
s
Np

=− N AT n ⊗ nAWI WJ det J(ξI1 , ξJ2 ) ,
I,J=1
11.2 FE Implementation of Anisotropic Adhesion-Friction Model 337
where NP is the number of integration points and WI , I = 1, 2, ..., NP are

weights of the chosen quadrature formula. The determinant of the Jacobian
det J(ξI1 , ξJ2 ) is computed for the surface segment – the contact segment – of
the body. For each integration point a set of history variables from the previ-
1 2
ous converged step (n) must be stored: convective coordinates ξ(n) , ξ(n) and
(n) (n)
tangential contact forces T1 , T2 . The tangent matrices for the tangential
part as defined in Sect. 7.3.1 and 7.3.1 can be derived in a similar fashion.
Remark 11.2. For this specific case all geometrical characteristics such as the
normal vector n, the tangent vector ρi as well as the anisotropic tensors are
taken from the rigid surface which is defined by analytical functions.
11.2.3 Node-to-Segment Approach. Deformable

Anisotropic Contact Surface
If a deformable body has a surface with anisotropic properties and the slave
body is discretized with linear finite element, then the node-to-segment ap-
proach can be applied. In this case, the corresponding contact segments cov-
ering the anisotropic surface are taken as master segments and slave node is
taken from the linear meshed slave body. A nodal displacement vector con-
tains then an additional (n + 1) slave node besides the first (n) nodes from
the master segment:
(1) (1) (1) (n) (n) (n) (n+1) (n+1) (n+1) T
uTe = {u1 , u2 , u3 , ..., u1 , u2 , u3 , u1 , u2 , u3 } . (11.7)
The approximation matrix A is constructed similar to the standard NTS

approach, see Sect. 8.1.1, as
⎡ ⎤
N1 0 0 N2 0 0 ... N(n) 0 0 −1 0 0
A = ⎣ 0 N1 0 0 N2 0 ... 0 N(n) 0 0 −1 0 ⎦ . (11.8)
0 0 N1 0 0 N2 ... 0 0 N(n) 0 0 −1
The components of the tangent matrices (normal and tangent vectors etc.)
are computed in the projection point of the master segment. The structure
is again algorithmic, e.g. a part for the normal contact has the form:
#
KN = − N AT n ⊗ nAds = (11.9)
s
= −N AT n ⊗ nA.
Here, the associated master segment contains the aforementioned history

variables.
Remark 11.3. In the case of arbitrary approximations for both master and
slave body, the Segment-To-Segment (STS) approach, see Sect. 8.1.2, should
be employed.
Mapping of Anisotropic Properties from the Surface to a Contact

Segment
It is expected, that the anisotropic properties are defined for the complete
surface, not only for a segment. Thus, the main problem is, how to properly
transfer the anisotropic properties from the surface to the contact segment.
In the case of a simple curvilinear rectangular patch this can be organized
as follows. Let s1 , s2 are convective coordinates defining the parametrization
of the patch, see Fig. 11.1, with 0 ≤ s1 , s2 ≤ 1. The anisotropic properties
are determined then via the tensor basis e1 (s1 , s2 ) ⊗ e2 (s1 , s2 ). The regular
numbering i = 1, ..., m and j = 1, ..., n is introduced according to the mapped
mesh on the patch, see Fig. 11.2. Therefore, a direct mapping of the convective
coordinate on the element ξ 1 can be defined as
⎫
j−1⎪
ξ 1 = −1 −→ s1 = ⎬ 2j − 1 + ξ 1
n =⇒ s1 = . (11.10)
j ⎪ ⎭ 2n
ξ 1 = 1 −→ s1 =
n
According to the introduced direct transformation, the backward transfor-

mation is defined according to the following algorithm:
do j = 1, n
j−1 j
if ≤ s1 ≤
n n
then k = j (11.11)
ξ 1 = 2ns1 − 2k + 1
endif
enddo
The second coordinate s2 is analogously computed.
r1
e2
e1
r2
s1
s2
Fig. 11.1 Curvilinear rectangular patch.

11.3 Effects of the Coupled Anisotropic Adhesion-Friction Model 339
m1 mn
2
ξ
1
ξ
in
ij
21 22
s2
11 12 1n
s1
Fig. 11.2 Curvilinear rectangular patch. Mapping scheme.
11.3 Numerical Examples Illustrating Effects of the

Coupled Anisotropic Adhesion-Friction Model
In this section we present numerical examples illustrating several types of
the orthotropy. As known, the orthotropic frictional properties of the surface
leads to changing of kinematical behavior of the contact bodies, see [22], [24],
[67], [82], [215], [216], [217], [206]; therefore, the set of numerical examples
is chosen to illustrate particular kinematical effects which appear due to the
presence of anisotropy for adhesion and friction. In the first example we
chose constant orthotropy on a plane represented by a tensor with spectral
decomposition. This model possesses a simple mechanical interpretation a
so-called rheological model. Constant orthotropy is thoroughly investigated
for the case with small displacements in order to show the development and
the distribution of the sticking-sliding zone for different types of orthotropy.
These cases are considered for the start of sliding as well as for large sliding
deformations in order to consider the trajectories of a block for different types
of orthotropy. Then a large displacement problem for a plane with polar
orthotropy is taken as an example for curvilinear orthotropy on the plane.
In order to show the robustness of the developed approach for curvilinear
surfaces, kinematical effects of a bolt connection are modeled with spiral
orthotropy defined on a cylinder.
11.3.1 Linear Constant Orthotropy on the Plane

In order to investigate the properties of the proposed model, a rectangular
block is considered on an orthotropic plane. The dimensions of the block are
10 × 10 × 4 with linear elastic properties: Young’s modulus E = 2.10 · 104 and
Poisson ratio ν = 0.3, assumed within a consistent dimension system. The
block (see Fig. 11.3) is located on the XOY plane and loaded by prescribing
displacements on the upper surface in two steps: 1) vertical loading with

w = 1.0·10−2, 2) incremental loading with Δu along the X axis. Contact with
regard to the constant orthotropic model, see Sect. 6.2.4, with the adhesion
tensor

B = [bij ] = − 1 , (11.12)
and the friction tensor

⎡ 1 1 1 1 ⎤
cos2 β + 2 sin2 β ( 2 − 2 ) sin β cos β
⎢ μ1
2 μ2 μ1 μ2 ⎥
⎢ ⎥
F = [fki ] = ⎢ ⎥ (11.13)
⎣ 1 1 1 1 ⎦
2 2
μ1 μ2 μ1 μ2
is specified between the plane and the block. To compare both approaches
contact is modeled with a point-to-analytical surface contact element as well
as with a node-to-segment approach. The rigid plane is taken as master seg-
ment within the latter approach. As an example coincident orthotropy angles
α = β are chosen. First, we will investigate the development of a sticking-
sliding zone for small displacements for various cases of the ratio rλ = με11 · με22 ,
see the mechanical interpretation in Sect. 6.3.7 and eqn. (6.138). Afterward,
we will consider the large displacement problem and investigate the trajec-
tories of the block depending on the surface properties.
Fig. 11.3 Geometry and loading of the rectangular in plane block.

Small Displacement Problem. Development of the

Sticking-Sliding Zone
In order to investigate the development of the sticking-sliding zone small
displacement increments with Δu = 1.0 · 10−4 along the X axis are applied.
The following cases are considered:
1. Isotropic case ; both, the friction and the adhesion tensors are isotropic
with: penalty parameter for the normal traction εN = 2.1·105, parameters
of the adhesion tensor ε1 = ε2 = 2.1·105, parameters of the friction tensor
μ1 = μ2 = 0.3. Orthotropy angles: α = β = 0o .
2. Geometrically isotropic case with: penalty parameter for the normal trac-
tion εN = 2.1 · 105 , parameters of the adhesion tensor ε1 = 3.0 · 105 ,
ε2 = 2.0 · 105 , parameters of the friction tensor μ1 = 0.3, μ2 = 0.2. Or-
thotropy angles: α = β = 45o . This case leads to the ratio of eigenvalues
eqn. (6.138) rλ = 1 and, thus, to a circular adhesion region.
3. Orthotropic case ; orthotropic adhesion, isotropic friction with: penalty
parameter for the normal traction εN = 2.1 · 105 , parameters of the
adhesion tensor ε1 = 2.0 · 105 , ε2 = 3.0 · 105 , parameters of the friction
tensor μ1 = μ2 = 0.3. Orthotropy angles: α = β = 45o . The eigenvalue
ratio is then rλ = 2/3.
4. Orthotropic case ; isotropic adhesion, orthotropic friction with: penalty
parameter for the normal traction εN = 2.1 · 105 , parameters of the
adhesion tensor ε1 = ε2 = 2.1 · 105 , parameters of the friction tensor
μ1 = 0.3, μ2 = 0.2. Orthotropy angles: α = β = 45o . The eigenvalue
ratio is then rλ = 2/3.
Y
11
10
9
8
7
6
5
4
3
2 ϕ
1
1 2 3 4 5 6 7 8 9 10 11
X
Fig. 11.4 Numbering of nodes on the lower surface and direction ϕ of the devel-
opment of a sliding zone.
Isotropic case (1). The results are depicted in the diagram in Figure 11.5
showing the developed sticking area (in grey color) for the applied horizontal
displacements on the lower contact surface in several states from the top view.
This area is identified by sticking nodes on the lower contact surface; these
nodes are inside the adhesion ellipse. For the results of the investigations we
use the node numbering and the inclination angle ϕ for the block as given in
Fig. 11.4. One can observe that the edge nodes are sliding from the beginning,
i.e. when only vertical loading is applied. This effect is due to the singularity
of stresses on the edges known from the analytical solution for a rigid punch
problem, see e.g. Johnson [77]. The symmetrical sticking region is vanishing
and the block begins to slide fully, once the applied horizontal displacements
are beyond the value u = 10.0 · 10−3 .
It is also interesting, especially for the forthcoming orthotropic cases, to
observe the beginning of full sliding. Both the initial configuration and the
scaled deformed configuration from the bottom view are depicted in Fig. 11.6
for loading u = 9.0 · 10−3 (the block begins to slide partially) as well as for
loading u = 20.0·10−3 (fully developed sliding of the block). The deformation
is symmetric as the horizontal axis and the current horizontal symmetry axis
are moving along the reference axis.
Geometrically isotropic case (2) . The results are depicted in the diagrams
in Figure 11.7 showing the development of the sticking area. It is interesting
to observe that the initial sticking area u = 0.0 in this case is symmetric
along the main orthotropy axes which are turned according to the references
coordinate system by the angles α = β = 45o .
We recall the geometrical interpretation to explain this phenomena. The
adhesion region becomes a circle in this particular example, see Fig. 11.9.
The elastic properties inside the circle are orthotropic according to the com-
putation of the trial force as Ttr = BΔx with the adhesion tensor computed
with α = 45o , see eqn. (11.12):

1 ε1 + ε2 ε1 − ε2 105 5 1
B = [bj ] = −
i
=− . (11.14)
2 ε1 − ε2 ε1 + ε2 2 1 5
Since, in the current example the stiffness in the second direction is smaller
than in the first one ε2 < ε1 , all points Ael tend to reach the boundary
of the adhesion circle in the direction of the ξ 2 -axis. This can explain the
asymmetric behavior of the initial sticking area in Fig. 11.7 (a). The direction
of the sliding force Tsl , which is starting to act from the boundary of the
circle, is defined by the matrix BFB, see eqn. (7.48), which in the current
example becomes:
⎡ 2 ⎤
ε1 ε22 ε21 ε22
+ − 6
1⎢ μ22 ⎥
2 μ22 μ21
⎢ μ1 ⎥ 10 0
BFB = ⎢ ⎥ = . (11.15)
2 ⎣ ε2 ε 2
ε 2
ε 2 ⎦ 0 106
1
− 22 12 + 22
μ21 μ2 μ1 μ2
This recovered isotropic behavior of the sliding force Tsl is depicted in

Fig. 11.9. This effect is depicted for the developed sliding in Fig. 11.8, where
the horizontal axis of symmetry of the deformed body is moving along the
reference axis.
(a) u = 0.0 (b) u = 1.0 (c) u = 2.0 (d) u = 3.0 (e) u = 4.0
(f) u = 5.0 (g) u = 6.0 (h) u = 7.0 (i) u = 8.0 (j) u = 9.0
Fig. 11.5 Isotropic case (1). Development (degeneration) of sticking zone for several
displacement states. Horizontally applied displacement u pointing into the right
direction (u is scaled by 10−3 ).
For completeness, we would like to present the results showing the conver-
gence of the sticking zone with regard to: 1) adhesion parameters; 2) mesh
density. Thus, Fig. 11.10 represents the comparison of the initial sticking
zones computed first for the set of parameters εN = 2.1 · 103 , ε1 = 3.0 · 103 ,
ε2 = 2.0 · 103 , then for the same set, but instead of 103 scaled sequentially
by the factors 104 , 105 and 106 . For the latter computation the vertical load-
ing has been provided in 10 load steps. The results serve also to show the
convergence of the results with increasing penalty parameters.
The influence of the mesh density on the sticking zone for the case of
parameters εN = 2.1 · 105, ε1 = 3.0 · 105, ε2 = 2.0 · 105 is shown in Fig. 11.11.
The uniform mesh was varied as 10 × 10 × 4, 16 × 16 × 6, 20 × 20 × 8 and
32 × 32 × 8 respectively, the third number always representing the number of
elements in thickness direction.
Orthotropic case (3). The results are depicted in the diagrams in Fig. 11.12
showing the development of a closed sticking area. The orthotropy of the ad-
hesion region results in the sticking region being turned by the angle 45o .
The development of a sliding zone starts at the upper right corner and con-
tinues unsymmetrically leading further to a parallel shifting of the block and
a straight trajectory inclined at an angle ϕ (see definition at Fig. 11.4) in the
case of large displacements. This effect is depicted in Fig. 11.13, a bottom
view.
Z X
Z X
(a) (b)
Fig. 11.6 Isotropic case (1). Initial and deformed configuration. Bottom view.
Displacements scaled by factor 150. Applied horizontal displacements on the upper
surface: (a) u = 9.0 · 10−3 (start of sliding); (b) u = 20.0 · 10−3 (fully developed
sliding).
Orthotropic case (4). The results are depicted in the diagrams in Fig. 11.14
showing the development of a closed sticking area. Now, isotropy for the ad-
hesion region together with a small difference between the friction coefficients
results in the obtained symmetric region. As known, large differences between
the coefficients lead to an unsymmetrical region, see e.g. the results for the
classical orthotropic friction model in [67]. The sticking region is diminishing
unsymmetrically resulting globally in a shifting of the block, see Fig. 11.15.
Large Displacement Problem. Investigation of Influence of

Adhesion Parameters on the Trajectory of the Block
The goal of this analysis is to show that with properly chosen parameters for
the adhesion tensor it is possible to achieve a predefined motion on the sur-
face. Namely, if the surface is uniformly orthotropic with a given orthotropy
angle α then it is possible to prescribe a straight trajectory of the block with
an angle ϕ ≈ α, see Fig. 11.4, keeping the driving force at a certain level.
Moreover, it is possible to obtain even the geometrically isotropic case, when
a trajectory is a straight line coinciding with the direction of the force, though
the friction tensor remains orthotropic leading to different global forces in the
different global directions.
(a) u = 0.0 (b) u = 1.0 (c) u = 2.0 (d) u = 3.0 (e) u = 4.0
(f) u = 5.0 (g) u = 6.0 (h) u = 7.0 (i) u = 8.0 (j) u = 9.0
Fig. 11.7 Geometrically isotropic case (2). Development of the sticking zone for
various displacement states. Horizontally applied displacement u pointing into the
right direction (u is scaled by 10−3 ).
Z X
Z X
(a) (b)
Fig. 11.8 Geometrically isotropic case (2). Initial and deformed configuration.
Bottom view. Displacements scaled by factor 150. Applied horizontal displacements
on the upper surface: (a) u = 9.0 · 10−3 (start of sliding); (b) u = 20.0 · 10−3 (fully
developed sliding).
In the numerical examples the penalty parameter for the normal trac-
tion is taken as εN = 2.1 · 104 for all cases; the orthotropy angles are given
as α = β = 45o . The horizontal incremental displacements are given as
X2
2 ξ1
ξ
ε1
ε2
el sl
α A T sl A
O 1
T tr X
Fig. 11.9 Adhesion region and direction of an elastic trial force and a sliding
force for specially chosen parameters leading to the geometrically isotropic case
(2). ε1 < ε2 , α = 45o .
(a) 103 (b) 104 (c) 105 (d) 106
Fig. 11.10 Geometrically isotropic case (2). Initial sticking zone. Variation of
penalty and adhesion tensor parameters. (a) scale factor 103 , one load step. (b)
scale factor 104 , one load step. (c) scale factor 105 , one load step. (d) scale factor
106 , 10 load steps.
Δu = 5.0 · 10−2 in order to reach a sliding state of the block from the begin-
ning. The other parameters are varied to achieve different cases as presented
in Table 11.1.
The trajectories of the block for all cases are depicted in Fig. 11.16. It
becomes obvious that the block tends to move into the direction of the
eigenvector with smaller eigenvalue λ. E.g. if rλ = ε1 /ε2 = 0, then the
trajectory is inclined at the angle of orthotropy ϕ ≈ α = 45o and vice
versa if rλ = ε1 /ε2 = ∞, i.e. ε2 = 0, then the block is inclined at angle
ϕ ≈ −45o . The other parameter variations lead to different trajectories with
angles −45o < ϕ < 45o . As a particular example, the geometrically isotropic
case is recovered for the ratio rλ = 1 leading to a circular adhesion region.
In this particular case the block is moving into the direction of the applied
force.
(a) 10 × 10 (b) 16 × 16 (c) 20 × 20 (d) 32 × 32
Fig. 11.11 Geometrically isotropic case (2). Initial sticking zone. Variation of mesh
density. (a) 10 × 10 elements in plane. (b) 16 × 16 elements in plane. (c) 20 × 20
elements in plane. (d) 32 × 32 elements in plane.
(a) u = 0.0 (b) u = 1.0 (c) u = 2.0 (d) u = 3.0 (e) u = 4.0
(f) u = 5.0 (g) u = 6.0 (h) u = 7.0 (i) u = 8.0 (j) u = 9.0
Fig. 11.12 Orthotropic case (3). Development (degeneration) of sticking zone for
several displacement states. Horizontally applied displacement u pointing into the
11.3.2 Polar Orthotropy on a Plane. Large

Displacement Problem
A more complex orthotropy is given by a polar orthotropy on a plane, see
derivation in Sect. 6.2.4. As in the previous example, we will show that it is
possible to define the orthotropic structure of a plane by using the adhesion
tensor. An elastic block with dimensions 1×1×0.25 and mesh 4×4×1 is sitting
on a rigid block, see Fig. 11.17. Linear elastic material is assumed within a
consistent dimension system: Young’s modulus E = 2.10 · 104 ; Poisson ratio
ν = 0.3. The loading is applied sequentially by prescribing displacements on
the upper surface in (1 + n) steps: 1) vertical loading with w = 1.0 · 10−2 , 2)
n steps with horizontal displacement increments Δu = 1.0 · 10−2 along the
X axis. The structure of the adhesion tensor is as follows,
Z X
Z X
(a) (b)
Fig. 11.13 Orthotropic case (3). Undeformed and deformed configuration. Bottom
view. Displacements scaled by factor 150. Applied horizontal displacements on the
upper surface: (a) u = 9.0·10−3 (start of sliding); (b) u = 20.0·10−3 (fully developed
sliding).
(a) u = 0.0 (b) u = 1.0 (c) u = 2.0 (d) u = 3.0 (e) u = 4.0
(f) u = 5.0 (g) u = 6.0 (h) u = 7.0 (i) u = 8.0 (j) u = 9.0
Fig. 11.14 Orthotropic case(4). Development (degeneration) of sticking zone for

several displacement states. Horizontally applied displacement u pointing into the

1 εr x2 + εϕ y 2 (εr − εϕ )xy
B=− . (11.16)
x + y2
2 (εr − εϕ )xy εr y 2 + εϕ x2
Z X
Z X
(a) (b)
Fig. 11.15 Orthotropic case (4). Undeformed and deformed configuration. Bottom
view. Displacements scaled by factor 150. Applied horizontal displacements on the
upper surface: (a) u = 9.0·10−3 (start of sliding); (b) u = 20.0·10−3 (fully developed
sliding).
Table 11.1 Sliding of a block on a plane. Large displacement problem. Variation

of orthotropy. Computed inclination angle of a sliding block ϕ.
Case Adhesion tensor Friction tensor Eigenvalue ratio Resulting angle ϕ

ε1 = 3.0 · 103
1 μ1 = μ2 = 0.3 ∞ −44.95o
ε2 = 0.0
ε1 = 3.0 · 103
2 μ1 = μ2 = 0.3 3/2 −21.01o
ε2 = 2.0 · 103
ε1 = 3.0 · 103 μ1 = 0.3
3 1 0o
ε2 = 2.0 · 103 μ2 = 0.2 geom. isotropy
ε1 = 2.0 · 103
4 μ1 = μ2 = 0.3 2/3 21.01o
ε2 = 3.0 · 103
ε1 = 0.0
5 μ1 = μ2 = 0.3 0 44.95o
ε2 = 3.0 · 103
The chosen contact surface parameters are: penalty parameter for the normal
traction εN = 2.1 · 105; isotropic friction tensor with μ1 = μ2 = 0.2. Adhesion
tensor with cases: a) εr = 100, εϕ = 1000, b) εr = 2000, εϕ = 5000, c) εr =
1000, εϕ = 1000, d) εr = 5000, εϕ = 2000, e) εr = 1000, εϕ = 100, f) εr =
1000, εϕ = 0.0.
ϕ = -44.95
ϕ = -21.01
0.4 ϕ = 0.00
ϕ = 21.01
ϕ = 44.95
0.2
OY
-0.2
-0.4
0 0.2 0.4 0.6 0.8 1

OX
Fig. 11.16 Trajectories of the block and inclination angle ϕ for various cases of
orthotropy with the eigenvalue ratios: rλ = ∞ =⇒ ϕ = −44.95o ; rλ = 3/2 =⇒ ϕ =
−21.01o ; rλ = 1.0 =⇒ ϕ = 0.00o ; rλ = 2/3 =⇒ ϕ = 21.01o ; rλ = 0.0 =⇒ ϕ =
44.95o .
In Fig. 11.18 the sequence of the motions for all cases leading to different
ratios rε = εϕ /εr is depicted. It is obvious, that the desired circular mo-
tion can be achieved by prescribing the corresponding eigenvalue to a small
value, e.g. rε = 0.0 in the current example. The last result can be derived
also analytically from the analysis of global motion of a block, see proof in
Sect. 6.4.3.
Remark 11.4. The trajectory of the block depends on the ratio λϕ /λr . How-
ever, the numerical computations show that it is possible to prescribe the de-
sired trajectory by only controlling the adhesion tensor parameters. Attempts
to achieve the desired trajectory controlling the friction tensor parameters,
e.g. taking μr ≈ 0 for the case (f), lead to disconvergence. A straightforward
conclusion is that the geometrical structure of the surface in the sense of
the desired trajectory can be defined via the adhesion tensor, while other
Fig. 11.17 Geometry and loading for the case of polar orthotropy.
mechanical characteristics such as the measured global forces are defined via
the friction tensor.
11.3.3 Spiral Orthotropy on the Cylinder

Another rather complex kinematical behavior of a curved contact surface, e.g.
a machined surface of a bolt can be described by means of controlling the
adhesion tensor parameters. The model of spiral orthotropy on the cylinder
developed in Sect. 6.2.4 allows to describe the kinematical behavior of a bolt
connection with a rather coarse mesh. In order to show this, we consider a
finite element model of the bolt, see Fig. 11.19.
The bolt is modeled with linear finite elements with elastic properties:
E = 2.1 · 104 , ν = 0.3. Contact is modeled with the point-to-analytical
surface contact element and specified on the cylindrical surface of the bolt,
i.e. each node on the bolt surface is a contact node. The important dimensions
of the example are the radius of a cylinder R = 3.0 and the distance H
between threads of a spiral line H = 3.3333. The central axis is constrained
to move along the OZ-axis and a rotation with an angle increment Δϕ = 1o
is applied to the head of the bolt. In order to supply contact with a rigid
external cylindrical surface an initial penetration is specified as 1.0 · 10−4
with a normal penalty parameter εN = 103 . The friction tensor is chosen
to be isotropic with μ1 = μ2 = 0.01. The adhesion tensor parameters ε1 , ε2
are chosen according to the following tensor representation (see derivation in
Sect. 6.2.4)
(a) rε = 10/1 (b) rε = 5/2 (c) rε = 1/1
(d) rε = 2/5 (e) rε = 1/10 (f) rε = 0
Fig. 11.18 Motion of the block in the case of polar orthotropy on the plane.
Varying adhesion tensor parameters: rε = εϕ /εr = 10/1, 5/2, 1, 2/5, 1/10, 0;
isotropic friction tensor parameters: μ1 = μ2 = 0.2. Loading by prescribed vertical
displacement w and incremental y-displacement Δu.
Table 11.2 Rotation of a bolt. Variation of parameters of the adhesion tensor.

