Discontinuum Mechanics Using Finite and Discrete Elements

'I
.1
'
/'l..
,:5. l
~--
/1.
., .....
,//.( 11
v :._~ u:.::.~ e,
Jl
if
il
u,.
J._,....; .
1
t
....
Discontinu:um Mechanics
Using Finite and Discrete Elements
-----.-----
WITPRESS
------- --------, -
..
......
WIT Press publishes \eaing hooks in Scic111.:1.~ am\ lh\Jnology.

Visit our website for new and currenl list of titlcs.
www.witpress.com
----------
WIT eLibrary
Making the latest research accessible, the WIT electronic-library foaturcs papers
presented at Wessex lnstitut~ ofTechnology's prestigious international conferences.
To access the library and view abstracts free of cbarge please visit www.witpress.com
..
-.
.... ....: . : ~ '
.,';
,~t!
~~~'!" . . .
. - -----
,,.,.
Discontinuum M echanics ,
Using Finite and Discrete Ele1nents
' ,
S. Mohammadi
University o/Tehran. Imn
(_',CJ
or>
/' -~,.~~i~E;~~}
WITPRESS
Southampton, Boston
Author: S. Mohammadi
~t'',
, r
Department of Civil Engineering

University of Tehran
Tehran, lran
Contents
\
Fariflav Engineering Research Company

Tchran, lran
.......
Published by
Chapter 1
Introduction
WITPress
Ashurst Lodge, Ashurst, Southampton, S040 7 AA, UK
Tel: 44 (O) 238 029 3223; Fax: 44 (O) 238 029 2853
E-Mail: witpress@witpress.com
http://www.witpress.com
1.1
Discontinuum Mechanics, Why? .
1.2
Alternative Approaches
1.3
Scope of the Book
For USA, Canada and Mexico
1.4
Notations
Computational Mechanics lnc

25 Bridge Street, Billerica, MA O1821, USA
Tel: 978 667 5841; Fax: 978 667 7582
E-Mail: infousa@witpress.com
Chapter 2
2.1
Discontinuum Mechanics - A Ueview
11
12
Geomechanical Applications .
2.1.1
Rock BlRSting . . . .
12
2.L~
Mltti111~
13
2.1.3
Failure of Cut Embankni.ent .
14
2.1.4
DDA Slope Stability Analysis
16,
2.1.5
Shear Band Slope Stability
16
2.1.6
Falling
2.1.7
Porous Media .
Hock lhm1tM
/\ Catalogue record for this book is av ai la ble

from the British Library
ISBN: 1-85312-959-3
Library ofCongressCatalog Card Number: 2002116429
No responsibility is assumed by the Publisher, the Editors and Authors for any injury
and/or damage to pcrsons or property as a matter ofproducts liability, negligencei or
otherwise, or from any use or operation of any methods, products, instructions or ideas
contained in the material herein.
2.2
Object~
17
,,
18
20
Granular Materials . . .
WIT Press 2003.
2.2.1
Granular Flow in Silos .
20
Printed in Great Britain by Athenaeum Press, Gateshead.
2.2.2
Superquadric Elements .
21
Ali rights reserved. No part of this publication may be reproduced, stored in a retrieval
systcm. or transmitted in any fonn or by any means, electronic, mechanical, photocopying,
rccording, or olhcrwisc, without thc prior writtcn pcrmission of the Publisher.
2.3
Impact Analysis (Progressive Fracturiug)

2.3.1
Penetration of a Missile
....
21
21;
2.11
2.5
2.3.2
Metal Cutting
22
4.2.4
Generalized Strain Tensors
73
2.3.3
Demolition of Structures .
24
4.2.5
Rate of Deforroation ..
74
2.3.4
Composites
24
4.2.6
Stress Measures . . . .
75 :' -
2.3.5
Masonry Structures
26
4.2. 7
Objective Stress Rates ..
77
2'.3.6
Repair Modellings
30
4.2.8
Multiplicative Decoroposition
79
80
32
4.3. l
Equilibrium Equation ..
80
34
4.3.2
Variational Forro . . . .
82
32
:~.4.1
Shockwa.ve' AnalyHiH
Other Applications .
Chapter 3
3.1
Initial/Bounda.ry Value...,Pr2blem
Particulate Flow
Constraint. Enforcing Methods
Introduction ..
35
35
3.2 Definition of a Constraint
35
3.3
Constraint Enforceroent
40
:J.a.1
1m ><nietrahility Constraiut
42
3.3.2
Penalty Method
..
43
3.3.3
Lagrangian Multiplier
47
:t.:\..l
l 'cq1.111hnd l111~;r1u1~l1111 rvlc1t.l111d
tm
3.3.5
Augmented Lagrangian Method.
Variational Forros .
56
3.5
The Penalty Method
58
3.5.1
60
Contact Instability
N onlinear FE Discretization
4.4
Finite Element Discretization
85
4.5
Explicit ,Time Integration . .
87
4.5.1
Central Difference Method.
87
4.5,2
Luroped Mass Matrix
88
4.5.3
Stable Time Step .
89
Normal Contact
93
Chapter 5
5.1
93\. ''
Introduction . . . . . . .
t.. ~ llydrntuth H.1111111latluH

1
52
3.4
Chapter 4
4.3
5.2.1
Original Algorithms .
5.2.2
DYNA 2D Formulation
'
'
'
95
9.9
;' :;;;;~i8~{~?\4k ;f:;
DYNA Penalty Forroulation ..
. 104
5.3.1
Search For the Master Segroent
. 104
69
5.3.2
Finding the Contact Point .
5.3.3
Contact Force . . . . ,, . . .
4.1
Introduction .
69
4.2
Large Defonnation Kinematics
69
4.2.1
The Motion .
'1.2.2
4.2.3
5.3
..
'
10~
..... 1oi<:,}'
2D Node to Edge Penalty Contact
. 108 y~f,>:
70
5.4.1
Contact Geometry . . . .
Ddonnation Gradiont
71
5.4.2
Global Senrch
Polar Decoroposition .
72
5.4.3
Local Search Algorithm .
5.4
"-.._
o:;.
Algorlth~
10~ ;,:~-0;
/,'}i_::,::
. 110 ','
5.4.4
Normal Contact Interaction . . . .
. 111
G.4.G
Com;istent Node to Edge Contact . .
. 115
r r
".,
Frictional Contact
121
6.1
Introduction . . . . . . . . .
. 121
6.2
Frictional Contact - The Basics
. 121
6.4
Discontinuum Contact Mechanics
181
Introduction . . . . . . . . . . .
. 181
7.1.1
Histmical Development
. 182
7.1.2
Applications
. 183 '
7.1.3
Specifications
. 183
7.2 Contact Detection
. 185
6.2.l
Frictional Contact Model
. 121
7.2.1
Contact Geometry
. 187
6.2.2
Basic Kinematics . . . . .
. 122
7.2.2
Global Search Algorithms
. 188'
o.2.:3
Plasticity Theory of Friction
. 126
7.2.3
Buffer Zone . . . . . .
. .)PQ
7.2.4 .Contact (Field) Zone .
.. 194'.
2D Frictional Contact Models Without Hardening
.. 135
6.3.1
Residual Force Vector
. 137
7.2.5
Binary Tree Structures .
n.~t2
St.ifflWHH Contrilmtion
. . 138
7.2.()
Spaco
(i.a.:~
Cu11HiHl.c11t Li11eu.r:u~t.io11
. 142
7.2.7
Alternating Digital Tree (AD'I') .
3D J:i!ictionaJ Canta.et Models Without Hardening
. 147
6.4.1
Preliminaries . . . . . . . . .
. 147
().4,
Pla.':it.icity Theory of Friction
. 147
6.4.3
Numerical Integration
. 150
6.4.4
FE Discretization . . .
. 152
6.5 3D Frictional Contact Modela With

Frictioun.l lla.rdening . . . . . . . . .
6.6
7.1
... 118
Chapter 6
6.:3
Chapter 7
. 155
. 155
7.3
n(~COUlJ>OHitlon
Object Representation
.196
<'
203
.#t
. 22tc\ ...
~;:;_.t
f_; '(;" ..
"'
7.3.1
Circular Disks
222~,:..
7.3.2
Disk Clusters .
. 230<::
7.3.3
Ellipse Shaped 'Particle~
.. 233;>:.
7.3.4
Superquadric Objects .
. 239."
7.3.5
Geometric Models for Permeah!P Soli1 Is
~,; .~t.fi' :.
.. 242 '
6.5.1
Physica9A.spects of Frictional Phenomena .
6.5.2
Numerical Simulation of 3D Work Hardening

Frictional Contact . . . . . . .
. 159
8.1
Intro<luction ..
. 245
6.5.3
Numerical Integration Scheme
. 163
8.2
Mesh Dependency
. 246
6.5.4
Consistent Tangent Modulus .
. 164
8.3
Fracture Mechanics .
. 247
Node to Face J:irictional Contact Models
. 166
8.4
Strain Softening Models
. 249
6.6.1
2D No'de to Edge Model . . . . .
. 166
8.5
Damage Mechanics . . .
. 253
6.6.2
3D Nade to Quadrilateral Face Model
. 170
8.5.1
. 257
Chapter 8
Remeshing Techniques
Higher Morles of Fracture
8.G
Ansotropic Material Models .
. 258
. 259
8.6.1
Tsai-Wu Model . . .
8.6.2
Hoffman Anisotropic Yield Criterion
8. 7 Localization and Crack Direction
8.8
.. 260
... 267
8.7.l
Loss of Uniqueness . . . .
. 269
8.7.2
Discontinuous Bifurcation
. 272
Hmneshing Algorithm .
Appen<lix A
UTDem code
Preface
. 276
279
A.1 Introd uet ion . . . . .
. 279
A.2 Data File Structure .
. 279
A.3 Sample Data File . .
. 285
'1 arn always obl-iged to a person who hu,:;
l1mylit
11111.
n smgle word'
The finite element method h88 become .a mature powerful appro~Ji<;iL:

numerical simulation of various engineering and industrial applicatic>ns.
With rapid de~elopment of the computing focilitie:4, the analysis of la.r~
scale real discontinnum problema has now become a reality.
A month after I returned to Iran, 1 was asked to arrange for a ne;W
state of the art course for the Ph.D. students studying struct~$.l.,
mechanics. My answer W88 'Computational Contact Mechanics. Aftet.,:
three years, a comprehensive book has now lmco11w n.vniln.ble, cover~gi;''
the discrete element methodology ami priuciplm; of discontinutu:U,Y\
mechanics. It is mainly prepared as a postgraduate textbook and :fe;}~,(.;
resei:i.rch institutions undergoing research on general contact rp,ech&itlCS,'C: :
and discontinuum problema. Here, I should mentio11 my studenta in th'f fri;.
U11iVNttlt.,v ol' '1't1hrn.11 1 wl1utm frnqmmf, (jl1t111f.i111111 lulp1d 11111 lo prttpMt'
l11111lt wll,11 1 l hupt1, Uw l111u.~t1 put_:j~fltj tWWl'~.
1:,:,
tbl.;:::>
.)
."
-,11J1:
The book can be classified into three parts. The first part is dedica.te4;
to the basic concepts of constraint definitions and various methods foi
constraint enforcement as well as a short di~(11sfo11 1111 the finite
discretization ~d necessary finite deformation theory as the bases uf
combined finite/discrete element method.
element
the .
The second part is devoted to the cont.a.ct interaction and detection

procedures. Physical and experimental observations, normal 8.nd
tangential contact formulations, theoretical considerations
computationo.l nlgorit.lm1s will be discussed ror b11th grn11nla.r..,flow and
deformable finite/ discrete element modelings.
and
The final part describes the progressive fracturing phenomena..

Stability and convergence discussions, an anisotropic material model,
J()calit,ation difficult.ies, reme1:1hing tedmiqum;

computational algorithms will be discussed.
and
their
asimciated
A limited version of the UTdem discrete element code has been

iuclwled in the accompanying CD. H iH primarily aimed at givihg the
n~a.d<:r ;Lu overall insght of how complex discrete element software
performs and how various control parameters may affect the progressive
cracking and granular flow simulatfons. A. sample mam1a.I . fot preparing
the input data file has been included as an appendix. I \v;<>uld like .to
gratefully appreciate my colleague Mr. A Najafi Amin for his outstanding
work on pre and postprocessing modulus of the UTdem.
The present textbook is a result of 'infinite' research works for many
years all over the world. I have tried to appropriately a.cknqw~edge t.lie .
achievements of corresponding researchers within the text; relevant
fig11nt1, t11.l1hi n11cl forurnln.c. I nm m1wl1 in<loht.nd to their 011tHf'.11.11ding
lt wutliri, 1111d 1111.\' ~hn1l.1'.ot1tl11t.i, lt

tu 1d1wmuly rng101.t,1~cl.
111r1,1t1
1n1f'f'il11111ly u.1lrnow)t!tl}1.,i11R
To:
f.11t11tl
First of all, I would like to express my sincere gratitude to Prof.

D.H...J. Owen who wa.s the first to open rny eyes beyond the continuum
harriers, introducing me to new challenging frontiers of the computational
science. Also to Dr. A.J .L. Crook, my first discrete element teacher and
to Prof. D. Ferie and the late Prof. E. Hinton for valuable discussions. To
this end, I have used and built on the experience and work of Dr. N.
Petrinic, Dr. A. Munjiza, Dr. J. Yu, Dr. G.Q. Liu and Dr. J. Macedo
since the first days I got involved with this new subject; I gratefully
admowledge aud a.pprecia.te them.
QJ
I would like to extend my best regards to Dr. S.S. Yazdi and to the
Farinkav Engineering Research Company for the software support. The
techn ical su pport of the Computer Centre, Department of Civil
EHgiueering of Univ<~rsity of Tehra11 for facilitating the preparation of the
nH1.1111script H also ndmowledged .
.F'inally, rny special thanks to my family for their never ending
support. I have spent many hours preparing this text; hours that had to
be spent for my wife and little 'sogol'. Without their understanding and
love this book may not ever have been written.
..
;i~ ;Mohammadi
Tehran
Mansoureh
&J
$ogol
Chapter 1
Introduction
1.1
Discontinuum Mechanics,
Why?
...~-
A mrnd hourglnHH, 1111~1 f(1r 11mtttrloH boeu ti Hl11qil rn1111111111 t it11 mmwurlng
tool as well as a relatively accurate reliable mt~asure fur f>cientific experiments. Nevertheless, until very recently, even at the age of supercomputers
and efficient numerical methods such as the finite element method, it has
not been possible to accurately model this simple phenomenon. A highly
varying geometry, and interacting objects (sands) were the main obsta.eles
to any efficient approach (see Figure 1.1).
An interesting set of problems which have recently attracted special at
tention, includes the general behaviour of granular materials. In this class of
problema, large number of interacting bcidies, usually simple rigid elementS, . ,
are interacting in a donmin which will govern tite p;t'11crnl re1-1ponse of the'.
medium through these individual interactious. 'l'l1t~ l11'sL 1'x11mple, me.y be,
the filling or emptying a silo with/from granular materials as depicted in
Figure 1.2 [1].
Figure 1.1: Sand flow in a typical hourglass.
Discontinuurn Mephanics
T1me:23.1811
Chapter 1. Introduction
......
Fragmentation
Delamination
Figure 1.3: Progressive fracturing and fragmentation in a typical composit~:

specimen subjected to impact loading.
Figure 1.2: Filling a silo with granular material [1].
is
The behaviour of any individual element usually governed by the clas..

sicalrigid hody dynamics, although n. gonora.l finite element formuln.tionmay
liP llti<\d l.o 1011Hid<r dnlimnnblo rnHpomm for tlw111. Howcwor, to l11cl11do t.lm
i11t<H'ucLio11 of LJw lmdius lnto the analysh:;, conventional contlnuum methods
can not be effectively used (if at all). AdditionaJly, more complex phenomena such as dynamic filling/emptying, arching behaviour, shockwaves dueto
sudden closure, thermal effects, and lateral seismic behaviour were almost
untouchable.
This lack of success was not only limited to that simple case; almost
a.nywhere in the industry and academic world, severa! applications could
have been found that analysts cea.sed to be able to accurately simulate.
One of the major deficiencies was in the field of new advanced materials
bcing subjected to dynamic and hazardous loadings.
Gradually replacingflonventional materials, advanced materials such as

new metallic alloys, composite laminates, ceramics and plastics are now
widely used in many applications involving dynamic loading such as machinery, pressure vessels, defence structures, vehicles, sports equipment and;..
uotauly aerospacc 8t.ructures. Industries such as the automotive, and recreational industries have also been placing increased reliance on high performance materials. The rea.son for this dramatic increase in the use of
new materials can be attributed to the superior properties that they offer;
high strength-to-weight ratio, good damping characteristics, high fatigue
strength, corrosion resistance and versatility to meet speci:fic requirements.
Even with the siguificant advances that aw 11ei11g 1111:1.de with these new ~..
materials, there is still a great deal unknown about their mechanical cha.rO:!.
teristics. These unknowns can be attributed to the large munber of different
ma.terials being used, to t.he contribution of 111inor11<wha11knl interaction1'
of difforout. compowmf.H t.o thu ovorall boluwi1111r 1111d t.lw vorlot;y of fallu~
mechanisms involvcd. lt is also well known t hnt tlw pn~::;euce or inhorri(;.i
geneities plays an important role in the behaviour of these materials.
One of the major problems that dominates the design procedure fr ;; ;,
structural applications is the strength of material to transverse impact which ;: .,
may cause substantial internal damage. Thes<~ impact. dama.ges are ofte~ ';\:Y
embedded inside the material, hardly detectable by the uaked eye (2, 3]. Ac.-. :\.::
cordingly, detection of impact damage is critically important in maintainfug :/'.'
the integrity of the structures n service. By examining the numerous cori- :..
tributions to this area of research [4, 5, 6, 7, 8}, it is evident that impact C'
loading represents a highly complex phenomenon cmnpused of several inter- ',
acting processcs.

Figure L3 represents the progressive fracturh1g and fragmentation ph~
nomena in a typical composite specimen subjected to impact loading. 'fh!s
schematic representation, is perhaps only related to the failure observed ..
in high velocity impact. For low velocity impnct., howPv<~r, while it is unlikely that extensive fragmentation will be observed, material fracture and
delamination will be the likely modes of failure that exist.
It has been shown that interna! cracking of a structure subj~cted to
low or high velocity impact loadings, are progressive phenomena which may .
rapidly propagate throughout the component. This rnight rc~mlt in the cnF
Discontinuum Mechanics
11
Possible fractured region
Chapter l. Introduction
Transltion Interface
Bondio Interlllce
Figure 1.4: Composite specimen subjected to impact loading.

Figure 1.5: Discrete element modelling of a cornposite plate.
n
LJ
ation of new totally separated parts, which interact with their surrounding
rngionH. Conseq11<~ntly; a powcrful scheme is required to be a.ble to monitor
the fracturing process and to effectively model both individual and interaction behaviour. Recent deyelopments of discrete element methods have_
prepared the ground for a new approach to modelling this behaviour based
on the concepts of the discontinuum mechanics.
The tradit.ioual a.pproach to thesimulation of i:>lress distributions in arbitrary shaped components under possible nonlinear geometric and material
conditions is by finite element techniques. However, the traditional finite element method (FEM) is rooted in the concepts of continuum mechanics and
is not suited to general fracture propagation problems since it necessitates
that discontinuities be propagated along the predefined element boundaries.
The.corresponding elasticity and fracture mechanics concepts are applicable oniy in situations dealing with a single crack or a low-fractured area
without any fragmentation [9]. In contrast, the discrete element method
(DEM) is specifically designed to solve problems that exhibit strong discontinuities in material and geometric behaviour (10]. The discrete element
rnetht)d idealizes the whole medium into an assemblage of individual bodies, which in addition to their own deformable response, interact with each
other (through a contact type interaction) to perform the same response as
the medium [11]. A far more natural and general approach is offered by a
combination of discrete element and finite element methods.
LJ
Considera test specimen subjected toan impact loading as depicted in

Figure 1.4. Early material cracks are likely to appear. near the position of
applied impact load. As the analysis advances, two separate regions can
be distinguished. The first one is a highly fractured region, and the second
one is the remainder of the body which presumably contains no fracture
patterns. The fract ured region is usually formed in the vicinity of impact
loading and may comprise further separate parts.
It should be emphasized that drawing such a concl usion <loes ndt necessarily require a fracture analysis. A relatively fast elastoplastic analysis can
be performed by simple 8hell or solid. elements t.o detmmine the different
rogl1111M 1H~cordi11g to HUlll<' offect.lv Htl'688 or nt 111i11 <rit.n1 iu. Tho prtt<llc!ted
fractured/delaminated regionR may then be exarnined i11 t.lw lnter stages of
the analysis through the combined finite/discrete element algorithm. The
regions' boundaries should be further extended if the cracking reaches
the boundaries of the fractured region [12, 13].
to
Figure 1.5 shows a typical section of the above uwntioned specimen

(here, a quarter of a composite plate). In a cornbined FE/DE method,
the possible fractured region is modelled using a discrete element mesh
and the remainder of the sp~imen is modelled by a standard finite element
mesh. It is also possible to mod~ the whole structure with discrete elements. , ~
In this case, the possibility ?f delamination is i11veHtigau~d t.hroughout the
structure. A combined mesh enables us to preveut 111111ect~ssary contact
detection and interaction calculations which comprise a major part of the
analysis time. It is worth noting that even by modelling the whofe structure
with discrete elements, we ar~ still using a combined finite/discrete element
approach, owing to the fact that"'finte elements are used for modelling
deformable behaviour of individual dlscrete elements.
For the particular case of composites, each ply or a group of similar
plies is modelled by one discrete element. Each discrete element will be
discretized by a finite element mesh and might have nonlinear material
properties or geometric nonlinearities (latge deformations). The interlaminar behaviour of discrete clements is governed by bond iHg laws, including _
contact and friction interactions for the post delamination phase.
ti
Interactions between finite elements (not those which are used for DEM
discretization) and discrete elements are modeHed .by.tra,n.siti()n interfaces.
A transition interface is define<f as a bonding interface wi~b Ye'l'Y high 'bonding strengths which prevent debonding under all stre~s conditions.
Table 1.1: Comparison of FEM aiid BE!'v1 charn.ct.cristics.
Ali interfaces, firstly, are monitored a.gainst the delamination criterion.

Oucc two layen; are delaminated, the corresponding interface will still be
capa.ble of further contact and friction interaction. However, there will be
no re-bonding after delamination.