Resulting distance of a longitudinal motion h after rotation of a bolt by 360o .
Case Adhesion tensor Ratio rε Resulting distance h

ε1 = 0.0
1 0.0 1.000 H
ε2 = 10.0
ε1 = 5.0
2 1/2 0.742 H
ε2 = 10.0
ε1 = 7.5
3 3/4 0.430 H
ε2 = 10.0
ε1 = 10.0
4 1.0 0.000 H
ε2 = 10.0 geom. isotropy
ε1 = 10.0
5 2/1 −0.487 H
ε2 = 5.0
ε1 = 10.0
6 4/3 −2.758 H
ε2 = 7.5
ε1 = 10.0
7 ∞ −31.98 H = Ĥ
ε2 = 0.0
Fig. 11.19 Finite element model of a bolt connection. Outer surface is rigid and
is illustrated with one element only.
⎡ ⎤
R2 H
gε R 2 (ε1 − ε2 )
⎢ 2π ⎥
1 ⎢ ⎥
B = bij ρi ⊗ ρj = −
2 ⎢
2 ⎥
H ⎣ 2
R H H ⎦
R2 + (ε 1 − ε 2 ) ε 1 + ε 2 R 2
2π 2π 2π
(11.17)

2
2 H
with gε = ε1 R + ε2 . The adhesion tensor parameters are varying
2π
according to Table 11.2. The results of the analyses are depicted in Fig. 11.20
showing the computed longitudinal motion of the bolt along the OZ-axis vs.
the applied rotation angle. The resulting distance of a longitudinal motion h
after the rotation of a bolt by 360o is also presented in Table 11.2. Case (1)
with ratio rε = 0 leads to a pure forward motion according to the motion
along the main spiral line. The bolt moves forward at the distance H while
the bolt is rotating at the full angle 360o . The geometrically isotropic case
(4) does not lead to any longitudinal motion. For ratios rε > 1 a backward
motion is obtained. Finally, the case (7) with ratio rε = ∞ (not shown in
Fig. 11.20) leads to a motion along the orthogonal spiral, i.e. the bolt moves
(2πR)2
backwards at the distance Ĥ = , see Remark 6.9.
H
(a)
4
case 1
case 2
H case 3
Observed motion Z
3 case 4
case 5
case 6
2
0
60 120 180 240 300 360
-1 Applied rotation angle

(b)
Fig. 11.20 a) Spiral orthotropy on a cylinder. b) Observed longitudinal motion vs.

applied rotation. Variation of ratio rε = ε1 /ε2 . Loading cases according to Table
11.2.
Remark 11.5. This example is only chosen to illustrate the possibility to de-
scribe machined surfaces from a kinematical point of view. The applicability
of the proposed model to stress analysis of the bolt connection requires, cer-
tainly, a more sophisticated analysis.
Remark 11.6. The unsymmetrical solver is employed here and necessarily

combined with an iterative Newton scheme.
11.4 Symmetrization – Augmented Lagrangian Approach 355
11.3.4 Discussion
The proposed model for contact interfaces is analyzed for constant orthotropy
on a plane, for polar orthotropy on a plane and for spiral orthotropy on a
cylinder. The reliability of the numerical results is controlled by checking the
convergence with increasing the adhesion parameters and by remeshing. A
classification of orthotropy based on the ratio of eigenvalues for the corre-
sponding orthotropic tensor is proposed and the kinematical effects for the
classified cases are numerically investigated. In particular, as a specific case
of complex anisotropy the geometrically isotropic kinematic behavior of the
contact bodies can be found despite the presence of both, anisotropy for adhe-
sion and friction. The possibility of modeling machined surfaces with the help
of the adhesion tensor is shown by numerical examples for polar orthotropy
and for the model of a bolt connection.
The presented approach shows an algorithmic inclusion in covariant form
only for linear elastic adhesion and for Coulomb’s friction law. Any combina-
tion of elastic, visco-elastic or nonlinear laws for the adhesion region together
with non-associative or associative friction laws can be derived in straight-
forward covariant manner based on the metrics of the contact surfaces.
11.4 Symmetrization of Various Friction Models Based

on an Augmented Lagrangian Approach
The standard implementation of the classical Coulomb friction model to-
gether with both the Newton iterative method, and the Augmented La-
grangian method leads to non-symmetric tangent matrices for sliding zones
of contact surfaces. This fact is known in literature as consequence of the
non-associativity of the friction law. Laursen and Simo [109] proposed a sym-
metrization procedure based on the nested Uzawa algorithm. This approach
is based on symmetrization of the corresponding tangent matrices for the so-
lution of the equilibrium equations, while the contact constraints have been
still iteratively satisfied within the external loop with a specified tolerance.
Considering anisotropic models for friction, especially including coupling
of adhesion and friction, an additional non-symmetry due to anisotropy is
arising.
The geometrical structure of the corresponding tangent matrices allows
us to construct very simple symmetric tangent matrices for the anisotropic
case. The isotropic case can then be defined as a reduction of the anisotropic
case. The numerical examples show the good convergence behavior for various
problems such as small and large sliding problems.
11.4.1 Structure of Matrices after the Linearization

Process
We are recalling here the result of the linearization for the coupled adhesion-
friction model developed in Sect. 7.3 in order to study inclusion of each part
into the symmetric Uzawa algorithm. The curvature parts in the following
are excluded further only for short.
Linearization of the Normal Part δWcN

We denote L(f ) as a linearization operator acting on a functional f in the
covariant form. Thus, linearization of the virtual work of the contact normal
traction N is given as:
# #
L(δWcN ) = L( N δξ ds) = − N (δrs − δρ) · (n ⊗ n)(vs − v)ds
3
s
s !
leads to KmN
(11.18a)
#

− δρ,j · aij (n ⊗ ρi )(vs − v) + (δrs − δρ) · aij (ρj ⊗ n)v,i ds.
N ξ 3
s !
leads to KrN
(11.18b)
Here, the first term (resp. the second term) after approximation of the geom-
etry leads to the main part of the contact matrix Km N (resp. the rotational
part of the contact matrix KrN ).
Linearization of the Tangential Part δWcT , Sticking Case

The sticking case is understood as a case where the tangential traction re-
mains in the elastic region and, therefore, is computed via the evolution
equations ii). Linearization leads to
#
D(δWcT ) = D( Titr δξ i ds) = (11.19)
s
#
= (δrs − δρ) · B(vs − v)ds−
s
!
leads to KmT ,st
#

− Ti (δrs − δρ) · ail ajk ρk ⊗ ρl vj + δρ,j · aik ajl ρk ⊗ ρl (vs − v) ds,
s
!
leads to KrT ,st
Linearization of the Tangential Part δWcT , Sliding Case

The sliding case is understood as a case where the yield condition iii) is ful-
filled and, therefore, the tangential traction is computed as a sliding traction
given in eqn. (6.109). The corresponding linearization leads to
#
D(δWc ) = D( Tisl δξ i ds) =
T
(11.20)
s
#

N BFTtr ⊗ n
= (δrs − δρ) · √ (vs − v) ds
BFTtr · FBFTtr
s !
leads to Km,1
T ,sl
#

|N | BFB
− (δrs − δρ) · √ (vs − v) ds
BFTtr · FBFTtr
s !
leads to Km,2
T ,sl
# - .
|N | BFTtr ⊗ (BFB)T FBFTtr
+ (δrs − δρ) · (vs − v) ds
s (BFTtr · FBFTtr )3
!
leads to Km,3
T ,sl
#
$ %
− Tisl (δrs − δρ) · ail ajk ρk ⊗ ρl vj + δρ,j · aik ajl ρk ⊗ ρl (vs − v) ds.
s
!
leads to KrT ,sl
Now we can summarize the results concerning the symmetry of the neces-
sary tangent matrices. As expected, all parts concerning non-frictional and
sticking contact, namely Km r m r
N , KN , KT,st and KT,st are symmetric. Atten-
tion should be paid to the sliding tangent matrix, because it contains both,
symmetric and non-symmetric parts:
Km,1
T,sl is the first main part due to the coupling of the normal and the sliding
tractions. It appears due to linearization of the normal traction N and it is
non-symmetric.
Km,2
T,sl is the second main part. It appears due to linearization of the normal
trial tangential traction Ttr and preserves symmetry.
Km,3
T,sl is the third main part and non-symmetric. It appears due to lin-
1
earization of the complex term √ reflecting the coupling
BFT · FBFTtr
tr
of anisotropy for adhesion and friction.
Finally, KrT,sl is the rotational, symmetric part reflecting rotation of the mas-
ter segment.
11.4.2 Augmented Lagrangian Method and

Symmetric Uzawa Algorithm
The main advantage of the Augmented Lagrangian in comparison to the
penalty method is the possibility to select a value of the penalty parameter
leading to well conditioned tangent matrices together with the enforcement
of the constraint conditions (non penetration and sticking) within a speci-
fied tolerance. The method is constructed as a nested algorithm known as
Uzawa algorithm and possesses linear convergence. The algorithm using the
full consistent tangent matrix is known as exact Uzawa algorithm , while
the inexact algorithm is exploiting a somehow simplified matrix, see results
on convergence in Stadler [167]. Laursen and Simo [109] developed a sym-
metrized Uzawa algorithm, where symmetric matrices for the sliding case
have been obtained under the assumption that the normal traction remains
constant for the solution of the equilibrium equations within an internal loop.
This preserves the quadratic rate of convergence for the internal loop. The
correct value is enforced then within the external loop where a linear rate of
convergence is preserved.
Limitations of the Augmented Lagrangian Approach for the

Coupled Anisotropic Model
The complexity for a generalization of the discussed algorithm into
anisotropic friction especially including anisotropic adhesion is that the slid-
ing tangent matrix is fully non-symmetric, moreover, the assumption of
isotropy together with the constant normal traction would lead only to the
rotational part of the tangent matrix. Since, this part has no influence for
small displacement problems as well as in the case when the master segment
has no rotation, we obtain only zero matrices. This makes it impossible to ob-
tain any solution. Fortunately, the covariant approach allows to estimate the
influence of the matrices part by part. Thus, we can construct a symmetric
algorithm and analyze it numerically.
Another problem for the coupled anisotropy is that the sticking case is
defined when the slave contact point lays inside the elliptic adhesion domain,
see Fig. 11.21. However, as is known from numerical results the ratio ab = μμ12 εε21
does not influence the convergence result in the penalty based approach and
can lead to correct kinematics even for large sliding problems. An enforcement
to put a slave contact point inside the adhesion ellipse via the Augmented
Lagrangian approach necessarily leads to dis-convergence in cases as ab → 0,
or ab → ∞. In computations due to the linear convergence of the Augmented
Lagrangian method the global number of iteration is proportional to the
ratio ab for a > b, e.g. a simply computable case with ab = 10 would lead
to a 10-times increase of the global number of iterations for the Augmented
Lagrangian method . Thus, one should judge the coupled anisotropic model
2
ξ
m ain
μN it d
o 0
b= ε2 il m ε 1=
2
μ1 N
sticking a= 000
111
ε1
O φ
11
00
00
11
00S
11
00
11
1S
0
0
1
ξ (0)
sliding
ξ1
1
0
0
1
Fig. 11.21 Allowable elastic region (adhesion domain).
as an interface model for tangential traction, rather than as a penalty based

approach.
Summarizing the discussion, we can define computable cases for the Aug-
mented Lagrangian method in application to the coupled anisotropic model
for:
1. Small sliding problems with low anisotropy. Small sliding can be numer-
ically defined as a case where it is important to compute the distribution
of the sticking-sliding zone. Low anisotropy is defined then as a case with
b ≈ 1.
a
In this case, the Lagrangian multipliers for both normal N , and tangential
traction T are augmented. Convergence for tangential displacements is
checked by the proximity to the initial contact point:

i − ξi j j
aij (ξ(I) (K−1) )(ξ(I) − ξ(K−1) ) ≤ T , (11.21)
where T is a specified tolerance for tangential displacements. Conver-

gence for normal displacements is checked as in the standard isotropic
case by the normal penetration
|ξ 3 | ≤ N , (11.22)
where N is a specified tolerance for the normal penetration.

2. Large sliding problem with arbitrary anisotropy. Large sliding can be
numerically defined as a case where sliding is reached within a single
load step. In this case, the global kinematical behavior of a contacting
body is of interest rather than the distribution of the sticking-sliding

zone.
In this case, the Lagrangian multiplier for the normal traction N only
needs to be updated because the tolerance of the sticking condition will
normally not influence the tolerance of the computed trajectory. Thus,
convergence is checked only for the normal displacement.
Table 11.3 Update scheme for normal traction N
1. Loop over applied in load incremental steps: K = K + 1

initialization of Lagrange multiplier λ0N = 0
2. Loop over augmented multipliers: L = L + 1
3. Iterative solution of
global equilibrium equations:
with normal force
N K,L,I = λL−1
N + εN ξI3
penetration ξI3 is computed in each iteration I.
Update multipliers
λL
N = N
K,L,I
Convergence is checked by the non-penetration condition:
|ξI3 | ≤ N , f or N K,L,I < 0,
where N - specified tolerance for normal displacements
The symmetrized algorithm in accordance with the inexact Uzawa ap-

proach which appears to be numerically effective and efficient for the current
anisotropic problem is constructed as follows. The internal loop, No. 3 in
Table 11.3 and 11.4, serves for the solution of the equilibrium equations.
The normal traction N K,L,I is computed via the augmented scheme and up-
dated inside loop 2. The trial tangential traction TK,L,I for small sliding
problems is computed via the augmented scheme. The update according to
the scheme ΔλL T = ΔλT
L−1
+ B(ξ (I) − ξ (K−1) ) inside loop 2 allows to en-
force sticking conditions similar to the normal penetration. The modification
for large sliding problems is as follows: the sticking condition is satisfied in
the penalty form, then the trial tangential traction TK,L,I is computed as
TK,L,I = TK−1,L,I + B(ξ (I) − ξ (K−1) ). The Lagrange multiplier for the tan-
gential traction vector ΔλL−1
T is not introduced and, therefore, an update
loop 2 does not exist.
Table 11.4 Update scheme for trial tangential traction T
1. Loop over load applied in incremental steps: K = K + 1

initial condition for tangential traction T0,0,0 = 0
initialization of Lagrange multiplier Δλ0T = 0
2. Loop over augmented multipliers: L = L + 1
with tangential traction
TK,L,I = TK−1,L,I + ΔλL−1

T + B(ξ(I) − ξ(K−1) )
ξ(I) is a projection point in each iteration I,
ξ(K−1) is a projection point in the previous load step.
Update multipliers
ΔλL L−1
T = ΔλT + B(ξ(I) − ξ(K−1) )
Convergence is checked by proximity

to the initial sticking point:

i
aij (ξ(I) − ξ(K−1)
i j
)(ξ(I) − ξ(K−1)
j
) ≤ T ,
where T - specified tolerance for tangential displacements
An important modification for symmetrization should be done for the

return-mapping scheme, see Table 11.5, where the sliding force instead of
eqn. (6.109) is computed via the augmented multiplier for the normal trac-
tion λL−1
N as
BFTtr
Tsl = − √ |λL−1
N |. (11.23)
BFTtr · FBFTtr
The yield function (6.46) is respectively modified as

Φλ = TK,L,Itr · FTK,L,I
tr − |λL−1
N |. (11.24)
This leads to a constant sliding force for the internal loop and, therefore, the
first main part of the sliding tangent matrix is zero Km,1T,sl = 0. Then, the full
tangent matrix for the sliding case becomes:
Kfslull = Km r m,2 m,3 r

N + KN + KT,sl + KT,sl + KT,sl . (11.25)
The part Km,3

T,sl is still nonsymmetric due to anisotropy. The matrix is fully
symmetric only for isotropic friction. As it was found in numerical computa-
tions we can exclude this part with only a small loss of efficiency. This finally
leads to the following matrix in the sliding case according to the inexact
Uzawa algorithm:
Kfslull = Km r m,2 r
N + KN + KT,sl + KT,sl . (11.26)
The tangent matrix for sticking remains symmetric:
Kfstull = Km r m r
N + KN + KT,st + KT,st . (11.27)
Table 11.5 Return-mapping scheme for the symmetric Augmented Lagrangian

method
1. Loop over load applied in incremental steps: K = K + 1

initial condition for tangential traction T0,0,0 = 0
initialization of Lagrange multiplier Δλ0T = 0
initialization of Lagrange multiplier Δλ0N = 0
2. Loop over load augmented multipliers: L = L + 1
a) compute trial tangential traction TK,L,I
tr
(Table 11.4)
b) compute trial yield function Φλ

Φλ = TK,L,I
tr · FTK,L,I
tr − |λL−1
N |.
c) return-mapping: real tangential traction TK,L,I
⎧ K,L,I
⎪
⎪ Ttr if Φλ < 0
⎨
K,L,I
T =
⎪
⎪ T̂
⎩ Tsl = − |λL−1
N | if Φλ ≥ 0
T̂ · FT̂
(11.28)
where T̂ = BFTK,L,I
tr
11.5 Numerical Examples for Augmented Lagrangian

Method
Two cases with different anisotropy have been selected to illustrate the
convergence for the proposed approach: constant orthotropy and polar or-
thotropy on the plane.
11.5 Numerical Examples for Augmented Lagrangian Method 363
11.5.1 Small Sliding Problem. Constant Orthotropy

We present here an example, which has been analyzed for the penalty based
approach in Sect. 11.3. The rectangular block, see Fig. 11.3, is considered on
an orthotropic plane. The dimensions of the block are 10 × 10 × 4 with linear
elastic properties: Young’s modulus E = 2.10 · 104 and Poisson ratio ν = 0.3,
assumed within a consistent dimension system.
The case of constant orthotropy is defined by the spectral representation
of both, the adhesion tensor B, and the friction tensor F as follows:

B = [bij ] = − 1 . (11.29)
⎡ 1 1 1 1 ⎤
cos2
β + sin2
β ( − ) sin β cos β
⎢ μ21 μ22 μ21 μ22 ⎥
⎢ ⎥
F = [fk ] = ⎢
i
⎥. (11.30)
⎣ 1 1 1 1 ⎦
2 2
μ1 μ2 μ1 μ2
The following parameters are taken for the computations: normal penalty
parameter: εN = 2.1 · 104 ; adhesion parameters: ε1 = 3.0 · 104 , ε1 = 2.0 · 104 ;
friction coefficients: μ1 = 3.0, μ2 = 2.0; tolerance for penetration: N =
1.0 · 10−5 ; tolerance for tangential displacement: T = 1.0 · 10−5 ; orthotropic
angles: α = 45o , β = 45o . This leads to a, so-called, geometrical isotropic case
because of the ratio εε12 μμ21 = 1.0.
The block (see Fig. 11.3) is located on the XOY plane and loaded by
vertically prescribed displacements on the upper surface w = 1.0 · 10−2 . The
penalty based approach gives convergence in 4 iterations, but both normal
and tangential displacements inside the sticking region do not satisfy the
prescribed tolerances. The results for the Augmented scheme are presented
in Table 11.6. The method shows linear convergence as can be seen from the
Tolerance column.
11.5.2 Large Sliding Problem. Polar Orthotropy

For the case with large sliding , we consider the block on a plane with polar
orthotropy see [90]. An elastic block with dimensions 1 × 1 × 0.25 and mesh
4×4×1 is positioned on a rigid block, see Fig. 11.17. Linear elastic material is
assumed within a consistent dimension system: Young’s modulus E = 2.10 ·
104 ; Poisson ratio ν = 0.3. The loading is applied sequentially by prescribing
displacements on the upper surface in (1 + n) steps: 1) vertical loading with
w = 1.0 · 10−2 , 2) n steps with horizontal displacement increments Δu =
1.0 · 10−2 along the X axis. The frictional tensor is isotropic with μ1 = μ2 =
0.2, but the adhesion tensor has the following structure in the Cartesian
coordinate system:

1 εr x2 + εϕ y 2 (εr − εϕ )xy
B=− 2 . (11.31)
x + y2 (εr − εϕ )xy εr y 2 + εϕ x2
Parameters for the adhesion tensor are chosen as εr = 1000, εϕ = 0.0, in or-
der to obtain the circular trajectory of the sliding block. Now only the normal
tractions N are updated and a tolerance for penetration N = 1.0·10−5 is cho-
sen. The gap for the penalty based approach is computed as ξ 3 = 7.26 · 10−3
if the normal penalty εN = 2.1 · 104 is chosen. The total number of iterations
(4556) compared with the penalty approach (1834) is influenced only by the
normal penetration. The final coordinates of the nodal point X = 0.292,
Y = 4.996 are compared with the results found with the penalty approach
X = 0.306, Y = 4.994 in order to show that the main kinematical effect of
the anisotropic surfaces is preserved for the Augmented Lagrangian method
. A comparison of both trajectories leads to a slight difference comparable to
the value of the initial penetration and the initial sticking displacements.
Table 11.6 Constant orthotropy on a plane. Convergence results for the symmetric
Augmented Lagrangian method.
Aug. No. Number of eq. iter. Tolerance

1 7 1.8742 · 10−3
2 14 6.0419 · 10−4
3 12 1.7536 · 10−4
4 12 4.0497 · 10−5
5 512 3.5972 · 10−6
56
11.5.3 Discussion
In this section a symmetrization of the stiffness matrix has been developed
within the Augmented Lagrangian method for anisotropic contact surfaces
including both, anisotropy for adhesion and anisotropy for friction domains.
It is shown that in general a fully coupled model necessarily leads to a fully
nonsymmetric matrix in the case of sliding, but the covariant approach allows
to estimate the structure of the tangent matrix part by part and, therefore,
allows to construct a symmetric matrix used in accordance with the inexact
Uzawa algorithm . However, as is shown, the Augmented Lagrangian method
can not be directly applied to arbitrary anisotropic surfaces due to conver-
gence problems. Thus, to create a robust algorithm, contact problems have
to be subdivided into small sliding problems with low anisotropy where the
distribution of the sticking-sliding zone is of interest, and into large sliding
problems where the trajectory of the sliding body is of interest.
11.5 Numerical Examples for Augmented Lagrangian Method 365
For the small sliding problems, both the normal, and the tangential con-
tact tractions are augmented within the nested Uzawa algorithm. This makes
it possible to enforce both normal and tangential sticking displacements to
satisfy a prescribed tolerance. For the large sliding problems only the normal
traction is augmented leading to the enforcement of only the normal dis-
placements to satisfy a prescribe tolerance. In both cases the return-mapping
scheme is exploited to obtain the real sliding tractions. Numerical examples
including constant as well as nonlinear orthotropy e.g. a polar orthotropy
showed the effectiveness of the proposed approach.
12
Experimental Validations of the
Coupled Anistropic Adhesion-Friction
Model
The coupled anisotropic adhesion-friction model studied numerically in chap-

ter 11 is verified in this chapter experimentally for some structured surfaces
(originally published in [98]). The necessity to apply a coupled contact inter-
face model including anisotropy for both adhesion and friction is shown in
the current chapter via a set of experiments for a rubber surface possessing
a periodical waviness, and therefore, an obvious anisotropic structure. The
focus of experimental investigations is placed upon the measurements of the
global macro characteristics such as global forces and trajectories of a sliding
block in order to validate the proposed computational model. The theoretical
results presented in Sect. 6.4 is taken here for validation.
12.1 Introduction
Smoothness and isotropy of contacting body surfaces can vary considerably
for different contact problems. Classifying the surfaces roughness two types
can be distinguished: a) surfaces with randomly distributed asperities, and
b) asperities with algorithmic structure, e.g. the considered surface shows
different macro properties in different directions.
Mechanical characteristics for the associated contact problems of the first
type a) are obtained via statistically distributed asperities. Statistical analy-
sis of a real rough surface and experimental aspects of its measurements have
been developed in a series of publications: Longuet-Higgins [116], Green-
wood and Williamson [49], Whitehouse and Archard [182] and more recently
Whitehouse and Phillips [184] and Greenwood [47]. A comparative analysis
of these surface models is presented in McCool [123]. These experimentally
proved models later have been incorporated into finite element models, see
e.g. Wriggers and Zavarise [197], [196], Buczkowski and Kleiber [24]. More
advanced numerical analyses including homogenization methods and multi-
scaled modeling are presented in Bandeira et. al. [9], [8]. Carbone and Man-
gialardi [26] derived contact tractions analytically for a particular example
368 12 Experimental Validations of the Coupled Anistropic Adhesion-Friction
with a rigid wavy surface with a sinusoidal profile, assuming the presence of
an adhesion hysteresis for 2D plane strain elasticity problem.
Constitutive modeling is applied for problems of the second type b). Such
models are based on the generalization of Coulomb’s friction law into the
anisotropic domain. One of the first models has been proposed by Michalowski
and Mroz [128] considered the sliding of a rigid block on an inclined surface.
A model of orthotropic friction has been analyzed and consistently developed
in Zmitrowicz [213], Curnier [31]. Various cases of anisotropy were presented
in He and Curnier [61] based on the theory of tensor function representations
and in Zmitrowicz [214] based on consideration of a relative sliding velocity.
In the latter contribution, a classification of anisotropic surfaces based on the
number of eigenvalues of the friction tensor has been proposed.
When looking at practical problems concerning contact interactions with
friction between bodies made of soft rubber-like materials there are some sit-
uations in which the tangential elasticity of the contact surfaces should be
taken into account. In these cases anisotropy for elastic forces (adhe-
sion) and frictional forces might be coupled. Such a model including
coupling of anisotropy for both friction and adhesion has been developed and
analyzed numerically in Konyukhov and Schweizerhof [89], [90]. In the current
contribution we discuss the validation of this model with a particular exper-
imental test. The contact surfaces are chosen to possess visual orthotropic
properties, thus a corrugated rubber mat is taken. The results of the ex-
periments show the necessity to use the coupled model including
anisotropy for both friction and adhesion. Some originally surprising
experimental phenomena, such as geometrically isotropic observed behavior
of a sliding block despite obvious physical anisotropies can be explained only
within the proposed model.
12.2 Experimental Investigation

A series of experimental tests are performed in order to investigate the global
characteristics of the system ”block on a rough surface ”. The rough surface
possesses visually a clear periodical structure and, therefore, the mechanical
constitutive model for an observable orthotropic structure can be applied.
Since we are trying to verify the average interface model, the measured val-
ues in experiments are intentionally chosen to be global, namely we mea-
sure global forces leading to the macro friction coefficients and trajectory of
a block instead of micro friction coefficients and corresponding stiffnesses
of asperities. The focus of the discussion is placed upon the kinematical
behavior of the block driven by a constant force together with the mea-
surement of force components leading to this motion. Therefore, the main
measurable characteristics during these experiments are global forces and
trajectories of the block, which create a main basis for further calibration of
models for orthotropic friction. For the judgment of the results Coulomb like
12.2 Experimental Investigation 369
models are assumed a-priori to be valid for the global behavior, i.e. that the
tangential driving global force F is proportional to the normal reaction N :
F = f (x, y)N , where a function f (x, y) describes the orthotropic properties
of a surface.
12.2.1 Experimental Setup

A massive block positioned on a plane is moved with constant velocity by a
sliding carriage guided by rods on both sides, see Fig. 12.1. The block made
from steel has dimensions 110 × 110 mm in plane and 20 mm height. The
mass is m = 1.875kg. The contact surface between the sliding carriage and
the block is covered with a T ef lon(R) strip to minimize the friction between
them due to relative sliding. In contrast to this, the contact surface of the
steel block is covered by a suede-like material with square 90 × 90 mm to
increase the interaction between the block and the basement. A constant
driving velocity is achieved by a step motor acting on a rack, which allows a
straight displacement of 500 mm. The contact force between the rack and the
sliding carriage is measured by a force sensor. The displacement of the block
during sliding is captured by an optoelectronic device which is installed on
a tripod above the surface. The corresponding LED (light-emitting diode) is
fixed on the block.
For the first set of experiments a rubber mat with rather stiff ripples has
been taken. Naturally it was represented by an aged corrugated rubber map.
The frictional orthotropy is given by the wavy profile of the contact rubber
Fig. 12.1 View on experimental setup.

surface with parallel ripples possessing in the cross section a periodical struc-
ture, see the CAD model in Fig. 12.2. The distance between the ripples is
rather small in comparison to the dimension of the contact area of the block
allowing approximately 30 ripples in contact area depending on orthotropy
angle in experiments. The orientation of the ripples with respect to the fixed
driving direction can be varied from 0o up to 90o by repositioning the mat,
see Fig. 12.3. Again we should mention that the aim of the experiments is to
show the necessity of a coupled anisotropic model for adhesion and friction,
therefore, we intentionally skip any measurement of ripple stiffnesses concen-
trating on finding the global macro characteristics of the coupled behavior.
Fig. 12.2 Geometrical structure of the corrugated rubber mat.

Wavy periodical profile, CAD model.
00000
11111
11111
00000
00000
11111
11111
00000 ϕ 000000
111111
111111
000000
v 00000
11111
00000
11111
00000
11111
v 00000
11111
00000
11111
00000
11111
v 000000
111111
000000
111111
000000
111111
00000
11111 00000
11111 000000
111111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
00000
11111
α 000000
111111
000000
111111
000000
111111
00000
11111 00000
11111 000000
111111
(a) (b) (c)
Fig. 12.3 Orientation of the orthotropy with respect to the fixed direction of the
velocity: a) α = 0o , b) α = 0o < α < 90o , c) α = 90o . The trajectory of the block
is a straight line declined at angle ϕ.
12.2.2 Experimental Results

At the beginning some experiments are performed to find out the global
macro mechanical properties of the system. All experiments were reproduced
with a driving velocity of the carriage v = 24.4 cm/sec. First of all, the
sliding carriage was moved without the block in order to define the internal
resistant force Fint . Then experiments have been performed with the sliding
12.2 Experimental Investigation 371
block to define the resulting driving force in the case α = 0o , corresponding

to the X1 -axis along the ripples, see CAD model in Fig. 12.2, and α = 90o ,
corresponding to the X2 -axis across the ripples, see CAD model in Fig. 12.2,
respectively. These measurements together with subtracting the internal force
lead to the definition of macro friction coefficients μ1 and μ2 corresponding
to angles α = 0o and α = 90o . Assuming a Coulomb friction law the friction
coefficients μi were computed as
Fα=0o − Fint 12.50 − 5.00
μ1 = = = 0.408 ≈ 0.41
N 1.875 · 9.806
Fα=90o − Fint 16.50 − 5.00
μ2 = = = 0.625 ≈ 0.63. (12.1)
N 1.875 · 9.806
We note for further references that the macro friction coefficient across the
ripples is found to be higher than the macro friction coefficient along the
ripples.
The second set of experiments to find macro parameters of the interface
model is made by setting the angle α varying from α = 0o up to α = 90o
with a step Δα = 5o . The focus lies on the definition of the trajectory of the
sliding block. In all cases, the trajectory was observed as a straight
line inclined with the angle ϕ = ϕ(α), see the sketch in Fig. 12.3(b).
The mean value ϕ of sequences of 10 experiments for each angle α is taken
for the later analysis.
The standard deviation has been found as s ≈ 0.3o
5N
i=1 (ϕi − ϕ) ). Combining all results leads to the
1 2
(defined by s = N −1
diagram in Fig. 12.4 showing the dependency of the inclination angle ϕ on
the orientation of the orthotropy given by the angle α of the ripples. The
maximum of the inclination angle ϕ is located in the range of small angles α
(the maximum is on the left side on the graph).
Geometrically Isotropic Observed Behavior of a Sliding Block

As a fairly surprising result detected in the experiments a large sensitivity
to the elastic properties of the rubber ripples was obtained. Thus, if the
rubber mat with highly elastic rubber ripples (e.g. a new mat) has been
taken for the second set of experiments, then the inclination angle ϕ was
only varying in a very small range about 0o − 2o with a standard deviation
s ≈ 0.5o independently from the angle α. Since an interval 3s of the mean
value defined as ϕ−3s ≤ ϕ ≤ ϕ+3s covers the zero value ϕ = 0, and assuming
a Gaussian error distribution, we can conclude that the inclination angle ϕ
does not depend on the orthotropy angle α. However, the measurement of
forces still showed the difference between the macro coefficients of friction μ1
and μ2 . We observe that macro friction orthotropy is still present, but the
kinematical effect of the orthotropy disappears.
As we show later, the orthotropic friction model is not capable to describe
this effect from the macro-model point of view, but the coupled orthotropic
14
Experiment, vel. = 24.4 cm/sec
12
10
Observed angle
8
0
0 10 20 30 40 50 60 70 80 90
Orthotropy angle
Fig. 12.4 Observed mean value of the inclination angle ϕ vs. orthotropy angle α.
Experimental results for different velocities of the block.
adhesion – orthotropic friction interface model allows to qualitatively describe

the observed phenomena.
12.3 Calibration of Parameters for Different Models

As a representative parameter we take the curve ϕ(α) known from the ex-
periment, see Fig. 12.4. In addition, we distinguish two orthotropy angles: α
– orthotropy angle for the adhesion tensor B and β – orthotropy angle for
the friction tensor F. In order to unify the computations we chose the or-
thotropic values as μ1 < μ2 and ε1 < ε2 . The following test computations are
performed for calibration purposes, based on the theoretical results developed
in Sect. 6.4:
1. The orthotropic friction model as discussed in Section 6.4.1.
2. The interface model including coupled orthotropy for both adhesion and
friction as discussed in Section 6.4.2 with the specific case of isotropic
adhesion B = −εE.
3. The interface model including coupled orthotropy for both adhesion and
1
friction with the specific case of isotropic friction F = 2 E.
μ
4. The interface model including orthotropy for both adhesion and friction
with the specific case of coinciding orthotropy angle α = β.
5. The interface model including orthotropy for both adhesion and friction
with the specification of the friction orthotropy angle β by 90o degrees
as β = α + π/2.
12.3 Calibration of Parameters for Different Models 373
12.3.1 Case 1
We start the validation from the simple model including only orthotropic fric-
tion as discussed in Sect. 6.4.1. Our aim is to find out a case describing qual-
itatively the experimental results. Therefore, we perform a test computation
with the following friction coefficients μ1 = 0.1, μ1 = 0.5. In Fig. 12.5 the re-
sults are depicted. In addition, at an extremal angle α = arctan μμ21 = (78.69o)
μ2 − μ21
the maximum value ϕmax = arctan 2 = (67.38o ) is computed by ana-
2μ1 μ2
lyzing the shape of a curve. It can be seen, that the point is moving into the
direction with a smaller friction coefficient, which contradicts the experimen-
tal curve, see Fig. 12.4.
70 case 1
60
Observed angle
50
40
30
20
10
0
0 10 20 30 40 50 60 70 80 90
Orthotropy angle
Fig. 12.5 Computed inclination angle vs. orthotropy angle.

Case 1: Purely orthotropic friction model.
12.3.2 Case 2
As a next step, we choose the orthotropic adhesion – orthotropic friction
interface model, but including orthotropy only for the friction tensor and
keeping the adhesion tensor to be isotropic, B = −εT E. In this case, a struc-
ture of the matrix in eqns. (6.149) and (6.152) is given as A = −ε2T F leading
to the inclination angle in eqn. (6.153) to be defined as
(μ21 − μ22 ) tan β

tan ϕ = . (12.2)
μ21 + μ22 tan2 β
The analysis with the values μ1 = 0.1 and μ2 = 0.5 gives the curve presented
in Fig. 12.6 with the extremal parameters βext = 11.31o and ϕmin = −67.38o.
In this case, the point tends to move into the direction with the larger friction
coefficient. This is also a contradiction to the experimental results.
0
case 2
-10
-20
Observed angle
-30
-40
-50
-60
-70
0 10 20 30 40 50 60 70 80 90
Orthotropy angle
Fig. 12.6 Orthotropic adhesion – orthotropic friction interface model. Computed

inclination angle vs. orthotropy angle. Case 2: isotropic adhesion – orthotropic
friction.
12.3.3 Case 3
As a next step, within the orthotropic adhesion – orthotropic friction interface
model orthotropy is assumed only for adhesion with parameters ε1 = 1 · 104
and ε2 = 5 · 104 . The following structure of the matrix A is obtained
1
A = BFB = B, (12.3)
μ2
leading to an inclination angle ϕ defined as
(ε22 − ε21 ) tan α

tan ϕ = . (12.4)
ε22 + ε21 tan2 α
The result is depicted in Fig. 12.7 with the extremal values as αext = 78.69o
and ϕmax = 67.38o. The curve shows a motion into the direction with the
smaller stiffness ε1 . This contradicts the experimental results as well.
Summary
We observed that neither the purely orthotropic friction model, nor the cou-
pled model with separately included orthotropy either for friction, or for ad-
hesion is unable to capture the phenomena even qualitatively.
12.3.4 Case 4
Now we keep the orthotropy for both the adhesion tensor and the friction
tensor with the same angle, namely α = β. The structure of the tensor A
is found from the spectral decomposition in eqns. (6.143) and (6.158) with
Qα ≡ Qβ
40 case 3
35
30
Observed angle
25
20
15
10
5
0
0 10 20 30 40 50 60 70 80 90
Orthotropy angle

inclination angle vs. orthotropy angle. Case 3: orthotropic adhesion – isotropic
friction.
A = BFB = Qα ΛB QTα Qβ ΛF QTβ Qα ΛB QTα = Qα Λ2B ΛF QTα , (12.5)

! !
E E
leading to the eigenvalues λi = εi /μi . This case gives a more comprehensive

information for the analysis of the physical experiments. The most important
issue is that by the combination of two orthotropic tensors an isotropic case
can be recovered. This case appears if we take the value in proportion με11 = με22
leading to λ1 = λ2 , e.g. with the combination of two previous cases with
ε1 = 104 and ε2 = 5 · 104 and μ1 = 0.1, μ2 = 0.5 isotropy of the motion is
recovered. For λ1 < λ2 we obtain a behavior similar to case 3; for λ1 > λ2 we
find a behavior similar to case 2. E.g. the computation with the parameters
ε1 = 0.5 · 104, ε2 = 5 · 104 and μ1 = 0.1, μ2 = 0.5 shows similarity to case 3 as
shown in Fig. 12.8. In this computational case, we assumed that the ripples
are softer in direction X1 , see Fig. 12.2, which is obviously not the case even
without measurements of the ripple stiffnesses.
12.3.5 Case 5
Finally, we can choose the last possible modification for the observed geo-
metrical orthotropy for surfaces as taken in the experiment. We define a new
angle β̂ as a main angle of surface asperities in the experiment, see Fig. 12.9.
The orthotropy angle β for the friction tensor is shifted by 90o degrees to
β = α + π/2 with respect to the orthotropy angle α for the adhesion tensor.
The structure of the tensor A is given according to eqn. (12.5), but now the
composition of the two orthogonal matrices QTα Qβ leads to:
40 case 4
35
30
Observed angle
25
20
15
10
5
0
0 10 20 30 40 50 60 70 80 90
Orthotropy angle

inclination angle vs. orthotropy angle. Case 4: orthotropic adhesion – orthotropic
friction with eigenvalues λ1 < λ2 .
QTα Qβ = (12.6)
T
cos α − sin α cos β − sin β
= · =
sin α cos α sin β cos β
T
cos α − sin α − sin α − cos α
= · =
sin α cos α cos α − sin α

0 −1
= .
1 0
Then, the matrix A in eqn. (12.5) is derived as
A = BFB = Qα ΛB QTα Qβ ΛF QTβ Qα ΛB QTα (12.7)

! !
⎡ 2 ⎤
ε1
⎢ μ2 0 ⎥ T
= Qα ⎢⎣
2 ⎥Q ,
ε22 ⎦ α
0 2
μ1
leading to the following eigenvalues λ1 = ε1 /μ2 and λ2 = ε2 /μ1 in

eqn. (6.152). The computation with ε1 = 104 and ε2 = 103 and μ1 = 0.1, μ2 =
0.5 gives the curve ϕ vs. β̂ depicted in Fig. 12.10, which quantitively has a
shape similar to the experimental one (maximum from the left side). The
extremal values are found as β̂ext = 26.56o and ϕmax = 36.87o.
Thus, summarizing the numerical investigations and focusing on the com-
parison to the experiments it becomes obvious that for the surface as given in
the experiment it is necessary to apply the orthotropic adhesion – orthotropic
friction interface model.
2
X
2 1
ξ μ
β
1 2
ξ μ
α X1
O
^
β
Fig. 12.9 Definition of the experimentally observed angle β̂, by an orthotropy angle
α for the adhesion tensor and an orthotropy angle β for the friction tensor.
40 case 5
35
30
Observed angle
25
20
15
10
5
0
0 10 20 30 40 50 60 70 80 90
Orthotropy angle

inclination angle ϕ vs. redefined orthotropy angle β̂. Case 5: orthotropic adhesion
– orthotropic friction with angles β = α + π2 .
In this case, the force T2 measured in direction X2 is larger, while the rip-
ples are softer in the same direction, which is visually observed in experiments
even without measurement of ripple stiffnesses, see Fig. 12.2. Moreover, as-
suming εε12 = μμ12 we obtain ϕ(α) = 0, i.e. the block is no longer inclined, and
we recover the geometrically isotropic behavior already observed in experi-
ments, see Sect. 12.2.2. In this case, orthotropy for friction is compensated
by orthotropy for adhesion leading to the observed isotropic behavior of the
block, though the reason for this remains uncovered.
12.3.6 Calibration of the Theoretical Curve by

Extremal Values
As found from the proposed model, the inclination angle ϕ depends only on
the ratio of eigenvalues λ1 /λ2 , see eqn. (6.153). This ratio contains informa-
tion also about the ratio of adhesion parameters ε1 /ε2 :
λ1 ε 1 μ1
ratioext = = · . (12.8)
λ2 ε 2 μ2
Since, the friction coefficients μ1 , μ2 are defined via the measurement of
forces in the experiment, the inclination angle ϕ depends only on the ratio
of eigenvalues ε1 /ε2 , see also the Remark in Section 6.4.2. This fact gives
the possibility to judge macro properties for ripple stiffnesses without mea-
surements. Thus, we will use eqn. (6.153) for a calibration of the model.
Calibration is provided according to following rules: 1) a maximum rule –
both theoretical and experimental curves must achieve the same maximum;
2) a least square fit method. The friction coefficients have been determined
previously to μ1 = 0.408, μ2 = 0.625.
According to the maximum rule the value for the angle ϕext defined in
eqn. (6.157) is used for calibration purposes. Taking e.g. the maximum angle
ϕmax = 11.5o measured for the velocity 12.5 cm/sec the ratio of the eigenval-
ues given in eqn. (6.157) becomes
λ1
ratioext = = −0.20345 ± 1.02048 = 0.817, (12.9)
λ2
where only the positive solution is taken. The ratio of the stiffness coefficient
is then obtained as
ε1 μ2 0.625
= · ratioext = · 0.817 = 1.251 (12.10)
ε2 μ1 0.408
A more mathematically precise least square fit method leads to the statement
derived from eqn. (6.153). The following sum must be minimized:
N
2
(λ21 − λ22 ) tan β (k)
tan ϕ(k) + =
k=1
λ21 tan2 β (k) + λ22
N 2
(rλ2 − 1) tan β (k)
= (k)
tan ϕ + 2 −→ min, (12.11)
k=1
rλ tan2 β (k) + 1
where ϕ(k) are measured declination angles vs. applied orthotropy angles α(k)
and β (k) = π/2 − α(k) . Minimization with regard to the variable rλ2 leads to
the following expression:
12.4 Discussion 379
5N 2 3
k=1 tan β (k) (1 + tan2 β (k) )(tan β (k) − tan ϕ(k) )
rλ2 = 5N 2 3. (12.12)
k=1 tan β (k) (1 + tan2 β (k) )(tan β (k) + tan ϕ(k) )
The cases including the computed parameters are shown in Fig. 12.11 and
do not exhibit a quantitively good correlation, though we reached the aim to
show the necessity of coupling both orthotropy for friction and orthotropy
for adhesion in addition to the general applicability of the proposed model.
Among the possible reasons for the bad correlations might be the following:
a) a too simple linear elastic model for adhesion region;
b) the simple Coulomb friction model for the friction region without e.g.
hysteresis etc.
Nevertheless, it seems to be important to consider fairly complex models
including the coupled orthotropy for the adhesion and the friction in order
to describe the observed phenomena more precisely.
14 Experiment
Maximum, theory
Least Square Fit, theory
12
10
Observed angle
0
0 10 20 30 40 50 60 70 80 90
Orthotropy angle
Fig. 12.11 Observed inclination angle ϕ vs. redefined orthotropy angle β̂. Calibra-
tion: (1) by the absolute maximum value for different velocities of the block; (2) by
the least square fit method.
12.4 Discussion
The necessity to apply a coupled model including anisotropy for both adhe-
sion and friction is shown for particular soft anisotropic surfaces, such as a
periodic wavy rubber profile. The macro characteristics of the proposed model
such as macro friction coefficients μ1 , μ2 can be defined via the measurement
of global forces, while the trajectory, namely, the observed sliding angle ϕ(α)
gives information about the ratio of orthotropic stiffnesses ε1 /ε2 . The model
does not show the dependence of particular values of stiffnesses ε1 and ε2
on the trajectory, which was also confirmed by experiments. The model al-
lows to skip measurements of surface microstiffnesses. Moreover, particular
phenomena when a sliding block shows geometrically isotropic behavior can

be described correctly with the proposed model.
The considered model contains only the coupling of linear models for ad-
hesion and friction, which is a possible explanation for the rather poor quan-
titative correlation with experiments. The key to achieve better correlation
– from our point of view – can be the coupling of a more general law for
adhesion as well as for friction. Thus, a more complex elastic law (e.g. Ogden
material law for 2D case) can be taken for the adhesion region together with a
more complex friction law for the friction region (e.g. see the proposals of He
and Curnier [61] and recently Zmitrowicz [217]). The calibration process, in
due course, can be provided by experimental investigations as well as by nu-
merical tests involving homogenization processes and multi-scale techniques,
for which the methodology for the isotropic case is shown in [9].
13
Various Aspects of Implementation of
the Curve-To-Curve Contact Model
There are several questions in further applications of the curve-to-curve ap-

proach to contact between bodies:
1. combination of classical 3D finite element discretization with the edge-
to-edge approach;
2. combination of finite beam elements with the beam-to-beam algorithm;
3. development of special “solid-beam” elements with curve-to-curve algo-
rithm which allow to take into account 3D effects also for beam type
elements.
Combination of 3D finite elements with an edge-to-edge contact approach is
rather straightforward. The continuum is discretized with solid or solid-shell
elements and the chosen edge-to-edge contact algorithm is applied. In the
current chapter we consider various types of finite element approximation
for the structure including solid-shell and beam type elements as well as
including iso-geometric approach together with the Curve-To-Curve type of
contact. The consistent application with an isogeometric approach for the 3D-
continuum would require the iso-geometrical technique, see Hughes et.al. [70],
or high order finite element techniques for shells, see Rank et.al. [153].
13.1 General Structure of Tangent Matrices for CTC

Contact
Several situations should be distinguished for the curve-to-curve contact el-
ement. The first situation is based on the definition of the penetration p for
the specific case either for edges, or for beams. Thus, for the case no contact
with positive penetration p > 0 the contact element is excluded from the com-
putation in the current iteration. The situations for tangential as well as for
rotational sticking and sliding should be distinguished for each line separately
leading to the algorithmization of as many as four different situations based
on the corresponding return-mapping algorithm. For the numerical examples
382 13 Various Aspects of Implementation
with linear solid-shell elements and edge-to-edge contact elements, the ma-
trices are simplified due to the linear shape functions and small penetration,
i.e. all curvatures kI = 0 and torsions κI = 0 as well as penetration r = 0
are zero. For the numerical example with curvilinear beam-to-beam contact
elements we are concentrating on the non-frictional case and show the full
tangent matrix for the normal part only, all parameters such as curvature,
torsions are computed as well and the coordinate r is also taken into account
because it contains implicitly the size of the cross sections of the beams.
13.1.1 Tangent Matrix for the Normal Force N