11
Om importa.11t n.i:ipect of this type of modelling, which diHt.inguish<'R

ll. l'roru ot.ht-r w11f.11d. busod ddamlnn.Uou algorltluus {14, 15], is that it
does not require any predefined interface element. Being free from the
restrictions of interface elements, provides major advantages. Firstly, there
is no need for the nades on different layers to match each other, which
ea.ses the way that data are prepared. This is essential in defining the
transition interfaces. Secondly, in progressive cracking, particularly material
fracturing, we may end up with new nodes, edges, and boundaries that could
destroy the compatibility required for these interface elements.
1.2
FEM
Category
Mesh generation
entire domain
Solut.ion
entirc domain
.._Approxii-aoi~----- -... ...governing PDE

Matrices
sparse
usually symmetric
easier to evaluate
widely acceptable
ea.sy
Integrals
Nonlinearity
Implementation
..
BE1vI
f.lw11 111. d11111Hi11 point.H
l.>urii1tar.v cuii<litions
fully populated
non-symmetric .
far harder to evaluate
only linear problema
much more difficult
Alternative Approaches
There are alternative approaches developed and implemented for special'

applications over the past few decades competing the finite element based
methods. Sorne of tbcm can be regarded as special solutions of very specific
pmhln111s, whi11\ ~;n111H ot.lwrn h11vo IH\flfl 11dnpt.11d fnr ll wlltt rnnw1 11f 11ppll
1
1d 111 111 1 l11 1'11t l l1nl1 1q11, 11111ui 111 lfoq11 111111 1111111wlJ1 wth th~ gamftllly uf tlw
Ji11He dc11w11t mcthod.
The boundary element method (BEM) is an important technique in the
computational solution of a number of physical or engineering problems.
The boundary element method has the important distinction that the <lis"'."
cretiiation only applies to the boundary of the domain .of interest. Hence
. the computational advantages of the BEM over other methods can be considerable. Table 1.1 compares sorne of the main characteristics of the finite
element and boundary element methods as presented by Hunter [16J.
A combination of the finite element method and the bounqary element

rnethod would enhauce the general performance by utilizing the distinguished f~at~res of both~~thods. for. simulating variou~ engineering problems. Th1s kmd of modMng, for mstance, may be effic1ently used in modclling the behaviour of tunnel lining and its surrounding rock medium subjectcd to seismic waves. As an example, Figure 1.6 illustrates a simple
crack a.nalysis studied by Kim [17] and the results achieved by a combined
finite/boundary element method.
. i.
E(x)
w-.a
w.a
=E0 exp ( Jl x )
E0 ~
V=
11
bou11Jaries
fi rst. on l,j 1<~ houndary
11'1
(w < 8)
0.3
Figure 1:6: A combined finite/boundary elt::rnent rnodelling [17].
.1 . ,J;;:;:..
,.
..
_,
...'.
-4,;;:
1.3
Scope ofthe Book
As mentioned earlier, this text is prepared as a textbook for introducing

the nrn.th<~nJaticaI and computational ooncepts of the c0ntact rrwchanics
whdi 1tf'I! IJ9~d n inr;rea.'lin~ jndustrial and academc applications of the
<.:ombined finite/ discrete element method. The goal has been set to make
sorne contribution in discussing newideas in a less clearly undrstood area,
and to highlight the main problems that are rnet when dealing with highly
complex phenomena of discontinuum mechanics.
n
u
Cha1Jter 2 pr(Jvides sorne representative results and data on various applicatious thu.t: involve the general concepts of the contact mechanics and
related topics.
Chapter 3, reviews the available methods for enforcing a constraint relation onto a set of simultaneous equations. A detailed study of the penalty
method provides basic information for use in other chapters.
In Chapter 4, the main general computational aspects of the analysis will
be discussed which includi!s a comprehensive discussion on large deformation
theory, as is usually the case in the practica! simulation of impact loading
of structures. Then the basics of the finite element discretization will be
cxpla.incd. A short review of tho cxplicit algorithm as the timo integratton
scheme adopted will close the chapter.
,_,
'
Chapter 5 is devoted to providing historical and state of the art algorithms for normal contact problems. The fundamental aspects of classical
theory of normal contaet will he diHcRi-mci and practica.1 ambip;nitie~ will be
itJdret>s<.xl.
The relatively complicated theory of frictional contact will be explained
in Chapter 6. It includes discussions on various frictional contact phenomena and provides computational procedures for updating material state
variables for hardening/softening conditions and in general non-associated
formulation.
Chapter 7 is dedicated to discontinuum approaches of contact mechanics.
It includes a state of the art discussion on contact detection procedures for
problems with a large number of low or highly dense interacting objects.
Advanced object representation methods wiU also be explained which are
closely associated to contact detection procedures.
Chapter 8 summarizes the remeshing techniques and principies of crack
initiation and propagation phenomena and discussions on the problem of
size effect in fracture analysis of solids, softening and damage mechanics
methodologieH n.s well as lntroducing a three dimensional anlsotropic mar
terial model which consists of an 'stress return algorithm and deriving the
consistent tangent modulus.
9.
Finally, Appendix A is devoted to describing the data file structwt:: fr a

2D finite discrete element code called UTdem. This is not primarily aimd
as a professional general purpose finite/discrete element code, but to provide
the user with _an overall insight of how a discrete element code may be.usad.
for modelling progressive fracture analysis of st.rnct.un's as well as granular
flow simulations.
1.4
Notations
In this section, the adopted vector/tensor nntntions are briefly explained.

These notations are used throughout the book unless it is explicitly stated
otherwise.
~
l. Both component and compact forms are used for representation of
vectors and tensors.
2. Only cartesian reference axes are used in general.

3. Vectors are bold faced lower case italic letters; such as n,
t, i.
4. Second order temmrs are represented us liold f11n~ upper case itallc
letters, such as D, N, A. A few Greek bo1dmath letters, however,
are used for stresses (u) and strains (e).
5. The transforma.tion of a. vector by a second order tensor into another

vector is repmmntnd by sitnpt juxtapo:-;it io11,
t =un
or
t = Tn
6. The transformation of a second order tensor into another second order

tensor by a fourth order tensor is representcd by simple juxtaposition,
a=Ee
or
H=DN
7. The inner product of t:wo vectors is represeuted by a ,

(j.
8. The cross (vector) product of two vectors is represented by a x,
axb
9. The tensor product of two vectors is a maJrix d<'li1H~d as
A=xy
Aij
XiYj
10. The inner product of two second rank

A: B = tr(ATB)
ternmrs i8
represented
by :
Chapter 2
Discontinuum Mechanics -
A Review
To attain a realistic ovorview of the extent a co11 t.acl. lin~a '( 1n.lgorithm can be
used for analysing variom; academic, engineeri11g awl iwlustriu.l problems,
a quick review of potential applications are provided. It is not intended,
primarily, to compare the results w.ith available data in the literature, as it
is usual in academic papers, but to illustrate to the reader the applicability
of the method to different applications that rnay be analysed by the use
of the computational cficoutinuum mechanics. It. is 1t!Ho airncd at sparklng
new ideas for further research and future challengcs in this subject.
This chapter reviews the following engineering applications, amongst
many others, which are currently being researclwd in many research instit.11tinm1 thro11p;hn11t. t.lw workl,
.
Geomechanical applications
Granular materia.Is
Impact analysis (progressfve fracturing)
Particulate flow
Computer graphics
11
I'..!
1
1
1
1
1
Discontl1iuu111 Mcd1auics
2.1
Geomechanical Applications
2.1.1
Rock Blasting
(,'Jwpter 2. Visrn11ti111111111 Moclmnics - A llcvil'11
"
13
.,~'+-:/+:*,'
.
,....
1
'
' ....
'~
'
'
Ho<'k blastiug is an interesting area for application of discrete element

111..ihrnl. Bt>s11lts shown in this scct.iou llrn takm1 from [18J, 1>11.'ied on UHi111'. a si111plilitd solid rock - ddo11atio11 gall iutemdion models.
1("')
4.100.
.... u.a. - 0.000
~
,..._.owi
l:SOO(m)
Ul(m)
711n)
111("")
-11.00
\11' ......
! ,...........
0000
4.000.
4.100..
~
lllltlmlOO:'!
l'Mldll0004
Figure 2.2: Fragmentation patterns in a multiple bench blasting simulation [18].
1
1
1
1
1
ruJ tun 0011
...a-0.0JJ
Figure 2.1: A 2D bench blasting simulation [18].
,,
\.
'
be an infinite boundary rnodelled by a nn-reftecting boundary condition.

These results may be combined with results obtained for bench in order
to estimate fragmentation pattern more accurately. Optima! fragmentatlon
can then be obtained by estimating the correct spadng, wliiC'h is one of the
important geomctry parameters in a practica! lw11d1 blasti11g problem.
2.1.2
Mining Rock Bursts
In ord(r to eliminate reflection of waves, sorne artificial damping has

been introd11ced close to the fixed bottom and left edges, thus modelling
them as non rdlecting infinite boundaries. Other edges are free surfaces.
Rock burst is an energy phcnomenon accompa11ri111~ ro('k fail11rn in the form

of brittle fracture induced by mining. Energy diangcs oc:cur during the
course of mining activities and a part of the energy stored in the rock mass
surrounding the excavation. The ultimate stress state is created by the
action of the pre-mining and mining induced stresses. The changes brought
about by mining merely trigger latent seismic l'H111ts that. derive mainly
from the strain energy produced by geological ditferences in the state of
stress.
Figure 2.:2 illustrates the fragmentation patterns in a multiple bench

blasting prohkm. In this problem, only the bottom edge is assumed to
The rock burst is a sudden manifestation of tlie release of strain energy

stored in the rock mass. The possible sources for the liberated energy a.re:
D111..!1 hla.o.;ting is a very frcq11ent rock blasting operation. In this examilc~ a :lD prublem is modelled. The initial geometrical configuration and
pro~essivP fracture patterns in successive times from the initial blast are
rlepi2ted in Figure 2.1.
14
Chapter 2. Discontinuum Mechanics - A Review
11
Figure 2.3: Exp<~rirnental failttre rcsults nnd numerical simulation of a simple rock bursting problem [20).
-1. StorPcl Htrnill Pnergy in
th~
1mrrounding mass of an opening.
2. Change in tle potential energy of the rock mass.

3. Minor slippage along rock contacts.
4. Gas extrusion.
5. Dynamic sources away from the opening.
The rock bursting process can well be simulated by the combined finite/discrete element technology [19]. Because of loss of many peoples Uves
after the mining rock b,umps, this field has remained a very important and
active part of mining 4ineering.
Figure 2.3 illustrates sorne experimental aml numerical simulation results reported by Hazzard et al. [20) based on [21, 22, 23].
2.1.3
....'
Failure of Cut Embankment
'
~-'
Figure 2.4: Failure of cut embankment [24].
In this example, a wall of 2600 equi-size spheres betweeri two vertical plates
was rnodelled by O'Connor (24]. Gravity was first applied for 1000 iterations
ata time step of one milliseconds. At t = ls, the left boundary was suddenly
removed. The ensuing collapse configurations are shown in Figure 2.4. The
Hirnula,tion WClH l'Ull for a total Of 3000 iteratiOllS [24].
15
16
Chapter 2. Discontinuum 1\-:fechanics - A Review
Figure 2.6: Remeshing procesa and the 45 shear band development in a'
tensile plate undergoing large lateral necking phenomenon.
Figure 2.5: DDA slope stability analysis for various discontinuity patterns
and friction angles [26].
2.1.4
DDA Slope Stability Analysis
The discontinuous deformation analysis (DDA) was dev~loped by Shi [25] ,,

to model the behaviour of discontinuum media. The DDA is based on dynarnic eqilibrium, which considers the kinematics of motion of individual
blocks [26]. In DDA, each discontinuity (joint) may be assigned a different friction angle, cohesion and tensile strength. The blocks are assumed
to be elastic deformable bodies contacting to neighbour blocks through the
penalty enforcement of the impenetrability constraint. The DDA may be
assurned as a special case of the more general finite/ discrete elernent technique.
Figure 2.7: Two shear band slope instability problrn1s [28, 29].
The concept of shear band deformation can be bcst understood from

Figure 2.6 which depicts the deformation process of a simple plate with an
initial circular hole subjected to a set of tensile forces [27].
t
Figure 2.5 illustrates the results of slope stability analysis for various
discontinuity patterns and friction angles based on the discontinuous cieformation method as reported by MacLaughlin et al. [26].
Figure 2. 7 illustrates two different examples of slope instability simulations performed by Stead et al. [28] and Cramer et al. [29] in two and
three dimensions, respectively. In the 2D case, an h-adaptivc finite element
method has been adopted, whereas in the 3D example, only large deformation theory has heen considered.
2.1.5
2.1.6
Shear Band Slope Stability
Falling Objects
lJ
In contrast to the DDA method, a shear band slope stability analysis may
be performed by using a fully deformable nonlinear finite element simulation. An adaptive remeshing scheme has to be employed to avoid excess
distortions of the finite elements close to the highly deformed shearing band.
Figure 2.8 shows the application -of the discrete element approach to a full
three dimensional simulation involving randomly shaped grains falling under
gravity into a box. Two sides of the box on which sorne particles have come
to rest have been removed for rendering purposes [30, 24].
-------------------------------
18
------
Chapter 2. Discontinuum Mechanics - A Revienr
Figure 2.9: Porous objects with holes [31].
Figure 2.8: Discrete elemcnt modelling of falling objects [24].
2 .1. 7
LJ
Porous Media
Anotltor internsting application of the contact mcchanics is the geometric

modelling of porous rnedia. Modelling materials such as permeable soil or
rock requires the creation of a geometric representation of the solid matrix
enclosing the interstices. For a permeable model, the interstitial spaces must
form networks that cross the boundaries ofthe bulk material. Th model,
ca.u then he ui;ed as the input model for numerical experimente ~uch as fluid
flow sirnulations [24].
The model is created by taking a solid cast of the pore spaces in a densely
packed assemblage of particles. Due to the connectivity of the initial particle
assemblage, the pore spaces of the cast can be guaranteed to connect across
tlw honndaries, ensuring a permeable model. In order to obtain valid void
Figure 2.10: 3D particulate packing in a porous medium [24].
19
r
20
Chapter 2. Disco11tinuum Mecbanics - A Review
21
:;{
{~':
geometries, the modelling process must first generate a ma.trix that is in

mechanical equilibrium. Only then can we take a cast of a foid space that
is geomctrically valid and exists as a stable mechanical syatIl1 [32}.
Porous solids such as models illustrated in Figure 2.9, can be used for
simulating particulate flow in porous media, as shown in Figure 2.10.
2.2
Granular Materials
2.2.1
Granular Flow in Silos
2.2.2
Superquadric Elements
Supcrquadrics are a f.mily of parame.tric funct.o11:-;, rffn11t ly proposed for.

use in multibody dynamic8 analysis [32]. Tlwir deiuiLio11 aml interesting
characteristics will be discussed later in this book. Here, a typical applicer
tion using superquadric elements is illustrated. Figure 2.12 depicts a deposition simulation using this kind of object representation technique within a
discrete element analysis. As it is clearly seen, various shapes of the objects
cu.u he obt;a.itwd wltilo n Hhuila.r formulation is wa!d for 11.ll 11f t.hom.
Silos represent a vital part of the industrial infrastructure. Failure of a silo

often causes great economic losses either by wasting the ensiled materials,
delaying production Unes C>r disrupting transportation plana. In this ex-
11.111plo, t.lin predict.iou of pressuro nnd flow in siloa has been iuvstlga.tod
11tiliiing t.ho discrnle element nwthod. Silo and granular material are both
modelled in this approach as depicted in Figure 2.11 [33, 32). The results of
typical cond ucted analyses may be used to guide the silo design procedures - ~
by pointing out any u.Jl8onticipated loading conditions and pressure distributions which might altse during operation, as well as phenomena such as
arching, different filling/emptying regimes, seismic loading, etc.
Ttae:7.34281
Figure 2.12: Discrete element modelling ofa typical deposilion problem [3~j.
TtH:U.1111
2.3
Impact Analysjs (Progressive Fracturing)
41
)'
f
,b'
,:..
"':
~.
Progressive fracturing and fragmentation analysis of impact or explosive

loadings of structures have benefited substantially from the development .
of the discrete element method. In the follow i11g sorne of the interesting
applications are briefly revimved.
2.3.1
u
Figure 2.11: Discrnte element modelling of granular flow in a typical silo
[33].
Penetration of a Missile
Structural design of a shelter, a.rmoured military Pqnipnwut and safety measures for bullet-proof vests may force a designer/a.nalyst: t.o check for impenetrability response of a structure subjected to a high velocity object.
A complete analysis of an object penetrating a structure and developing
22
Discontinuurn. Mecha.pies
23'
Chapter 2. Discontinuum Mechanics '":'A Review
cxtensive damage in it, has only become possible by the use of combined
finitc/dii-;crnte clement techniques.
Figure 2.13, illustrates how the crack patterns are propagated within a
typical ceramic plate as a bullet penetrates the plate in different time steps.
Figure 2.14: Shear band development and chip creation in a metal cuti.i~s
application [34].
, Fig11rc~ 2.1 :~: Progrcssive fracturinp; in n structurn impa.cted by a high velocity h11lld.,
2.3.2
Metal Cutting
Metallic alloys are particularly susceptible to adiabatic shear localizatidh

due to their very low thermal diffusivity and associated high sensitivity of
the yield stress with respect to temperature. These effects are amplified in
high speed machining dueto the high rate of energy generation (34].
1'
i:
1'
l__.,J
Tlw present cxample simulates machining of a Ti-6Al-4V titanium alloy

sampl_e. Adopting an error estimator based on the rate of fracture indicators,
has resulted in thc evolution of the mesh refinement process as illustrated
in Figure 2.14. Plastic deformation first takes place near the free surface
of the workpicce. 'l'he advancing tool ca.uses the development of the shear
lrnud. At lat.cr 8tages, friction between thc chip and tool causes an increastf
of plastic deformation near the tool/ chip interface which may give rise to a
secondary shear zone.
Figure 2.15 illustrates another example of simulation of a blanking proccss. In this exarnplc, an adaptive remeHhing procedure has allowed ~or the
\\
\\
L'
Figure 2.15: High speed metal cutting problem [35].
24
25.
Rimulat:ion of tlw whole cutting process from the first contact until the onset
of total separation [35].
'
2.3.3
Demolition of Structures
DcmoliUun of structures is among the group of problems that we require a

realistic assessment of fractured regions and to what extent the scattered
materials are spread; to ensure the safety of neighboring structures. 'J:h!s
kind of problem always exhibits extensive fracturing and fragmentation.
Figure 2.16 givos an exam1)Je of a ta.U concrete chimney tower. A mass

of specific explosive is detonated within a tunnel in the base of the tower.
The problem was simulated by Munjiza [18). At first, the basis of the tower
is fractured, then the tower starts falling. Due to inertia forces it breaks
in two pieces and continues falling until a crash to the ground happens;
resulting in further fragmentation of the tower (18, 36).
2.3.4
0.000
0.200
Composites
Composite lamina.tes are now widely used in many applications involving

dynarnic loadings notably in the aerospace industry. One of the major
problenis that affects'. the design and performance of composite materials
for structural applications is their vulnerability to transverse impact which- . . . ._
may cause substantialinternal damage of the component due to matrix
cracking, fibre failure !nd delamination. There is, however, agreement that
the ni.ost dorninant causes of damage during impact are matrix cracking
coupled strongly with complex mode delamination mechanisms (37, 38].
Delaminated sublaminates are also particularly susceptible to dynamic load
buckling when subjected to impact loading. lt is evident that in addition to
the slenderness of each sublaminate in comparison to the laminate itself, the
local transverse impacts by adjacent plies will increase the risk of buckling
(see Figure 2.17).
As another example, the results of a 3D delaminatin analysis of a composite plate subjected to impact loading are presented. Figure 2.18 illustrates the cornbined finite/discrete element modelling [40]. The composite
ply pattern is set to [90n, On, 90n, On, 90n) The impact loading is simulated
by a triangular load with a variable peak force.
Figure 2 .19 ill ustrates the de bonding patterns at different layer interfaces
for two different stages of the loading. Delamination patterns are clearly
developing from the central region of the plate, i.e. the impacted zone,
towards the eclges of the plate. These figures depict only the DE part of the
whole mesh.
real time= 10.400
real time
=11.200
Figure 2.16: Demolition of a chimney tower subjected to explosive load-;

ing [18].
26
27
Interface between layers 1 and 2.
Buckllng of a delamlnated !ayer.

Mu.t.1!rlal crn.cking anc) fru.f.(ruontatiun.
Figure 2.17: Different failure modes in composites subjected to impact loading [39].
11n Mesh
Interface
betweerr1a~rs
2 and :3.
Figure 2.18: Combined FE/DE mesh of the composite plate.
It should be noted that these results were achieved without considering

a material fracture a.11alyHis, and only the bonding fracture was activated.
In a practica! test, however, the illustrated large deformation will certainly
involve extensive material fracture.
2.3.5
Masonry Structures
Severa! interesting implementations of the discrete element method have

been proposed for predicting the bahaviou~ of masonry structures [1, 41,
42, 43]. Nevertheless, the predictive modelling of the nonlinear behaviour
of manomy structures remains a challenge, due to' their semi-discrete and
composite nature.
t\) T = 0.00006 1mc.
h) T
0.00012 sec.
Figure 2.19: Delamination patterns at layer interfaces at different time

steps (39J.
'28
Discontinuum Mecban.ics
29.'
FiKlll'U :J.20 llhwt.rn.t.nH t.hn ld11nll~ud prohJ0111 ol' f.110 f11il11n~ of 11 t.wo tept\ll
bridge studied and simulated by Petrlnic [1). Tliu prolilrnn was treated as
a combined finite/discrete element problem with the masonry blocks being

represented by deformable discrete elements in frictional contact, the fill mB.-'. '
terial by spherical discrete elements and the foundation region was modelled
as a Mohr Coulomb material by conventional finite elements. The failure ;,
mode brought about by settlement of the central pier aml the traversing of '
an abnormal two point load is indicated.
As a real simulation, Figure 2.21 shows a 60-year old railroad two span}
masonry bridge and the finite/discrete element model. The bridge wes ::
incrementally loaded in place until severe cracking and la.rge bridge key '
deformation were observed as reported by Marefat et a.l. [44].
A combined finite/discrete element simulation was performed to simulate

the failure behaviour of the structure. Cracking pattems similar to the test
observations were predicted according to Figure 2.22 (45).
Figure 2.20: Discrete element modelling of masonry arch failure [1].
Figure 2.21: A two span masonry bridge and the finite element model [44).
30
Chapter 2. Discontinuum Mechanics - A Revienr
..1 .....
Figure 2.22: Crack propagation patterns at different times [44).

Figure 2.23: An experimental collapse test for a masonry bridge.
2.3.6
Repair Modellings
Strathmashie Bridge, 150 years old, was of rubble masonry, in reasonable

condition and showing little distortion, but there seemed to be very little
mortar in parts of the arch.
An experimental test was performed to assess the performance of the m~

sonry bridge until the collapse of the structure, Figure 2.23. A numerical
simulation was performed by Klerck [43] based on a combined finite/discrete
element technique, as depicted in Figure 2.24. The failure modes are interestingly similar to one observed in experimental test [46].
A repair plan was proposed based on the use of anchors to reinforce the
existing bridge (Figure 2.25). A combined finite/discrete element method
is capable of modelling the whole structure (old masonry structure and
new anchor elements), taking into account the bonding phenomena that
umy exist bot;ween anchors and masonry units [43]. Figure 2.26 depicts the
simulation of a pull out test, which is one of the standard design tests for
anchors.
Figure 2.24: Finite element ~imulation of tlw collapsing bridge (43).
Figure 2.25: Strengthening the ~asonry bridge by anchoring.
31
---
- - - - - -------------------
Cbapter 2. Discontinuum Mechanics - A Review
Figure 2.26: Finite element simulation of anchor pull out test [43J.
2.4
Particulate Flow
2.4.1
Shockwave Analysis
A wall of 2GOO cqui-sizc spheres is constructed by placing the spheres between two vertical plates. Gravity was first applied for 1000 iterations at
a time step of one millisecond. A stream of slightly denser particles of the
same size- are dropped under gravity from above the wall. On impact, the
stream creates a series of stress waves which are shown in Figure 2.27. Later
in the simulation, the wavefront has been reflected off the boundary walls
and interferes with thc waves still originating at the source of disturbance
[24J.
Figure 2.27: Shockwave propagation in particulate fluid [24].
33
2.5
Other Applications
1t
1
Chapter 3
The discrete element methodology is increasingly being utilized for a wide

variety of new industrial applications. Grouting, shotcrete modelling, supporting systems, shot peening process [4 7], hot forging problems, .printing and applications involving smart materials are amongst the engineering
problems while animation, virtual reality [48] and artificial intelligence technique are among various computer science applications that have benefited
from the development of the contact mechanics and the discrete element
rnethodology. An important area of research is the development of efficient
parallel processing finite/discrete element algorithrns for modelling of large
scale discontinuum problerns [49].
Constraint Enforcing
Methods
1
1
3.1
./.;:'
Introduction
:,; . r
Many different methods have been developed for enforcing a constraint con,- :e:
dition on the governing equation of a well established physical behaviour. ~: .
In this chapter, the following four rnethods for enforcernent of constraints
within a finite element analysis are reviewed:
Penalty method
Lagrange multiplier method
Perturbed Lagrangian rnethod
Augmented Lagrangian method
1
The final part of the chapter describes the peualLy mcthod in more detall ,
and explains the stability of'method for contact analysis and the associated
contact damping techniques.
3. 2
Definition of a Constraint
A constraint either prescribes a value for a freedom (single point constraint)

or a relationship between two or more freedoms (multipoint constraint).
LJ
35
36
37
Chapter 3. Constraint Enforcing Methods
Bodyl
Body2
Figure 3.2: Nodes A, B and C to stay in a straight line.
-UB
UA
{ UA.-UO
Contact
Constraint
=o
=0
Rearranging it into the standard format (3.1)
Figure 3.1: Impenetrability constraint for two contacting bodies.
-1
{3.3);
o
Fiv;11re :3.1 repro:.;entH a. t~~cal impcnetrability comitraint between two contacting bodies. This cons~aint defines the necessary condition~ to prevent
the bodies from penetrating each other.
where
[~
The mathematical description of a constraint equation may be written

in the form
Cu=Q
-1
(3.4)
is the matrix of constraints for this particular problem.

(3.1)
where C is a matrix of constraints, u is the vector of freedom and Q is a

vector of constants. Q in many cases may becomc a rmll vector.
To further clarify the numerical interpretation of (3.1), three constraint

examples are discussed.
Example 1
According to Figure 3.2, a series of nodes, i.e. A, B and C, are to be
constrained to stay in a straight line, e.g. for symmetry reasons. The
constraint equations may then be formed as:
Example 2
Figure 3.3 depicts a model combined of a continuum m<sh anda structural
mesh. To define the connection between a structural element to continuum
elements, i.e. to link the rotational freedom of the structural element, </>2
in Figure 3.4, to the displacement freedoms u 1 , u2 and u3 of the continuum
model, the rotational equilibrium about node 2 can be cowddered:
-lu1 + lus - </>2
=O
Therefore, the constraint equation can be forn1ed
(3.5)'
38
nodes 2 and 3 are
Rigid Rotational Link
39
just in contact
u =+0.J
~1--------2~3-------4k1 =JO
k1 =JO
Continuum
Element
Figure 3.5: A simple two bar mode!.
Structural
Element
l
;
.
[-l
-1)
U ]
~:
(3.6)
Figure 3.3: A combined continuum/structural model.
Example 3
Two straight bars which are just in contact are depicted in Figure 3.5.
Each node has a single degree of freedom along the bar directfon. A o.i
unidirectional rightward displacement is applied to node 1 of the left bar. ;
Each bar behaves as a linear spring, so
-10
10
(3.7)
l
The assembled system of equations will be
10
-10
_ _....... U3
[
:
o
o
-10
10
o
o
10
0
-10
-10
10
l[ l [ l
U U2 0.1
U;l
U4
=- ()
t
0
(3.8)
f.1
Figure 3.4: Rotational equilibrium about node 2 should be satisfied.

since the equations are uncoupled, the results will be:
U2
{ U3
0.1
=0
(3.9)
40
41
Chapter 3. Constraint Enforcing Metbods
0.1
1-----i
l-----..2
Figure 3.6: Uncoupled solution for the two bar problem.
n
u
Figure 3.6 shows the deformed shapes of the bars for this analysis, which
clearly shows overlapping the elements.
Body2
To avoid this, the following constraint equation should be enforced:
(3.10)
We will later use this simple example to verify the methods adopted as
constraint enforcing methods.
3.3
(3.12)
Constraint Enforcement
Equation (3.1) should be added to the conventional equations of the system

and solved simultaneously. Different approaches have been proposed for
solving this set of equations which will be briefl.y reviewed and compared.
One approach, which has been widely used by many researchers is the
concept of n~inimization of the total potential energy for deriving the nec~
cssary n<Jmtttons. Thn total potential mwrgy of a lin<~ar elastic RYHtem subjecte<l to static loading and consisting of two discrete bodies
and n
2
.
(
' 1
may b e written as Figure 3. 7)
Figure 3. 7: A system consisted of two interacting bodies.
where u is the nodal displacement vector, K is the system stiffness matrix

and R is the force vector. Without additional co11strai11t equation, bodies
n1 and n2 do not interact and the system is uncoupled.
In an alternative approach, imposing the constraint equation is performed by means of variational principies [50]. In this approach, we can
introduce the constraint by forming a constraint functional, ncon' in addition to the functional of the system (51)
= Il{u) + ncon(c, u, ... )
l
1
'
l_J
. (3.11)
Using a standard discretization procedure based on appropriate tria!

functions
where Il{u) is the functional of the unconstrained syi:;tcm. The variational

~ '.. ;
form of (3.13) will then be
fI = il + ncon
(3.14). :;
'.,:'
'J~i~~,,k;
i,
which is valid providing TI = O and the constraint is imposed, simultaneously. Further development of the formulation requires a clear definition of
the constraint equation, e, and the constraint functional, rrcon' in equations
(8.13), (3.14).
e= sign(n)[l
42
43
- 1]
(3.17)
(3.18)
3.3.1
Impenetrability Constraint
This constraint defines the necessary conditions requirE;?d for preventing the
contacting bodies from entering into each others boundaries. In this section
only a frictionless contact is considered.
Figure 3.8 shows two bodies, 1 and 2, at two successive time steps. At
time t, the bodies are yet to make contact and there is no penetration.
However in the next time step, the deformation states that without any
additional constraint the bodies will penetrate each other.
deformation,
(1)
It is important to completely clarify the nmnerical irnportance of the g

function. In a general contact problem, if the constraint equation Cu = q
is exactly satisfied,, we will have Cu - Q = O. However, in almost any
numerical analysis of a contact problem there will be sorne penetration
which could be interpreted as the g function
Augmented Lagrangian Method
be written as
re= r 1 n r 2
Lagrangian Multiplier Method

Perturbed Lagrangian Method
Figure 3 .8: Potential penetration of two contacting bodies.
3.3.2
(3.15).
where re is the common contact surface of two bodies, x 1 and x 2 are the
<Mormed positions of bodies 1 and 2, respectively, nis the normal to body
2 aml g is the gap function. Equation (3.15) specifies the gap in normal
direction.
In a one dimensional contact between two nodes of the bodies, the equations are sirnplified
g=Cu
(3.19)'
Penalty Method
-g-(2)
on
=o
tem and solving the resulting coupled set of equations. Hen\ four methods
whlch lmvo IHi<'n froq11011t.ly llf.16d ln vn.rlous co11t.1w1. 1w11l.vN1,~1 1m1 closcrlbad:
timet+dt
-.:::--..-----~--&
...~
~ay
=:u 2 -u 1
There are severa! methods for enforcing a constraint condition to a sys-
~
~.
The necessary constraint equation
(3.16)
Penalty Method
The Penalty method was probably the first approach adopted for a constraint enforcing method. It was developed by Hallquist and his colleagues
in Lawrence Livermore National Laboratory during the late seventies for
modelling impact/contact problema.
1
To obtain the necessary equations, and comparing to the first term in

Equation (3.12), the total potential energy for a constraiued problem can
be written as
1 T
TI( u)= II(u) + '2g ag
{3.21)
44
LJ
where a is a normal contact stiffness, called penalty number, and in general

is a diagonal matrix: of penalty terms for each degree of freedom. g is
the normal gap vector and for g = O the constraints are fully satisfied;
(u) = Il(u).
(3.22)
- [IT
-+gT a8gl
- fo
To maintain equilibrium,
8u
Bu
should be equal to
Remarks on Penalty Method

The main features of this method are:
The constraints are only satisfied in an approxima.te manner and the

correct range of penalty numbers have to be chmmu. If n is too low,
the constraints are poorly satisfied, while if et i::; too large, the stiffness
matrix becomes poorly conditioned {the difference between in and out
of diagonal terms becomes very high). Asan initial estimate for a
(3.23)
0.5E
The first term on the right hand side of (3.23) is the well known stiffness
equation
an
au
(3.24)
and for the second term, we have
g=Cu-Q
(3.29)
For explicit dynamic applications, large values of a rnay result in a

reduction in the critical time step. Large penalty values simulate stiff
constraint spring, increasing the global stiffness and so reducing the
required critica! time step.
ag corresponds to the penalty force required to enforce the constraint..
An example
(3.25)
To illustrate how to use the penalty method for enforcing a constraint equation, Example 3 of Section 3.2 is re-considered (Figure 3.9).
(3.26)
(3.27}
Therefore, the modified stiffness equation will be
[K + CTac] u= R+CTaQ
< a < 2.0E
where E is the young modu!!is gf the contacting bodies.
O.
-=Ku-R
45
Enforcement of constraints requires no extra equations.
Minimization of the total potential energy will result to
il=
Chapter 3. Constraint Enforcing Metbods
(3.28)
The term, Gr aC should be added to the system stiffness matrix to

incorporate the irnpenetrability constraint stiffness.
,nodes 2 and 3 are

just in contact
~
-1------2-3----4-:
ki =JO
k2 =JO
l
Figure 3.9: A simple two bar model.

1
46
47
The comitraint equation may be written as
0.0333
(3.30)
with
(3.31)
0.0667
...
for a constant value of a

Figure 3.10: Re~aining penetration in a two bar contact problem using the .
penalty method.
aCT C=a [ _1
1
{3.32)
the results show the existence of sorne penetrat.ion of bar 1 into the bar 2 ., ,
(See Figure 3.10).
which is similar to the stiffness of a spring attached to the bars.

The assembled system t>f equations will then be
10
-~10
Table 3.1 summarizes the results obtained for the same equation using
different penalty numbers. It is dearly seen that by increasing the penalty
number the solution converges to the exact solution.
-10
o
10 + a . -a
-a
lO+a
(3.33)
g_,-10
solving for unknown
-a
u2
and
-a
10 + O:
l = l
u2 ]
U3
1
0
(3.34)
3.3.3
u3
An undetermined Lagrange multipliers method can be used to find the min

imum of a multi-variable function.
(3.35)
U3
0.0333
0.0476
0.0497
0.0500
Table 3.1: Results for different penalty numbers.
To continue the solution procedure only consider the active part

10 + a
U2
0.0667
0.0523
0.0502
0.0500
100
1000
10000
Equation (3.33) is a coupled equation due to the existence of nondiagonal a terms.
a
10
Here, the constraint is introduced by adding a new term to the total

potential energy
To get sorne numbers, for a = 10
?/,'},
[
'll3
l=
_1 [ 20 10
300 10 20
l [o l
1
(3.37)
= [ 0.0667 ]
0.0333
{3.-36)
in whfoh ..\is the vector of Lagrangian multiplien;. ..\ could be interpreted as
48
Cllapter 3. Constraint Enforcing Methods
the force required to enforce the constraint. In this method, new variables
L_)
Similar methods, such as the projection technique, ha.ve also been

developed to solve the Lagrange multiplier equatiou approximately
are added to the system of equations and the constraints a.re satisfied if
g = O for which = II.
(53].
Lagrange multiplier method did not necessarily preserve a smooth
force distribution across interfaces. The lack of a smooth force field
has excited the zero energy modes in sorne of the prirnitive elements
Minimizing the total potential energy with respect to u and A
-= an-au + TY]
au
.X -
ull
8.xfi = g.X
[54].
(3.38)
Example
(3.39)
Again we return to Example 3 of Section 3.2 (Figure 3.9). The constraint

equation has already been derived
ancl using a standard discrefzation technique will result in

(3.40)
()11,
(3.43)
wlt.h
(a.41)
Equation (3.41) ensures achieving an exact solution (complete constraint

enforcement). The final set of equations could be written in a matrix form
(3.44}
The assembled system of equations will then be constrnctcd from (3.40)

(3.42)
Remarks on Lagrange Method
Thr. main fcaturm; of this rnethod are:

The coustraints are satisfied exactly.
-10
10
10
-10
-10
10
~l
-1
10
-10
l[~~ l [~ l
U
U4
0.1
= ()
f.1
(3.45)
Solving for the active part
'-'.
Additioual unknowns are introduced. For general contact problems,

thc uumbcr of extra equations is often continually changing.
{3.46)
The equations associated with the Lagrange multipliers have to be

interlaced with the equations of the nodal freedoms to prevent large
increases in the bandwidth of the equations.
L.J
Thc extra equations associated with the Lagrange multipliers have

zero terms on the diagonals. Therefore, the constraint equation can
not be eliminated until at least one of the system variables associated
with the constraint have been eliminated. As a result, the bandwidth
minimization, assembly and elimination procedures are much more
complex [52].
the unknown variables
u2, U3
and arthen calculated
as
(3.47)
which are the exact values for
u2
and ua.
50
.
3.3.4
111111111111111
Cl1apter 3.
Constl'l:tint Enforcing
1888J'988674'2
..,a JJl,a 41...,bS
Perturbed Lagrangian Method
[~ ~][~]=[~]
'l'hiH rnethod iH un extenslon of tho clusHical Lagrange multlpller method
Methods
where an additional tertQis added to the total potential energy of system
Note that because of A no exact solution ca...'1 he found.
lJ
(3.48)
where ~,\TA,\ is added to prevent zero diagonal terms, and A is a diagonal

matrix -of the form
1
Aii = -
{3.49)
Cli
where ai are penalty parameters associated with each constraint equation.

Minimizing the total potential energy with respect to
u and ,\
-= [Il
8gl 8u
-+,\
-Bu
Bu
1'
uIT
8,\fi
[g - A..\) 8,\
(3.50)
Remarks on Perturbed Lagrangian Method
The main features of this method are:

Similar to the penalty method, direct solution of the syst~m' by this '
method will.not result in exact enforcement of the constraint equa.- '
tions.
The approach is always stable as long as the a term are not infinitesimally small numbers (so there will be no zero diagonal terms).
In penalty method, II(u) is minimized with the constraint g = O.
However, in perturbed Lagrangian method, I1(u) is minimized under
the condition (g - A,\= O). The algorithms are similar as a-+ oo.
As a --+ oo, the solution approaches to the exact solution similar to

the Lagrange multiplier method.
(3.51).
In perturbed Lagrangian method, ,\ is au explicil ddiuition of the

contact force
a.nd using a standard discretization technique will result in
(3.55)
Ku-R+C T ..\=O
(3.52)
--
Example
Cu-Q-A,\ =0
w Itere
the followiug rclations have been used
(3.53)
Taking up the Example 3 of Section 3.2 (Figure 3.9), the assembled system
of eq uations will be
an -K
au
u- R
[ -10
10
g=Cu-Q
!!JL
au =C
Expressing Equations (3.52, 3.53) in a matrix forro
o
o
o
-10
10
o
o
o
o
o
o
10
-10
0
-1
1
-10
10
-1
with active part
.=.!
a
l [ll
[U=
U2 0.1
U3
U4
=0
,\
(3.56)- '
~o
1o1
-1
~1 J [, ~~A J = . [ O~ ]
=-!.
minimizing the with the constraint g = O where the penalty term

is treated as another stiffness contribution.
(3.57)
Minimization of the total potential energy (3.58) will result in
. Table 3.2 summarizes the results obtained for the same equation using
chfferent penalty numbers. lt is clearly seen that by increasing the penalty
number the solution converges to the exact solution.
u
Q
10
100
1000
10000
U2
U3
0.0667
0.0523
0.0502
0.0500
0.0333
0.0476
0.0497
0.0500
A
0.0333
0.4762
0.4975
0.4998
fJg + o7 o <>" ] fo,

,
n. (:~;. [an
- +~ r -8u
l)u
tlu
1
or in a discretized matrix form
K:.u - R+ CTA + CT aCu- CT o.Q =O
(3.62) .
or
Table 3.2: Results for different penalty numbers.
(K
a.:J.5
53 :
Cha.pter 3. Constraint Enforcing Methods
+ cT nC) u = R -
C'I' (nQ . . . .X)
(3.63)
Angmouted Lagrunglan Mctbod
The objective. of this method, which was developed in the early nineties
is to minimize the disadvantages of the penalty and Lagrangian multiplie;
methods by combining the two methods. The method is independent of
the magnitude of penalty number and introduces no extra equation to the
system.
If it is desirable to avoid adding any new equation to the system, then

Equation (3.63) should be solved for both u and ..\ simultaneously. This is
only possible with an appropriate iterative solution. For insta.nce, beginning
with A O and using the iterative form
(3.64)
The total potential energy of a constrained system can be written as

where
-
IT(u, .\) = Il(u)
1
+ tXT g + ?,YT
ag
(3.58)
~A
is updated from the nodal variables in cach iterntion.
Following the above strategy, the total potential energy of the system in
the Lagrange multiplier form can be written as
The solution may be interpreted from two points of view

(3.65)
As a da..';.;Hical peuu.lty method

Equating (3.65) to the augmented Lagrangian fonn (3.G8) results in
-
1 T
Il(u, .\) = Il(u) + -g ag
minimizing the
(3.59)
~3.66)
with the constraint g = O.

This suggests that a suitable relationship will be of the form
As a Lagrange method
1
(3.60)
(3.67)
Discontinuum MechMics
54
Box 3.1- Iterative procedure for Augmented Lagrangian Algorithm [52].
=O
1.0
Initialize Al
2.0
For each iteration k

2.1
2.2
+ CT aC)u =
R-CT(aQ-Ak)
Compute penalty forces
Example
fe= 6..X = ag
2.3
Check for convergence
Working on the same Example 3 of Section 3.2 (Figure 3.9), the active
TF 16.-\I ~ TOL THEN END

2.1 Updn.te Lu.graugo mult;iplicrH
Ak+i
2.5
An iterative solution procedure is required. However, for contact problems an iterative procedure will usually be needrnl due to other con;:
siderations.
The algorithm can be used effectively for applications where the con~
tact pressures become very large in comparison with the material elastic parameters, i.e. with rubber materials.
Solve for displacements
(K
HyHt.nrn of equnt.iom1 will lw
= Ak + 6A
10 + a
-a
GOTO 2.0
-a
10 + a
l[
u2 ]
u3
=[1]
(3.68)
- .\ [ -1 ]
1
Assuming a value of 10.0 for a

When gis satisfied, 6.X will be zero. The procedure may be viewed asan
iterative transfer of the constraint enforcement from a penalty to Lagrange
form.
20
[. -10
20
U2 ]
U3
= [ 1+A
-.\
2+,\
30
Starting the iterative solution with the initial value of )q
Iteration 1
The- main features of this method are:

The augmented Lagrangian approach uses a penalty ~iffness but iterati vely updates the contact traction to impose the contact constraints
with a.specified precision.
No additional equations are required.
Large penalty values are not required; avoiding the ill conditioning of
the stiffness matrices.
The constraints are satisfied within a user defined required tolerance
.A1 == O
(3. 69 )
----+ {
(3.67)
----+
U2
U3
Iteration 2
= 0.06~7
= 0.0333
Li,X = ag = 10(0.0333 - 0.0667)

,\ 2 = .\ 1
+ Li.A =O -
( 3.69 )
--+ {. U2
= 0.0556
U3
= 0.0444
(3.67}
----+
Iteration 3
( 3.69 )
Li,X
::::=
--().:333
0.3333 = -0.333
= ag = ld(0.0444 ~ 0.0556) :.-~
--0.1117
A3 = .A2 +A;\= -0.3333 - 0.1117 = -0.445
--+ {
U2
U3
= 0.05183
= 0.0482
(3.70)
1- ,\
(3.64).
Remarks on Augmented Lagrangian Method
(3.69)
[ u~-]:..!_ [ l
Box 3.1 summarizes an iterative procedure for augmented Lagrahgian

algoritJun [52]. The itcration begins with solving for u from (3.63) and
computing the contact force using a simple penalty method. If this force,
which is equal to 6..X, is not small enough to ensure a sufficient convergence,
the procedure will be repeated using an updated A according to Equation
-10 ] [
=O
56
(3.67)
---4
Iteration 4
(3 . 69 )
A4
~{
= ,\3 +
U2.