The tangent matrix for normal forces is constructed from one part with the
normal force N representing the geometrical nonlinearity as presented in
Sect. 7.5.1 and from another part containing εr from Section 7.5.2. The tan-
gent matrix represents the linearization of the normal force which is always
present for contact for both, sticking and sliding cases. Thus, eqn. (7.65c)
together with the part in eqn. (7.73) containing a normal stiffness εr should
be discretized, namely
(δρ2 − δρ1 ) · (εr e1 ⊗ e1 ) (v2 − v1t ) (13.1a)
+N Sym{ (δρ2 − δρ1 ) · [k1 cos ϕ1 (1 − rk1 cos ϕ1 )ṡ1 δs1
−r(κ1 ṡ1 + ϕ̇1 )(κ1 δs1 + δϕ1 )] (v2 − v1t )} (13.1b)

Rates and variations of coordinates are transformed using eqns. (4.194) for
arbitrary geometry. Following the transformation pattern given in eqn. (7.67)
and using the approximation operator A in eqn. (13.20) we can obtain the
following matrix:
KN = εr AT [e1 ⊗ e1 ] A (13.2)

k1 cos ϕ1 1
+ N AT [Sym τ 1 ⊗ τ 1 − g1 ⊗ g1 ]A. (13.3)
(1 − rk1 cos ϕ1 ) r
This matrix consists of a main part eqn. (13.2) together with a curvature and
a rotational part given in eqn. (13.3). In the case of a linear approximation
and edge-to-edge contact approach the matrix is simplified only to the main
part as
KN = εr AT [e1 ⊗ e1 ] A. (13.4)
The matrix represents both segments.
13.1 General Structure of Tangent Matrices for CTC Contact 383
13.1.2 Tangent Matrices for Tangential Sticking

This part is computed and then added to the global matrix if the tangential
sticking case is determined within the corresponding return-mapping scheme
for the certain I-th segment:
KT,
I
st
= εI AT [τ I ⊗ τ I ] A. (13.5)
The matrix is taken only for the I-th tangentially sticking segment.
13.1.3 Tangent Matrices for Tangential Sliding

This part, see eqn. (7.79b), is computed and then added to the global matrix if
the tangential sliding is determined within the corresponding return-mapping
scheme for the I-th segment:
KT,
I
sl
= −εr μ1 AT [τ 1 ⊗ e1 ] A. (13.6)
The matrix is then taken only for the I-th tangentially sliding segment.
13.1.4 Tangent Matrices for Rotational Sticking

This matrix includes a part representing the geometrical nonlinearity from
eqn. (7.66d) for the linear segment ( kI ≡ 0, κ1 ≡ 0 ) and a constitutive part
for the rotational sticking from eqn. (7.73). This part is computed and then
added to the global matrix if the rotational sticking is determined within the
corresponding return-mapping scheme for the I-th segment:

M, st M1
KI = A ε1 g1 ⊗ g1 − 2 (e1 ⊗ g1 + g1 ⊗ e1 ) A
T
(13.7)
r
The matrix is then only for the I-th rotationally sticking segment.
Here, ε1 represents a rotational stiffness, see Remark in Sect. 7.5.2.
13.1.5 Tangent Matrices for Rotational Sliding

These matrices are zero, because the sliding contribution is computed ac-
cording to the Tresca model independent of other parameters and remains
constant, see proof in Section 7.5.4.
Remark 13.1. The tangent matrices are representing only the contact kine-
matics according to the developed theory and, therefore, must be added to
the global tangent matrices. Thus, finite element models can be enriched with
a contact possibility. The full finite element model must contain then finite
elements describing the behavior of a continuum according to the selected
model, e.g. shell, solid-shell and beam elements.
13.2 Residual Vector

The residual vector represents discretization of the weak form. For the consid-
ered linear finite element the weak form in eqn. (5.32) should be discretized
with special attention to its symmetrical structure. We consider this proce-
dure separately for each part for normal, tangential and moment interactions.
13.2.1 Part for Normal Interaction

The normal part δWN according to the symmetry 1 ↔ 2 is fully written as
δWN = Sym {N (δρ2 − δρ1 ) · e1 } =

1
= {N (δρ2 − δρ1 ) · e1 + N (δρ1 − δρ2 ) · e2 }. (13.8)
2
We note that the normal force N is mutual for both curves and vectors defin-
ing the shortest distance are satisfying e2 = −e1 , see definition in eqn. (3.46).
Introducing the approximation operator A we obtain:
1
δWN = {N (Aδx) · e1 − N (Aδx) · (−e1 )}
2
= N (Aδx) · e1
= {δx}T N [A]T {e1 }. (13.9)
The last line in eqn. (13.9) is written using the square and figure brackets
to emphasize the vector-matrix notation after the discretization. Thus, this
part of the residual for the normal interaction becomes then
{RN } = N [A]T {e1 } (13.10)
13.2.2 Part for Tangential Interaction

The tangential part is written as
δWT = Sym{T1(δρ2 − δρ1 ) · τ 1 }

1
= {T1 (δρ2 − δρ1 ) · τ 1 + T2 (δρ1 − δρ2 ) · τ 2 }. (13.11)
2
Now the tangential forces T1 , T2 are defined independently according to the
corresponding return-mapping algorithms . The tangent vectors τ 1 and τ 2
are individual for each segment as well. Thus, the discretization leads to
13.3 Combination with Various Finite Element Models with CTC contact 385
1
δWT = {T1 (δρ2 − δρ1 ) · τ 1 + T2 (δρ1 − δρ2 ) · τ 2 }
2
1
= {T1 (Aδx ) · τ 1 − T2 (Aδx ) · τ 2 }
2
1
= {δx}T [A]T (T1 {τ 1 } − T2 {τ 2 }). (13.12)
2
The residual for the tangtential part is written then as
1
{RT } = [A]T (T1 {τ 1 } − T2 {τ 2 }). (13.13)
2
13.2.3 Part for Moment (Rotational) Interaction

The part for the moment interaction starting from the equation

M1
δWM = Sym (δρ2 − δρ1 ) · g1 (13.14)
r
is discretized completely similar to the tangential part. The residual for the
rotational part is written then as
1
{RM } = [A]T (M1 {g1 } − M2 {g2 }). (13.15)
2r
The full residual vector is finally constructed as the sum of all parts:
{R} = {RN } + {RT } + {RM } (13.16)
13.3 Combination with Various Finite Element Models

of the Continuum for Curve-To-Curve Contact
13.3.1 Implementation of Curve-To-Curve Contact
Algorithm – Linear Element for Edge-to-Edge
Contact
In this section we present the structure of the iterative solution scheme for
a Newton type method and describe the implementation of the developed
theory based on the simplest finite element with linear shape functions. Let
us consider two edges from contacting bodies covered with a finite element
mesh possessing linear approximations. We construct a possible contact pair
(1) (2)
taking a linear approximation through nodes r1 and r1 for the edge ρ1 (ξ 1 )
(1) (2)
from one body and a linear approximation through nodes r2 and r2 for
the edge ρ2 (ξ 2 ) from another body. A pair of ρ1 (ξ 1 ) and ρ2 (ξ 2 ) defined by
four nodes, see Fig. 13.1 gives us a contact element:
1 − ξI (1) 1 + ξI (2)
ρI (ξI ) = rI + rI , I = 1, 2, −1 ≤ ξ I ≤ 1. (13.17)
2 2
A nodal vector for the contact element is defined as follows:
' 6T
(1) (1) (1) (2) (2) (2) (1) (1) (1) (2) (2) (2)
{x}T = x1 , y1 , z1 , x1 , y1 , z1 , x2 , y2 , z2 , x2 , y2 , z2 (13.18)
r1(2)
ξ 1
r1 (1)
r2(1) r2 (2)
ξ2
.
Fig. 13.1 Two linear edges define a contact pair – a contact finite element with
linear approximations.
In almost all parameters the relative displacement vector ρ2 − ρ1 is required,

therefore, a matrix approximation operator A(ξ1 , ξ2 ) introduced as
⎡
1 − ξ1 1 + ξ1
− 0 0 − 0 0
⎢ 2 2
⎢ 1 − ξ1 1 + ξ1
A(ξ1 , ξ2 ) = ⎢ 0 − 0 0 − 0
⎣ 2 2
1 − ξ1 1 + ξ1
0 0 − 0 0 −
2 2
⎤
1 − ξ2 1 + ξ2
0 0 0 0
2 2 ⎥
1 − ξ2 1 + ξ2 ⎥
⎥(13.19)
0 0 0 0 ⎥
2
1 − ξ2
2
1 + ξ2 ⎦
0 0 0 0
2 2
gives us the approximation of the aforementioned vector as
ρ2 (ξ2 ) − ρ1 (ξ1 ) = A x. (13.20)

For arbitrarily curved lines and arbitrarily large displacements it is necessary

to compute the following geometrical parameters of the contact:
1. The closest point projection (CPP) procedure for convective variables,
see in eqn. (3.58). The procedure gives as results the ξp1 , ξp2 as projection
points and the first local coordinates. The corresponding arc-lengths s1 , s2
can be computed using the Jacobian J(I) .
2. The Jacobian for the corresponding transformation becomes ξI → sI

3

dsI ∂xj ∂xj
J(I) = = ρI,ξI · ρI,ξI = .
dξI j=1
∂ξI ∂ξI
3. Local coordinate vectors of the Serret-Frenet frame :

• a unit tangent vector τ I
∂ρI ∂ρI 1
τI = =
∂sI ∂ξI J(I)
Then a curvature kI in eqn. (2.44) and a torsion κI in eqn. (2.45) can be

computed.
• a binormal vector from eqn. (2.44) as
ρ˙I × ρÏ
βI =
|ρ˙I × ρÏ |
• a normal vector as
ν I = βI × τ I
4. The distance between curves at projection points as a second coordi-
nate r:
r = |ρ2 (ξ2p ) − ρ1 (ξ1p )|

5. Coordinate vectors for the local coordinate system :
ρ2 (ξ2p ) − ρ1 (ξ1p )
e1 = , g1 = τ 1 × e1
r
6. The angle ϕI as the third coordinate defined via
cos ϕI = (eI · ν I ), sin ϕI = (eI · β I ).

For the element with linear approximations (both, curvature k and torsion κ
are zero) the procedure is simplified.
1. Tangent vectors for the corresponding linear segments
(2) (1)
r1 − r1 ρ1
ρ̇1 (ξ1 ) = −→ τ 1 =
2 |ρ1 |
(2) (1)
r2 − r2 ρ2
ρ̇2 (ξ2 ) = −→ τ 2 =
2 |ρ2 |
2. The coordinate vector e1 is derived by its definition to be orthogonal to
both tangent vectors , therefore, via the cross product
e1 = τ 1 × τ 2 .
3. The coordinate vector g1 is derived via a cross product
g1 = τ 1 × e1 .
4. The closest point projection procedure is resolved exactly. The projection
points ξ1p , ξ2p (first coordinates) are defined via the following linear system:
⎧ 1 (1)
⎪ (2) (1) (2)
⎨ (ρ̇1 · ρ̇1 )ξ1 − (ρ̇1 · ρ̇2 )ξ2 = 2 (r2 + r2 − r1 − r1 ) · ρ̇1
⎪
⎪
⎪
⎩ −(ρ̇ · ρ̇ )ξ + (ρ̇ · ρ̇ )ξ = 1 (r(1) + r(2) − r(1) − r(2) ) · ρ̇
2 1 1 2 2 2 2
2 2 2 1 1
The determinant of the system can be transformed as
det = (ρ̇1 · ρ̇1 )(ρ̇2 · ρ̇2 ) − (ρ̇1 · ρ̇2 )2 = |ρ̇1 × ρ̇2 | = 0
leading to the known condition “tangent vectors should not be parallel”,

see Sect. 3.4.2.
5. Local searching procedure .
If any of projection points is laying outside the corresponding segment i.e.
if ξ1p ∈
/ [−1, 1], or ξ2p ∈
/ [−1, 1] then
a contact element is excluded from the computation at the current itera-

tion.
6. The closest distance between lines (the second coordinate r) is defined as
1 (1) (2) (1) (2)
(r + r2 − r1 − r1 ) · e1
r=
2 2
7. Definition of the coordinate increments. Since the straight line has an
infinite number of normal vectors, the third coordinate ϕI can be defined
arbitrarily, e.g. via “the orientation node“ similar to the definition of
linear finite element for beams. However, all coordinate increments are
defined via displacement increment vectors.
Defining a displacement increment vector for the first curve as Δu1
and resp. Δu2 for the second curve all increments for convective coordi-
nates are computed from the incremental analog of eqn. (4.193)
⎧
⎨ Δs1 = (Δu2 − Δu1 ) · τ 1
⎪
Δr1 = (Δu2 − Δu1 ) · e1 1 ↔ 2.
⎪
⎩ Δϕ1 = (Δu2 − Δu1 ) · g1
r
They are directly used for the corresponding return-mapping schemes in
eqns. (6.182), (6.183), (6.185), (6.186).
13.3.2 Combination of Finite Beam Elements

with the Curve-To-Curve Contact Algorithm
For a beam-to-beam algorithm we have to select a special finite beam ele-
ment formulation allowing both finite rotations and enrichment with arbi-
trary curvilinear geometry. An extensive development is given e.g. in a com-
plete series of articles by Ibrahimbegovic [72], Ibrahimbegovic [73], Ibrahim-
begovic and Taylor [74]. The kinematics of deformation is subjected to the
Reissner or the Timoshenko hypothesis; finite rotations are represented via
quaternions. A short overview of the necessary equations is given in the fol-
lowing. The variational equation is given by
# #
1 1
Wint = n+κ m ds0 =
T T
T ΛCn ΛT +κT ΛCm ΛT κ ds0 , (13.21)
2 L 2 L
where is a strain measure for the mid-line of the curvilinear beam, κ is

a bending strain for the curvilinear beam. Finite rotations are represented
by the matrix Λ given by the Rodrigues formula. In the current approach,
the matrix Λ, the update of Λ as well as κ within the iteration scheme are
computed via quaternions operating with an incremental rotation vector Δw,
see details in [72]. The matrix Cn represents the stiffnesses of the mid-line
with regard to a load vector n containing axial and two shear forces. The
matrix Cm represents the bending stiffnesses with regard to a moment vector
m containing torsional and two bending moments. In a beam coordinate
system they have the following form:
⎡ ⎤ ⎡ ⎤
EA 0 0 GIt 0 0
Cn = ⎣ 0 GAs1 0 ⎦ , Cm = ⎣ 0 EI2 0 ⎦ , (13.22)
0 0 GAs2 0 0 EI3
where EA is the axial stiffness, GAs1 , GAs2 are shear stiffnesses, GIt is the
torsional stiffness, EI2 , EI3 are bending stiffnesses. The finite element then
contains six degrees of freedom: 3 displacements and 3 rotations. Thus, for
the numerical algorithm, both a displacement vector u and an incremental
rotation Δw have to be approximated.
C 1 -Continuous Isogeometric Approximation of the Curve

For further implementation, we are using two types of approximation:
• C 1 -continuous cubic spline in Hermite form ;
• a special NURB spline representing the circular geometry exactly.
The Hermite type interpolation leads to the following finite element:
x(t) = x(1) H03 (t) + m(1) H13 (t) + m(2) H23 (t) + x(2) H33 (t), (13.23)
where Hi3 (t), 0 ≤ t ≤ 1 are Hermite polynomials, see Table 8.3. x(1) , x(2) are
two nodes and m(1) , m(2) are tangent vectors at corresponding nodes, see
more in Sect. 8.2.4. Defining then the tangent vectors via a chord interpola-
tion from nodes of the neighboring elements we obtain the following structure
of the finite element, see Fig. 13.2:

H 3 (t)
x(t) = x(1) H03 (t) − 2
2

3

H (t)
+ x(2) H33 (t) + 1
2
3
H (t) H 3 (t)
− x(3) 1 + x(4) 3 (13.24)
2 2
Finally, the finite element is defined by nodes x(1) and x(2) i.e. all parameters
such as forces, moments etc. are computed for the parameter 0 ≤ t ≤ 1 , see
Fig. 13.2. Nodes x(3) and x(4) are taken from the neighboring element to keep
C 1 -continuity on element boundaries. The displacement vector u(t) and the
vector of infinitesimal rotation Δw are approximated in the same fashion.
Equation (13.24) gives an approximation which is valid only for inner
points of the curve. For boundary nodes the corresponding tangent vector
m(i) should be given according to the geometry.
t 1
0
x (2) x (4)
x (1)
x (3)
Fig. 13.2 Approximation for the spline beam element. The beam element is given
.
by nodes x(1) and x(2) . Nodes x(3) and x(4) are taken from the neighboring element
1
to keep C -continuity on element boundaries.
w2 =2
1 w1 =1
w0 =1
0 1
Fig. 13.3 Quadratic NURBS approximation for the quarter of a circle. Weights
are w0 = 1, w1 = 1, w2 = 2.
This type of approximation is falling into a simplest type of NURBS ap-

proximation written in a very general form via Bernstein’s polynomials of
order n
w0 b0 B0n (t) + w1 b1 B1n (t) + ...wn bn Bnn (t)
x(u) = . (13.25)
w0 B0n (t) + w1 B1n (t) + ...wn Bnn (t)
This description of a spline has many advantages due to the additional de-
grees of freedom because of additional weights wi . It leads to multiple possi-
bilities to control the geometry, e.g. to represent a certain geometry exactly. A
large number of special literature, however, is available containing all NURBS
properties in detail, see e.g. Farin [38], Piegl and Tiller [144]. An example of
a special NURB spline of the second order representing exact geometry of a
circle is used in further implementation. For the quarter of a circle it has the
following weights w0 = 1, w1 = 1, w2 = 2, see Fig. 13.3.
A NURBS type contact is performed as two spline segments from the po-
tentially contacting curves defining a beam-to-beam contact pair leading to
a contact element possessing twice as many nodes, e.g. a contact element
for the 4-nodes spline element is the 8-nodes curve-to-curve contact finite
element.
13.3.3 Development of Special “Solid-Beam”

Elements for Beam-to-Beam Contact
Beam finite element models are representing the dimension reduction from
the 3D continuum into a 1D manifold. The costs of this are additional rota-
tional degrees of freedom and as a consequence for large rotations a costly
update procedure for finite rotation parameters as well as for bending strains.
ϕ 2
1
r
ξ
6
4
Fig. 13.4 “Solid-Beam” with elliptic cross-sections defined by 6 nodes. Definition

of local variables.
For the cases of special beam-to-beam contact like cables possessing a particu-
lar elliptic geometry for their cross-sections it is rather attractive to construct
a special 3D finite beam element . This element can be constructed as follows,
see Fig. 13.4:
1. the mid-line of the beam is taken, first, with e.g. linear approximation.
It is defined then by nodes 1 and 2;
2. the left cross-section is defined to be elliptic with the reference main axes
defined by nodes 1-3 and 1-4;
3. the right cross-section is defined to be elliptic with the reference main
axes defined by nodes 2-5 and 2-6;
Thus, an approximation of the 6-nodes “solid-beam” element is defined by
the following shape functions:

6
x= N (ξ, r, ϕ)xi = (13.26)
i=1
= x1 (1 − ξ)[1 − r(cos ϕ + sin ϕ)]
+ x2 ξ[1 − r(cos ϕ + sin ϕ)]
+ x3 (1 − ξ)r cos ϕ
+ x4 (1 − ξ)r sin ϕ
+ x5 ξr cos ϕ + x6 ξr sin ϕ
with 0 ≤ ξ ≤ 1; 0 ≤ r ≤ 1; 0 ≤ ϕ ≤ 2π (13.27)
13.4 Numerical Examples for Curve-To-Curve Contact 393
The nodal locations xi = (xi , yi , zi ) are given in a global Cartesian coordinate

system.
Contact for the solid-beam element is performed as follows: two mid-line
segments from the potentially contacting solid-beam element are defining
a beam-to-beam contact pair leading to the 4-nodes beam-to-beam contact
finite element.
Remark 13.2. Extension of the solid-beam approximation into a NURBS type

approximation can be obtained, if the linear shape functions defining the
approximation of a mid-line and depending on the variable ξ in eqn. (13.26)
will be changed into corresponding shape functions of NURBS type, e.g. into
those defined via Hermite functions in eqn. (13.24).
13.4 Numerical Examples for Curve-To-Curve Contact

In this section, first, a comparison of the non-frictional edge-to-edge contact
formulation, see Section 6.7.1, with a special case of Equilibrium of elastica
problem is presented for the verification. This case is also compared with the
beam-to-beam contact formulation, see Section 6.7.1. Afterwards, an example
with edge-to-edge contact for two intersecting beams is analyzed in details.
A “solid-shell” bilinear finite element, see [58], is chosen to model the be-
havior of the beams, thus modeling beams with rectangular cross section
directly. A contact element with two linear segments, see implementation de-
tails in Section 13.3.1, is used to enrich the model with edge-to-edge contact.
In order to study beam-to-beam contact in detail, it is necessary, first to
choose a proper finite element model for the beam itself and discuss then
the consistency with our model concerning the rotational moment M1 , i.e. it
should be consistently transfered to the corresponding rotational degrees of
freedom. The consistent application of the developed contact theory for the
curvilinear beam contact requires the application of high order finite element
techniques, see e.g. [153], or as an alternative the iso-geometrical technique,
see e.g. [70]. However, the development of such a curvilinear finite beam el-
ement together with a curvilinear beam-to-beam contact element especially
for the arbitrary frictional case as well as further numerical analysis are out-
side of the scope of the current discussion. Some details can be summarized
as follows:
1. The finite beam element model is based on Reissner theory . The finite
element then contains six degrees of freedom per node: 3 displacements
and 3 rotations. The stiffnesses of the mid-line Cn with regard to a load
vector containing axial and two shear forces and the stiffnesses Cm with
regard to a moment vector containing torsional and two bending moments
represent the mechanical properties of the beam. In a beam coordinate
system they have the following form:
⎡ ⎤ ⎡ ⎤
EA 0 0 GIt 0 0
Cn = ⎣ 0 GAs1 0 ⎦ , Cm = ⎣ 0 EI2 0 ⎦ , (13.28)
0 0 GAs2 0 0 EI3
where EA is the axial stiffness, GAs1 , GAs2 are shear stiffnesses, GIt is
the torsional stiffness, EI2 , EI3 are bending stiffnesses.
2. Finite rotations are taken into account within the finite element formu-
lation according to Ibrahimbegovic [72].
3. The iso-geometrical technique, see [70], has been exploited to obtain C1-
continuous approximations for the curvilinear beams.
4. Beam-to-beam contact elements are obtained using the same approxima-
tion functions for the contact pairs in consistent fashion.
Three approaches to model continuum including solid-beam element, beam
element with finite rotation and “solid-beam“ are finally compared numeri-
cally.
13.4.1 Bending of a Flexible Beam by a Rigid Beam.