U3
(3.67) . -.....:... ~-X
Iteration 5
=10(0.0482 -
~,\ === ng
As
(3.69) ~ {
:e-.::
= -0.03663
Penalty Method
l1 = -0.445 - 0.03663 = -0.4816
In a penalty based method, the const1'a.i.nt functional is constructed based

on the fa.et that CT C is always a positive quat1tity and is equal to zero
when the constraints are satisfied. Therefore,
= 0.05061
= .04939
ng = 10(0.04939 - 0.05061) = -0.01222
= ,\4 + 6,\ =
7.t2
0.05183)
(3.73) ..
-0.4816 - 0001222 = -0.4938

and the constraint functional can now be written as
= 0.0502
n:1 ~:=
0.0498
(3.74)
(3.67) ~ 6,\ = o:g = 10(0.0498 - 0.0502) = -0.0041

If the achieved tolerance is acceptable, the iterative approach is terminated, otherwise it will be further continued u'ntil the necessary tolerance is
achieved.
Here, the constraint term is introduced by the following functional
3.4
LJ
-Variational Forms
(3. 75)
Using a variational approach to derive the necessary equations for enforcing

constraint equations will lead to same results as previously obtained by the
minirnization of the total potential energy of the system.
The fundamental aspects of the adopted variational formulation will be

discuHRed in Cha.pter '1. To perform imposing thc constraint equation by
means of variational principies we can introduce the constraint by forn:fing
a constraint functional, ncon' in addition to the functional of the system
The vaxiation of the new functional is now
(3.76}
Perturbed Lagrangian Metho'Cl The constraint functional is defied by
= Il(u)
+ nc
n(c, u, ... )
(3.71)
(3.77)
where TI( u) is the functional of the unconstrained system. The variation of
the new functional leads to the general variational forrn of the constrained
system
8Il +mcon
(3.72)
Here, only a very brief review of the main functionals for each method
is provided.
Augmented Lagrangian Method

The constraint functional for this method is co11Hucted from the penalty
method and the Lagrange multiplier method
(3.78)'.
58
59
Lm1st Squares Method

time=t
In this method, the constraint functional is defined by setting II =
using a similar approach as for the penalty method,
CT(u)pC(u)dO
Oand
(3.79)
where p is a diagonal matrix of positive valued weighting functions. This

fonn is sometimcs convenient as it may atta.ch different importance to the
satisfaction of individual components of the constraint equation through the
selection of the weighting terms.
3.5
The development and implenentation of the penalty method for contact

applications may be attributed to the work by Hallquist [54] in the late
seventies.
The gmwral a.spect.s of the penalty method for imposing a constraint
equation has been discussed in the previous section. Here, further details
of the scheme as a contact interaction algorithm are discussed. In a contact
mechanics analysis, the constraint condition is the impenetrability of the
contacting objects. The impenetrability constraint equation for two nodes
in direct contact may be expressed from equation (3.1)
- U1
= 0
b) Possible normal mu/ u111gt111ia/ gaps
Figure 3.11: Contact force based on impenetrability.
on
re
(3.81)
where g is the gap function, x 1 and x 2 are the deformed configuration of

body 1 and 2, respectively, n is the normal to the body at the contacf
. . r e -- r 1 n' ' ...f" 2 .
sur f ace, and r e .is t he contact' domam,
Therefore, the va.ria.tional form of the constrniut cquat.ion (3.74~ may l>e
explicitly expressed as
owcon
={
ro
agg(u) da
(3.82).
(3.80}
In somo applications, the exact impenetrability is strictly sought. For

instance, in simulations of molecular dynamics or animations. These cases
usually comprise sparse populations of bodies moving around at high speed
and interact by collision. The collisions are brief and can be modelled as
instantaneous exchanges of momentum, in which energy may or may not be
conscrved by the particle pair (55).
a) Be/ore contact
g = [a::2 - x 1] n 2: O
The Penalty Method
C :.= U2
Target segment
In a penalty method, :penetration of the contactar object is used to

establish the contact for .,between contacting objects at any given time
(see Figure 3.11) .
The general form of equation (3.80) for canta.et between two bodies may
tJwu be defi11ed by [5G, 52)
Equation (3.83) may be,re-written in terms of the contact force vector
wcon = {
con
0
og u da
OU
Attention is now focused on a single boundary node in conta.ct to formulate the residual contribution of conta.ct coni;trnint., r'~. The component
form of the virtual work of the contact forces associated to the contact node:
is then given by [57]:
()U
Discontlnuum Mechanlcs

' :-~
'' '
1.;'.:i',
a} Bondtd lciyer.1 ut tlmr. t
t~.
satisfied by this method. This is clearly observed from Equatlon (3.21) '. {:
where a non-zero term is added to the total potential energy of the system, . ,
Without any additional treatment, the penalty method will cause the sy&- <
tem to gain energy artificially, although sometimes this extra energy is '
compensation for the loss of deformation energy dueto contact penetration..
//'
b} Dtbonded layer.r at time t+dt
In a central diference one dimensional cont-.a.ct analysis ha.sed on the

penalty method, the contact force may be defined as
e} Projectlon onto a rtfertnce .1y.vtem
Figure 3.12: Normal and tangential gaps.

if g
where k = n, t and i = x, y, and ui is the the i-component of displacement

vector n.t norfo .<1, g : : : (gn,gt) is the relativo motiou (gap) vector in normal
and tangential directions, respectively, and con is the contact force vector
over the contact area Ac,
>o
otherwise
{3.88)
where g is the normal penetration. One may expcct tlw central dU,ference
scheme to be stable when applied with
J(
(3.86)
where a is the penalty term matrix, which can vary for normal and tangential gaps and even between single contact nodes. The corresponding
recovered residual force is then evaluated as:
~l !!llt2 < 4
m
(3.89)
However, numerical tests show that in sorne cases considerable energy is

added to the system because the central difference scheme becomes unstable
(58] (For details of the central difference method see Section 4.4).
To clarify the problem, consider the impact of a material point (hall) on
to a rigid wall as depicted in Figure 3.13.
(3.87)
The partial derivative part of equation (3.87) defines the direction and
distributiou of normal and tangential contact forces. The calculated contact
force has then to be distributed to the target and contactar nodes.
The potential normal and tangential gaps for each contacting couple are
evaluated by monitoring the coordinates of contacting couple nodes in ~ach
time step. Then by projecting the coordinates in the current and previous
time steps to a ref~ence con:figuration, the possible gaps are calculated
(Figure 3.12). Note that for a debonding problem, both penetration and
release normal gaps (positive and negative) are valid.
1
,,
a) The ball penetrating into the wall.
3.5.1
Contact Instability
One of the concerns of using the penalty method as a numerical constraint

scheme for explicit analysis is its stability. The rea.son can be attributed to
the fact that the impenetrability condition (g = O) is only approximately
b) The ball escaping out ofthe wall.
Figure 3.13: A material point entering and leaving a. ri;id wnll at successive
time steps [58].
Ata time step when the material point is about to enter the wall boundaries, for sorne time t 1 < ll.t no contact force exists to resist penetration.
62
Thn rns11lt.ing penetration genera.tes contact force and stiffness for subsequent time steps .. When the material is leaving the waH boundaries, the co~
tact force continues to be ptishing out the material point from the boundary
for sorne time t2 < .t, although there is no penetration any longer. Consequently, each time the hall enters and leaves the wall boundary, it gains
so me additional energy [58].
Figure 3.14 mustrates a simple contact problem in which a material
point of mass m = 0.lkg is moving between two parallel walls with an
initial velocity of v = lOm/ s. The walls are on the distance l = 7.5m from
each other and the penalty coefficient a: = le + 06.
63
50 --------"" ...- ...----- ...... ----....----;-
45
35
+------.+-----
{ 30
25
i> 20
1
1
1
1
100
300
400
500
600
700
n(number of colllslons)
v=JO mis
----
200
/ .5m
1
Figure 3.15: Velocity of the material point after it passes n times from its
original, position.
A Modifted Central Difference Scheine

Figure 3.14: A material point moving between parallel walls.
Iu this method, at each Lime step the conta.ct 1<111<~ d11n t.o t.ho peuetra.tlon
is calculated from the modified version of (3.88) [10]
Figure 3.15 shows the velocity of the material point after it has passed
through the starting position n times. The results are obtained by unmodified central difference time integration scheme (3.88)[59].
o
if gt > O and gt+b.t < O
LJ
. It is clearly seen that as the Ll.t is increased, the velocity of the material
point is also rapidly increased, i.e. the system becomes faster and faster,
representing the system gaining extra energy during the time steps. Larger
values of D.t leads to larger values of ti < D.t for which larger energy values
are added to the system.
LJ
'
L_J
con=
(3.90)
To avoid instability, the evaluation of the contact force for the specified
domains has to be modified using a damping scheme. In the following, four
different sch<mrns whkh ha.ve been used to sta.bilize the conta.ct analysis, are
where v is a user defined constant and a negative value of gt means no

2
< 0.5 and 11 -< 1.0, numerical tests have shown_
P enetration. For K = .li.At
m
sta.bility and at the su..1m1 time the damping cau1:1ed by d1:-;crilir~d modification
briefiy reviewed.
is negligible [lOJ.
64
65
A N onlinear Contact Force
where mi; = min( mi, mi) is the minimum mass of the contactirig bodies i
and j. (are input parameters controlling the penalty term, usually taken
It is sometimes convenient to assume that the contact force is not linearly

proportiona.l to thc va.lue of the penetration g. A natural way to define
a nonlinear g - f relation is by assuming that the value of penetration is
limited to a maximum, called field :z;one. Penetrations larger tP,an the s~ze
of the field zone are usually ignored. For the exact definition and further
details of the field zone, see Section 7.2.4.
0.04.
e,
The nonlinear contact force can then be written in terms of the g and
the field zone,
con= a
f ield -
(3.91)
In order to obtain a stable central difference scheme, at time t, the

contact force f is calculated as
It is proved that the central difference time integrnt.ion scheme is stable

for any value of !:i.t if the following condition is satisfied:
.
{3.96).
In Equation (3.96), mmin and mma:i: are the mnimum and maximum
mass for all the bodies, respectively. However, if the penalty value or d&Dlping coefficient is too large, this acherne might not work properly.
Figure 3.16 compares the velocity of material point. of previous example
for a range of damping coefficient for two different time steps. The ~
sults have been obtained from Equation (3.93) and clearly show substantia.f
improvements relative to an undamped solution [59].
,
(3.92)
The stability of the acherne is controlled by parameter OS (J
< 1; proper
B has to be chosen in order to avoid instability due to nonlinear contact

force. 8 = 0.6 proved to give a stable scheme in a variety ofthe discrete
element applications [58].
Relative Velocity Dsliping
Defending Node Method

In an alternative approach, called the defending no de method, the contact.. .
force is directly evaluated from the explicit integration equation (60, 42) I(
may be used to provide a very good trial value of the contact force of ea.ch :
nodal point independently. Given xn; Vn and Rn as the current positfon,.
velocity and interna! force, respectively, the gap in the next step can. be
calculated as,
A simple damping scheme is provided by applying an additional damping
force according to the relative velocity [v],
an+l
(3.97)
con = ag + c[v}
.where e is the contact damping coefficient. For normal
= (Rn + /1~) /m
(3.93)
Xn+l = Xn
c~ntact
+ Vn+1dt
then,
(3.94)
LJ
{3.98) .
with
an = emi;/ !l.t 2
{
(3.95)
e = (mii / l:i.t2
The contact force f,~" is determined hy selecting gn+ 1 . An obvlo&

choice for impenetrability is gn+l = O. However, to avo id the system gainlng
extra energy another choice is to let Vn+i = O to stop further penetration.
This conserves the energy of the system and as long as the initial penetration
()()
Chapter 3. Constraut E11forcir1g Methods
67'
is small, a stable and reliable solution is achieved. a11 +1 is calculated from

Equation (3.97b ), and ~on will then be evaluat(~d fron1 Equation (3.97a).
50 .... - ....................................... _. ____ --------- ---------
~---
In a multibody system, equation (3.98) leads to a set of equations that

can not be solved explicitly.
.... ~-- - - - -
~t=l x10' se
35
30
>.
20 ' - - - - - + - - - - - - ! - ...
25
......... '
-.---=-+----
--
. , C-0.5
........
,.
. <-.----~~------.;-.-
.......... - - --..,...., ...- - - - h..........~
15
..-t.~.-: - :,..: _.
, """'
10
100
.... -
,.... -- - ..... ...- - -
300
200
400
i- -
C-0,8
C-0.95
500
600
700
n(number or colllslons)
25 - ...- ........---........... .,__.,
---
-------
t=Sxl~'.
20
----
100
..- .
C-0
----+----l---+----1----,,,......:;.-+----l
..-
200
300
400
500
800
700
n(number of colllslons)
Figure 3. 16: Effects of relative velocity damping on the velocity of the

material point after it passes n times from its original position.
Chapter 4
N onlinear FE
Discretization
4.1
Introduction
Plasticity theory is applicable primarily to those solids that can experience

inelastic deformations considerably greater than their elastic }imit. How".'
ever, the resulting total deformations and rotations accompanyirtg them,
may still be sufficiently small so that many problems can be solved with
small-displacement kinematics. However, when strain or rotations become
significant, analysts must resort to the theory of large or finite deforID&tions [61].
In this chapter, after a brief review of the large deformation principies;
the basic concepts of the boundary value problem and flite element discretization are explained. A more comprehensive discussion on the explicit
time integration scheme and its related energy balance considerations which
are extremely irnportant in a sccessful discrete element analysis, a.re provided.
4.2
Large Deformation Kinematics
In this section the general principies. of a finite deformation formulation in

the dynamic analysis of solids are discussed.
69
70
L ___;
Chapter 4. Nonlinear FE Discretizatiou
71
for example, a rotation of the given body in space. This kind of inva.riance
may, accordingly, be used to establish material isotropy [61 J.
time=t
4.2.2
time=t-+<ft
Deformation Gradient
Deformation gradient, the most basic definition in a. large deformation analysis, is defined as,
~,x
F=
ax.
ax
or
ax
F=-=l+Vu
(4.3)
ax
where the displacement field u is introduced as

1
Figure 4.1: Large deformation problem.
(4.4).
u=rn-X
4.2.1
The Motion
According to Figure 4.1, the relation between the refercnce and current
coordinates may then be expressed as
Consider the motion of a body during its deformation as depicted in Figure 4.1. The deformation of the continuum can be described by a mapping X
which gives the current coordinates, x, in terms of the material coordinates
in the undeformed configuration, X,
= X(X,t)
(4.1)
(4.5)
--
where e is a material tensor quantity and b is a spatial tensor quantity

defined as
.
or,
Right Cauchy-Green dformation tensor e= /l'p

x
= x(X,t)
(4.2)
There are two classes of description to express various quantities within ,_

a large deforrnation analysis. Lagrangian or material description which describes the quantity in terms of where the body was be/ore deformation,
and Eulerian or spatial description which describes the quantity in terms of
where the body is after deformation [62].
Here, a fundamental difference between the Lagrangian and the Eulerian deformation and strain tensors is emphasized. The Lagrangian tensora
transform according to the usual rules under a rotation of the Lagrangian
basis, while they are invariant under a rotation of the Eulerian b~is. Sine~
the Eulerian basis is, in a sense, fixed in space, invariance under its fOtation
is related to the history of the space itself, and is invoked in the formulation
of fundn.m1~11t.n.l physical principies. Converscly, Eulerian tensora are invariant under a rotation of the Lagrangian basis, such a rotation may represent,
----------
--- -- --------------------
(4.6)
Left Cauchy-Green deformation tensor
b= ppT
Following these definition& two frequently used strain mea.sures may be

defined
(Lagrangian) Green strain tensor
E=
( Eulerian) Almansi strain tensor
He -1)
= ~ (J
(4.7}
- b-
In the literature, the Almansi strain tensor iH regar<led a.s. the push Jorward of the Green strain tens<;>r. Conyersely, E might be interpretad as
the pull back of e. These two strain tensors measure the total deformation.
According to equations (4.6) and (4.7), however, the relationship between
t.heso strains and displacemcnts is nonlinear, whirh ndcls t.o t.lm complexlty
of the numerical algorithm [63].
Discontlnuum Mechanics
72
4. 2.3
Chapter 4. Nonlinear FE Discretization
Polar Decomposition
73
where
Deformation gradient, F, transforms a vector in the material configurar

tion dX into a spatial vector dx. It is often convenient to decompose this
transforrnation into a stretch and a rotational part,
\,'.
Ai = >..1,
i,l=l,2,3
i,l=l,2,3
F=RU
or
F=VR
(4.16)
(4.17)
(4.8)
where R is an orthogonal rotation matrix and U and V are symmetric

stretch rnatrices. These two matrices might
evaluated from the right
and left Cauchy-Green deformation tensors,
be
which implies that the tensor R rotates the material vector triad N 1 into the
spatial triad n. Analogously, the deformation gradient and strain measure8
may be expressed in a similar way as,
3
U 2 =e
and
y2 =b
F =
(4.9)
L >-1n1N1
1=1
and the corresponding strain tensors are
L 2(.X~ - l)N1N1
E=
l=l
,_j
1 . 2
E= -(U -I)
2
(4.10)
1(
-2)
e=-1-V
2
(4.11)
4.2.4
Generalized Strain Tensors
By using an eigenvalulanalysis, one may obtain
A general formulation for the Lagra.nglan a.nd E11lorin11 strnin tensora

e= .:>..~N1N1
(4.12)
f=l
E<n) =
= L:>..~nini
1.1&cy"
be defined by generalization. of equations (4.10, 4.11) ami (4.19, 4.20),
.!.cu" - I)
n
(4.21)
(4.13)
(4.22)
i=l
which leads to,
and
3
U= L>..1N1N1
(4.14)
E<n> =
l=l
J=l
V= L..\inini
l=l
: .!:.c.x - I)N1N
(4.23)
(4.15)
e<n)
=E .!.(1 - xn}n10n1
'=l n
(4.24)
..
Discont.(nuum Mechanics
74
which implies
(4.25)
In the particular
sors,.
casen~
O, this provides the Logarithmic strain ten-
3
E(o)
= Lln..\1N1@N1
= lnU
(4.26)
75
current rate of deformation, but gives no information about the total deformation of the continuum. Furthermore, its integral in t.ime for a material
point does not yield a well defined, path dependent tensor so that infor-.
mation regarding phenomena such as total stretching (or uut..'<imum strain)
are not available. Therefore, to obtain a measure of total deformation, the
velocity strain has to be. transformed to some other strain rate that can be
integrated to yield a measure of the total deformation [63].
The spin tensor is then defined as the skew-symmetric part of the velocity
gradient matrix
1=1
l=w+d
(4.33) .
e< 0 > = 1n..\in1n1 = ln V
(4.27)
where
i=l
4.2.5
= 8v(x, t) = Vv
8v
F= =lF
8X
= ilRr + ~R (u- 1 - u- 1 u) nr
(4.35)
The second term vanishes in rigid body motion for which d =O.
Therefore, the rate of deformation gradient may be expressed as
(4.28)
8x
(4.34)
and
The velocity gradient matrix is defined as,
w = -(l - l )
Rate of Deformation
(4.29)
4.2.6
(4.30)
The Cauchy stress, which is also called the true stress, is defined as a spatial
tensor, u, which provides the nQcessary and sufficient condition for satisfaction of the momentum ha.lance la.w. This syrn111etrk tmHor relates the
normal vector n to the tracti~n vector t
The rate of deformation matrix (also called the velocity strain, or stretching) is defined according to the velocity gradient matrix, l,
Stress Measures
t(n) =un
1
T
d= -(l+l )
(4.31)
'"'
(4.32)
dln..\ 1
= L., --n1n1
1=1
dt
where d gives the rate of extension per unit current length of a line element
having a current direction defined by n. The velocity strain measures the
and satisfies the general equation of dyna.mic motion

divu + body
= piJ
,.,
where n is the unit normal to the boundary of thc current configuratlon~\\;~,:~L

p is the density at the current configuration, v is the acceleration of the .: ;'!lf'~'':
body and body is the body force measured per unit volume of the current. )~':1;:
.....
\,".
,.
'i~~J~: ':
Discontinuum Mechanlcs
76
confip;11mtion. Thti variational form of equation (4.37) leads to the principal

of virtual work (lhr moru d(1ta.HH 1:mn 1mcUon 4.3). The lnterna.l vlrt;uu.l work
4.2. 7
rnay then be expr<:'.s::;e<l us,
An important aspect of the measure of stress i::; the o/Jjectivity (or jrame
invariance) of its rates. An objective rate must transfonn according to the
law for a second order tensor when the spatial refereuce frame is rotated.
This is important because constitutive equations are often expressed in re.te
form, so if the rates are not objective, the material behaviour will depend
on the choice of the spatial coordinate system, which is obviously physically
unrealistic and must be avoided.
'5Wint
u : ddv
(4.38)
Pairs such as u and d are said to be work conjugate with respect to

the current deformed volume. Therefore, appropriate work conjugate stress
measure is required for different strain measures. The objective in defining
them, in fact, is to express the interna! virtual work in terms of an integral
over a volume that is known and can incrementally decompose the stresses
and strains in an effective manner [64].
Sometimes it is e.sier to <leal with the reference configuration than with
the current configuration. The Kirchhoff stress is defined per unit area of
thc reforence configuration
77
Chapter 4. Nonlinear FE Discretizatlon
T=J<r
= detF
Objective Stress Rates
Although the Cauchy stress tensor is an objective tensor, its rate does
not satisfy the objectivity condition. Therefore, for a general constitutive
material model 'Ye are bound to use other stress rntes. Only the m0$t
popular ones are briefly reviewed here.
The Truesdell Stress Rate is constructed directly from the terms.required
for the objectivity of Cauchy stress rate,
(4.39)
(4.44)
(4.40)
The Green-Naghdi stress rate is defined by ignoring the stretch compo'

nent of F in the Truesdell stress rate [66],
which is work conjugate to the rate of deformation tensor with respect to

the initial volume.
(4.45)
If the un-symmetric part,
RRT, is approximated by
the spin tensor
'
U,
(4.41)
<Tv = t!r Another stress measure is the first Piola-Kirchhoffstress which is defined
at the reference configuration
p = Jup-1'
tTW -
wcr
(4.46)
the J aumann stress rate is obtained. This tensor has two important prop- . , ' .:
erties:
'
:<..
(4.42)
(4.47)
and is work conjugate to F. The second Piola-Kirchhoff stress tensor, another frequently used stress measure, is defined as
(4.48)
(4.43)
and is conjugate to the Aterial strain rate tensor, E, with respect to the
initial volume. Finally, the logarithmic strain measure (ln U), defined in
(4.26), is work conjugate to the rotated Kirchhoff stress, RTTR [65].
and if <rv is a deviatoric tensor, then (t7V)rotated is also deviatoriC.

Any frame invariant stress and strai11 me9.sures may be used in the con
struction of a constitutive equation. One of the mosL pupuiar raLt; fr~~ is
to express the Jaumann rate in terms of the velocity strain
78
uv = Cd
(4.49)
The above form can be used for nonlinear materials in that C can be a
function of objective measures of stress or strain, such as u, u V d or other
'
Htat<~ va.ria.bles. Noto thn.t it can not be a function of ir or d because they
11w 1101. frnrno it1voriu.11t..
Cha.pter 4. Nonlinear FE Discretization
4.2.8
79
Multiplicative Decomposition
and
The additive decomposition of the infinitesimal strain tensor into elastic

parts, is basic to virtually all the subsequently developed theory of
plast1c1ty,
inel~t~c
(4.54)
When a material is anisotropic, the C matrix changes with time as the
material rotates. Therefore, in addition to the change in C that results

from the change in the constitutive law of the material, it is necessary to
accrnrnt for tlu~ dfoct of the rigid hody rot.ation on the components of C
wlirn1 it. iH expreHH<\d in a fixed coordinatc systcm. This is not necessury lf
the constitutive law is expressed in a corotational framework,
(4.5Q)
If the corotational coordinate is updated by the rotation of the material

based on the pola: decomposition theorem, the rotation of the principle
strain ra.tes, then ir corresponds to the Green-Naghdi stress rate [67]. Note
that such corotational formulations are much more suitable for anisotropic
rna.terials than Jaumann or 'fruesdell rate forrnulations. When the latter
are uscd with anisotropic materials, C must be updated continuously to
reflect the rotation of the material.
Therefore, the corotational constitutive equation may be constructed in
a rotated format [68],
T=Cep:D
(4.51}
where T is the material rate of the rotated Cauchy stress tensor and D is
the rotated rate of deformation,
.
For strains that are too large to be treated as infinitesimal the decomposition still works, at least for longitudinal strain, when this is,taken as the
logn.rithmic Htrnin. In cmltH' to formula.te a tlwury of pl11HI i<'it.y for large deformation, we hav..e to establish the approprlate dccomposit.iuu fr arbitrary
states of deformation.
An ad.ditive decomposition of the logarithmic strain is equivalent to a
multiplicative decomposition of the deformation gradient F,
I'
; (4 56) : <>
F=FeFp
. . ,r;n:,'.~/
which was first explicitly formulated by Kroner [D], ami fnrther developed\
by Lee (70), Nemat-Nasser [71) and others on t.ht! hnsiH nf consideration of
the behaviour of crystals. The reasoning behind t.his, is comprehenslvely
presented, among others, by Lubliner (61].
.
The transformations involved in the decomposition are shown schemai.;
(:i:, x + dx) is usually sa.id to be
m an intermediate configuration.
~cally ~n Figure. 4.2. The unloaded element
The velocity gradient then becomes
where
T=RtFR
(4.52)
-1
-1
(4.57)
le= FeFe
D=RTdR
Then, the Cauchy stress tensor can be evaluated from the expression,
lp = FeFp
(4.58)
and finally the Eulerian rate of deformation is

O'n+1
= Rn+l
( +l
O'n
tn+l
tn
Tdt R~+l
(4.53)
d =de+ (FelpF; 1 )
'(4.59)
.11,
80
Box 4.1 - Stress integration procedure for a huge defonnution pl~ticity.

lntermodlate
Configurar.ton
Compute current deformation gradient
= l+Vu
Fn+l
_ Elastic predictor phase
(F~+itial
= Fn+1 (F;rt =
R~i+i (u~+ 1 t'ial
'nial elastic stress is evaluated.

Check far yielding
IF (q,n+l) trial :S O
lnilia)(Rcfcrencc)
Conliguration
Elastics update
Set ( ) n+ 1 = ( () n+l) trial
Figure 4.2: Multiplicative decomposition of the deformation gradient.
n
i
LJ
ELSE
Plastic corrector phase
Box 4.1 summarizes th stress integration procedure for a large deforr

mation plasticity.
Perform small stra.in stress update

un+l un+l ~ '
Update intermediate configuration
4.3
Initial/Boundary Value Problem
F;+i
(Fn+ir1 F~+i
= (Fn+1r1 R~ttiu~t+i
ENDIF
4.3.1
Equilibrium Equation
Consider a bod, B, occupying a region f2 with a boundary r subject to

body forces f bo v throughout its domain
Here, the boundary is assumed
to consist of a pa.rt with prescribed displacement Ui 1 r uc, and a part With
prescribed traction force J:ur f' r U (Figure 4.3). The boundary conditions
may then be described as
n.
CTn =
x=x
j8urf
on
on
ru
ru
(4.60)
and for a state of dynamic equilibrium,
(4.62)
Applying the divergence theorem to the first term iu the above equa.tlon ;
and using equation (4.60), the following is obtained
where u represents the Cauchy stress tensor and n representa the unit ohter
normal u.long
{ rur f da
(1'
Far this body to be in a state of static equilibrium, the following condition must be satisfied
. (4.61)
lr
= f O'nda = { d'ivCTdv
lr,,
Jn
Thus, the equilibrium condition for the body B becomes
fo
(divo- + body) dv
pdv
(4.63)
82
83
Cha.pter 4. Nonlinear FE Discretization
As this condition holds for any arbitrarily chosen subdomains, the integrands should be equal pointwise,
.
(l'tVO'
lJ
+ bml', : : : pu..
(4.65)
which represents the dynamic equilibrium condition at a poi~t withi~ the_.

body. It is also called the strong form of the linear momentum balance (or
translational equilibrium) written in spatial description.
4.3.2
Variational Form
Here, a weak form of the equilibrium equation is derived, since this is utilized
as the basis of the Finite Element procedure. Thereby, the above constraints
are not applie<l pointwise, but rather in an average sense, by a weighted
residual scheme,
r (<huO" +
Jn
fbody -
p) wdv + lr~
r (rurf - un) wda =o'
Figure 4.3: Description of the boundary value problem.
In addition, it is assumed that a part of boundary, re, may be in contact

with another body (Figure 4.3) according to the contact boundary condi-.
tions (72, 56]
(4.66)
un= O
un= con
where the weighting functions w satisfy the natural boundary condition over
r u, i.e. w = O. Applying Green's theorem and integrating by parts, results
in the following identity
where gN is the gap between the bodies. By denoting
V:= {5u:
- fn u: Vwdv +In fbodywdv - In pwdv
+ fr unwda +Ir
(rurf - un) wda =O
(4.69)
=o
on
r uJ
(4.67)
the space of admissible variatiop.s, the variational (weak) form of the dy-.
namic initial/boundary valu~ problem may be Pxprn:-;sed as (73, 74)
which is simplified to
{a: Vwdv
Jn
Jn
pwdv =
lo
ldYwdv +
lr~
aurf.wda
(4.70)
(4.68)
According to the Galerkin weighted residual approach for solving the

boundary value problem, the weighting functions are chosen as the field of
virtual displacements u, and the weak form of the equilibrium conditions
represented in equation (4.68) is equivalent to the principle of virtual work.
More details may be fo~ in Zienkiewicz et al. [51].
where
Wint(u, tt) =
f O E(u) : u(u)dv
M(u, u)=
'Updv
0
84
85
L_.:
wext(u) =
uJbodydv
ururfda
4.4
(4.74)
r~
A standard methodology according to Zienkiewicz [51] and Hughes (75] is

followed. The rate forms derived in section 4.3 may be exploited with appropriate linearization to generate incremental fonus. Li1warization of the
functional W(x, 1J) at the known configuration X = X cau be obtained
through the directional derivative formula
(4.75)
denote, respectively, the virtual work of interna! forces, the inertia forces
contribution, the virtual work of ex~ernal forces and the virtual work of
contact forces. Here u is the Cauchy stress tensor, e is the strain tensor,
u is t}:e displacement vector, while g representa the contact gap vector.
Observe that in the present formulation the contact terms correspond to a
penalty formulation of contact interaction.
u
1
Finite Element Discretization
DW(x, 11)u
ddv
L [W(x, 71)] = W(x, 11) + DW(x, 71)u
(4.76)
(4.82}
'v'71 E V
the residual force ma.y theu be described as (Sl'P ( 1. 71)),
(4.83)
Alternative expressions may be given for other conjugate stress and

strain measures,
j no r:
ddV
where wcon is the virtual work of the contact forces and w(~xt is the virtual
work of external loading
(4.77)
(4.84)
where ,,- = Ja is the Kirchhoff stress tensor and !lo is the initial (reference)
volume of the body, and
wint
wint
P :.FdV
(4.81)
Equilibrium, of course, must be satisfied for every configuration X
which d is the virtual rate of deformation.
wint
(4.80)
in the direction u,resulting in the linear part of the functional W(x, 17)
W(x,11f=0
(J':
W(:X, 71)
E=O
To incorporate appropriate stress and strain measures in a finite strain

analysis, the virtual terms may be expressed in terms of an arbitra.ry virtual
vclocity from the current position of the body. Using the same analogy as
before, and concentrating on the virtual work of the interna! forces, one
may write
wint
= dd 1
(4.78)
Now, consider the space Vh e V as afinite dimensional approximation to ...

V described in (4.70). Furthermore, let subspace Vh be generated through
spatial finite element discretiz~tion in .the current configuration
Oo
Ne
no
S : EdV
for the first and second Piola-Kirchhoff stress tensors, respectively.
(4.79)
x(O)
= U
e=l
x(ne) with x(!la)lx(nb)
0 if a i= o
where Ne is the number of finite elements. So tlmt over the typica.l element'
x(le) one has the interpolation
86
N .. ,.
uhlxcne)
= 2: N a(x)tta
(4.86)
F=
Fnodal
a=l
wlwre
N~n
is the numher of nodes of each individual finite element and

Nu(:i:). i.11 n. Ht.1111d1t.nl 1:1hn.po funct.lon ovm cuch i;ulmpo.co, th1:tt tranaposes.
the c~ntmuum problem to a finite element problem. We may now represent
the d1screte gradient operator in the following form
'
87
Ne
Nw
e=l
n=l
A(/e)+ A(/~n)
where Nen is the number of nodes in contact.

The finite element discretization .of the incrt.ia tcrm in the bondary
value equation (4.71) results in the consistent mass matrix
(4.94)
Nen
vsymuhlX(2e)
=L
(4.93)
Batta
(4.87)
a=l
Ba = vsvmNa
{4.88)
The effects of the inertia term is directly considered in the ~X.P.!~cit time
integration scheme to evalute the
displacement, velocity and
celerati0"e1Cls:
. .......... ......
"-~~---.....
ciirreiit
' .... _.... ~ .. '""""''"'""
T'hernfore, the dl!:lcrete version of (4.83), (4.84) supplies the element

resi<l ual force vector
"''~''
e.e:-:
':
, .. """ r, """"
:.~ . '
:! T.:!~:. ;,
: ''',_;./
4.5
Explicit Time Integration

',i>,;,r(;.,
{4.89)
wbcre
The standard finite element discretization of t.he va.riational form (4.71).':tf

resulte in the discreto snt of o.lgebralc time dupoudonl. cq11nt.lotlM whlch may !;2,:~.,
be expressed, in matrix form, as
.
M(t)
+ int(u,t) = ext{t) + con(t)
(4.95).
where t is the time, int(u, t) the interna.1 force vector, rxt(t) the externa!.
force vector, rn(t) the contact force vector while A1 denot~s the mass
matrix.
4.5.1
are H~ c~n.l.led elenwnt; intemal and extemal force vectors, respectively. The
contnbut1on of the contact forces is consi<lered directly through the associated contact nodes,
n = Acon
n . ( ng )
! con
(4.92)
Central Difference Method
The velocity v = u and acceleration ii = iJ are approximated by using the

central difference method with variable time steps, which is appropriate. if
the response of the continuum is strongly nonlinea.r. Ldting a variable with
subscript n denote the numerical solution at t.imu Htn.t.io11 t.= tn, we have .
(4.96).
in which a penalty based algorithm is adopted to evaluate contact forces.

A~ion i~ the contact area associated to each contact node, and a is the penalty
coefficient.
The global residual vector is then obtained,
Vn-1/2
'Un - Un-1
A
utn-1
(4.97)
Vn
= 2(Vn-1/2 + Vn+i/2)
,:__ Vn+l/2 - Vn-1/2
Vn -
.tn
(4.98)
In addition to providing uncoupled equations and generally more accurate

results than consistent mass matrices in explicit integra.tion, lumped mass
matrices provide for larger stable time steps [77].
( 4.99)
There are severa! ways of constructing the lumped mn.ss ma.trix, which
are briefly reviewed here: [75]
l. The same proportion of the masa is associat.c1l with rnHh node.
(4.100)
'1
~rth.er,
the mass matrix is assumed to be diagonal, so as to avoid

solvmg s1multaneous equations. Making use of the above approximations
we obtain for the i-th degree of freedom,
'
{4.101)
!J
LJ
89
88
:'here mi is the i-th.diagonal ter;n of M. Time incrementation is then readIly pe:formed by evaluating displacement, velocity and acceleration using
cquat1ons (4.96), (4.98) and (4.99), respectively.
. If a damping matrix pl'oportional to the mass matrix. is assumed the
solving of eql,l~tions is .still avoided and the equa~!ii ~(4\9if
is generahzed to
.___ . ., ...".
~1multane?us
2. Terms in each row of the conslstent m!Uis rnatrix un~ ::;ummed and
lumped on the diagonal. This technique has sometimes a tendency
to generate zero and negative masses, that is imacceptable in practice
and from a pl,!.ysical point of view.
3. Setting the terms of the lumped mass matrix proportional t the diagonal terms of the consistent mass matrix with the constant of proportionality selected to conserve the total element mass. [78}
4. To match eigenvalue analysis of single unconstrnined element with

the same quantities of a consistent mass matrix. The improved lumped
mass matrix will be the one which has approximatdy the samf;' spectral characteristics as those produced whtm 11si11; t.hc con~istent masa
matrix. [79}
The first and third schemes are the most frequently employed in practice
for 1\rbltrn.ry uloumnt.H. Howover, for certe.ln t'lt11w11t.H

2 - f3i6.tn
v,n+112 = 2 + /3i6.tn vi,n-112
2~tnm'.':" 1
+ 2 + /3idtn (U:Xt)n + Ufn)n - Ult)n)
~,
and
ci
is the i-th diagonal term of the damping matrix.
There are alterna~ive approaches, such as Cranck-Nicholson method (76]

etc., to the central d1fference explicit time integration technique. However
they generally lack the simplicity and robustness of the central differenc~
approach in a general contact mechanics problem within the combined finite/discrete element methodology.
4.5.2
4.5.3
Stable Time Step
The stable time step for the central difference method is given by
At
~ min :n
( y'1
+e~ - en)
~tstab
(4.103)
where Wn are .the natural frequencies of the mesh ami ~n is the fractlon
of critica! damping in the n-th mode. Since strm:tural systems are usually
underda.mped, the mnimum of (4.103) is usually given by the maximU!i
Lumped Mass Matrix
Al~hough the cor~sistent mass matrix, equation
of theao can
be directly employed. once the finite element cbu.rn.deriHticH of the elementl!I

being used in this study have been discussed.
{4.102)
where /3i
111 1it.hN
(4.94), leads to optimal error
estnnates, there 1s, however, a strong motivation for employing diagonal or

~umped .rnass matrices because they lead to sorne especially attractive time
mtegrat10n schemes, i.e. e~plicit methods, as seen from equation (4.101).
frequency,
Wma.~
(4.104)
Dlscontlnuum. Mechanics
90
where
is the fraction of critica! damping in the. maximum frequency, and
(4.105)
It is generally quite inconvenient to determine a stable time step by

actually finding the rnaximum eigenvalue of the system. This is partic1.llarly true for. nonlinear problema where eigenvalues change as the solution
evolvc:::;. In finite element methods, the maximum eigenvalue of the system
may be bounded with sufficient accuracy by the ma.ximum eigenvalue of any
(e)
e1ement, Wmax
Chapter 4. Nonllnear FE Dlscretlzatlon
u:;,in
which is suitable for discrete element contact problems. The:m energy terms
may be expressed as,
.w.ne:i:t --
Therefore, the stable time step can be determined within the time integration loop, while the maximum frequency, w~lm is computed within the
element loop.
11
LJ
w.aur J
n
+ wbody
n
w~urf = { [ r~n rurf dx] da
It is in fact usually notl~actical to compute even the ma.ximum eigenvalues of the element directly, which would undoubtedly increase the cost of
the computation considerably. !nstead, formulas which give the maximum
olmnont
or upper bounds on the maximum element eigenvalues ,
w eigenvalues
.
Amax, are used.
(4.110)
Jr Jreo
(4.111)
Another way of dealing with simple (low order) elements is to employ a

wave equation approach. The physical interpretation of this method is that
tit rnust be small enough that lnformation does not propagate across more
than one element pcr time step,
D.t < ~
- e
u:tr
where w.ext is the work of externa! forces,

is the kinetic energy,
is
t.lw Atrai~ energy, W,~o.m iH the dissipa.tOll energy d11n to Work by damping
forces, 6 ls the specitied allowe.ble tolere.nce of t.110 n.1ml.v~;i:, n.ud Wn ls some
norm of energy'
(4.112)
u~r =
1. [t ud] dv
(4.113)
(4.106)
(4.114)
[f,
where e is the wave speed, e =

and le is the effective element size (77].
le is usually approximated by the minimum size of the elements throughout
the finite element mesh. Equation (4.106) is valid for linear elements, though
an updated version of it could be used for nonlinear problems.
Within the context of a step-by-step finite element solution, and applying

trapezoidal integration rule, the following expressions can be derived,
Energy Ealance
(4.115)
N umerical instabilities are normally associated with a large growth of energy. Therefore, monitoring the stability and accuracy of the solution can
be performed by continuously checking the energy balance of the system.
The energy balance equation at time tn can be expressed as [1],
wext
- ukin
n
n
'
ustr
n
wdam
<
l';r'
1
n
Yl'n
(4.107)
(4.116)
which has to be determined based on midpoints velocities
(4.117)
Chapter 5
a.ncl
N1p
Unstr
1 "' { i
}T
. .
= ustr
n-1 + 2 L_. O'n-1 +O'~
~e~ V'
(4.118)
Continuum Based Normal

Contact Algorithrns
i=l
{4.119)
.wl~en~ N nnd iV_p u.re t.he numbor of no<les an<l the nmuber of integration
~omts
the g1ve~ ~ody, respectively, C = cM is the linear viscous dampmg matnx, and V Is the volume associated to the integration point i.
fo:
n
LJ
5.1
Introduction
The slideline methodology, based on the finite difference approach, was

developed in early 1960's for arbitrary contact problems. Thc codes were
rnostly developed for simulation of explicit dynarnic applicntions for military
purposes. They adopted simple masterwslave contad algoritluns to represent
sliding interfaces, i.e. slidelines. Figure 5.1 shows a typical sliding interface
and the basic terminology that is used for terming various contributing
components.
In this method, the slave nodes are constrained to slide on or close to

the master line unless a tensile interface force develops. It is worth noting
that designation of master and sla.ve lines was required as input for the
slideline definitions, and a truly symi"etric treatment in the early slideline
algorithms was not ever reported.
By the development of the ,penalty method for enforring the impenetrability constraint in late seventies, the need for specific definition of master
and slave lines vanished, giving perfect symmetric results unaffected by the
choice of master and slave lines.
In mid 1970's, node to node contact algorithms were developed based on

contact/ga.p elements or via the interface elemeut.::>. While thcse elements
were performing very well in small deformation analyses, the problem of
excess distortion of the interface elements dueto large relative deformation,
a common behaviour in many contact a.pplications, extensively limited their .
93
1
slave element s
slave segment s
slavenode s
slave line
95 .
Chapter 5. Continuum Based Normal Contact Algorithms
The hydrocode algorithm was first implemented in HEMP finite dilfer"'\

ence code by Wilkins [80] in 1964, and later modified by Maenchen [81] and
Bartholf [82] at Sandi National Laboratory, developing TOODY and TENSOR finite difference codes. Here, only the general aspects of the method
are briefl.;r explained based on the material published by Hallquist et al. (54).
5.2.1
Original Algorithms
In the original HEMP code, a pressure boundary cm1dition is applied to each

segment of the master surface based on the stress state in the slave elem~.1.&tci
across from the centre of 'the corresponding master element (Figure 5.2).
After repeating this procedure for all master segments, the entire surface is
accelerated as a free surface with a pressure boundary conclition.
The contact pressure Pm for the master segment m is defined by:

master element m
(5.1)
Figure 5. !: Slideline definition.
where a~ und a:1+ 1 a.re the normal stress componentH in slave elements
and s + 1, respectively, ami (is o. dimensionlef-1. diHtnrnn foct.or,
performance and application.

Further developments were based on the ground breaking work by DYNA
team, to implement the penalty method within a practica! general purpose
dyrnunic impac:t/co11tnct n.nalysis program. The method was later adopted
111 altuHI.. ev<'l'Y ol.lwl' HlrnHu.r codes.
By deriving consistent tangent modulus for penalty algorithms in mid

1980's, together with the development of augmented Lagrangian method in
early 1990's, the method found the prospects for wider range of applications.
l'
(5.2)
/"-~
1;,-
l,,,+r
slave element s+ 1
slave element s-
In this chapter, following the historical developments of the normal contact algorithrns, a brief review of the hydrocode based methods, as the first
developed contact method, is provided. Then, a comprehensive discussion
on the penalty based normal contact ~lgorithm is given which inclt.Idf.'s inaterials about the original DYNA approach and also the present state of the
art on the subject.
1
5.2
Hydrocode Formulation
master element m
In this section a brief review of the hydrocode algorithm is provided to

introduce to the readers the basic ideas of the first ever known developed
eontact algorithm.
Figure 5.2: Slideline definition to compute the interface pressure.
96
97 ,
Chapter 5. Continulim Based Normal Contact Algorithms
l~,s+t
.