Non Frictional Case
The first flexible beam, see Fig. 13.5 modeled with 50 solid-shell finite ele-
ments is positioned in the XOY -plane and clamped at the left end. Material
parameters: Linear Hooke’s material with Young’s modulus E = 2.1 · 104 and
Poisson ratio ν = 0.3. Geometrical parameters: length 1.0 and square cross
section 0.02 × 0.02 – units are consistent. The second beam with the same
cross section is rigid (only one solid-shell element) and is positioned paral-
lel to the OY -axis under the right end of the first beam. The first flexible
beam is turned at 45o along the OX-axis such that during further load-
ing it is contacting with the rigid beam only along the lower edge. Thus,
only edge-to-edge contact is realized during the deformation process. In the
case of modeling with bi-linear solid-shell elements, the displacement vector
u = {−1.0000, 0.0000, 0.6366}T is applied with 1000 equal increments (load
steps) to all nodes of the rigid beam. The contact between beams is assumed
to be non-frictional with a penalty parameter εN = 2.1 · 105 . All nodes of
the flexible beam are constrained along the OY -axis to prevent bouncing
along the rigid beam. During the loading process the first flexible beam is
sliding along the rigid beam. Thus, a quasi 2D-deformation is realized, see
the diagram of deformation in Fig. 13.6.
Verification – Equilibrium of the Elastica Problem

The analytical solution of the current evolution contact problem is hardly
possible, however, the final deformed state (for non-frictional problem) can
be formulated as an equilibrium of the elastica problem which was a subject
of interest since centuries for many researchers starting with Leonhard Euler.
The equilibrium of the elastica problem in 2D can be formulated via the

famous equation:
dϕ
EJ = M, (13.29)
ds
where ϕ is angle between a tangent line and the X-axis, s is the standard arc-
length parameter, and M is a bending moment. Eqn. (13.29) shows the well
dϕ
known proportionality of the curvature κ = to the bending moment M
ds
with bending stiffness. EJ as a proportionality coefficient. The definition of
tangent angle ϕ and the differential dependence of a shear normal force Q and
and a tangential force N on the moment M should be taken into account in
order to obtain then the deformed line. Though, several analytical approaches
to solve the equilibrium of the elastica problem are known in literature, e.g.
solutions via elliptic integrals see e.g. in Popov [146], we are going to solve
the problem directly via a finite difference scheme. Our special problem can
be formulated as a Boundary Value Problem (BVP) (we adapt the stiffness
to our “solid-shell” formulation by setting EJ/(1 − ν 2 ) instead of EJ) for the
following system of Ordinary Differential Equation (ODE):
⎧
⎪
⎪
EJ dϕ
= M + N x cos α − N y sin α
⎪
⎪
⎪
⎪ 1 − ν 2 ds
⎨
dx
= cos ϕ (13.30)
⎪
⎪ ds
⎪
⎪
⎪
⎪ dy
⎩ = sin ϕ
ds
with boundary conditions
⎧
⎪
⎪ ϕ(0) = 0; ϕ(sc ) = α
⎪
⎨
x(0) = 0; x(sc ) = xc (13.31)
⎪
⎪
⎪
⎩ y(0) = 0; y(s ) = y ,
c c 0 ≤ s ≤ 1.
where coordinate system is chosen at the clamped end; N is a contact force

between beams, x(sc ) and y(sc ) are Cartesian coordinate of the contact point
appeared at length sc ; α is the angle between the beam tangent vector and
the X-axis. In order to solve the system (13.30) the aforementioned differ-
ential dependencies between forces and bending moment M should be taken
account. However, in order to avoid the direct solution of BVP we will solve
it approximately by reformulation BVP into the Initial Value Problem (IVP)
Moreover, some values we will use from the finite element solution. Namely,
the following steps of verification are fulfilled for comparison of the solutions:
1. From the finite element solution a final force N and an angle α at the
contact point (xc , yc ) are defined. For the current problem N = 2.8092
and α = 35.5346o are obtained at xc = 0 yc = 0.6366.
2. The system (13.30) is solved with given N and α as the Initial Value
Problem with the following initial conditions
⎧
⎪
⎪ ϕ(0) = 0;
⎪
⎨
x(0) = 0; (13.32)
⎪
⎪
⎪
⎩ y(0) = 0, 0 ≤ s ≤ 1.
3. The obtained deformed line is then compared with deformed line from
FE solution.
Remark 13.3. The current approach of solution for BVP as IVP allows to
satisfy the boundary conditions at the contact point x(sc ) = xc and y(sc ) =
yc only approximately.
The simple explicit Euler finite difference scheme is applied then to solve
the problem with Δs = 0.001. Comparison of the deformed central-lines is
presented in Fig. 13.9. As mentioned in Remark 13.3, the deformed line is
not passing through the contact point xc , yc .
Computations with Curvilinear Beam-to-Beam Contact

For a further comparison of the final deformed configuration the beam fi-
nite element model shortly described at the beginning of the Section 13.4 is
used. Only four C 1 -spline continuous finite elements are used to model the
elastic beam. The same material data is taken. In order to obtain a model
mechanically equivalent to the solid-shell one, the same stiffness character-
istics as given in eqn. (13.28) have been taken. Namely, the cross section A
and all area moments of inertia in eqn. (13.28) are taken to be the same as
for the square section of the solid-shell finite element model. However, the
cross section for the beam is taken to be circular for the contact algorithm
as shown in eqn. (6.179) with constant radius for both beams R = R1 = R2 .
The radius R is computed from the equivalency of the cross section area for
the solid-shell and the circular section. Now, the displacement vector u is
applied only with 100 load steps to all nodes of the rigid beam, see Fig. 13.7.
The verification with the analytical solution is also presented in Fig. 13.9.
The problem is analyzed with both tangent matrices for normal interaction
and presented in Table 13.1 for the following cases:
• Solid-shell FE model with “edge-to-edge” contact elements;
• Curvilinear beam FE model with “beam-to-beam” contact. Normal con-
tact matrix for arbitrary large distance r and curvature with parts in
eqns. (13.2) and (13.3).
• Curvilinear beam FE model with “beam-to-beam” contact for linear con-
tact elements with only one part as given in eqn. (13.4).
During the loading process the flexible beam is undergoing large deformations
(from a straight beam into a curved beam) as well as large sliding, especially
Table 13.1 Comparison of solution for 1) linear solid-shell elements; 2) curvilinear

beam elements with full contact matrix; 3) curvilinear beam elements with “linear”
contact matrix.
FE model No. of elem. No. of load steps Global No. of iterations

1 50 1000 3986
2 4 100 396
3 4 100 428
at the end of the loading process, see diagram in Fig. 13.6. This leads, on
one side, to the necessity of a relatively large number of bi-linear solid-shell
elements (50 elements) and, on the other side, to a relatively small load step
size (resp. a large number of load steps) in order to describe the sliding over
segments correctly. Steps 7 − 10 presented in Fig. 13.6 are crucial during the
loading process because the beam undergoes large deformations and large
sliding. Namely this part of loading requires small load steps for linear ele-
ments. For the beam-to-beam element the number of equilibrium iterations
per load step is increasing from 4 to 11 during the last load steps. A rela-
tively small advantage using the full matrix compared with the reduced one
is observed in the global number of iterations – 396 against 428 iterations,
see Table 13.1. This can be explained by the fact that the beam possesses
a small curvature at the contact point together with a small radius for the
circular cross section. A crucial example can be constructed by taking “a
soft” beam, scaling all shear moduli in eqn. (13.28) with a factor 10−4 . Then
displacements are now applied in only 75 load steps of the same size as in
the previous example. The final configuration is presented in Fig. 13.8 (visual
non-smoothness is just the result of incomplete visualization – only element-
wise tube surfaces are constructed during post-processing via Matlab which is
used here). In the example shear softening leads to high curvatures at contact
points, thus the influence of the curvature part should be more pronounced.
That is essentially true!!! – no convergence after the 70th load step has been
found with using the linear matrix given in eqn. (13.4) and 297 global itera-
tions are necessary taking all parts into account in eqns. (13.2) and (13.3).
The advantage of C 1 -spline continuous beam finite elements compared
to the linear solid-shell finite elements is obvious – only four elements are
sufficient to describe efficiently the highly deformed configuration.
The small differences between deformed lines obtained for the solutions
using “solid-shell” and beam finite elements as well as using the a finite
difference scheme for the Equilibrium of the elastica problem are explained by
the different kinematical hypotheses which are the basis of the corresponding
theories for the three compared solutions and by the application of different
numerical approaches.
13.4.2 Analysis of Contact for Intersecting Beams

Two clamped beams of unit length intersecting each other are chosen to study
various combinations of parameters within the developed theory such as the
tangential and the rotational sticking including various sticking and sliding
cases of two lines. The “first” lower beam is parallel to the OY -axis and the
“second” upper beam is parallel to the OX-axis at the initial configuration,
see the position in Fig. 13.10. Both beams are turned at 45o along their
central-lines in order to provide initial contact only along the edges at a
distance 0.75 from both clamped ends. Now two beams are flexible, made of
the same material: linear Hooke’s material with Young’s modulus E = 2.1·104
and Poisson ratio ν = 0.3; and possessing the same geometrical parameters:
length L = 1.0 and square cross section A = 0.02 × 0.02. The beams are
modeled with 50 solid shell elements and edge-to-edge contact elements are
supplied.
The prescribed displacement uz = 0.5 in vertical OZ-direction is pre-
scribed at the upper node of the lower beam, see Fig. 13.10. The vertical
displacement is applied in 100 increments. In the following always the top
view along the OZ-axis for both initial and deformed final configurations are
presented for different combinations of parameters.
Non-Frictional Contact
For non-frictional contact only the normal tangent matrix in eqn. (13.4)
is taken into account together with the corresponding normal force N in
eqn. (6.180). The penalty parameter is taken as εN = 2.1 · 106 . Convergence
is reached with 391 global iterations for 100 load steps. Both, relative slid-
ing along the “first” lower beam increasing ξ 1 and along the “second” upper
beam increasing ξ 2 are observed, see Fig. 13.12.
Full Sticking (Tied) Contact

Full sticking, or tied contact, is enforced using all tangential and rotational
sticking conditions (corresponding penalty for both lines ε1 = ε2 = 2.1 · 106 )
without the return-mapping scheme. Convergence is reached with 375 global
iterations for 100 load steps. The result is presented in Fig. 13.13. To study the
influence of various parts for the enforcement of full sticking (tied) condition
the following cases are computed:
1. Tied contact enforced by tangential and rotational sticking .
2. Tied contact enforced only by tangential sticking .
3. Tied contact enforced only by rotational sticking .
It is interesting to note that the same final configuration presented in
Fig. 13.13 is reached for all three cases. To emphasize this effect, an an-
gle between the vectors a and b defining the edge directions for elements as
depicted in Fig. 13.10 is studied. This angle is equal to 45o at the initial un-
deformed state and is then changing during the deformation, see Fig. 13.11.
However, one can see that the angle remains the same for all kinds of enforce-
ment. The explanations for this effect appear, however, obvious: full sticking
(tied contact) can be enforced only by rotational sticking because the corre-
sponding tangent matrix g ⊗ g responsible for the g direction is orthogonal
to the vector e direction, and therefore, is leading to the sticking of the in-
tersecting lines. The reason for the constant angle lays in the application of
solid-shell finite elements which are by construction not capable to transmit
the moment applied to the edge. However, the large opportunities for cou-
pling this moment between curves either with classical beams or with classical
shell finite elements possessing both rotational degrees of freedom and, in due
course, moments remain open.
Partial Tied Contact – Sticking Only Along One Line
Now, partial tied contact is modeled along only one line. First, partial tan-
gential sticking is enforced along the upper edge of the “first” lower beam by
setting the tangential penalty parameter ε1 = 2.1 · 105 without applying the
return-mapping scheme, while the non-frictional contact is assumed for the
lower edge of the “second” upper beam by setting ε1 = 0.0, see Fig. 13.14.
Convergence is reached with 386 global iterations for 100 load steps. Then,
partial tangential contact is enforced along the second line with the same
values only changing the beam 1 ↔ 2, see Fig. 13.15. Convergence is reached
with 379 global iterations for 100 load steps in the latter case.
One can notice that during the loading for the non-frictional case, see
Fig. 13.12, the “first” lower beam is undergoing large relative sliding along the
“second” beam increasing ξ 2 , while the “second” beam is undergoing only a
small relative sliding – expressed in tangential coordinates – ξ 2 >> ξ 1 . Thus,
partial tied contact with a locked variable ξ 1 supplying ε1 = 2.1 · 105 looks
similar to the non-frictional case – compare Fig. 13.12 and Fig. 13.14. In due
course, partial tied contact with a locked variable ξ 2 supplying ε2 = 2.1 · 105
looks similar to the full sticking case – compare Fig. 13.13 and Fig. 13.15.
Sliding Contact
Finally, the four considered cases non-frictional, full sticking and two partial
sticking cases lead to “an envelope” in which frictional cases with all possible
coefficients of friction μ1 and μ2 are contained. In Fig. 13.16 a top view of
this envelope (shadowed) and the “first” lower beam positioned inside this
envelope are given. The computation is provided for tangential sliding with
coefficients of friction: μ1 = 0.001 and μ2 = 0.05. Convergence is reached with
419 global iterations for 100 load steps. The choice of rather small coefficients
of friction is found to show that two beams are visually positioned inside the
envelope constructed by the deformed configurations for four extremal cases.
Fig. 13.5 Bending of a flexible beam by a rigid beam – initial and final configura-
tions applying solid-shell finite elements and edge-to-edge contact elements
8 7
6
9
5
10
Fig. 13.6 Diagram of deformation for equal load steps. Displacements of the rigid
beam are prescribed.
0.5
0.4
0.3
0.2
0.1
−0.1
−0.2
−0.3
−0.4
−0.5
0.6
0.4
1
0.2 0.8
0 0.6
−0.2 0.4
0.2
−0.4 0
−0.2
Fig. 13.7 Bending of a flexible beam by a rigid beam applying beam finite elements
and beam-to-beam contact elements
.
0.5
0.4
0.3
0.2
0.1
−0.1
−0.2
−0.3
−0.4
−0.5
0.6
0.4
0.2
0
1
−0.2 0.8
0.6
−0.4 0.4
0.2
0
−0.2
Fig. 13.8 Bending of a flexible “soft” beam by a rigid beam applying beam finite
elements and beam-to-beam contact elements
.
Edge-To-Edge
0.3 Finite Difference
Beam-To-Beam
0.2
0.1
Z
-0.1
-0.2
-0.3
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4

X
Fig. 13.9 Comparison of the deformed central-lines for a) 50 bilinear solid-shell
finite elements together with edge-to-edge contact; b) finite difference solution for
“the equilibrium of the elastica problem”; c) 4 C 1 -continuous curvilinear Reissner
beam finite elements and beam-to-beam contact
applied displacement
Fig. 13.10 Intersecting deformable beams. Initial configuration and loading

condition.
45
Full stick
Tangential stick
40 Rotational stick
Non frictional
35
Z 30
25
20
15
10
0 10 20 30 40 50
X
Fig. 13.11 Evolution of the angle between edges with vectors a and b for contacting
elements during deformation. Tied contact enforced: a) by tangential and rotational
sticking (Full stick); b) only by tangential sticking (Tangential stick); c) only by
rotational sticking (Rotational stick). d) Non-frictional contact.
Fig. 13.12 Non-frictional contact. Undeformed and deformed configurations.
We have to note that in the sliding case especially for large deformations
considerable convergence difficulties appear for large coefficients of friction
and the application of the Augmented Lagrangian scheme has shown to be
more efficient in this case.
Remark 13.4. The considered effect with partial sticking and sliding can
be considered as a result appearing during interaction between two beams
Fig. 13.13 Full sticking (tied) contact. Undeformed and deformed configurations.
Fig. 13.14 Partial tied contact. Tangential sticking is enforced along the upper
beam with a penalty of ε1 = 2.1 · 105 , however, non-frictional contact is supplied
for the lower beam. Undeformed and deformed configurations.
possessing anisotropic surfaces with different properties (e.g. one beam has a
very rough surface while the other has a very smooth surface). Another pos-
sibility to apply the developed curve-to-curve contact approach is to model
a hinge slider joint between two beams. In this joint the partial sticking
as well as different elastic and friction properties can be prescribed by the
construction.
Fig. 13.15 Partial tied contact. Tangential sticking is enforced along the lower
beam with a penalty of ε2 = 2.1 · 105 , however, non-frictional contact is supplied
for the upper beam. Undeformed and deformed configurations.
13.4.3 Discussion
Several achievements appear to be novel for the curve-to-curve contact de-
scription:
1. consideration of any relative motion including normal, tangential and
rotational components separately for each curve is possible;
2. rotational interactions including corresponding rotational moments be-
tween curves can be considered consistently.
The verification of the CTC algorithm is provided by the comparison between
beam-to-beam and edge-to-edge finite element models as well as verification
with a famous “Equilibrium of Euler elastica problem” computed via finite
difference scheme. The further numerical examples are illustrating the abil-
ity to describe various kinematics for curve-to-curve contact situations e.g.
partial sticking of a single curve. The necessity to use the full matrices for
the cases with high curvature at contact points has been shown in the case
of the “soft” beam.
13.4.4 Bending of a Flexible Beam by a Rigid Beam

– Comparison of Three Finite Element Models
Here the three finite element models for the continuum (solid-shell, beam
and solid-beam finite elements) are analyzed with the example “Bending of
a flexible beam by a rigid beam” computed in Sect. 13.4.1, see Fig. 13.5.
Fig. 13.16 Sliding contact—tangential sliding with coefficients of friction: μ1 =

0.001 and μ2 = 0.05 is considered. The “lower” beam lays inside the envelope
(shadowed) constructed by four extremal cases: 1) non-frictional contact, 2) full
sticking (tied) contact, 3) partial tied contact for the “lower” beam and 4) partial
tied contact for the “upper” beam.
The following cases are computed now representing various combinations

discussed in Section 13.3 (originally published in [95]):
1. 50 solid-shell finite elements and edge-to-edge contact;
2. 4 C 1 -continuous isogeometric beam elements and beam-to-beam contact;
3. 20 linear “solid-beam” element and beam-to-beam contact.
For the first case the flexible beam (with a square cross section 0.02 × 0.02)
is turned at 45o along the OX-axis such that during further loading it is
contacting with the rigid beam only along the lower edge. Thus, only edge-
to-edge contact is realized during the deformation process.
For the second case a mechanically equivalent beam model with the same
stiffness characteristics is taken. Namely, the cross section A and all area
moments of inertia in eqn. (13.22) are taken to be the same as for the square
section of the linear solid-shell finite element model. However, the cross sec-
tion for the beam is taken to be circular for the contact algorithm as shown
in eqn. (6.179) with constant radius for both beams R = R1 = R2 . The
radius R is computed from the equivalency of the cross section area for the
solid-shell and the circular section.
The third case is geometrically equivalent to the second one, i.e. the cross
section is circular, however, the stiffness properties will be computed directly
for the solid-beam element with its stiffness matrices.
Table 13.2 Comparison of solution for 1) linear solid-shell elements and edge-
to-edge algorithm; 2) curvilinear beam elements and beam-to-beam algorithm; 3)
linear solid-beam elements and beam-to-beam algorithm.

1 50 1000 3986
2 4 100 396
3 20 100 752
Remark 13.5. An energy norm together with the relative tolerance εconv =
10−12 has been used as a convergence criterion in all computations.
During the loading process the flexible beam is subjected to large deforma-
tions, especially at the end of the loading process. This leads, on one side, to
the necessity of a relatively large number of bilinear solid-shell elements (50
elements) and, on the other side, to a relatively small load step size (resp.
a large number of load steps) in order to describe the sliding over segments
correctly, see Table 13.2. For the solid-beam case the local contact searching
is performed according to the beam-to-beam case (larger area than edge-to-
edge) – this allows to work with less elements (20 FE) than for the solid-shell
elements (50 FE). However, 20 linear solid-beams elements are resulting in a
stiffer behavior than the first two models, see Fig. 13.9. This leads to a fairly
large number of global iterations – 752. The number of iterations per load
step is increasing towards the end of the deformation process to 18 per load
step. We have to note again that in Fig. 13.9 the beam-to-beam case is shown
for the mid-line; but both edge-to-edge and solid-beam cases are shown for
contact lines (the point of intersection of both lines is at x = 0).
Bending of a Flexible Beam (Smooth “Soft” Cable) by a Rigid

Beam
Convergence for a “soft” cable depending on the type of approximation has
been studied in [95], where shear softening has been defined by setting the
shear stiffnesses kGA with a scale coefficient k = 10−4 . This leads to “the
softening” of the shear forces. For the solid-beam approach this effect can
be defined as orthotropy inherited with the curvilinear coordinate system
ξ, r, ϕ. In this case both orthotropic shear moduli Gξr , Gξϕ are scaled with
the same factor k = 10−4 . The result is presented in Fig. 13.17 for the
-0.05
Beam-To-Beam
Solid-Beam
-0.1
-0.15
Z
-0.2
-0.25
-0.3
-0.35
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X
Fig. 13.17 “Soft” cable. Comparison of the deformed lines for a) 4 C 1 -continuous
curvilinear beam finite elements and beam-to-beam contact; b) 20 linear solid-beam
finite elements together with beam-to-beam contact.
displacement vector u = {−0.3000, 0.0000, 0.1910}T (only 30% compared to

the previous case). The figures shows a good correlation, however, linear solid-
beam elements are requiring (again due to stiffer behavior) a larger number
of iterations, see Table 13.3.
Table 13.3 Comparison of solutions for “soft” cable case with curvilinear beam
elements together with beam-to-beam algorithm (FE model 1) and linear solid-
beam elements together with beam-to-beam algorithm (FE model 2).

1 4 30 93
2 20 30 260
13.4.5 Contact between Rings

Two equal rings represented by two intersecting circles in orthogonal planes
are contacting each other, see Fig. 13.18. The distance between both is se-
lected so that each circle is passing through the center of curvature of the
other one. The radius of the midlines for rings is R0 = 1.0, the radius of a
circular cross-section is r = 0, 2. The material is linear elastic with E = 200
and ν = 0.3. The cross section of the left ring is fixed – all displacements
and rotations of the last left node resp. cross section are fixed. Both rings
are meshed with 5 NURBS-elements each, the corresponding linear skeleton
is shown in Fig. 13.18. The situation is more general than the simplest node-
to-node contact – the spline elements are positioned along the ring so that
in the reference configuration the shortest distance between rings is found
between the middles of the spline elements. The assumption of non-frictional
contact leads further to oscillations at the contact point, therefore, frictional
sticking contact is assumed with high coefficients of friction μ1 = μ2 = 1.0
enforcing sticking behavior at the contact point.
The displacement u = 1.273 in the last right node of the right ring
is applied with increments Δu1 = 0, 001. The rings are contacting with
u = 1.05 − 0.2 − 0.2 = 0.6. The penalty parameter for contact is taken as
108 . The final configuration before disconvergence is presented in Fig. 13.18.
Computation with such small load steps is motivated to reach the highly de-
formed final configuration, however, even in this definitely not yet strongly
deformed case severe convergence problems are observed with total number of
iterations 8713. The possible reason for this is the rather small number of ele-
ments and that a linear material law taken into account. The ill-conditioning
of the tangent matrix due to the large penalty in comparison with the elastic
module is also influencing convergence.
13.4.6 Tying of a Knot

A knot is a method for fastening or securing linear material such as rope by
tying or interweaving. Wikipedia
A knot is a perfect example requiring both a robust smooth cable element,

and a robust curve-to-curve contact algorithm. Since many centuries knots
have been in professional usage for climbers, fishermen, sailors and others.
Among the huge number of knots, see the Ashley book of knots [7], there are
binding knots and bends. The binding knots are used to constrict and hold
objects together, and the bends serve to unite or join two ropes. In order
to fulfill the security conditions such type of knots must not show relative
sliding of ropes and must not spill or capsize (destroying the shape of a knot)
under tension.
From our knowledge there is no investigation yet of such a problem in the
computational mechanics community, therefore the current example is a trial
to begin to study this problem. We start with the most common Reef Knot,
or Square Knot, see Fig. 13.19. This knot however is reported to be dangerous
– There have probably been more lives lost as a result of using a Square Knot
as a bend to tie two ropes together than from the failure of any other half
dozen knots combined [7, p. 258].
u Y
1.5
1
0.5
0.5
0 0
u −0.5
−0.5
−1
−1
−1.5
1.5
1
1
0 0.5 0
0.5
0.5
0.5
0 1 0
1 1.5
1.5 2 −0.5
−0.5
2.5
2 3 −1
2.5 −1 3.5 −1.5
3 4
Fig. 13.18 Contact between rings. a) reference linear skeleton; b) reference

smooth (real) configuration after visualization; c) deformed configuration after
visualization.
The knot is modeled as follows, see the visualization via Matlab in