+
is the distance between the element centres of slave elements s and

s + 1, and l~,m is the distance from the centre of slave element s to the
intersection of the normal to master segment m drawn from the centre of
the master element m with the Une drawn between the slave element centres
(see Figure 5.2) [54].
phony element p
To include the contribution of the mass of the slave surface, a constant

factor Km is defined for the entire slideline:
K.m
'"'nm
Mm
+ "ns
M i8
L--i=l
i
L--i=l
'"'~m M~n
L.Ji=l
slave element s
(5.3)
where nm and ns are the total number of master and slave elements along
the slideline, Mri and Mi are the total mass of the ith master and slave
elements, respectively.
1
1
master element m
1
r1
1
-----
(-----
~--- ---------
Only the normal component of the master node acceleration is scaled

with
to a.ccount for the added mass of the slave surface. The global
accelerations are then computed by rotating the scaled normal acceleration
1
1
1
1
dvJ--- ---- --"". ___ ....1'1
\.---d~--i
1
---l
---1
"'m
and un-scaled tangential acceleration back to the global system.
Figure 5.3: Defining phony element for a master/slave slideline.
To update the motion of the slave surface, the slave nodes are projected
onto the up<lated master surface. Velocities for the slave nodes are then
found by dividing the change in coordinates by the time step size.
To incorpora.te effects of oth-er slave elements for computation of interface pressure, TOODY finite difference code, developed in 1968, assumed
a phony element approach. A phony element is defined during ea.ch iteration, overlaying the slave ele.ments and matching the mesh density of the
corresponding master surfacl Figure 5.3 shows how such an elenent can
be defined for a typical slideline. The position of node q in Figure 5.3,
is determined in the same proportional distances of node p of the phony
element:
where lm denotes the length of the master segment m, and
if . rm,s . tm > o
if rm,s tm $ O
Tm,s tm
1
""
,-
(5.6}
if rm ..~-t-1 tm < lm
if rm,s+l . tm 2: lm
.
_ { Tm,s+l tm
dm,s+l lm
(5.7)
where r m,a is the vector connecting master node m to slavc node s.
(5.4)
If inequality (5.6b) is not satisfied, the contribution of the next over-

lapped slave element to the stress tensor is computed. The proce,dure continues until (5.6b) is satisfied. The stress tensor for phony element P then
becomes:
1
where l 8 ,p is the distance between nodes s and p.
LJ
The stress in the phony element p is computed based on a length weighted

interpolation along the master surface. The stress contribution to phony
slave element p made by the stress in slave element s, <T 8 , is designated cr~
and is defined by:
O"~
= (dm,s+l -
CTB
dm,s) lm
(5.5)
noe-1
qP=
(5.8)
i=O
where noe is the number of overlapping elements. The procedure will then
be repeated for the next phony element.
98
Chapter 5.
Continuum Da.sed Normal Contact. Algorii.hm"
Avm-1,m+l
____
n _ __
m+l
m -
(time=t)
(5.12).
lm-1,m+l
The expressions (5.11) and (5.12) can be solved for Pm,

m
n =
(time=t+M)
(!::' - !PCm)
(5.13)
Mm
(5.14)
m
(5.15)
Figure 5.4: Updating the sla.ve surfoco motion.
where Mm and Mp are the lumped masses assodated to nodes m and p,

respectively, and !::;" and f'!:, are the normal components of the nodal forces
at nodes m and p due to the intern~ stress states.
In updating the slave surface motion, TOODY projects the slave nodes
normally onto the updated master surface, and then moves them tangentially according to their tangential velocities and accelerations (see Figure
5.4),
This scheme, repeats the foregoing process of creating phony nodes and
elements to accelerate the slave surface. After updating the geometry, slave
nodes are projected normally to the surface, an<l the normal cornponents of
velocity are then computed by inter..Polating from the master side.
(5.9)
5.2.2
The next improvement was provided by TENSOR finite difference code

wit.h closeHt results to a symmetric algorithm. The approach adopted dynamic equations of rigid body rnotion to more accurately derive the necessary constraint equation for the normal acceleration.
Referring to Figure 5.3, point p can be treated as a particle that has no
uonal velocity relative to a rotating coordinate system attached to master
node m. The normal acceleration a~, is then defined by:
DYN A 2D .implemented the previously developed hydrocode algorithrn in a

systernatic manner, clearing sorne of the ambiguities existing in the original
.. ,
approach [54]. The following steps are followed:
(5.10)
where the normal direction at node m is assumed to be perpendicular to a

line connecting master nodes m - 1 and m + 1 ( tangential direction), and
ac is the Coriolis acceleration,
DYNA 2D Formulation
Slave N ode Search

The location of each slave node is updated by identifyng the first master
node of the master segment on which it lies, or by identifying its closest
master node if it is not in contact with the master smface.
According to Figure 5.5, the slave node s lies on the master segment m -1 .
if,
LJ
(5.16)
(5.11)
and
Wm
is the angular velocity at node m,
otherwise on the master segment m. r,,J is tlw position wctor

points i and j.
.. ;, < ..
connect~'.,;.:.
Chapter 5. Continuum Based Normal Canta.et Algorithms
100
101
'1"
Mass Distribution
The mass of the slave elements a.long the master surfacc is attributed to
the appropriate master nodes. One half the mass of each slave element is
distributed to the corresponding master nodes if both slave nodes of the
slave segment are in contact with the eiement. If only oue slave node is
in contact, one half the nmsH lying between the ('()11tact.i11g 11ode and the
centre of the slave segment of the slave element is distril.Htt.cd. No masa is
distributed, if neither of the slave nodes are in contact.
The procedure of adding the contribution of the mass of slave node, M 8 ,
to the mass of master nodes can be summarized according to the terminology
of Figure 5.6,
Figure 5.5: Slave node search.
Slave Element Search

For each master segrnent, the overlapping slave elements are identifi.ed. Only
the overlapping slave elements contribute to the contact pressure loading
on the master segment. Again by checking the sign of the inner product of
the segment's tangential vector tm and different vectors connecting mas~er
nades m and m + 1 to slave nodes s, s - 1 and s - 2, the overlapping sla've
elements are identified, if any.
m
Figure 5.6: Distribution of the slave element ma.ss to the master nodes.
(5.19}
Tensile Interface Force

Each slave node that lies in contact with the master surface, is checked for
the existence of a tensile interface force. In absence of any adhesion, if the
normal acceleration of the slave node relative to the master segment, a:/m,
is posi tive, the slave node is set free and the normal acceleration is released.
_ d's,m+l -d'1,m ,_ 1
l,
i~ s
(5.20)
m+l -
s
M m+2
d's,m+2 -d's,m+l ll,f

la
ll s
(5.21}
M:n
(5.17)
i
where
is the contribution of the mass of slave node s into the mass of
master node m. ls = ra,s+1 ta is the length of slave element s, t 8' is the
slave taugential vector, and
where ( 8 is a local coordim,ite,

1
t = { r;,m
s,m
O
(5.18)
d'
If the slave node s lies very close to the master node m or m + 1, the
angle Bm-l,m+l or Bm,m+2 will be used to determine thenormal component
of the accelerations in (5.17).
t,
- { r;,m+l . t,
s,m+l -
la
if
if
if
i/
r~,m.
ts
>o
r~,m t 8 'SO
r~,m+l . ts < ls
r~,m+l . ts ~ ls
(5.22)
(5.23}
and vector r~,m. connects the node s to the middle of master segment m.
102
Moment um Balance
_ dm,s+2 - dm,s+l s+l

lm
ef f
Ps+l -
Using the principle of conservation of momentum, the post-impact velocity

of the master node m can be obtained,
(5.24)
103
(5.28)
and if the iilequality (rm,s+2 tm ;::: lm) is not satisfied, .s is incremented by

one and the procedure continues untff the inequality is satisfied. The total
pressure for master segment mis then computed,
ncs-1
ncsn
M::i = l\f~ + L M:n
(5.25)
Pm =
t=O
Pz+i
(5.29).
~:. ' - . ' ' 1 ;
i=l
where ncsn is the number of contributing slave nodes, Mm is the mass at

node m, Mfn is the contribution of mass of the slave node i to master node
m, ancl superscripts and b denote post (after) and pre (before) impact
q11n11t.it.i<~H, n~Hpnct;ivdy. v~ n.nd v:i are tho normal velocity of master node
111. urn 1 Hlav<f 110dc t1, n!Hpctlvly. Tho tuugeutla.l componont of tlw volodty
vector remains unchanged du:ring the momentum balance.
where ncs is the tqtal number of contributing slave segments. The pressure.
is then distributed to the master nodes m and m + 1.
Using_ the TENSOR algorithm, the normal interface force at master node
m is defined by:
Contact Pressure
From the state of stress in the slave elements that are in contact with
the master surface, a contact pressure and its equivalent nodal forces are
computed for each master segment, based on the approach given by TOODY
code.
With mfornnce to Figure 5.3, the contribution of pressure to master

Hcgmcnt m from the sla.ve element .<1 1 p~, is defined by:
. (5.30)
where ac is the coriolis acceleration, F![' == Mma:~b and f 11in is the normal:
internal force in master element m. The normai ncceleration of the master
node mis then computed:
(5.31)
m
Ps
dm,s+l - dm,s O'

=
lm
eff
(5.26)
where <J! is. the effective stress in slave element s perpendicular to master
1
segrnent m,
'
s
O'ef f
s s+l
f n'
(5.27)
l,,
where f~ and ~+ 1 are the internal normal forces in slave elements s and
s + 1 respectively.
pr:
If the inequality (rm,a+l tm ~ lm) is satisfied,

becomes the master
scgli1ent pressure, otherwise the contribution to the pressure of the next
ov1~rln.ppiug Hlavc~ dt~1.nnnt is computed accordingly,
Updating the Slave Motion

The motion of slave nodes not in contact with the master surface is not
changed, whereas the motion of those slave nodes in contact with the master
surface must be adjusted. In DYN.A, the slave velocities and accelerations
normal to the master lineare reset, not imposing tlw slavc t.o rcmain exactly
on the master surface.
For the slave nades in contact with the master segme11t ni, the tangentia.1.
velocity and acceleration, vf anda:, are computcd from the slave element
global velocity and acceleration vectors, and the normal components a.re
interpolated from thc relevant master nodes,
104
Chapter 5. Continuum Based Normal Contact Algoritlims
105
(5.32)
(5.33)
Then, the new velocities are the combination of the tangential components from the slave side and normal components from the master side,
while the new accelerations can be computed from the dynamic equations
of relative motion.
5.3
DYNA Penalty Formulation
The first practica! penalty based finite element conta.ct analysis was developed by the DYNA team in the eighties. It included both explicit and
implicit versions. In implicit code; a contact interface is treated like any
other element class, whereas in. the explicit code, the implementation of the
penalty method is straightforward. The main advantage of the method to
the earlier developed hydrocode algorithms was in maintaining the symmetry of the approach, eliminating any biaE. in choosing the slave and master
surfaces [54].
A general interface treatment using penalty method may be outlined as .

follows:
.,~
Figure 5.7: Node mis the closest master node to the slave node s.
For each slave node s, locate the closest master node m, and corre-
(5.34)
sponding master segments that include m.

Locate the position of the slave node on the master surface.
If s has penetrated the master segment, compute an interface force
vector and an interface stiffness matrix.
The procedure is repeated for all master nodes.
5.3.1
Search For the Master Segment
Figure 5.7 shows a typical contact problem in which four master segmenta
can harbour the slave node s, and node mis the closest master node.
If m and s do not coincide, s can usually be shown to lie in a master
segment i via the following tests:
(5.35)
where rm,i is the position vector along the common edge of the master
segments i and i - 1 and outward from node m, t 8 is the projection of the
g vector, connecting points m and s,
ta =g-(gn)n
n = rm,i
lrm,i
rm,i+l
X rm,1+il
(5.36)
(5.37)
since the sliding constraints keep s close but not necessarily on the master
surface, and since s may lie near or even on the intersection of two master
106
L_;
n
LJ
segments, the inequalities of (5.34) and (5.35) may be inconclusive, failing

to be positive or giving more than one positive result (54]. In this case, s is
assumed to lie along the intersection which yields the maximum value for
the quantity
107
with,
/i(~, r) =
L N1x~i)
(5.40)
j=l
grm,i
lrm,il '
5.3.2
i = 1, 2, ...
(5.38)
(5.41)
Finding the Contact Point
where Ni are the standard shape functions, and xJl) is the nodal coordinate
of the jth node in the ith di~ection. The normal to the surface is non-zero,
A contact point is defined as the closest point on master segment to the

slave nade s. If a quadrilateral finite element discretization is used, each
point on the master surface segment i, can be represented, by the position
vector r (sce Figure 5.8),
(5.39)
(5.42}
The contact point coordinates (ec, TJc) must satisfy the following conditions:
(5.43)
(5.44)
slave node
In DYNA 3D, equations (5.43) and (5.44) are solved numerically to
compute the contact point coordinates (54).
5.3.3
Contact Force
If the slave node penetrates through its master segment, an interface torc~
is applied between the slave nade and its contact point, with a magnitude
proportional to the amount of penetration g,
g
= n [r, -
r(ec, TJc)]
(5.45)
u
!~ = -a:gn
Figure 5.8: Finding the closest master point e to the slave node s.
i/
<o
(5.46}
where J~ is the contact force acting on the slave node ::;. An equal and
opposite force is distributed over the master segment nodes,
1.08
if
<o
5.4.1
(5.47)
In DYNA, t.lw penalty stiffness factor, ~ is given in terma of the bulk

rnod11lm1, K, volume V, and tlm facc ama A of the master element,
eKA 2
V
Chapter 5. Continuum Based Normal Contact Alp,Drithrns
109
Contact Geometry
Figure 5.10 shows a typical slideline between two contacting bodies, defining
1.he contacting nodes nnd H(!gments.
(5.48)
a=--
where f. is a scalc factor for the interface stiffness, usually taken as 0.10.
Larger values reported to cause instabilities unless the time step size is
reduced.
11
LJ
lJ
5.4
2D Node to Edge Penalty Contact
In this section, a simple and logical approach for two dimensional nod~ .,to edge normal contact based on the penalty method is discussed. The approach was originally <leveloped in 1980 by Circa {83), and later documented
by Crook [52).
In this approach, the canta.et surfaces are explicitly defined. Two surfaces are then linked to forma slideline ora contact pair (or,contact couple).
A simple global search algorithm will be followed by an accurate local search
method to identify u.ll coi:itacting couples. A penalty based formulation is
adopted for contact inter.lon and evaluation of contact forces.and forming _
the non-consistent and consistent tangent matrices.
Both contact surfaces may be searched to give a symmetric treatment of
the contact surface, as opposed to earlier master-slave algorithms. To verify
the problem, in Figure 5.9, if point 1 moves up into edge e, then point 2
will penetrate edge e2. Therefore, we need to search for both surfaces at
each slideline.
2
o
Figure 5.9: Two surface search for a slideline.
segmentm
Figure 5.10: Definition of a contact geometry.
Th~ contact surfaces are defined by a list of element edges (segmenta)

stored m order from the start of the surface. The order is defined by insisting
that the outward normal is constructed with a right hand screw rule (Figure
5.11).
'
n (normal)
Figure 5.11: Definition of normal and tangential vectors.
Ea.ch segment nades of cach surface will be dwcked For pt~netration of

the segments of the opposite surface.
110
5.4.2
111
Global Search Algorithm
segment m-1
s
t., \
The first step is to find the closest node m on surface 2 (master) to node
s on surface 1 (slave) (see Figure 5.12). This is achieved by evaluating the
distance to each node, and finding the minimum value:
m-1
Figure 5.13: Finding the segment in potential contact with node s.

-
(2)
m - 1,nnode
(5.49)1
(5.50)
where n~~de is the number of nodes on surface 2.
The global search is numerically very expensive, and it should be avoided

unless it is proved to be necessary. Fortunately, for many applications such
as most metal forming processes, the global search algorithm can only be
performed once. After one global search, the next contact node is simply
found by knowing the direction of motion and examining the next node on
the list.
then there is a
pot~ntial
contact with segment m, and if
(5.51)
there will be a potential contact with segment m - l. Figure 5.13 depicts
the above mentioned cases based on the sign of the inner product of the
position and tangential vectors.
If only one potential contact is encountered for point. s we use the stan
dard case of contact with single segment, otherwiHe t.lie :;pedal case of contact near a corner has to be used.
m-1
segment m-1
5.4.4
Normal Contact Interactfon
Contact With a Single Segment

To compute tlie normal contact force for preven:ing the penetration of a
conta.et node into the associated eontact segment (target), a. standard steJJ
by step proced ure is followed ..
Figure 5.12: Finding the closest node to node s.
The first step is to check for penetration. The normal vector is de:fined
by:
5.4.3
LJ
Local Search Algorithm
After identifying the closest point to a contact node s within the global
search algorithm, the associated contact segment on surface 2 is located
(Figure 5.12). This is done by verifying the sign of inner product of the
position vector r ms nd the segments tangential vectors tm and tm-1
If the following condition is satisfied:
n =ea x t
"(5.52)
and the penetration distance is computed

Un= rm,a n
and the penetration occurs if 9n < O.
(5.53)
112
Chapter 5. Continuum Based Normal Contact Algoritl11w-1
'l'ho confod co11Htrnl11t. wlll lnvolve the <~011l.1wl.t 1r

uodc8 ( 1. tm<l 2) ou the tw:gct segment, leadlng t.o:
lit 1d1~
113
.1
11.1u l
th(l two
(5.55)
t
1
where
(5.56)
r----1.
-1
The magnitude of the contact force in the local system is defined by:
Figure 5.14: Local coordinate system and projected conta.ct point.
e= Un
The second step is to locate the contact point. By defining a local
natural coor<linate Rystem E [O, 1], anda local segment numbering system
according to Figure 5.14, the !ocal contact point is defined as
{5.57)
where a is the normal penalty coefficient. In terms of global variables,

becomes:
(5.58)
(5.54)
The last step is to compute the contact force and its associated stiffness. The model is constructed by inserting the penalty spring between the
contactor node of body l and the target segment of body 2, as depicted in
(5.59)
Figure 5; 15.
e.
where Ni = 1 and N2 =
are the target segment shape function8 .
Figure 5.16 shows the geometric representation of the contact forces. It is
worth noting that in any case, the angular morne11t.11m of the body should .
not be changed.
.
The contact stiffness is then defined by:
Kc
' T
= aN,N,
(5.60)'
The penalty stiffness matrix, defined in (5.60) is not consistent; i.e. it

does not contain the second order terms. Therefore, the described formulation does not converge quadratically, or at all in sorne circumstances.
I
Figure
Contact Near a Node
5.J:t,; Equivalent penalty linear spring.
A node to edge ha.sed canta.et problem which happens near to a corner

node, or in more general form of multi-node <:11ntacts near the cornera, .'
114
115
this is not a practica! way in general problems, thus perfonning a costly

accurate local search is unavoidable.
5.4.5
!/
Figure 5.16: Distribution of the contact forces.
Consistent Node to Edge Contact
lt luu; n.lrmtdy btmll pointod out thttt the 8lmplo w111dt:v 11wt.l1<1i I wlll le"d to
a stiffness matrix which is not in a consistent form. By usi11g thc prlnclple
of minimization of the total potential energy of the system, including the
contact constraint, a consistent form for the stiffness matrix can be derived.
As described earlier in Chapter 3, the equilibrium condition can be expressed as:
an T o:a;, =0
-+g
8u
8u
Body2
;f
Rody I
.,
....
(5.61)
Considering the contribution from the conta.ct constraints alone, the

tangent matrix is obtained by further differentiatiou of (5.61) with respect
to the nodal variables. Therefore, the force vector and the tangent stiffness
matrix for a single element are defined by:
JI
o
l)g
(5.62)
~-"00-
8u
Figure 5.17: Oscillatory behaviour around the corners.
Ke
pose numerical arnbiguities as well as numerical instability in particular
conditions.
For instance, there are. always oscillations between two comer nodes in
iterations. Actually, we may have. oscillations even without ;having any real
penetration. Figure 5.1 7 shows a typical situation where during the analysis,
poiat. 3 may soem to be. penetrated through its opposite segment, causing
the contact forces to be generated to return the node to an unpenetrated
position. At the same time another force with opposite direction is acting
011 node 1 to pm;h it out of the regiC>n occupied by body 3 near the comer
point 2. This causes oscillation of the systems around the corner points.
g)T + au
ag + gTo: fJua ( 8u,
og )]
= o: [( fJu
(5.63)
The first term in (5.63) corresponds to the stiffness matrix (5.60).

Differentiation of the gap function
To evaluate Equations (5.62, 5.63}, derivative of tlH~ ga.p f1111ct.ion g has to
be computed. Here, the same methodology as deHcribed by Crook [52) is
followed. Referring to Figure 5.18, the gap function is defined as:
(5.64)
There exist remedies to treat this set of problems which may frequently
happen in progressive fracturing phenomena, for which several small parts
of a fractured and fragmented body may interact with each other simultaneously. One simple remedy is to slightly enlarge the end segments, so a
11ode to (~dge contad can be performed without any ambiguities. However,
The first variation of {5.64} is:

8g
= n (fo, - u1) + 6n (xs -
x1)
(5.65}
Discon~inuum
116
Mechanics
Chapter 5. Continuum Based Normal Conte.ct, Algol'itllms
117
Using (a b)c = (e b)a, will lead to:
1..............................................
and finally:
s
~
(5.74)
Figure 5.18: Defining the gap function.

Substitutions _of (5.74) in (5.67) and then in (5.65) gives,
It is worth noting that we did not consider the change of n (n) during the contact in previous approach. n can be expressed as:
(5.75)
= e3 x t
(5.66)
Noting that
ec =
t(xs - ::r:1). t, (5.75) becu11H~H:
(5.67)
(5.76).
and for the tangential vector t,
or,
(5.68)
(5.77)
and
(5.69)
Therefore, an expression identical to (5.58) is clerived for evaluation of.

the contact force vector:
To evaluate l, we have
(5.78)
(5.70)
The first term in the consistent tangent matrix (5.63} becomes:

(5.71)
uK c~u
so,
(5. 72)
uKc.u =
o! )
= a6u [( 8
1
'
a ] ~u
0~
au [(6u,, - N16u1 - N2bu2) (n 0 n)

(u,, - N16u1 - N2u2)]
(5.79)
(5.~l>:::
-\~::/('
118
Chapter 5. Continuum Based Normal Contact .Algoritillw;
By using the definition of N 8 shape function, the linear part of the

stiffrwss nmtrix is tJwn defined by:
Klinear -
aN
NT
. s s
(5.81)
which is identical to (5.60).

To derive an expreHHion for tho nonlinear stiffness term, the second variation of (5.75) is derived, which will eventually result in the following consistent form:
(5.82) (
where
(5.83)
Figure 5.19: An edge to edge contact.
--
(5.84)
'. (5.85)
allows for a numerical solution of edge to edge contact formulation. The

contact traction of two interacting line segments can be approximated by
considering each body as a strip loaded elastic half space.
Petrinic (1) has adopted and successfully implemented this methodology
to analyse various solid/ granular interaction prohlems. According to [1] the
vertical displacement of a general surface point i:-; defined by
Note that for z~ro gap, the nonlinear term is zero. Also, as g becomes
smaller, the significance of the last term in (5.82) becomes smaller, however,
for the sake of convergence we need this term.
(5.86)
where Pn is the intensity of the normal surface pressure at the other end of
the contact surface, b is half the width of the contact surface and
5.5
Edge to Edge Contact
Figure 5.19 illustrates a simple contact problem defined in terms of edge/ edge
contact couples. Although it is notas popular as the node to edge contact,
it has heen used by many research groups working on general crack propagation problems by discrete element technique.
Utilizing the Boussinesq solution for the concentrated force acting on an
elastic half space in combination with the Winkler elastic foundation the'ory
w(b, x) = 2b2 ln(2b) - -~ ln(x) + (~

-2.
2
.
k1
k2
= ia~t
2b2 ln(2b - J') + x(2b - x). (5.87) .
(;~
(5.88)'':'. .' :'
120
Chapter 6
Frictional Contact
Figure 5.20: Description of an edge to edge contact.
6.1
where G represents the shear modulus and v is the Poisson's ratio. The
normal contact stiffness can also be derived
(5.89)
and finally the normal contact force can be evaluated
F~"
Pn A =
= b (PnA + PnB)
(5.90)
In this chapter, the main features of computational frictional contact algorithms are discussed. The chapter is divided into five parts. The first
section, explains the basics of frictional contact problems including the kinematics and principies af the plasticity theory of friction. The second sootion, provides a detailed formulation for two dimensional frictional modela
without frictional hardening. The thh:d section generalizes the approach
to three dimensional models and the fourth section will introduce frictional
hardening within the 3D modela. Finally, a 2D node to edge model and a
3D node to quadrilateral face model for Coulomb frictional contact wi1Gho~ut
hardening e.re presented.
6.2
xconU Q
UnA
n
n UnA+UnB
Introduction
Frictional Contact - The Basics
(5.91)
In this section, sorne of the fundamental aspects of a frictional contact. behaviour are discussed and sorne f the pri11cip~.l issues regarding the math-.
ematical formulation of the computational methods are addresse<l.
6.2.1
Frictional Contact Model
Because of microscopic nature of frictional coutuct, includiug the mlcroscopic geometrical conditions shown in Figure 6.1, friction can not b~ easily
modelled adequately in the macroscopic numerical analysis of many. applications such as metal forming processes.
c__.I
121
122
Tool
',
LJ
lnltlal
... ..
con,pgurutlon
>-.::~~r~~~~::::
.
' ..
~
123
Chapter 6. Frictional Contact
'
'
Master
<
Current
Configuratlon
Figure 6.1: Microscopic nature of frictional canta.et.
The frictional sliding is simulated by assuming mathematical relationships between contatt stresses and other related state variables, which describe the conta.ct conditions. These relationships can be detived from the
micromechanical models of friction and then incorporated into an appropriate constitutive theory [84). At first, only a quasi static type of friction is
considered, which occurs, for example, when two metallic surfaces are slowly
pressed together and relatively displaced in a quasi static equilibrium [85].
The most conspicuous mechanisms contributing to this type of friction
are:
plastic deformation ~f asperities.
configurations. In the .reference or initial configuration, the objects are yet

to make any contact, while the masterand slave objects are in contact in the
current configuration. In a metal forming application, thc tool is regarded
as a rigid master bo<ly and the workpiece is t.rcat:cd ns n deformable slave
body. It should be noted that, however, in a general contact algorithm,
. the role of master and slave bodies are interchangeable and th~ term. is
considered following the historical development of the contact algorithms.
To begin with deriving the kinematic relations, Figure 6.3 depicts the
master and slave bodies at a current configurntion, showing both relative
normal and tangential displacements.
formation of elastopl;istic junctions on the contact surface.

possible fracture of ju11ctions.
strong coupling between normal. and tangential plastic defo~mations.
Figure 6.2: Cont.act of a rigid master and a deforma.ble slave solid.
In this figure:
dissipative mech~nisins caused by the junction deformation.
Thermal coupling aspects of friction, which recently received considerable attention in the computational.literature, will not be considered. Their
inclusion, however, is possible without any major modification in the context of the general thermo-mechanical coupling strategy.
6.2.2
e
Yn
Yt
outward normal on the master surface

tangent orient~tion
directional coordinate along the master surface
normal gap
tangential gap
where E is measured from a fixed point on thc master, and Yn and 9t

two independent kinematic variables.
Basic Kinematics
are
At the current configuration, the position of slave node S, ros, is defined

by:
In this section, basic kinematics and geometric assumptions required for

deriviug tlie necessary e'quations governing a frictional contact behaviour
are explained. Figure 6.2 shows a typical metal forming problem in two
l;'(I
ros=Xs+us
(6.l)J: .
l . . -'
-~;_ji~l~f-~"
124
125
where Mo is the master surface initial point of first uninterrupted contact

with the slave node S,
......
(6.7)
where to is the time associated with the first contact of slave point So and
master node Mo.
Defining the tangential gap in the curvilinear coorclinate system (Equar
tion 6.6) constitutes one of the main features of described kinematics. It is
a nonlinear basis which is not in accordance with the main cartesian basis
in which the equilibriurri equations are to be stated.
The relative displacement of the slave node with respect to the master,
gr, can be obtained from the known master movement
Figure 6.3: Relative normal and tangential displacements of two contacting

bodies.
J
1
LJ
gr=
where res and X s denote the current and initial positions of slave node
S, respectively, and us represents the deformation from initial to current
configuration.
1
t
to
'
dgt(t) =
1t
dg1(t)t(t)
(6.9)
to
In general [86],
Point M geometrically represents the closest projection of S onto the

master surface r m:
.
(6.10)
(6.2)
Also,
The normal and tangential gaps are then defined as:
,r
-~
''
,.:.;
(6.11)+\
Yn = 9n n
9n
LJ
= (xs-XM) n
9t = 9t t
t<i
(6.3)
(6.4)
(6.5)
(6.6)
'-...
In general, due to the variations o n a.nd t, the decomposition g"

9n + 9t is not valid. For flat master surfaces, however, this u~l:omp!l-qition
is valid, and it may also be extended for the cases of small amounts of slip
and nearly flat and rigid contact surfaces.
( ~Jmptdt 6. Mlct.Jmml
DJ..,contlnuum Muduudcs
121
C011t.11d
Stick-Slip Decomposition
The present model defines contact kinematics perfectly and unambigu1

ously for:
The starting point, motivated by similar decornposition of the strain into

elastic and plastic parts in the classical theory of plasticity, is the additive
decomp9sition of the gap into two parts; stick (adherence) and slip.
any amount of slip.

any shape of rigid master surface.
(6.14)
any possible friction law.

(6.15)
It should be noted that extension of the present concept of unilateral
constrnint to thc frictional contact of two deformable bodies with nonlinear
geometries is not straightforward and would gravely complicate the numerical formulation.
n
LJ
6.2.3
and in scalar form
.:,
(6.16)
'(6.17)
where e and P superscripts denote stick (elastic part of gaps) and slip (plasti~ : .
part of gaps), respectively. The stick gap is assumed to be reversible, while
the slip is irreversible, or permanent.
The objective of a computational theory of friction is to provide a computational description of the relationship between contact stress and stick/slip
phenomena at the interface of contacting objects.
In the presence of high pressures, some plastic deformations me.y take ;

place on the contact surfaces in the normal din'diou. 011 a macroscoplc.
level, however, the influences of these phenommm are rnprcsented by ples-''
tification of the contacting materials. Therefore, irreversible behaviour can
be simply neglected in the normal component of t.he gap:
In a pioneering work, Fredriksson (87) noticed that the theoretical deHcri ption of frictional contact behaviour can be derived from the classical
theory of plasticity [88, 86, 89). Table 6.1 indicates the analogy between
various frictional contact and plasticity features.
In the following sections, the basic features of the plasticity theory of

friction are examined in more detail.
Un =g~
Table 6.1: Plasticity theory of friction versus classical elastoplastic theory

Plasticity theory of friction
LJ
Stick-slip decomposition
Stick res.use
Slip criterion
Wear/tear rules
Slip rule
Classical elastoplastic theory

+----+
+----+
+----+
+----+
+----+
Elastoplastic decomposition
Elastic response
Yield criterion
Hardening/ softening rules
Flow rule
Stick Response
.,
Sticking is caused by the elastic (reversible) deformation of the asperities
on the contact surfaces. The sticking iaw C$ln be suitably described by
decoupling the normal and the tangential directions.
The surface traction (force per unit area) can be represented by tfft,
and r as the work conjugates of the normal and tangential gaps, 9n ~el
128
g t, respectively. un and T can be. easily included in the mathematical

equilibrium expression for a deformable body.
(6.24)
The directions of the surface tractions are uniquely defined in terms of

the same direction vectors n and tas for the gaps Yn and 9t As a result,
the following scalar expressions are sufficient for the description of sticking
behaviour:
Numerical experiments ha.ve indlcated that a lincnrized interface law

<loes not change the global solution [92). Nevertheless, it should be noted
that the described linear and nonlinear physical models are not compatible
with the more usual impenetrability approximation of the contact interface
kinematics, expressed as the:
(()~20)
{6.25)
(6.21)
If the impenetrability constraint is assumed, the variational principie~

the penalized fon11 beco~es
129
= r(gn
Equation (6.20) which is relevant for the normal direction, can also be
regarded as the frictionless contact law.
(6.26)
The microscopic contact area differs from the apparent, macroscopic

one due to the fact that the 'two bodies come into a contact only at the
asperities of the rough surfaces. Severa! experimental based formula.e have
been proposed for the pressure-gap relation, where the pressure is an average
normal contact traction and the gap is a chosen geometrical parameter,
usually the true contact a.rea or the normal gap (90].
which will be exactly the same formulation as Equation (6.24) if the nu

merically convenient l~rge penalty number an is equal to the parameter
One of the most common pressure-gap relations is defined by:
where en is a contact stiffness parameter; a material property obtained on

physical grounds. For most metallic materials, Cn is proportional to the
elastic. modulus of material, and mis set to 2.
For higher contact pressures, an exponential law is expected to better . ~
suit the experimental data [91}:
= a. exp( blgn 1)
The sticking behaviour in the tangential direction, equation (6.21), can

be defined in the same way,
(6.22)
O'n
Cn
=Ct
Uf
(6.27}
The linear law of (6.27) is a reasonable approximatiou, acceptable on ml-.

croscopic level and convenient for numerical implementation. Nevertheless,
it is in contrast to the following experimental observations:
1
Dependence of the tangenJ;ial traction 011 the irrcvernible part of the
tangential microdispla~ements is slightly nonlinear [93].
The bodies in contact show relative tangential microdisplacements to
the contact area, even befare sliding starts. This is in contrast with
the Coulomb friction law that assumes no tangential movement occurs
until the maximum tangential force is reached.
(6.23)
It is worth noting that these nonlinear contact laws are primarily important for the accurate description of contact behaviour on the microscopic
level. However, the normal micr~displacements are practically negligible
compared with the overall macro displacements. Therefore, only a linear
sticking law in the normal contact direction will be considered in the present
model:
Finally, .similar to the normal sticking law, a penalized variational

mulation for tangential sticking can be derived
which is similar to (6.27) if Clt = Ct
130
Chapter 6. Fi'ictional Contact
Slip Criterion
LJ
131
2) Constant Shear Law
A slip ( friction) criterion is used to indicate whether tangential sticking or

slipping occurs along the contact surfaces. It can be generally stated as:
TSL
Tm
= sUs
[0, 1)
{6.34)
where P, 8 is the constant shear coefficient of friction, a11d
<P.(uc w) = { < O :
'
=O
st~ck
slip
(6.29)
'
{6.35):
where uc = { <Jn, T} contains all contact tractions and w indicates an interna! variable for describing the tear and wear phenomena (which will not be
included in this section). For 3D applications, tangential traction T can be
expressed in terms of two independent directions t 1 and h in the tangential
plane:
(6.30)
u a is the current shear strength of the deforming material with <1'1J yield
stress. This model is m01;e suited to cases of high contact pressures such a8
bulk forming operations.
3) Parabolic Law
In this model, the frictional resistance is constantly rising, but with a decreasing rate (92):
Assuming isotropy for general 3D friction,
(6.36) .
(6.31)
a, {3 and n depend on surface properties such a.s roughness or hardness. For
and far the 2D model:
(6.32)
Norm of the tangential traction, llrll, in a contact analysis is similar
to the equhalent stress function, f(u), in an elastoplastic analysis. In the
same way; the maximum allowable tangential traction (slipping stress), TsL,
is analogous to the yielding stress, u y.
= 1 and a + f3 =
the Coulomb law is obtained.
4) Avltzar Model
A combination of Coulomb and constant shear laws has been derived by ari
upper bound analysis of the interaction between the rigid and deformable
asperities of two contacting surfaces [94):
t
for
Here, the most commfl:>. choices of friction law are reviewed. Any choice
for the friction law of a specific application has to be experimentally verified.
(6.37)
for
'>
5) Sinusoidal Modification
1) Coulomb Law
TSL
= lunl
E (0, 1)
(6.33)
where is the coefficient of friction. Studies have shown that this model
is adequately applicable to the frictional conditions with moderate normal
contact tractions, such as sheet forming problems.
ThiH model provides a 1m1ooth transltion to th1~

constantly changing slope (un):
1011:-;t.11111. slarn1.r
regime wlth
.
! or lcrn < JI2 Im..

to
or
Jcr
>
JI!=
!
n 2 o
1
(6.38)
132
The way two contacting bdies slide against one another can be classified
to two distinguished classes of problems. In the first cla.ss 1 the bodies are, .
rubbed against each other in an arbitrary direction, creating a.n lsotroplc ,
wear phenomenon. While in the second class, the sliding is orientt3d along
a prefer~ntial direction. Thus, an anisotropic tearing phenomenon occurs~
Wear and tear phenomena perform as isotropic and anisotropic softening
processes, respectively and reduce the overall frid.ional strength of the con
tact surfaces.
6) Parabolic Approximation
Auother model with a smooth transition to the constant shear regime is :
f or . lanl < 2~
for
lanl ~ 2~
(6.39)
Any defined wear and tear model may be included through the appropriate interna! variables w in the general slip criterion
7) Trigonometrical approximation
To avoid singularities of the original and modified approximations of Avitzar

rnodel, a trigonometric approximation can be defined which covers the whole
range of O'n, but reaches Tm only in the limiting case [95]:
nLJ
133
<l>(<Tc w) {
'
< O : sti~k
=O : slip
It should be noted that based on sorne theoretical considerations a.nd experimental observations, an internal scalar variable w = w( aw), describing
wear softening, is more work dependent
(6.40)
Figure 6.4 compares 2D representation of these friction laws.
....
..-(1)
than slip dependent (84) .
...
(6.43)
Slip Rule
1
1
p_.
aw
g - 'Y Bue
lJ
Figure 6.4: 2D representation of various friction laws.
Wear and Tear Rules

Asan analogy to the hardening/softening theory of plasticity, the kinematic
variables of directional slip and accumulated slip can be assumed as work
conjugates with the associated forces of wear and tear, respectively [86].
;.
Analogous to the flow rule of plasticity, direction of the irreversibi~ slip is

defined by a slip rule. This is'8.ccomplished by introducing a slip potential
'11, whose gradient uniquely defines the irreversible slip direction,
(6.44)
where 'Y is the proportionality constant. The component form for a general
3D slip rule becomes:
.:J'
Ytl
ltt_
='Y'Hii
(6.45)
134
Chapter 6. Hictio11al Contact
135
Associated slip rules has }ee.Jl used to a.void the non-symmetric tangent modulus despite its unjustified phy:rn.l chnrnct.l'l'Htics [87).
1t1
Here, a non-associated slip rule with isotropic frictional conditions is employed and consistently linearized. The slip potentials can then be defined
as:
(-a )
'1f(T)
= llrll
for 3D
\Jl(r)
= lrl
for 2D
(6.46)
If the slip potentials (fi.46) is utilized to

slip direction is defined as:
exm~~s
the Hlip rule (6.45), the

'
P-'A
9t
- "'( llTll
Figure 6.5: Associated and non-associated slip rules.
gf ='Y sign(r)
where subscripts
and
' = 12
!
for 3D
for 2D
, (a.41Y:.
are orthogonal bases in the tangential plane.
The third part of Equation (6.45), states that an irreversible normal gap
increment may exist for a frictional slip rule. In other words, even in a pure
frictional slip, an inelastic normal gap increment may be generated when
the slip criterion (6.45) is satisfied. This unacceptable physical behaviour
is in contradiction with the previously discussed basic assumption that no
slip occurs in the normal direction.
Therefore, the slip potential must not be a function of O'n As a result,
an analogy with the classical associated flow rule of metal plasticity is not
applicable, and a u n-independent non-associated slip rule must be adopted.
An associated slip rule would lead to uplifting in the normal direction under
frictional conditions where un is always compressive. Figure 6.5 compares
the associated and non-associated slip rules.
Thc non-associated slip rule will lead to generation of non-symmetric
tangent stiffness modulus. Two methods have been proposed to avoid creation of non-symmetric modulus:
The first appro~ is to utilize a two step algorithm, where in the
firnt step, the stress normal to the interface, O'n, is obtained from
equilibrimn conditions. In the second step, the slip rule is specified
only in the tangential plai:te (88). The linearization in this approach is
not consistent, and hence it <loes not provide quadratically convergent
solutions.
Equations (6.47) satisfy the condition that the direction of frictional slip '
is determined by tangential frictional contact traction and is independent
of the normal contact traction.
6.3
2D Frictional Contact Models Without;

Hardening
In this section, detailed formul~tions for two dimensional frictional contact

models are provided, which are based on the work by Schonauer (84). These . ~'.,,..
models include straight slave segment contacting straight, conve:X and con;,;. tt;
cave are master segments. We will start with geometrical definitions and ~k:.
kinematics of the three mentioned cases, shown in Figure 6.6. It is worth. ,. .
noting that the relative slave node position with respect to the master configuration at the end of previous step, So, must also be stored in the code
database by means of 9to and 9n0
.
In order to obtain nodal forces from the surface tractions, which have,
already beeri defined by decoupled normal a.mi t:rn;ential constitutive laws,
an area must be associated to each slave node:
(6.64)
136
137
where t' is the thickness at the slave node.
Fo1 notationu.l conveuieuce, we define:
a)
a;M'l _ a;M1
-lla;_M_<J___a;_,,M-=-1""'-="11
(6.48)
Current contact traction
(6.49)
Current gap values
9n =
(6.50)
Ut
(xs - a;M). n
= (xs - a;Mi). t
(6:51)
t=
= e3 x
Incremental displacement
uc = (T,un)T
,,..
= (gt, Un)T
= ( Uz, Uy) T
Contact force
6.3.1
b)
a;S -
(6.52)
= n x e3
(6.53)
[y'x --xcy]
(6.54)
t
M,
Mt
a=tan- 1
Un=
feo
= llres -zll
---
llx 8 - zll 9t
(6.55)
= R(a ~ a1)
(6.56)
(6.57)
Aflt :..-: R(a - ao)
Residual Force Vector
A contact mechanics problem can be classified as a minimization problem

with constraint. Details of available constraint enforcing methods and associated variational formulations have been disc11s::;ed in Chapter 3 and will
not be repeated.
The global residual equation is formed by adding the contribution of the

virtual work of the boundary nodes contact forces, Re, to the standard FE
residual vector Rb:
R=Rb+Rc=
nel
nen
e=l
c=l
Are+ Are
{6.65)
e)
z 0 -z 8
n=
llz -msll
(6.58)
t=n
(6.59)
x eg
[y' - y]
M1
a=tan- 1 -
x' -xc
9n = R-
llzs -zll
9t = R(a1 - a)
6.gt == R(ao - a)
~j
(6.60)
where nel is the number of elements, nen is the number of boundary nades
which are in contact and R is th~ total global residual force vector.
The residual contribution for a single boundary nade in contact, S, be-.
comes
(6.61)
(6.62)
(6.63)
Figure 6 .6: Contact kinematics for a slave node S ~n contact with different
master segments; a) straight segment, b) convex cll'cular are segment, a.nd
e) concave circular are segment (84].
(6.66) ''
where T is the transformation matrix,
138
6.3.2 Stiffness C~tribution
[ 8u8~
ag
T2n
Equation (Je = Auc) represents a dependence of contact force to the associated contact area. According to Equation {6.64), the contact area itself
depends on the position of the slave node S and its neighboring boundary
points N 1 and N 2 As a result, variatins of the residual force vector (6.66)
has to be considered with respect to both u and uN.
,_.
Therefore, linearization of (r~ = TT e) yields not only to a (K") stiff11css matrix, but ah;o a (Kn~, i = 1, 2) stiffness contributiqn, which must all
be included at appropriate positions in an assembled global stiffness matrix .
[84]:
(6.68)
~ K
8r~
(6.69)
Kn' = 8r~
(6.70)
8u 8
{)uN
Finding partial derivative of (r~
= TT r = TT Auc) with respect to U
'
"'
"'
~
"'
11
~.
"'
a2g
!/
8u;a~~
''
(6.75).
and the stiffness contribution from two neighboring slave boundary nodes:
= 1,2
'(6.76)
Therefore, we would n~ed to find the first and second gap derivatives, as
well as area derivatives, and to define an appropriate constitutive law for a
slipping regime.
For .two deformable bodies in contact, stiffness terms resulting from
second arder gap variations are essential to achieve quadratic convergence
[96, 97], whereas for a rigid master/deformable slave contact, numerical tests
have indicated that the second order gap variations can practically be neglected. Therefore, the. second bracket in the contact node stiffness matrix
(6.73), which represents the variations of T, need not he computed, saving
computational time, while practically retaining tlie quadratic convergence
convergence rate [84).