Fig. 13.19:
• A spline forming a loop is passing through 33 characteristic nodes. Thus,
each cable is modeled with 32 C1-smooth spline beam elements.
• Two loops are positioned initially without contact to form an opened Reef
Knot, see Fig. 13.19.
• Material is linear elastic with E = 200 and ν = 0.3.
• Cross section of both cables is circular with the radius r = 1.
• Dirichlet boundary conditions are applied at points A and B in order to
supply the symmetry boundary conditions in plane XOZ.
• The displacement vector along the OX-axis is applied incrementally at
both ends of both cables.
Different loading situations are studied:
1. beams are modeled as “soft” cables with a coefficient k = 10−4 in
eqn. (13.22), see Sect. 13.4.4. and displacements are free along the OY -
axis, see Fig. 13.19;
symmetry BC
20
u 15
A
10
B u
Z 5
0 −20
−15
X −5 −10
−5
Y O −10 0
5
−15
10
15
−20
20
−25 −20 −15 −10 −5 0 5 10 15 20 25
Fig. 13.19 Geometry of a knot. Boundary conditions of symmetry and loading by

prescribed displacement u.
Fig. 13.20 Both ends are free in the XOZ plane. Opening of the knot is observed
though both ends are pulled.
Fig. 13.21 Additional displacements along the Y -axis are applied to tie the knot.
2. beams are modeled as a “soft” cables, displacements are prescribed along

the OY -axis in order to tie the knot, see Fig. 13.19.
Remark 13.6. During modeling it has been found that only C1-smooth spline
finite elements are capable to represent the result. Higher order finite elements
(quadratic, cubic), but without C1-continuity are showing disconvergence
during the first crossing of element boundaries.
If both ends are free to move in the XOZ plane as in case 1 then the stretching
of them leads to opening the knot, namely both ends of the knots are moving
outwards, see Fig. 13.20. A proper tying of the knot is thus enforced by
additional application of the tying displacement along the OY -axis in case 2,
see Fig. 13.21.
The result is rather showing “an unusual behavior”. A more careful anal-
ysis recovers that the effect of untying the knot is caused by the CPP pro-
cedure. In the final position shown in the picture the tangents of the cables
at contacting points are becoming almost parallel. In this position the CPP
procedure becomes unstable because the corresponding Jacobian is close to
zero. This leads to the necessity to develop an improved contact algorithm
for the problem of parallel tangent vectors.
13.4.7 Discussion
The bending of a flexible beam by a rigid beam example is considered together
with several finite element models such as 3D solid-shell finite elements and
finite beam elements. A special solid-beam finite element for elliptic cross-
sections is developed to consider 3D continuum behavior in beam-to-beam
contact. It is clear now that a soft cable behavior is described within the
orthotropy inherited with a curvilinear coordinate system by setting very
small value for corresponding shear modulus. Various finite element tech-
niques including isogeometric techniques have been employed. The numerical
examples are selected to illustrate the possibility of the geometrically exact
curve-to-curve contact formulation to work together with various combina-
tions of finite element formulations. Interesting examples are knots for which
the C 1 continuous isogeometric elements proved to be favorable.
14
Frictional Interaction of a Spiral Rope
and a Cylinder – 3D-Generalization of
the Euler-Eytelwein Formula
Considering Pitch
Historically, following e.g. Holzer (2005) the problem of the definition of fric-
tional rope forces appearing in a rope sliding over a cylinder has been for-
mulated in the dissertation of Gautier (1717). The solution of this problem
was reported by Euler in his Remarks on the effect of friction on equilibrium
published by the Berlin Academy of science, see Euler (1769). Since the first
time publishing the Euler solution by Eytelwein (1808) in his Handbuch der
Statik fester Körper in 1808 the problem is spread through practical appli-
cations and became known as Euler-Eytelwein problem in standard books of
technical mechanics see e.g. in Gross et.al. (2004), and is also known as belt
or coil friction formula see e.g. in Maurer and Roark (1944), or as capstan
friction problem, see e.g. in Meriam (1978).
The Euler-Eytelwein problem is usually formulated as 2D problem de-
scribing the interaction between a rope and a circle, see Fig. 14.1 with the
well-known solution for forces T1 and T2 as
T1 = T2 e μϕ , (14.1)
where μ is a coefficient of Coulomb friction between the cylinder and the rope
wrapped over the cylinder with an angle ϕ. However, in practical applications
one can observe a pitch between coils, thus leading to a spiral line or helix
wrapped over the cylinder. In this chapter an analytical solution of this case
is presented.
14.1 Equilibrium Equations for the Rope in Space

We consider the elementary part of the rope with length ds, see Fig. 14.2. The
center line of this part is assumed to be in equilibrium under resultant forces
R and R + dR acting on the center line resulting from the stresses on the
cross-sections, and under resulting moments M and M + dM acting on the
center line resulting from the moments of stresses on the cross-sections, with
414 14 3D-Generalization of the Euler-Eytelwein Formula Considering Pitch
μ
T1
T2
ϕ
Fig. 14.1 Classical representation of the Euler-Eytelwein problem known also as

belt friction problem, or capstan problem: A rope wrapped over a cylinder in 2D.
R
q
−( R + d R )
τ ds
β
(−M+dM)
ν
M
Fig. 14.2 Center line of a rope wrapped over a cylinder: geometry and loading.
distributed load q and distributed moment m. The equilibrium conditions

can then be written as the classical vector equations of the equilibrium of
elastica:
a) for forces
R − (R + dR) + qds = 0 (14.2)
b) for moments
r × R − (r + dr) × (R + dR) + M − (M + dM) + r × qds + mds = 0 (14.3)
This leads to a system of ordinary differential equations:

dR
=q (14.4)
ds
dM
+ τ × R = m. (14.5)
ds
14.1 Equilibrium Equations for the Rope in Space 415
Here, the equilibrium of forces in eqn. (14.4) and the definition of the unit tan-
dr
gent vector τ = are taken into account in order to formulate eqn. (14.5).
ds
However, in comparison with classical elastica ropes have no moment in-
teraction M acting on the cross sections and are not capable to transfer
distributed moments m. Thus, only the force equation (14.4) is left, see Mad-
docks and Keller [120].
The force vector R can be expanded in the natural Serret-Frenet
coordinate system used in the differential geometry of curves, see e.g.
in Kreyszig (1991), built by the unit tangent τ , the unit normal ν and the
unit bi-normal β vectors
R = T τ + N ν + Bβ, (14.6)
with T – a tangential force in direction of the rope, N – a normal force acting

perpendicular to both the rope and the surface, and B – a bi-normal force,
acting perpendicular to the rope but tangential to the surface, see further
example for a cylindrical surface and a rope formed as the spiral line in
Fig. 14.3. The full derivative of the force vector includes also a part due to
changing the basis vectors τ , ν, β along the curve and can be written as
dR d̂R
= + ΩR (14.7)
ds ds
d̂R
where include derivatives of vector components in eqn. (14.6) with respect
ds
to s and Ω is the rotation matrix of the local basis τ , ν, β and is given by
the Serret-Frenet formula for an arbitrary 3D curve
⎛ ⎞
0 k 0
Ω = ⎝ −k 0 κ ⎠ . (14.8)
0 −κ 0
Here, k is the curvature and κ is the torsion of the center line of the curve
which are in general also functions of s.
The projection of eqn. (14.4) onto the axes τ , ν, β results then in a system
of ordinary differential equations:
⎧
⎪
⎪ dT
= qτ – equilibrium in τ direction,
⎪
⎪ + kN
⎪
⎪ ds
⎪
⎨
dN
− kT + κB = qν – equilibrium in ν direction, (14.9)
⎪
⎪ ds
⎪
⎪
⎪
⎪
⎪
⎩ dB − κN = qβ – equilibrium in β direction.
ds
14.2 Solution of the Equilibrium Equation for a Spiral

Line (Helix)
Now we consider a special case with the following conditions:
1. the rope is positioned on the rigid surface.
2. the rope is loaded only by tangential forces T0 τ 0 and T1 τ 1 at both ends.
3. the rope is beginning to slide preserving its geometrical shape, i.e. only a
motion along τ is possible. Thus e.g. the form of the helix is conserved.
4. sliding is subjected to Coulomb friction such that T (s) = μN (s), where
μ is a coefficient of friction.
Thus, due to condition (3), the first equilibrium equation in (14.9) is trans-
ferred into an equation of motion according to D’Alemberts principle with
∂ 2u
qτ = −ρ , (14.10)
∂t2
where u is a tangential component of the displacement and ρ is the linear
mass density. However, equilibrium along the ν and β axes is fulfilled and
we also consider the absence of distributed forces, thus qν = 0, qβ = 0. The
problem is then described by the system:
⎧
⎪ dN
⎪
⎨ ds − kT + κB = 0
(14.11)
⎪
⎪
⎩ dB − κN =0
ds
with the following initial conditions at s = 0 resulting from condition (2):
T (0) = T0 , N (0) = N0 , B(0) = 0. (14.12)
Now, we consider a case with a constant curvature k and constant torsion κ of

the curve. This is a case of the spiral line – a helix – on a cylinder. The system
of equations (14.11) is transformed into the single differential equation of
second order by taking the derivative of the first equation in (14.11):
d2 N dT dB
2
−k +κ =0 (14.13)
ds ds ds
and then using the second equation in (14.11) we obtain
d2 N dT
−k + κ 2 N = 0. (14.14)
ds2 ds
14.2 Solution of the Equilibrium Equation for a Spiral Line (Helix) 417
Now taking into account the sliding condition (4) we obtain a second order
differential equation:
d2 N dN
2
− μk + κ2 N = 0 (14.15)
ds ds
with the initial conditions:
dN (0)
N (0) = N0 , = kT0 . (14.16)
ds
Solutions of the ordinary differential equation (14.15) are obtained by using
the characteristic equation
λ2 − μkλ + κ 2 = 0 (14.17)

μk ± μ2 k 2 − 4κ 2
λ1, 2 = (14.18)
2
Depending on the discriminant D = μ2 k 2 − 4κ 2 three solutions are possible.
We consider these values for the spiral line (helix). The geometrical charac-
teristics for this line are defined as follows, see Fig. 14.3:
Bβ
D Nν
B Tτ
H
A ϕ
α Y
X
Fig. 14.3 Spiral line (helix) with a pitch H. Geometry and acting rope forces T τ ,
N ν and Bβ.
• Parametrization of the spiral line (helix):

⎧ ⎫
⎪
⎪ R cos ϕ ⎪
⎨ R sin ϕ ⎪ ⎬
r= . (14.19)
⎪
⎪ ⎪
⎩ H ϕ + const ⎪⎭
2π
Within 0 ≤ ϕ ≤ 2π the line is making a full turn around the OZ-axis
raising by a pitch z = H with the following geometrical items.
• Arc-length of the spiral line (helix)

2
H
s= R + 2 ϕ, ϕ ∈ [0, ...) (14.20)
2π
• Curvature of the helix

R
k=
2 (14.21)
2 H
R +
2π
• Torsion of the spiral line (helix)
H/2π
κ=
2 . (14.22)
2
H
R +
2π
Then positivity of the discriminant can be expressed via the geometrical

parameters of the helix as
H
D ≥ 0 =⇒ μk ≥ 2κ =⇒ μ ≥ (14.23)
πR
Remark 14.1. Having noted, see Fig. 14.3, that the tangent of the descent
angle α is defined as
H
tan α = , (14.24)
2πR
we can redefine the discriminant via the critical descent angle αcrit as
H
D = tan αcrit = , (14.25)
πR
H
Case 1. Positive discriminant D > 0, μ > .
πR
Since with H = 0 the discriminant is positive D > 0, then this case is
representing a spiral line with a very small pitch H in comparison with the
radius R.
The solution is given by the exponential functions:
N = C1 eλ1 s + C2 eλ2 s . (14.26)

Satisfying the initial conditions in eqn. (14.16) we obtain the system:

C1 + C2 = N0
(14.27)
C1 λ1 + C2 λ2 = kT0 .
The solution of this system is:

kT0 − N0 λ2 N0 λ1 − kT0
C1 = , C2 = . (14.28)
λ1 − λ2 λ1 − λ2
Then the normal force N can be written as
1 $ %
N= (kT0 − N0 λ2 )eλ1 s + (N0 λ1 − kT0 )eλ2 s . (14.29)
λ1 − λ2
If the rope is sliding at each of its points then
T (s) = μN (s), s ≥ 0, −→ T0 = μN0 . (14.30)
Taking into account this condition and dividing eqn. (14.29) by N0 we obtain
the form:
N T 1 $ %
= = (kμ − λ2 )eλ1 s + (λ1 − kμ)eλ2 s . (14.31)
N0 T0 λ1 − λ2
In order to write down the solution in eqn. (14.31) via the geometrical pa-
rameters of the spiral line given by eqns. (14.20, 14.21, 14.22), we introduce
an additional value ω:

2
H
μ2 R 2 −
μ2 k 2 − 4κ 2 π
ω= = -
2 . (14.32)
2 H
2 R + 2
2π
and will use the hyperbolic sinus and cosine functions:
ews − e−ws ews + e−ws

sinh ωs = , cosh ωs = .
2 2
Then the force ratio in eqn. (14.31) gets the form:

μk
T μk s
= sinh ωs + cosh ωs e 2 , (14.33)
T0 2ω
or is written analogously via R, H and ϕ

⎡ ⎛

2 ⎞
H
⎢ ⎜ 2 2 ⎟
T ⎢ μR ⎜ϕ μ R − π ⎟
⎢
= ⎢ ⎜
2 ⎟
T0
2 sinh ⎜ 2 ·
2
⎟+
⎣ H ⎝ H ⎠
μ R −
2 2 R +
π 2π
μRϕ
⎛
2 ⎞⎤

2
⎜ μ2 R 2 − H ⎟⎥ 2 · R 2 + H
⎜ϕ π ⎟⎥
+ cosh ⎜
2 ⎟ ⎥ 2π
⎜2 · ⎟⎥ e . (14.34)
⎝ 2 H ⎠⎦
R +
2π
If we consider the 2D case with H = 0 in eqn. (14.34) the classical Euler-

Eytelwein for the rope on the cylinder is recovered.
T = μϕ μϕ 0 μϕ
= sinh + cosh e 2 = eμϕ (14.35)
T0 2 2
H
Case 2. Negative discriminant D < 0, μ < .
πR
This case is representing either a case with a spiral line with large pitch H,
or with a small coefficient of friction μ.
The solution is given by the trigonometrical and exponential functions:
μk
N =e 2 s [C1 cos ω̃s + C2 sin ω̃s] (14.36)
with ω̃
2
H
− μ2 R 2
4κ 2 − μ2 k 2 π
ω̃ = = -
2 . . (14.37)
2 H
2
2 R +
2π
Satisfying the initial conditions in eqn. (14.16) we obtain the system:

"
C1 = N0
μk (14.38)
C1 + C2 ω̃ = kT0 .
2
The solution of this system is:
μk
kT0 − N0
C1 = N0 , C2 = 2 . (14.39)
ω̃
Then the normal force N can be written as

⎡ ⎤
μk μk
s⎢ kT − N
0
2
0
⎥
N = e 2 ⎣N0 cos ω̃s + sin ω̃s⎦ . (14.40)
ω̃
Taking into account the sliding conditions in eqn. (14.29) and also in
eqn. (14.30) we can transform eqn. (14.40) after division by N0 into the
following form:
⎡ ⎤
μk μk
N T s kμ − 2
= = e 2 ⎣cos ω̃s + sin ω̃s⎦ (14.41)
N0 T0 ω̃
μk
kμ s
= sin ω̃s + cos ω̃s e 2 .
2ω̃
H
Limit case 3. Discriminant is zero D = 0, μ = .
πR
This case can be obtained either by taking the solution in the form
N = (C1 + C2 s)eλcrit s (14.42)
and satisfying then the corresponding initial conditions, or just by the limit
process with ω → 0 with the solution in eqn. (14.33) or in eqn. (14.41) as
μk
T kμ s
lim = s+1 e 2 (14.43)
ω→0 T0 2
The computation for the cases with D > 0, D = 0, D < 0 and the classical
Euler formula with H = 0 is presented in Fig. 14.4 for the angle ϕ ∈ [0, 2π].
The computation is given with a coefficient of friction μ = 0.3 and a radius
R = 1.0 for the following cases:
• Classical Euler case H = 0;
• Positive discriminant D > 0 with H = 0.75
• Zero discriminant D = 0 with H = μπR = 0.9424
• Negative discriminant D < 0 with H = 1.2
One can see, that with increasing pitch H the ratio of the forces T /T0 is
decreasing, thus, the Euler case is representing the upper limit of this ratio.
Remark 14.2. The assumption of constant curvature k and torsion κ leads
to a differential equation of second order with constant parameters. This
allows to define the analytical solution using trigonometric and exponential
functions in a fairly straightforward fashion. As is known, solutions of the
differential equation of second order appearing e.g. for other surfaces with
8
H=0
7 D>0
D=0
6 D<0
5
T/To
4
3
2
1
0
0 1 2 3 4 5 6 7 8 9
ϕ
Fig. 14.4 Comparison of all cases with discriminant D > 0, D = 0, D < 0 and the
classical Euler formula with H = 0.
non-constant parameters can be obtained via special functions such as Bessel

functions, Legendre polynomials or others. It would be thus a further step
to identify the geometry of the corresponding line sliding on corresponding
surfaces.
14.3 Conclusions
The solution of the classical Euler-Eytelwein problem formulated as 2D prob-
lem is generalized in the current contribution into the 3D case under the
restriction that the rope is represented by the spiral line and the sliding is
appearing only in tangential direction according to the Coulomb friction law
T = μN . Any motion along the cylindrical axis, thus, changing the shape of
the spiral line is restricted. Another restriction is given by the assumption
of an anisotropic
√ friction model represented by the T = μN , but not by the
T = μ N 2 + B2.
The generalized solution also shows that considering the pitch H leads to
a reduction of the force ratio T /T0 in comparison to the 2D Euler-Eytelwein
formula.
References
1. Abaqus. theory manual (2007)

2. Al-Dojayli, M., Meguid, S.: Accurate modeling of contact using cubic splines.
Finite Elements in Analysis and Design 38, 337–352 (2002)
3. Alart, P., Curnier, A.: A mixed formulation for frictional contact problems
prone to Newton like solution methods. Computer Methods in Applied Me-
chanics and Engineering 92, 353–375 (1991)
4. Alart, P., Heege, A.: Consistent tangent matrices of curved contact operators
involving anisotropic friction. Revue Europeenne des Elements Finis 4, 183–
207 (1995)
5. Araki, Y., Hjelmstad, K.D.: Rate-dependent projection operators for frictional
contact constraints. International Journal for Numerical Methods in Engineer-
ing 57, 923–954 (2003)
6. Armero, F., Petoecz, E.: Formulation and analysis of conserving algorithms for
frictionless dynamic contact-impact problems. Computer Methods in Applied
Mechanics and Engineering 158, 269–300 (1998)
7. Ashley, C.W.: The Ashley book of knots, 12th edn. Farber and Farber Limited,
London (1993); Reprinted with amendments by Geoffrey Budworth
8. Bandeira, A.A., Pimenta, P.M., Wriggers, P.: A study of contact surface mi-
crostructure using a homogenization procedure considering elastoplastic be-
haviour of the asperities. In: 4th Contact Mechanics International Symposium,
CMIS 2005 (2005)
9. Bandeira, A.A., Wriggers, P., Pimenta, P.M.: Numerical derivation of contact
mechanics interface laws using a finite element approach for large 3d deforma-
tion. International Journal for Numerical Methods in Engineering 59, 173–195
(2004)
10. Belytschko, T., Neal, O.: Contact-impact by the pinball algorithm with
penalty and Lagrangian methods. International Journal for Numerical Meth-
ods in Engineering 31, 547–572 (1991)
11. Benson, D.J., Hallquist, J.Q.: A single surface contact algorithm for the post-
buckling analysis of shell structures. Computer Methods in Applied Mechanics
and Engineering 78, 141–163 (1990)
12. Bertsekas, D.: Constrained optimization and Lagrange multiplier methods.
Acad. Pr., New York (1982)
13. Bertsekas, D.: Convex analysis and optimization. Athena Scientific (2003)
424 References
14. Betsch, P.: The discrete null space method for the energy consistent integration
of constrained mechanical systems. part I: Holonomic constraints. Computer
Methods in Applied Mechanics and Engineering 194, 5159–5190 (2005)
15. Betsch, P., Hesch, C.: Energy-momentum conserving schemes for frictionless
dynamic contact problems. part I: NTS method. In: Wriggers, P., Nackenhorst,
U. (eds.) IUTAM Symposium on Computational Contact Mechanics. IUTAM
Bookseries, pp. 77–96. Springer (2007)
16. Bonet, J., Wood, R.: Nonlinear continuum mechanics for finite element anal-
ysis. Cambridge Univ. Press (2000)
17. Borodich, F.M., Galanov, B.A.: Non-direct estimations of adhesive and elastic
properties of materials by depth-sensing indentation. Proc. of the Royal Soci-
ety of London. Series A, Mathematical and Physical Sciences 464, 2759–2776
(2008)
18. Borodich, F.M., Keer, L.M.: Evaluation of elastic modulus of materials by
adhesive (no–slip) nano–indentatio. Proc. of the Royal Society of London.
Series A, Mathematical and Physical Sciences 460, 507–514 (2004)
19. Borwein, J.M., Lewis, A.S.: Convex analysis and nonlinear optimization. The-
ory and examples. Springer, Heidelberg (2000)
20. Bowden, F., Rowe, G.: The adhesion of clean metals. Proc. of the Royal Society
of London. Series A, Mathematical and Physical Sciences 233, 429–442 (1956)
21. Bruneel, H.C.J., Rycke, I.D.: Quicktrace: a fast algorithm to detect contact. In-
ternational Journal for Numerical Methods in Engineering 54, 299–316 (2002)
22. Buczkowski, R., Kleiber, M.: Elasto-plastic interface model for 3d-frictional
orthotropic contact problems. International Journal for Numerical Methods
in Engineering 40, 599–619 (1997)
23. Buczkowski, R., Kleiber, M.: A stochastic model of rough surfaces for finite
element contact analysis. Computer Methods in Applied Mechanics and En-
gineering 199, 43–59 (1999)
24. Buczkowski, R., Kleiber, M.: Statistical model of strongly anisotropic rough
surfaces for finite element contact analysis. International Journal for Numer-
ical Methods in Engineering 49, 1169–1189 (2000)
25. Campbell, J., Libersky, L.D., Vignjevic, R.: A contact algorithm for smoothed
particle hydrodynamics. Computer Methods in Applied Mechanics and Engi-
neering 184, 49–65 (2000)
26. Carbone, G., Mangialardi, L.: Adhesion and friction of an elastic half-space
in contact with a slightly wavy rigid surface. Journal of the Mechanics and
Physics of Solids 52, 1267–1287 (2004)
27. Cescotto, S., Charlier, R.: Frictional contact finite elements based on mixed
variational principles. International Journal for Numerical Methods in Engi-
neering 36, 1681–1701 (1993)
28. Chawla, V., Laursen, T.A.: Energy consistent algorithms for frictional contact
problems. International Journal for Numerical Methods in Engineering 42,
799–827 (1998)
29. Cocou, M., Raous, M., Schryve, M.: Analysis of a dynamic contact problem
with adhesion and friction in viscoelasticity. In: ECCM-III, Conference Pro-
ceedings, June 5-8 (2006)
30. Crisfield, M.A.: Re-visiting the contact patch test. International Journal for
Numerical Methods in Engineering 48, 435–449 (2000)
31. Curnier, A.: A theory of friction. International Journal of Solids and Struc-
tures 20, 637–647 (1984)
References 425
32. Davis, P., Rabinowitz, P.: Methods of numerical integration, 3rd edn. Aca-
demic Press (1984)
33. Dubrovin, B., Fomenko, T., Novikov, S.: Modern geometry – methods and
applications. Part 1. Springer (1984)
34. Duester, A., Broeker, H., Rank, E.: The p-version of the finite element method
for three-dimensional curved thin walled structrures. International Journal for
35. Eck, C., Jarusek, J.: Existence result for the static contact problem with
Coulomb friction. Mathematical Models and Methods in Applied Science 8,
445–468 (1998)
36. El-Abbasi, N., Bathe, K.: Stability and patch test performance of contact
discretization and new solution algorithm. Computer and Structures 79, 1473–
1486 (2001)
37. El-Abbasi, N., Meguid, S., Czekanski, A.: On the modeling of smooth contact
surfaces using cubic splines. International Journal for Numerical Methods in
Engineering 50, 953–967 (2001)
38. Farin, G.: NURBS: from projective geometry to practical use. Peters (1999)
39. Feng, Y.T., Owen, D.R.J.: An augmented spatial digital tree algorithm for
contact detection in computational mechanics. International Journal for Nu-
merical Methods in Engineering 55, 159–176 (2002)
40. Fischer, K.A., Wriggers, P.: Mortar based frictional contact formulation for
higher order interpolations using the moving friction cone. Computer Methods
in Applied Mechanics and Engineering 195, 5020–5036 (2006)
41. Fortin, M., Glowinski, R.: Augmented Lagrangian methods: applications to
the numerical solutions of boundary-value problems. North-Holland, Amster-
dam (1983)
42. Francavilla, A., Zienkiewicz, O.C.: A note on numerical computation of elastic
contact problems. International Journal for Numerical Methods in Engineer-
ing 9, 913–924 (1975)
43. Fuller, K., Tabor, D.: The effect of surface roughness on the adhesion of elastic
solids. Proc. of the Royal Society of London. Series A, Mathematical and
Physical Sciences 345, 327–342 (1975)
44. Giannakopoulos, A.E.: The return mapping method for the integration of
friction constitutive relations. Computers and Structures 32, 157–167 (1989)
45. Glowinski, R., Lions, J.-L., Tremolieres, R.: Numerical analysis of variational
inequalities. North-Holland, Amsterdam (1981) rev. edition ed.
46. Gray, A.: Modern Differential Geometry of Curves and Surfaces. CRC Press,
Boca Raton (1993)
47. Greenwood, J.A.: A unified theory of surface roughness. Proc. of the Royal
Society of London. Series A, Mathematical and Physical Sciences 393, 133–157
(1984)
48. Greenwood, J.A.: Adhesion of elastic spheres. Proc. of the Royal Society of
London. Series A, Mathematical and Physical Sciences 453, 1277–1297 (1997)
49. Greenwood, J.A., Williamson, J.B.P.: Contact of nominally flat surfaces. Proc.
of the Royal Society of London. Series A, Mathematical and Physical Sci-
ences 295, 300–319 (1966)
50. Gurtin, M.E., Weissmueller, J., Larche, F.: A general theory of curved de-
formable interfaces in solids at equilibrium. Philosophical Magazine A 78,
1093–1109 (1998)
426 References
51. Hallquist, J.Q.: LS-DYNA. Theoretical manual (2007),

http://www.lstc.com/
52. Hallquist, J.Q., Goudreau, G.L., Benson, D.J.: Sliding interface with contact-
impact in large-scale Lagrangian computations. Computer Methods in Applied
Mechanics and Engineering 51, 107–137 (1985)
53. Han, W., Sofonea, M.: Quasistatic Contact Problems in Viscoelasticity and
Viscoplasticity, American Mathematical Society. International Press (2002)
54. Hardy, W., Nottage, M.: Studies in adhesion. I. Proc. of the Royal Society of
55. Hardy, W., Nottage, M.: Studies in adhesion. II. Proc. of the Royal Society of
56. Harnau, M., Konyukhov, A., Schweizerhof, K.: Algorithmic aspects in large de-
formation contact analysis using ’solid-shell’ elements. Computers and Struc-
tures 83, 1804–1823 (2005)
57. Hartmann, S., Oliver, J., Weyler, R., Cante, J., Hernández, J.: A contact do-
main method for large deformation frictional contact problems. part 2: Numer-
ical aspects. Computer Methods in Applied Mechanics and Engineering 198,
2607–2631 (2009)
58. Hauptmann, R., Doll, S., Harnau, M., Schweizerhof, K.: “Solid-shell” elements
with linear and quadratic shape functions at large deformations with nearly
incompressible materials. Computers and Structures 79, 1671–1685 (2001)
59. Hauptmann, R., Schweizerhof, K.: A systematic development of “solid-shell”
element formulations for linear and non-linear analysis employing only dis-
placement degrees of freedom. International Journal for Numerical Methods
in Engineering 42, 49–69 (1998)
60. Hauptmann, R., Schweizerhof, K., Doll, S.: Extension of the solid-shell concept
for large elastic and large elastoplastic deformations. International Journal for
61. He, Q., Curnier, A.: Anisotropic dry friction between two orthotropic surfaces
undergoing large displacements. European Journal of Mechanics A-Solids 12,
631–666 (1993)
62. Heegaard, J.H., Curnier, A.: An Augmented Lagrangian method for discrete
large-slip contact problems. International Journal for Numerical Methods in
Engineering 36, 569–593 (1993)
63. Heegaard, J.H., Curnier, A.: Geometric properties of 2d and 3d unilateral
large slip contact operators. Computer Methods in Applied Mechanics and
Engineering 131, 263–286 (1996)
64. Heege, A., Alart, P.: A frictional contact element for strongly curved contact
problems. International Journal for Numerical Methods in Engineering 39,
165–184 (1996)
65. Heinstein, M.W., Laursen, T.A.: A three dimensional surface-to-surface pro-
jection algorithm for non-coincident domains. Communications in Numerical
Methods in Engineering 19, 421–432 (2003)
66. Heintz, P., Hansbo, P.: Stabilized Lagrange multiplier methods for bilateral
elastic contact with friction. Computer Methods in Applied Mechanics and
Engineering 195, 4323–4333 (2006)
67. Hjiaj, M., Feng, Z., de Saxce, G., Mroz, Z.: Three-dimensional finite element
computations for frictional contact problems with non-associated sliding rule.
International Journal for Numerical Methods in Engineering 60, 2045–2076
(2004)
References 427
68. Hueeber, S., Stadler, G., Wohlmuth, B.I.: A primal-dual active set algorithm
for three-dimensional contact problems with coulomb friction. Preprint, Inst
Angewandte Anal. und Num. Simulation, Stuttgart (2006)
69. Hueeber, S., Wohlmuth, B.I.: Large deformation frictional contact mechan-
ics: continuum formulation and augmented lagrangian treatment. Computer
70. Hughes, T., Cottrell, J., Bazilevs, Y.: Isogeometric analysis: CAD, finite el-
ements, nurbs, exact geometry and mesh refinement. Computer Methods in
Applied Mechanics and Engineering 194, 4135–4195 (2005)
71. Hughes, T.J.R., Taylor, R.L., Sackman, J.L., Curnier, A., Kanoknukulchai,
W.: A finite element method for a class of contact-impact problems. Computer
72. Ibrahimbegovic, A.: On finite element implementation of geometrically non-
linear reissner’s beam theory: three-dimensional curved beam elements. Com-
puter Methods in Applied Mechanics and Engineering 122, 11–26 (1995)
73. Ibrahimbegovic, A.: On the choice of finite rotation parameters. Computer
74. Ibrahimbegovic, A., Taylor, R.L.: On the role of frame-invariance in structural
mechanics models at finite rotations. Computer Methods in Applied Mechan-
ics and Engineering 191, 5159–5176 (2002)
75. Izi, R., Konyukhov, A., Schweizerhof, K.: Large penetration algorithm for 3d
frictionless contact problems based on a covariant form. Computer Methods
in Applied Mechanics and Engineering 217-220, 186–196 (2012)
76. Johnson, K.L.: Adhesion and friction between a smooth elastic spherical as-
perity and a plane surface. Proc. of the Royal Society of London. Series A,
Mathematical and Physical Sciences 453, 163–179 (1997)
77. Johnson, K.L.: Contact mechanics. Cambridge Univ. Press, Cambridge (2001)
78. Johnson, K.L., Kendall, K., Roberts, A.D.: Surface energy and the contact of
elastic solids. Proc. of the Royal Society of London. Series A, Mathematical
and Physical Sciences 324, 301–313 (1971)
79. Jones, R.E., Papadopoulos, P.: A yield-limited Lagrange multiplier formula-
tion for frictional contact. International Journal for Numerical Methods in
Engineering 48, 1127–1149 (2000)
80. Jones, R.E., Papadopoulos, P.: A novel three-dimensional contact finite ele-
ment based on smooth pressure interpolations. International Journal for Nu-
81. Jones, R.E., Papadopoulos, P.: A geometric interpretation of frictional contact
mechanics. Zeitschrift für Angewandte Mathematik und Physik (ZAMP) 57,
1025–1041 (2006)
82. Jones, R.E., Papadopoulos, P.: Simulating anisotropic frictional response using
smoothly interpolated traction fields. Computer Methods in Applied Mechan-
ics and Engineering 195, 588–613 (2006)
83. Kane, C., Reppetto, E.A., Ortiz, M., Marsden, J.E.: Finite element analysis
of nonsmooth contact. Formulation and application to two-dimensional prob-
lems. Computer Methods in Applied Mechanics and Engineering 180, 263–297
(1999)
84. Kikuchi, N., Oden, J.T.: Contact problems in elasticity: a study of variational
inequalities and finite element methods. SIAM (1988)
85. Kim, H., Kim, W., Falk, M., Rigney, D.: MD simulations of microstructure
evolution during high-velocity sliding between crystalline materials. Tribology
Letters 28, 299–306 (2007)
428 References
86. Konyukhov, A.: Geometrically exact theory for contact interactions. KIT Sci-
entific Publishing, Karlsruhe (2011)
87. Konyukhov, A., Schweizerhof, K.: Contact formulation via a velocity descrip-
tion allowing efficiency improvements in frictionless contact analysis. Compu-
tational Mechanics 33, 165–173 (2004)
88. Konyukhov, A., Schweizerhof, K.: Covariant description for frictional contact
problems. Computational Mechanics 35, 190–213 (2005)
89. Konyukhov, A., Schweizerhof, K.: Covariant description of contact interfaces
considering anisotropy for adhesion and friction: Part 1. formulation and anal-
ysis of the computational model. Computer Methods in Applied Mechanics
90. Konyukhov, A., Schweizerhof, K.: Covariant description of contact interfaces
considering anisotropy for adhesion and friction: Part 2. Linearization, finite
element implementation and numerical analysis of the model. Computer Meth-
ods in Applied Mechanics and Engineering 196, 289–303 (2006)
91. Konyukhov, A., Schweizerhof, K.: A special focus on 2d formulations for con-
tact problems using a covariant description. International Journal for Numer-
92. Konyukhov, A., Schweizerhof, K.: On a continuous transfer of history variables
for frictional contact problems based on interpretations of covariant derivatives
as a parallel translation. In: IUTAM Symposium -“Multiscale problems in
Multibody System Contacts”, IUTAM Bookseries (2007)
93. Konyukhov, A., Schweizerhof, K.: Symmetrization of various friction models
based on an Augmented Lagrangian approach. IUTAM Bookseries, pp. 97–
111. Springer (2007)
94. Konyukhov, A., Schweizerhof, K.: On the solvability of closest point projection
procedures in contact analysis: Analysis and solution strategy for surfaces of
arbitrary geometry. Computer Methods in Applied Mechanics and Engineer-
ing 197, 3045–3056 (2008)
95. Konyukhov, A., Schweizerhof, K.: Geometrically exact covariant approach for
contact between curves representing beam and cable type structures, pp. 00–
56 (2009)
96. Konyukhov, A., Schweizerhof, K.: Incorporation of contact for high-order finite
elements in covariant form. Computer Methods in Applied Mechanics and
Engineering 198, 1213–1223 (2009); HOFEM 2007- International Workshop
on High-Order Finite Element Methods (2007)
97. Konyukhov, A., Schweizerhof, K., Vielsack, P.: On models of contact surfaces
including anisotropy for friction and adhesion and their experimental vali-
dations. In: ECCM-III, Conference Proceedings, III European Conference on
Computational Mechanics. Solid, Structures and Coupled Problems in Engi-
neering, June 5-9 (2006)
98. Konyukhov, A., Vielsack, P., Schweizerhof, K.: On coupled models of
anisotropic contact surfaces and their experimental validation. Wear 264, 579–
588 (2008)
99. Kreyszig, E.: Differential geometry, Mathematical expositions. Dover Publi-
cations, New York (1991)
100. Krstulovic-Opara, L., Wriggers, P.: Convergence studies for 2d smooth con-
tact elements. In: ECCM, European Conference on Computational Mechanics
(2001)
References 429
101. Krstulovic-Opara, L., Wriggers, P.: A two-dimensional C1-continuous contact

element based on the moving friction cone description. In: WCCM V. Fifth
World Congress on Computational Mechanics (2002)
102. Krstulovic-Opara, L., Wriggers, P., Korelc, J.: A C1-continuous formulation
for 3d finite deformation frictional contact. Computational Mechanics 29, 27–
42 (2002)
103. Lang, S.: Fundamentals of differential geometry. Springer, New York (1999)
104. Laursen, T.A.: Formulation and Treatment of Frictional Contact Problems
using Finite Elements, PhD thesis, Stanford University (1992)
105. Laursen, T.A.: Convected description in large deformation frictional contact
problems. International Journal of Solids and Structures 31, 669–681 (1994)
106. Laursen, T.A.: Computational Contact and Impact Mechanics. Fundamen-
tals of Modeling Interfacial Phenomena in Nonlinear Finite Element Analysis.
Springer, Heidelberg (2002)
107. Laursen, T.A., Chawla, V.: Design of energy conserving algorithms for friction-
less dynamic contact problems. International Journal for Numerical Methods
in Engineering 40, 863–886 (1997)
108. Laursen, T.A., Heinstein, M.W.: Consistent mesh tying methods for topolog-
ically distinct discretized surfaces in non-linear solid mechanics. International
Journal for Numerical Methods in Engineering 57, 1197–1242 (2003)
109. Laursen, T.A., Simo, J.C.: A continuum-based finite element formulation for
the implicit solution of multibody large deformation frictional contact prob-
lems. International Journal for Numerical Methods in Engineering 35, 3451–
3485 (1993)
110. Libersky, L.D., Petschek, A.G.: Smooth particle hydrodynamics with strength
of materials. Advances in the Free-Lagrange Method, pp. 248–257. Springer,
New York (1991)
111. Litewka, P.: Hermite polynomial smoothing in beam-to-beam frictional con-
tact. Computational Mechanics 40, 815–826 (2007)
112. Litewka, P.: Smooth frictional contact between beams in 3d. In: Nackenhorst,
U., Wriggers, P. (eds.) IUTAM Symposium on Computational Contact Me-
chanics. IUTAM Bookseries, pp. 157–176. Springer (2007)
113. Litewka, P., Wriggers, P.: Frictional beam-to-beam contact by the Lagrange
multipliers method. In: VII International Conference on Computational Plas-
ticity, COMPLAS (2003)
114. Litewka, P., Wriggers, P.: Contact between 3d beams with rectangular cross-
sections. International Journal for Numerical Methods in Engineering 53,
2019–2042 (2002)
115. Litewka, P., Wriggers, P.: Frictional contact between 3d beams. Computa-
tional Mechanics 28, 26–39 (2002)
116. Longuet-Higgins, M.S.: The statistical analysis of a random, moving surface.
Philosophical Transaction of the Royal Society of London, Series A, Mathe-
matical and Physical Sciences 249, 321–387 (1957)
117. Longuet-Higgins, M.S.: Statistical properties of an isotropic random surface.
Philosophical Transaction of the Royal Society of London, Mathematical and
118. Luding, S.: Contact models for very loose granular materials. In: Eberhard,
P. (ed.) IUTAM Bookseries, IUTAM Symposium on Multiscale Problems in
Multibody System Contact (2007)
430 References
119. Luenberger, D.: Linear and Nonlinear programming, 2nd edn. Addison-Wesley
(1984)
120. Maddocks, J., Keller, J.: Ropes in equilibrium. SIAM Journal of Applied
Mathematics 47, 1185–1200 (1987)
121. Maker, B.N., Laursen, T.A.: A finite element formulation for rod-continuum
interactions: the one-dimensional slideline. International Journal for Numeri-
cal Methods in Engineering 37, 1–18 (1994)
122. Marsden, J.E., Hughes, T.J.R.: Mathematical Foundations of Elasticity.
Dover, New York (1994)
123. McCool, John, I.: Comparison of models for the contact of rough surfaces.
Wear 107, 37–60 (1986)
124. McDevitt, T.W., Laursen, T.A.: A mortar-finite element formulation for fric-
tional contact problems. International Journal for Numerical Methods in En-
gineering 48, 1525–1547 (2000)
125. McFarlane, J., Tabor, D.: Adhesion of solids and the effect of surface films.
Proc. of the Royal Society of London. Series A, Mathematical and Physical
Sciences 202, 224–243 (1950)
126. McFarlane, J., Tabor, D.: Relation between friction and adhesion. Proc. of the
Royal Society of London. Series A, Mathematical and Physical Sciences 202,
244–253 (1950)
127. Merkle, A., Marks, L.: A predictive analytical friction model from basic the-
ories of interfaces, contacts and dislocations. Tribology Letters 26(1), 73–84
(2007)
128. Michalowski, R., Mroz, Z.: Associated and non-associated sliding rules in
contact friction problems. Archives of Mechanics (Archiwum mechaniki
stosowanej), Polish Academy of Sciences 30, 259–276 (1978)
129. Montmitonnet, P., Hasquin, A.: Implementation of an anisotropic friction law
in a 3d finite element model of hot rolling. In: Shen, S.-F., Dawson, P. (eds.)
Simulation of Materials Processing: Theory, Methods and Applications, Proc.
of NUMIFORM 1995, pp. 301–306 (1995)
130. Mroz, Z., Stupkiewicz, S.: An anisotropic friction and wear model. Interna-
tional Journal of Solids and Structures 31, 1113–1131 (1994)
131. Munjiza, A.: The Combined Finite-Discrete Element Method. Wiley (2004)
132. Oancea, V., Laursen, T.A.: On the constitutive modeling and finite element
computation of rate-dependent frictional sliding in large deformations. Com-
puter Methods in Applied Mechanics and Engineering 143, 197–227 (1997)
133. Oliver, J., Hartmann, S., Cante, J., Weyler, R., Hernández, J.: A contact do-
main method for large deformation frictional contact problems. part 1: Theo-
retical basis. Computer Methods in Applied Mechanics and Engineering 198,
2591–2606 (2009)
134. Paczelt, I., Szabo, B., Szabo, S.T.: Solution of contact problem using the
hp-version of the finite element methods. Computers and Mathematics with
Applications 38, 49–69 (1999)
135. Padmanabhan, V., Laursen, T.A.: A framework for development of surface
smoothing procedures in large deformation frictional contact analysis. Finite
Elements in Analysis and Design 37, 173–198 (2001)
136. Paggi, M., Carpinteri, A., Zavarise, G.: A unified interface constitutive law
for the study of fracture and contact problems in heterogeneous materials.
In: Wriggers, P., Nackenhorst, U. (eds.) Analysis and Simulation of Contact
Problems. LNACM, pp. 297–304. Springer (2005)
References 431
137. Pandolfi, A., Kane, C., Marsden, J.E., Ortiz, M.: Time-discreized variational
formulation of non-smooth frictional contact. International Journal for Nu-
138. Parisch, H.: A consistent tangent stiffness matrix for three-dimensional non-
linear contact analysis. International Journal for Numerical Methods in Engi-
neering 28, 1803–1812 (1989)
139. Parisch, H., Luebbing, C.: A formulation of arbitrarily shaped surface elements
for three-dimensional large deformation contact with friction. International
140. Peric, D., Owen, D.R.J.: Computational model for 3-d contact problems with
friction based on the penalty method. International Journal for Numerical
Methods in Engineering 35, 1289–1309 (1992)
141. Persson, B.N.J.: Sliding Friction. Physical Principles and Applications, 2nd
edn (2000)
142. Persson, B.N.J., Tosatti, E.: The effect of surface roughness on the adhesion
of elastic solids. Journal of Chemical Physics 115, 5597–5610 (2001)
143. Persson, P.: Parallel Numerical Procedures for the Solution of Contact-Impact
Problems, PhD thesis, Linkeoping University (2000)
144. Piegl, L., Tiller, W.: The NURBS Book (Monographs in Visual Communica-
tions), 2nd edn. Springer (1997)
145. Pietrzak, G., Curnier, A.: Large deformation frictional contact mechanics: con-
tinuum formulation and Augmented Lagrangian treatment. Computer Meth-
146. Popov, V., Gerve, A., Kehrwald, B., Smolin, I.: Simulation of wear in com-
bustion engines. Computational Materials Science 19, 285–291 (2000)
147. Popov, V., Psakhie, S.: Numerical simulation methods in tribology. Tribology
International 40(6), 916–923 (2007)
148. Puso, M.A.: A 3d mortar method for solid mechanics. International Journal
for Numerical Methods in Engineering 59, 315–336 (2004)
149. Puso, M.A., Laursen, T.A.: A 3d contact smoothing method using Gregory
patches. International Journal for Numerical Methods in Engineering 54,
1161–1194 (2002)
150. Puso, M.A., Laursen, T.A.: Mesh tying on curved interface in 3d. Engineering
Computations 20, 305–319 (2003)
151. Puso, M.A., Laursen, T.A.: A mortar segment-to-segment contact method for
large deformation solid mechanics. Computer Methods in Applied Mechanics
152. Puso, M.A., Laursen, T.A.: A mortar segment-to-segment frictional contact
method for large deformations. Computer Methods in Applied Mechanics and
Engineering 193, 4891–4913 (2004)
153. Rank, E., Düster, A., Nübel, V., Preusch, K., Bruhns, O.: High order finite
elements for shells. Computer Methods in Applied Mechanics and Engineer-
ing 194, 2494–2512 (2005)
154. Raous, M., Cangemi, L., Cocou, A.: A consistent model coupling adhesion,
friction, and unilateral contact. Computer Methods in Applied Mechanics and
Engineering 177, 383–399 (1999)
155. Schoen, R., Yau, S.: Lectures on differential geometry. International Press,
Boston (1994)
156. Schweizerhof, K., Hallquist, J.Q.: Efficiency refinements of contact strategies
and algorithms in explicit finite element programming. In: Owen, D.R.J.,
Onate, E. (eds.) Computational Plasticity, pp. 457–482. Pineridge Press (1992)
432 References
157. Sextro, W.: Dynamical contact problems with friction: models, methods, ex-
periments, and applications. Springer, Heidelberg (2002)
158. Shillor, M., Sofonea, M., Telega, J.J.: Models and Analysis of Quasistatic
Contact: variational methods. Springer (2004)
159. Shimizu, T., Sano, T.: An application of a penalty method contact and fric-
tion algorithm to a 3-dimensional tool surface expressed by a b-spline patch.
Journal of Material Processing Technology 48, 207–213 (1995)
160. Simo, J.: The (symmetric) Hessian for geometrically nonlinear models in solid
mechanics: Intrinsic definition and geometric interpretation. Computer Meth-
161. Simo, J.C., Hughes, T.J.R.: Elastoplasticity and Viscoplasticity: computa-
tional aspects. Springer, Berlin (1992)
162. Simo, J.C., Laursen, T.A.: An Augmented Lagrangian treatment of contact
problems involving friction. Computers and Structures 42, 97–116 (1992)
163. Simo, J.C., Wriggers, P., Schweizerhof, K., Taylor, R.L.: Finite deformation
post-buckling analysis involving inelasticity and contact constraints. Interna-
tional Journal for Numerical Methods in Engineering 23, 779–800 (1986)
164. Solberg, J.M., Papadopoulos, P.: An analysis of dual formulations for the
finite element solution of two-body contact problems. Computer Methods in
Applied Mechanics and Engineering 194, 2734–2780 (2005)
165. Solin, P., Segeth, K., Dolezel, I.: Higher-order finite element methods. Chap-
man and Hall (2004)
166. Souza de Cursi, J.E.: Stress unilateral analysis of mooring cables. International
167. Stadler, G.: Infinite-dimensional semi-smooth Newton and Augmented La-
grangian methods for friction and contact problems in elasticity, PhD Thesis,
University of Graz, Graz (2004)
168. Stadler, G.: Path-following and Augmented Lagrangian methods for contact
problems in linear elasticity. Journal on Computational and Applied Mathe-
matics 2, 533–547 (2007)
169. Stadler, M., Holzapfel, G.A.: Subdivision schemes for smooth contact surfaces
of arbitrary mesh topology in 3d. International Journal for Numerical Methods
in Engineering 60, 1161–1195 (2004)
170. Stadler, M., Holzapfel, G.A., Korelc, J.: Cn continuous modeling of smooth
contact surfaces using nurbs and application to 2d problems. International
171. Strack, O.D.L., Cundall, P.A.: A discrete numerical model for granular assem-
blies. Geotechnique 29, 47–65 (1979)
172. Tan, D.: Mesh matching and contact patch test. Computational Mechanics 31,
135–152 (2003)
173. Taylor, R.L.: FEAP. A Finite Element Analysis Program, FEAP - Mechanik
- Karlsruhe, Institut fuer Mechanik, Universitaet Karlsruhe (1987)
174. Taylor, R.L., Papadopoulos, P.: On a patch test for contact problems in two
dimensions. In: Wriggers, P., Wagner, W. (eds.) Computational Methods in
Nonlinear Mechanics, pp. 690–702. Springer (1991)
175. Temizer, I., Wriggers, P., Hughes, T.: Contact treatment in isogeometric analy-
sis with nurbs. Computer Methods in Applied Mechanics and Engineering 200,
1100–1112 (2011)
References 433
176. Temizer, I., Wriggers, P., Hughes, T.: Three-dimensional mortar-based fric-
tional contact treatment in isogeometric analysis with nurbs. Computer Meth-
ods in Applied Mechanics and Engineering 209-212, 115–128 (2012)
177. Tomas, J.: Micromechanics of particle adhesion. In: Eberhard, P. (ed.) IUTAM
Symposium on Multiscale Problems in Multibody System Contacts. IUTAM
Bookseries. Springer (2007)
178. Wang, F., Cheng, J., Zhenhan, Y.: FFS contact searching algorithm for dy-
namic finite element analysis. International Journal for Numerical Methods in
Engineering 52, 655–672 (2001)
179. Wang, S.P., Makinouchi, A.: Contact search strategies for FEM simulation
of the blow molding process. International Journal for Numerical Methods in
Engineering 48, 501–521 (2000)
180. Wang, S.P., Nakamachi, E.: The inside-outside contact search algorithm for
finite element analysis. International Journal for Numerical Methods in Engi-
neering 40, 3665–3685 (1997)
181. Wellmann, C., Lillie, C., Wriggers, P.: A contact detection algorithm for su-
perellipsoids based on the common-normal concept. Engineering Computa-
tions 25, 432–442 (2008)
182. Whitehouse, D.J., Archard, J.F.: The properties of random surfaces of sig-
nificance in their contact. Proc. of the Royal Society of London. Series A,
Mathematical and Physical Sciences 316, 97–121 (1970)
183. Whitehouse, D.J., Phillips, M.J.: Discrete properties of random surfaces.
Philosophical Transaction of the Royal Society of London, Mathematical and
184. Whitehouse, D.J., Phillips, M.J.: Two-dimensional discrete properties of ran-
dom surfaces. Proc. of the Royal Society of London. Series A, Mathematical
and Physical Sciences 305, 441–468 (1982)
185. Willner, K.: Elasto-plastic contact of rough surfaces. In: Proc. of International
Conference, Contact Mechanics III (1997)
186. Willner, K., Gaul, L.: A penalty approach for contact description by fem based
on interface physics. In: Proc. of International Conference, Contact Mechanics
II (1995)
187. Willner, K., Gaul, L.: Contact description by FEM based on interface physics.
In: Proc. of International Conference (1997)
188. Willner, K., Goerke, D.: Contact of rough surfaces. a comparison of experi-
mental and numerical results. In: ECCM III. Solid, Structures and Coupled
Problems, Proc. of International Conference (2006)
189. Wriggers, P.: Finite element algorithms for contact problems. Archives of Com-
putational Methods in Engineering 24, 1–49 (1995)
190. Wriggers, P.: Computational Contact Mechanics. John Wiley and Sons (2002)
191. Wriggers, P.: Computational Contact Mechanics. Springer (2006)
192. Wriggers, P., Krstulovic-Opara, L., Korelc, J.: Smooth c1-interpolations for
two-dimensional frictional contact problem. International Journal for Numer-
193. Wriggers, P., Scherf, O.: Different a posteriori error estimators and indicators
for contact problems. Mathematical and Computer Modelling 28, 437–447
(1998)
434 References
194. Wriggers, P., Simo, J.C.: A note on tangent stiffness for fully nonlinear contact
problems. Communications in Applied Numerical Methods 1, 199–203 (1985)
195. Wriggers, P., Van, V., Stein, E.: Finite element formulation of large defor-
mation impact-contact problem with friction. Computers and Structures 37,
319–331 (1990)
196. Wriggers, P., Zavarise, G.: On the application of Augmented Lagrangian tech-
niques for nonlinear constitutive laws in contact interface. Communications in
Applied Numerical Methods 9, 815–824 (1993)
197. Wriggers, P., Zavarise, G.: Thermomechanical contact – a rigorous but simple
numerical approach. Computer and Structures 46, 47–53 (1993)
198. Wriggers, P., Zavarise, G.: On contact between three-dimensional beams un-
dergoing large deflection. Communications in Numerical Methods in Engineer-
ing 13, 429–438 (1997)
199. Wriggers, P., Zavarise, G.: A formulation for frictionless contact problems
using a weak form introduced by Nitsche. Computational Mechanics 41, 407–
420 (2008)
200. Zavarise, G., Lorenzis, L.D., Taylor, R.: On regularization of the convergence
path for the implicit solution of contact problems. In: Recent Developments
and Innovative Applications in Computational Mechanics, pp. 17–23. Springer
(2011)
201. Zavarise, G., Lorenzis, L.D., Taylor, R.L.: A non-consistent start-up procedure
for contact problems with large load-steps. Computer Methods in Applied
Mechanics and Engineering 205208, 91–109 (2012); Special Issue on Advances
in Computational Methods in Contact Mechanics, dedicated to the memory
of Professor J.A.C. Martins
202. Zavarise, G., Wriggers, P.: A segment-to-segment contact strategy. Mathemat-
ical and Computer Modelling 28, 497–515 (1998)
203. Zavarise, G., Wriggers, P.: Contact with friction between beams in 3-d space.
International Journal for Numerical Methods in Engineering 49, 977–1006
(2000)
204. Zavarise, G., Wriggers, P., Stein, E., Schrefler, B.A.: A numerical model for
thermomechanical contact based on microscopic interface laws. Mechanics Re-
search Communications 19, 173–182 (1992)
205. Zavarise, G., Wriggers, P., Stein, E., Schrefler, B.A.: Real contact mechanisms
and finite element formulation a coupled thermomechanical approach. Inter-
national Journal for Numerical Methods in Engineering 35, 767–785 (1992)
206. Zhang, H.W., He, S.Y., Li, X.S., Wriggers, P.: A new algorithm for numerical
solution of 3d elastoplastic contact problems with orthotropic friction law.
Computational Mechanics 34, 1–14 (2004)
207. Zheng, Q.-S.: Theory of representation for tensor function – a unified invari-
ant approach to constitutive equations. Applied Mechanics Reviews, ASME
International 47, 545–587 (1994)
208. Zhong, Z.H.: Finite element procedures for contact-impact problems. Oxford
Univ. Press, Oxford (1993)
209. Zhong, Z.H., Nilsson, L.: A contact searching algorithm for general contact
problems. Computer and Structures 33, 197–209 (1989)
210. Zhong, Z.H., Nilsson, L.: A unified contact algorithm based on the territory
concept. Computer Methods in Applied Mechanics and Engineering 130, 1–16
(1996)
References 435
211. Zhou, B., Accorsi, M.L., Leonard, J.W.: Finite element formulation for mod-
eling sliding cable elements. Computers and Structures 82, 271–280 (2004)
212. Zienkiewicz, O.C., Taylor, R.L.: Finite Element Method: The Basis, 5th edn.,
vol. 1. Butterworth-Heinemann, New York (2000)
213. Zmitrowicz, A.: A theoretical model of anisotropic dry friction. Wear 73, 9–39
(1981)
214. Zmitrowicz, A.: Mathematical description of anisotropic friction. International
Journal of Solids and Structures 25, 837–862 (1989)
215. Zmitrowicz, A.: An equation of anisotropic friction with sliding path curvature
effects. International Journal of Solids and Structures 36, 2825–2848 (1999)
216. Zmitrowicz, A.: Illustrative examples of anisotropic friction with sliding path
curvature effects. International Journal of Solids and Structures 36, 2849–2863
(1999)
217. Zmitrowicz, A.: Models of kinematics dependent anisotropic and heterogenous
friction. International Journal of Solids and Structures 43, 4407–4451 (2006)
List of Languages
Index
C 0 -Continuous Surface in 3D, 47 Babuska-Brezzi (BB), 22, 322

C 1 -Continuity, 46 Backward Euler Scheme, 140, 161, 323
C 2 -Continuity, 45 Backward Loading, 306
2D Coordinate System, 64 Backward Scheme, 140
2D Case, 194 Beam Coordinate System, 389, 393
2D Contact Element, 85 Beam-To-Beam, 21, 24, 123, 381, 389,
2D Contact Interaction, 122 391, 392, 396, 406
2D Contact Kinematics, 91 Beam-to-Beam, 177
2D Plane Geometry, 44 Belt Friction Problem, 413
3D Local Coordinate System, 64 Bernstein Polynomials, 218
Bezier Form, 222
Bi-Normal Vector, 34, 387
Isotropic case , 341
Bilinear Solid-Shell Elements, 277
Orthotropic case, 341
Biquadratic Solid-Shell Elements, 279
Block on a Rough Surface, 368
Adhesion Domain, 165
Adhesion Models, 7 Calibration, 378
Adhesion Tensor, 373, 374 Christoffel Symbols, 31, 68, 69, 71, 139,
Allowable Elastic Region, 165 186
Angle, 27 Circular Motion, 172
Anisotropic Evolution Equations, 145 CLCL, 319, 320
Anisotropic Friction Coefficients, 148 Closed Form Solution for Penetration,
Anisotropic Plane, 152 98, 214
Anisotropic Tensor, 150 Closest Distance, 388
Anisotropic Yield Function, 146 Closest Point Projection, 18, 35, 38, 44,
Approximation Matrix, 210, 212, 214, 50, 53, 54, 84, 121, 186, 212, 295,
295, 317, 337 387, 388
Arbitrary Anisotropic Properties, 150 CLRS, 319, 325
Arbitrary Curves, 62 Complementary Condition, 137
Arbitrary Parameterization, 52 Complex Contact Interface Law, 143
Arc-Length, 33, 34, 54 Composite Surface, 223
Area, 27 Concave Part, 42
Augmented Lagrangian, 331, 358, 359, Constant Orthotropic Friction, 149
364 Constant Orthotropy in the Plane, 148
440 Index
Constitutive Equations, 137, 138 Curve in 3D, 33

Constitutive Relations, 5 Curve-To-Curve, 19, 53, 62, 107, 109,
Contact Between Beams, 3 123, 175, 202, 385, 393
Contact Between Particles, 4 Curve-To-Rigid Surface, 20, 24, 115,
Contact Between Surfaces, 3 118, 127, 182, 183, 207
Contact Element, 210, 211 Curve-To-Rigid-Surface, 22
Contact Interaction , 83 Curve-To-Surface, 19
Contact Layer - Contact Layer, 319, Curvilinear Coordinate System, 56
320, 323
Contact Layer - Rigid Surface, 319, Deep Drawing, 281, 282
323, 325 Direct Approach, 13
Contact Layer Finite Element, 315, Distance, 102, 103
317, 323 Distance Between Curves, 387
Contact Patch Test, 209, 269, 271, 323
Continuous Integration Scheme, 299 Edge-To-Edge, 21, 24, 381, 385, 396,
Contravariant Components, 27, 28 406
Contravariant Metrics Components, 67 Edge-to-Edge, 177
Elastic Part, 157
Convective Coordinates, 112
Elastica Problem, 394
Convective Normal Velocity, 71
Elliptic Point, 29, 41, 42
Convective Tangent Velocities, 72, 92
Enforcement of Contact Conditions, 9
Convective Variables, 4
Euler-Eytelwein Problem, 413
Convective Variation δξ, 94
Euler-Eytelwein problem, 24
Convex, 39
Evolution Equations, 138, 140, 145,
Convex Part, 41
151, 173
Coordinate System, 95, 109
Coordinate Vector, 388 Finite Beam Element, 392
Coulomb Friction, 138, 144, 178 Finite Difference Scheme, 163
Coulomb Type Yield Functions, 147 Finite Rotations, 389
Coupled Adhesion-Friction Model, 21, Forward Loading, 305
22 Frechet Derivative, 190
Coupled Adhesion-Friction model, 23 Friction Models, Phenomenological, 6
Coupled Orthotropy, 372 Friction Models, Statistical, 5
Coupled Adhesion-Friction Model, 7 Friction Tensor, 373, 374
Covariant Approach, 13 Frictional Cone Description, 143
Covariant Basis Vectors, 87 Frictional Contact, 249, 250
Covariant Components, 27 Full Displacement Vector, 158
Covariant Derivative, 68, 74, 89, 139 Full Newton Method, 334
Covariant Derivative for the Covariant Full Normal Tangent Matrix for Normal
Component, 33 Interaction, 189
Covariant Derivative of the Contravari- Full Tangent Matrix, 230
ant Component, 32 Full Time Derivative, 75
Covariant Form for Sliding, 182 Fundamental Tensors, 26
Covariant Form for Sticking, 182
Covariant Metrics Components, 67 Gauss Point-Wise Substituted
CPP, 18, 53, 54, 62, 212, 214, 295 Formulation, 287
CTRS, 20, 24 Gauss Points, 267
Curvature, 34, 382, 387 Gauss Quadrature, 265
Curvature Part, 230 Gauss-Codazzi Formula, 31, 88
Curvature Tensor, 27, 40 Gaussian Curvature, 30
Index 441
General Interface Model, 144 Large Sliding, 363

Generalization of Coulomb Friction Length, 27
Law, 143 Lie Time Derivative, 68–70
Generalized Eigenvalue Problem, 28 Linear Approximation, 214, 231, 250,
Generalized Projection Procedure, 52 269
Geometrical Approaches, 17 Linear Contact Element, 294
Geometrical Interpretation, 166, 167 Linearization, 12, 77, 79, 92, 94, 185,
Geometrically Exact Theory, 1 194, 356
Geometrically isotropic case, 341, 342 Linearization of the Normal Contact,
Global Searching Algorithms, 12 187
Lobatto Quadrature Formula, 213, 263
Heaviside Function, 137 Local Coordinate System, 387
Hermite Form, 390 Local Searching Procedure, 388
Hermite Functions, 229 Local Surface Structure, 28
Hermite Patch, 223
Hermite Polynomials, 222, 390 Main Part, 230
Hermite Type Interpolation, 390 Master Segment, 211, 212
Hertz Contact Problem, 22, 315, 323 Master Surface, 210, 212
Hertz Hypothesis, 325 Mean Curvature, 30
Hyperbolic Paraboloid, 29 Measure, 4
Hyperbolic Point, 42, 43 Measure of the Normal Interaction, 66
Hysteresis Curve, 305, 306 Measure of the Tangential Interaction,
66
Incremental Formulation, 158
Measures of Contact Interaction, 109
Indifferent Covariant Derivative, 32
Mechanical Interpretation, 168
Integration of Evolution Equations, 142
Mechanics of Knots, 24
Integration Points, 211
Metric Tensor, 27, 40, 66, 111
Integration Rule, 213
Mixed Components, 27, 28
Integration with Subdomains, 265,
Moment (Rotational) Interaction, 385
269–272, 323
Mortar, 22, 263, 268
Interpolation in Hermite Form, 221
Isogeometric, 209, 381
Natural Coordinate, 54
Isogeometric Beam Elements, 406
Nitsche, 21, 128, 129, 132, 287
Isotropic Coulomb Friction, 136, 145
Nodal Vector, 210
Isotropic Equations, 143
Node-To-Segment, 22, 121, 210, 230,
Jacobian, 34, 54, 387 293, 294, 337
Node-To-Surface, 209
Kinematics, 2 Non-Allowable Domains, 48
Kronecker Delta, 27 Non-Frictional Case, 136, 230
Kuhn-Tucker Condition, 160, 162, 170 Non-Linear Approximation Operator,
189
Lagrange, 129 Non-Penetration, 137
Lagrange Function, 159, 170 Non-Penetration Terms, 130
Lagrange Multiplier, 129–131, 137, 334 Normal Contact Conditions, 136
Lagrangian Formulation, 21 Normal Force, 382
Lagrangian Method, 322 Normal Interaction, 117, 174, 384
Large Penetration Algorithm, 236 Normal Part, 356
Large Penetration Algorithm, Normal Vector, 34, 387
Enhanced, 237 NTS, 22, 210, 230, 293, 294, 325
442 Index
Numerical Solution, 11 Reversible Loading, 298, 304

NURBS, 218, 222, 390, 391, 393 Rheological Model, 167
Rigid Cone, 103
Orthotropic Adhesion, 374 Rigid Cylinder, 101
Orthotropic case, 341, 343, 344 Rigid Plane, 98
Orthotropic Coulomb Friction Law, 169 Rigid Sphere, 99
Orthotropic Friction, 372, 374 Rigid Surface, 95, 97
Orthotropic Surface in Polar Coordi- Rigid Surface is a “Master”, 97, 99,
nates, 153 101, 103
Orthotropy Angles, 378 Rigid Surface is a “Slave”, 95, 98, 105
Orthotropy Axes, 149 Rigid Torus, 102
Osculating Circles, 60 Rotational, 382
Rotational Contact, 180
Parabolic Cylinder, 29, 105 Rotational Friction, 181
parabolic cylinder, 273, 274 Rotational Part, 230
Parabolic Point, 43 Rotational Sliding, 206, 383
Parabolic point, 42 Rotational Sticking, 383, 398
Parallel Translation, 75
Rotational Sticking , 180
Penalty, 129
Penalty Method, 137
Segment, 95
Penalty Term, 131
Segment-To-Analytical Surface, 21, 95,
Penetration, 66, 98, 136
213, 272
Planar Point, 30, 43
Segment-To-Segment, 22, 209, 211, 264,
Point-To-Analytical Surface, 335, 336
268
Point-To-Corner, 48
Semi-Infinite Domain, 51
Point-To-Curve, 19, 50, 105, 122, 174,
Serret-Frenet, 34, 55
202
Serret-Frenet Coordinate System, 53,
Point-To-Edge, 48
387
Point-To-Point, 20
Serret-Frenet Formulas, 34, 51, 56, 88
Point-To-Surface, 19, 121
Serret-Frenet System, 34
Positive Definite Matrix, 40
Shape Function Derivatives, 210, 212,
Power of the Sliding Forces, 158
213
Principal Curvatures, 30, 41
Principle of Maximum Dissipation, 156, Singular Determinant, 61
158 Slave Node, 210
Projection Domain, 39, 41–43, 58 Slave Point, 71, 211, 212, 264
Proximity Criteria, 41 Slave Segment, 211, 212, 264
Proximity Domain, 47 Sliding, 139, 144, 148, 199, 357
PTC, 105 Sliding Force, 160
Smooth Contact Surface, 224
Quadratic Approximation, 232, 257 Smooth Surfaces, 271
Solid-Beam, 381, 392, 406
Rate of the Penetration, 92 Solid-Shell, 406
Rate-Independent, 145 Solvability of the Minimization
Reissner Theory, 393 Problem, 40
Relative Displacement Vector, 114 Spatial Basis Vectors, 66
Relative Velocity, 157 Spatial Christoffel Symbols, 68
Residual Vector, 384, 385 Spatial Coordinate System, 38, 40
Return-Mapping, 140–142, 180, 296, Spectral Decomposition, 171
383, 384 Spectral Representation, 148, 149, 170
Index 443
Spiral Motion, 274 Tangent Vectors, 388

Spiral orthotropy, 351 Tangential Contact, 190
Spiral Orthotropy on a Cylindrical Tangential Contact Conditions, 138
Surface, 154 Tangential Force, 157
spiral orthotropy on the cylinder, 351 Tangential Interaction, 117, 384
Spline Interpolation of Curves, 218 Tangential Part, 356, 357
Spline Surfaces Interpolation, 222 Tangential Sliding, 205, 383
STAS, 21, 95, 213, 272 Tangential Sticking, 383, 398
Sticking, 144, 147, 190, 198, 204, 356 Torsion, 34, 387
Sticking Zone, 166 Trial Force, 158, 161
STS, 22, 211, 268 Trial Step, 141
Surface, 25 Trial Tractions, 139
Surface Convective Coordinates, 66
Surface Coordinate System, 40, 149 Update of the Sliding Displacement,
Surface of Revolution, 96 298
Surface Smoothing, 217 Uzawa Algorithm, 355, 358, 364
Surface-To-Surface, 19, 63, 65, 119, 185 Uzawa Approach, 360
Sylvester Criterion, 40, 41, 57
Symmetrization, 355 Variation δξ 3 , 77
Variation δξ i , 79
Tangent Matrices, 188, 382 Variation Parameters, 106
Tangent Matrix for Tangential Sticking, Variational Equation, 389
192 Variations, 113
Tangent Matrix for the Normal Part, Velocity Vector, 106, 112
195
Tangent Matrix for the Tangential Weingarten Formula, 31, 88
Part, 196
Tangent Vector, 34 Yield Function, 144, 152

Computational Contact Mechanics - Konyukhov

Uploaded by

Copyright:

Available Formats

Computational Contact Mechanics - Konyukhov

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computational Contact Mechanics - Konyukhov

Uploaded by

Copyright:

Available Formats

Lecture Notes in Applied

and Computational Mechanics 67

Prof. Dr. Peter Wriggers

For further volumes:

Geometrically Exact Theory for Arbitrary

Alexander Konyukhov and Karl Schweizerhof

ISSN 1613-7736 e-ISSN 1860-0816

c Springer-Verlag Berlin Heidelberg 2013

Printed on acid-free paper

Who understands geometry understands anything in this world.

1564-1642 Galileo Galilei

Geometry is founded on mechanical practice and is nothing other

The Prinicipia: Mathematical Principles of Natural Philosophy.

Geometrical ideas correspond to more or less exact objects in nature,

Relativity: The Special and General Theory.

is undoubtedly observed for many branches of mechanics: the combination

analytical results for the veriﬁcation of contact algorithms. A systematic anal-

Karlsruhe Institute of Technology, Alexander Konyukhov

I owed my geometrical ideas to my education as well as PhD study at Kazan

Karlsruhe Institute of Technology, Alexander Konyukhov

2 Diﬀerential Geometry of Surfaces and Curves . . . . . . . . . . . 25

3 Closest Point Projection Procedure and Corresponding

3.2.2 Proximity Domain for Globally C 0 -Continuous

4 Geometry and Kinematics of Contact . . . . . . . . . . . . . . . . . . . . 63

4.5.4 Linearization in a Covariant Form of Variations for

5 Weak Formulation of Contact Conditions . . . . . . . . . . . . . . . . 119

6 Contact Constraints and Constitutive Equations for

6.3.6 Geometrical Interpretation of the Solution Process . . . 164

7 Linearization of the Weak Forms – Tangent Matrices

7.3 Linearization of the Weak Form for the Surface-To-Surface

8 Surface-To-Surface Contact – Various Aspects for

8.4.5 Bending of a Beam over a Rigid Cylinder . . . . . . . . . . . 243

9 Special Case of Implementation – Reduction into

9.1.4 Treatment of Special Cases . . . . . . . . . . . . . . . . . . . . . . . 297

10 Implementation of Contact Algorithms with High

11 Anisotropic Adhesion-Friction Models – Some

11.2.2 Point-to-Analytical Surface Contact Element.

12 Experimental Validations of the Coupled Anistropic

13 Various Aspects of Implementation of the

13.1.4 Tangent Matrices for Rotational Sticking . . . . . . . . . . . 383

14 Frictional Interaction of a Spiral Rope and a Cylinder

List of Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437

Computational contact mechanics has been developed into a separated

1.1 Overview of Approaches to Model Contact

In computational contact mechanics the modeling process depends gener-

Step 1: Kinematics of Contact Interactions

rs − ρ −→ min . (1.2)

The “master-slave” approach is treating the normal contact such that a

This measure can be deﬁned diﬀerently depending either on the geometry of

Then two convective variables Δξ 1 , Δξ 2 in a surface covariant basis ρ1 , ρ2

This approach has many advantages:

Step 2: Constitutive Relations for Contact Conditions

Statistical Friction Models

Statistical analysis of a real rough surface and experimental aspects of its

Phenomenological Friction Models

Lagrangian multiplier methods. Parametric quadratic programming was used

In the earliest publications adhesion phenomena have been introduced in nor-

Coupled Adhesion-Friction Models

A constitutive model of tangential adhesion as a reversible elastic force has

Motivation of Interface Surface Models from Computational Tribology