It should be noted that Ni is not considered a contact no de ( at least while
considering the slave contact node S) and its associated contact stiffness
term, K~s, must be set to zero. As a result
(6.71)
1
(6.77)
(6.72)
and therefore, the global stiffness matrix Kc (6.G8) will be non-symmetric
(6.73)
Numerical investigations have shown that the stiffness terms arising from
the area variation, can be neglected in the present model without any significant influence on the quadratic convergence rate. Thercfore, as an acceptable approximation for K~, the second term in the first bracket of K:
(6.73) and the whole of K~n (6.76), can be negh~cted
Therefore,
where
(6.78)
(6.74)
Eflg
~
"'
11
The following sections provide closed form sol u tions originally provided
by Schonauer [84] for the first and second gap dt~rivativcs ns well as the
necessary area derivatives for ali three cases defined in Figure 6.6.
Discontinuum Mcchanics
Ghnpter 6. Frictiomil Cont;nct
141
First gap derivatives
(6.88)
Closed form solutions for the first gap derivatives of a straight master segment can be defined by
_a_2U_t_ = -2R..;...(x__-_x_c....;;.)...;.;(y'--_-__;:_Yc~)
8u~8u~
llx -
zc 114,
(6.89)
(6.79) .
(6.90}
(6.80)
. fJ2gn
{6.81)
'.
(y - yc)2
8u~8u~ - lla: - a:cll3
{6.91)
(6.82)
(6.92)
and for convex circular are segments,
(x' - xc)(y - ye)
llx -
(6.83)
(6.84)
8gn
8u~ =
X8 -
xc
llx - xcll
a:cll3
Again, for a concave circular are segment, equations (6.88-6.93) multiplied by -1 can be used.
Area derivatives
(6.85)
Area derivatives, in their general forms may be written as

(6.86)
8A
au.
:::a
[A
8Al
au~ ' oui
For a concave circular are segment, the same equations {6.83-6.86) multiplied by -1 are used.
Second gap derivatives

For a straight segment (with {3
(6.93}.
(6.94)
(6.95).
which can be simplified for problema with constant thickness t,
= n, t and i, j = x, y),
(6.96)
()2gp
--=O
8uf8uj
t.
and for a convex circular are segment, we get
(6.87}
(6.97)
142
1 (x ~ xN)
N
2 llx - X ; 11
aA
lJ
--t
--.:- =
OUx;
= 1, 2
(6.98)
1 2
(6.99)
A =-~t (y"-yN')
OUyN i
11 X -
X N i 11 '
'
ve= [ t0 Ono ]
= 211"X
(6.104)
'fo derive au oxp1e::;::!lo11 for thc slip opera.tor, ~;i~\1rul 1111lhodH hu.ve boon
proposed. One approach considera the terms in (6.103) separately under
slipping conditions and then combines the final results to derive the slip
operator [92, 98). In an alternative approach, the slip operator may be
obtained by considering the whole matrix ~ in the context of return
mapping algorithm which arises from the integration algorithm employed
to up<late the nodal contact traction for the given gap iur.rements [84].
For axisymmetric problems, additional terms may arise due to the thickness variation. If the axis of symmetry coincides with axis y, an assumptibn
is made that the thickness associated.to the contact node is that at the slave
node itself [84]:
Here, the strategy of fully implicit (Euler backward) integration is utilized. For any iteration within the increment j, the elastic trial contact
traction vector is predicted as:
(6.100)
143
Chapter 6. Ftictional Contact
and the only necessary change will be:

(6.105)
= uC.3-l
( uC.)trial
J
.
n
1
Consistent Linearization
Tn thir-; Rcction, the integrntion algorithm and the consistent linearization are
presented for the present frictinal contact constitutive law. The consistent
linearization will provide the consistent tangent slip operator needed for the
contact node stiffness matrix (6.73):
(6.107)
If the slip criterion is satisfied; i.e. the traction point lies in the sticking
domain, the correct traction is set to be equal to the trial traction
(6.102)
where the ~ term must be defined for both sticking and slipping conta.ct.
conditions, denoted by
and DP respectively,
ve
De
if
~<O:
(6.106)
This trial traction vector ( uc.)trial

is checked against violation of the slip
3
criterion
LJ
6.3.3
+ (f1uC.)'1'ial
J
(6.108)
However, if the slip surface is violated, the contact traction return algorithm must be performed. According to the definition of slip criterion and
slip potential,
Slip criterion:
stick operator
<P(un, r) =
lrl -
rsL(an)
(6.109)
(6.103)
DP
if
~=o:
slip operator
Slip potential:
The stick operator is readily obtained from the linear penalty contact
constitutive law,
\ll(r) = lrl
explicit expressions for slip derivatives can be obtained:
(6.110)
Discontinuum
144
8\J!
[ 8\J! 8\J!
(bcf = 8uc = 8r' 8un
l= .
[sign(r), O)
Mech~ics
Chapter 6. Frictional Contact - -
145
(6.111)
(6.120)
(6.112)
which vanishes for the present 2D case as it can be clearly seen from Equa
tion (6.U2).
Following the backward Euler approach, the following residual equations

are obtained:
(6.113)
for 2D
(6.121)
For 30 models, however, the slip potential \JI( r) = llrll includes tangential traction r defined as,
\
(6.114)
(6.122)
Linearization about the known solution (6uc, 6')') is performed by applying the Newton's method,,
The direction of the tangential resultant t = t 1 +t 2 is nota priori known

for the tangential plane in 3D applications. Therefore, B a" will contribute
to the consistent tangent operator (99, 56).
(6.115)
Back to the present 2D model, the following local iterative improvements

for the unknown 6 and 6uc are obtained:
with
(6.116)
(6.117)
The unknowns. are then pdated according to:
(6.118)
(6.125).
(6.119)
{6.126)
No wear and tear rule has been considered in deriving the Equation (6.119).
where i denotes local iteration on the level of each gauss point. The whole
iterative procedure is repeated starting from the residual calculation (6.113,
6.114) and terminated once the selected norms of the residuals rei,smaller
than a prescribed tolerance (84).
_ 8r2 = 0
86
The second term of the right hand side of Equation (6.116) is termed as
Bue
14
Consistent Tangent Slip Modulus
6.4
The corn1iRtent tangent Hlip operator can be derived from expression (6.124)
(6.127)
The simple forms of the expressions for ac, be and De, allows for an
explicit expression for the non-symmetric consistent tangent slip operator:
(6.128)
This is generally valid for any proposed frictional constitutive law. The
only unknown variable a~n can readily be derived from
(6.129)
Table 6.2, defines this variable for seven previously defined slip criteria
(Section 6.2.3)[57].
n
Law
(1)
(2)
(3)
(4)
(5)
(6)
---
(7)
acu ..
<1?
lrl - J.tlanl
lrl-Tm
lrl - alanln - ,Blanl
lrl - fLlanl
lrl-Tm
lrl - Tm sin ( ~lanl)
lrl-rm
lrl - {Lolanl + ~lanl 2
lrl-rm
lrl - Tm~ tan-
~~lanl)
nalan-n-i
3D Frictional Contact Models Without :

Hardening
In this ~ction, a general three dimensional frictional contact model using the
penalty method is presented, which is based on the work by Peric and Owen
[56]. By using the penalty method, successful applications are expected for
problems not involving high normal forces. For high normal force problema,.
other constraint enforcing methods shciuld be adopted.
In this model, kinematical nonlinearity associated with surface curvar
tures is avoided through discretization of the rigid body surface by a sufticient number of t~iangulai facet elements [56].
.
6.4.1
Preliminaries
Consider two bodies, ns and nm, which are in a state of unilateral contact
problem. The distance. between the bodies are determined by:
(6.130)
where 9n is the gap between the bodies, n is the normal vector on the
master surface, and x(0 8 ) and x(nm) are the configurnt.iou of the slave and.
master bodies, respectively.
The contact condition can be defined as:
J.b
147
+ ,8
9n ~O
/n ~O ,
fn9n =O
(6.131)
where fn is the normal contact force acting on the slave body, and the two
unilateral constraints are the'kinematic condition of no penetration and the
static condition of compressive normal force.
o cos ( ~lunl)
o - ..&un
21'm
/LO f1 +
(H;;1"4r
IJ
Tablo 6.2: Slip criteria and normal traction derivatives: (1) Coulomb law;
(2) coustant shear law; (3) parabolic law; (4) Avitzar model; (5) sinusoidal
trausitional approximation; (6) parabolic transitional approximation;(7)
trigonometric approximation for the whole range [57].
6.4.2
Following the standard formulation of the theory of plasticity, additive decomposition of the tangential gap velocity at the contact surface is adopted;
(6.132)
where
--
148
Yt =(1-nn)u
(6.133)
Assuming a perfect friction law, which states that the friction force is
proportional to the normal force and is independent of other state variables,
leads to the following slip criterion
cI>
= llftll +llf nll-c
149
<ll ~
o ' t ~ o ' i'cl> = o
(6.142)
A standard procedure of rate independent plasticity is followed. Therefore, under conditions of frictional slip without frict:ional hardening, the rate
form of the constitutive equation is obt.ined:
(6.134)
wherc e characterizes the adhesion, is the friction coefficient, Jn and

are the normal and tangential contact force vectors, respectively.
Jt
(6.143)
where
Consequently, simple classical elastoplasticity format of constitutive frictional contact equations can be derived,
(6.144)
g,.
ilt = 9 + iJf
(6.135)
= Dn'U~
(6.136)
fn
where Dep is the slip modulus tensor. Box 6.1 summarizes the
for constitutive model for the plasticity theory of friction [56].
proce<lti:~
Box 6.1 - Constitutive model for the plasticity theory of friction [56].
ft
P _
e_;
= Dtuf
,Yt - 'Y
8\J! (f, e)
fJf t
(6.137)
Additive decomposition of the tangential displacement rate
ilt = g~ + gf
Linear constitutive equation
(6.138)
Dn
= -annn
Dt = -at(I- nn)
e= i'h(gf, e)
(6.139)
with,
in= Dnu~
it = Dtf
Slip criterion and nonlinear adhesion hardening law
Dt
= -at(l-nn)
(6.140)
cI> = llftll + \lf nll - e

e= i'h(gf, e)
Slip potential and non-associated slip rule

Dn = -annn
(6.141)
where Dt and Dn represent the tangential and normal parts of the elastic
modulus tensor, respectively, \J!(/, e) is the slip potential and h(gf, e) is the
hardening/softening law for the adhesion e (and not frictiori).
!
Loading and unloading conditions may then be formula.ted in the standard form,
'1!(f,c) =
11/cll -
gf = iT ,
T =
ff.tt
Loading/unloading condition&
q> ~
o' i' ~ o' "fcl> = o
150
As a result of applying the non-associated slip rule (6.138}, the slip

modulus tensor will no longer remain symmetric.
151
where the vector t;+i defines the slip direction

trial
t.
6.4.3
N umerical Integration
N umerical integration of the constitutive equations..,.ief> frictional contact

problems may be performed using the standard techniques of the classical
theory of efastoplasticity [56, 100, 101, 102).
A single step Euler backward scheme is adopted (for j-th step):
Ligt = !1g~ + !1gf
+1
- f t,j+l
1
- 111tJ~ 1ll
(6.152) '
Consistent Tangent Operator

The numerical updating procedure, lead.s to the following incremental ~
sponse function f,
'(6.153)
(6.145)
Applying the chain rule of differentiation,
(6.146}
(6.154)
LI
(6.147)
(6.148)
and utilizing the relationship
and the constraint equation must be.satisfied:
8t;+1 _
{}/trial t,j+l
{6.149)
11/trial
11
t,j+l
(I _ t.
.
3+1
t.
J+l
--'(;'
(6.155) ,if,:.
/_'
the consistent tangent modulus for frictional slip is obtained:

. ep_
De -
Stress U pdate Algorithm
A two phase elastic predW;or/plastic corrector algorithm is utilized to update the frictional contacl9stress state according to slip potential and nonassociated slip rule.
(6.156)
1
-antj+l nj+1 - Onn;+1 0 n;+1

1
where
If the radial return method is used, in the elastic predictor phase, incremctits of displacements in the contact region are calculated:
trial
= f j +D
f i+l
'Uj+l
(6.157)
(6.150)
Then, th'e plastic corrector phase begins with radially projecting the trial
elastic stress f ~~~t onto the slip surface
(6.151)
Comparing consistent modulus (6.156) with the continuum tange~t modulus (6.144), the stiffness in the tangential plane perpendicular to the slip
direction is reduced by the fMtor at,c For largc displacement increments,
the continuum tangent modulus (6.144) does not provide quadratic rate of
convergence due to the fact that Ot can become considerably larger than
Ot,c
Dlscontlnuum Mechanlcs
152
153
Chapter 6. Frlctional Contact
:i:, fJ E v and v is the space of admissible variations. a is a fourth order

tensor given by
Box. 6.2 - Numerical integration procedure for frictional plasticity [56).
Update configuration
Xj+l
( i)
Xj+l
(i)
(6.159)
+ Uj+l
Elastic (trial) prediction phase

trial
J D ' Uj+ 1
JJ+
1 = j +
(6.160)
Check for slipping criterion
IF 4>}~1t
= llftJ+'1ll + ll/:::~~111 :5 O
and hep is the standard el~toplastic constitutive modulus. The terms on the
right-hand side of (6.158) are linear in u and provide the tangent stiffness.
. Set ()Ht = ()}~1 1 then EXIT

ELSE
Finally, the following expressions for the material, p;mrn1otric and frie .
tional contact tangent stiffness tensora, Km, K 9 and K e, respectively, can
be readily derived:
Plastic corrector phase

tr(a.I
tH1
= nt!~m,,.,
f t,j+l
:::;;
11illftJt'1iltJ+1
(6.161)
Consistent frictional slip tangent modulus
_ a
t,c -
D~P =
11/ t,H1'1
t11J:~if1t11
K9
-at,c(I - tJ+l tj+1 - n;+1 n;+1)
1.
Vu : (u 1) : V' r,dv
{6.162)
X(O)
-ant;+1 ni+l - a:nni+t ni+l
ENDIF
(6.163)
A summary of described numerical integration procedure is provided in

Box 6.2 (56).
where '\1 8 is the symmetric part of v. For the residual force we obtain,
(6.164)
6.4.4
FE Discretization
Following the procedur~ extensively discussed in Chapter 4, the linearized

form of the weak type of the boundary value problem at the known configuration x = x in the spatial description will be (103, 69]
-r(x,17)
= f
lx(n)
\i'u: a: \i'rdv +
lx(roc)
(6.165)
where b is the body force vector,

the frictional contact force.
u: D; rda
(6.158)
1
where r( x, 'Tf) can be interpreted as an unbalanced force at the configuratin
t is the surfacc
traction vector and
is
In a finite element simulation, the subspace vh e v is set as an apprax:imation to v space, where vh is J0D_!'rated through spatial finite element
discretization
154
155
(6.175)
nel
x(O)
= LJ x(Oe)
wth
x(Oa) n x(Ob) =O f a 1= b
(6.166)
where
e=l
Using standard shape functions Na for a typical element e,
(6.176)
nen
uhlx(Oe)
= 2: N aUa
(6.167)
a=l
(6.177)
wlwre ncl and nen are the number of clements, and the number of element
nodes, respectively. By defining the discrete gradient operators,
..... .-.
nen
V'uhlx(O.:)
==
L Baua
and finally the global residual force vector,
""'ne
(6.168)
F = F nodal +
a=l
nen
Ar + A
e=l
cn=l
c,cn
(6.178}
nen
"Vuhlx(Oe)
=L
GaUa
(6.169)
a=l
6.5
the global stiffness matrix can then be defined:
(6.170)
whern nen iH t.he nurnl>Cr of contact no<les ami
In this section, the formulation of three dimensional frictional contact models is extended to include frictional hardening or softening behaviour, based
on the work by Hashimoto et al. (72].
6. 5 .1
(6.171)
(6.172)
(6.173)
3D Frictional Contact Models With

Frictional Hardening
Physical Aspects of Frictional Pheno1nena
According to the classical Cou~omb friction law, the friction force is independent of the apparent area of the two contacting bodies, and is proportional
to the normal contact force between them. It is g<1wrn1ly accepted that the
kinetic friction is almost independent of the speed of sliding.
If A is the real contact area, and H is the hardness of the softer material
in contact, according to the adhesion theory, the normal contact force, /n
is defined by:
fn=AH
(6.174)
and the frictional force required for sh~aring of the junction of asperities,
and the element residual force vector,
''
156
Discontinuum Mech8Ilics
(6.182)
Table 6.3: General categories of dynamic frictiona.l effects (72].
Type
Category
Quasi-static, dry friction
Dynamic, Sli<ling friction
Charactetistic
Initial stage of sliding
Small amount of sliding distance
Middle/late~,stages of sliding
Wear and plowing
Middle/later stages of sliding
where d is the contribution of deformation of surface asperities, !'f' is the

contribution of plowing by wear particles and hard particles, and finally , 0
is the contribution of adhesion of the flat portion of the sliding surface t.o
the overall coefficient of friction . In practice, accurn1( ~ experimental data
are required to evaluate these contributions.
..
Is =Ar
Figure 6.7 depicts various mechanisms of fransitional changes encountered on contact surface with regard to several surface coated sheets. The
main features of each type of frictional phenomena, may be summarized as:
Types (a): ,
. Progressive sliding leads to Hattening of asperities.
. Lower tangential force is required for asperity deformation.
(6.180)
where r is the shear stress of the softer material in contact. The coefficient
of friction may then he defined:
n
LJ
. , becomes smaller than 0.10 at final stagcs.

. The lubricant works 88 a weak layer t.hat rcd11ces
s
= n
T
= H(Rj
0.2
for most metals)
{6.181)
Type {b):
Softer surface coat than types (a) and (b).
. Larger asperity defrnation is expected than those of types
(a) and (d) with the same frictional work.
Larger frictional tangential force is required once the friction
condition is changed from the mixed lubricant condition to the
boundary lubricant condition
. Larger force is required for shearing the junctions
The main differ?nce of this section with Section 6.4, is that is no longer
a constant and has a hardening or softening. behaviour.
U nder very high .normal force, the real contact area no longer follows
fn. =.AH due to work hardening of materials. Other real industrial apphcat1o~s
may also exist that the conditions of the classica.l laws can not
be apphed. Table 6.3 compares different categories of frictiona.l effects in
dynamic friction problems (85, 72]:
Type 1 is related to quasi-static contacts with dry friction under relatively small sliding distance of an order of magnitude of t~n microns.
The adhesion theory may be effectively adopted for analysing this type
of problem.
Type 2. takes place in dynamic loadings with sliding friction. It shows
large rigid relative motions. The plasticity theo~y of friction based
o~ deco~position of displacement into elastic deformation and sliding
d1stance is used as the analytical approach.
1
Lhe required
shear forc~ at contact junction.
Type'3 includes wearing and plowing, frequently encountered in sheet

forming applications, especially for surface coated steel sheets.
As an example, a model for dynamic friction phenomena, for types 2
and 3, may be defined as [104]:
.
Type (e):
Larger asperity deformation
. Rapid increase of ehe actual contact area
. Wear particles lmprove the lubricant condition
. Relatively stable frictiona.l behaviour unt.il occurrence of contact with the bulle metal.
Plowing of the zinc layer because of low resistance to shear.
Types (d):
. Progressive sliding leads to Hattening of coating layer ..
. The lubricant works 88 a weak layer t.hat reduces the required
shear force at contact junction.
In summary, transitional chang~ in the coefficient of friction tan be~

tributed to the h~dness of surface coat, thickness of surface coat, resistande .
to shear, surface roughness, and lubricant conditiori.
158
Discontinuum MechMics
_.
159
Frictional Work Parameter
LJ
(Mixed Lubrlcant Condition)

tool
tool
The frictional work parameter w is introduced as a state variable in order

to take into account the non-constant frictional behaviour in dynamic/quasi
static friction. This evolving frictional behaviour can be attributed to the
changes in interface condition due to asperity dcfonnnt.iou, wear particles
and lubricant conditions. The frictional work parameter may be defined as:
a) Cold rolling.
w=
J.' -f,(J,dt= J.' -f,dg,
Therefore, ~an be defined as a fnction of
b)
~nd
w,
= (Jn,w)
(6.184) .
Several experimental based definitions of Equation (.184} have been

proposed. For example, Suh et al. [104] has proposed the following polyno
mial:
Electro galvanized coating.

(thin, aoft
f n and
(6.183)
layered pure zinc coat)
(Mxed Lubrlcant Condition)

. tool 1
:J
e)
tool 1
where Ai coeflicients are experimentally determined.

Zn '---1-teel_ _ _ _ _ _____,
Hot-dip galvanized coating.

(thick and aoft pure zinc coat plus an oxide film on top)
....._ _ _ _ _ _ _to_o_I_.1
(Mxed Lubrlcant Condition)

1
tool 1
~__, Zn;~~::c~:;oco~~=~]j~;:cog>''~
,__e_st_ee_
______
d)
(6.185)
6.5.2
Numerical Simulation of 3D Work Hardening

Frictional Contact
Following the same procedure as the classical elu...'itoplmitic nnalysis and frictional contat behaviour without hardening, we begin with the additlve
decomposition rule of the displacement rate.
.
.. _--.1
(6.186)
Galvannealed coating.
(Zn-Fe alloy coat harder than the bulk material)
Figure 6. 7: Four mechanisms of transitional contact surface changes for

various typical surface coated sheets (72].
Ut
= (1-n n) u
where superscripts e and P, represent the elastic (adherence

plastic (slip) parts of the displacement rates.
(6.187)
o~ stick) and:
160
161
The constitutive laws for normal and tangential components, based on

the penalty method can pe written as:
non-associated
(6.188)
it =
--
Dt e = --at(I - n n) e
a) Linear friction la.w.
(6.i89)
where f n and f t are normal and tangential contact loads, respectively,

Dn and Dt are the normal and tangential parts of the elastic modulus
tensor of friction, and an and at are the normal and tangential penalty
nurnbers, respectively. The loading/unloading condition is expressed by
kinetic constraint of impenetrability and static condition of compressive
normal load:
gn ~O
,,
fn
~O
n 9n
=O
(6.190)
slip direction
b) Slip criteria.
Slip Criteria
Figure 6.8 illustrates the associated and non-associated slip rules for linear
and nonlinear friction laws [72]. It provides sorne general understanding
of the expansion or shrinkage in the radial direction of the siip surface
according to the amount of frictional work parameter w. The slip criterion
rnay be defined by:
l?(/' w) =
11/tll -
(J n111)ll/ nll
~ ~ :!!~k
(6.191)
non-auociated
Assuming the slip criterion (6.191), the associated slip rule generates a
normal force, causing normal separation of the surfaces. In order to avoid
this slave body separation from the contacting surface, a non-associated slip
rule is typically adopted [86].
The slip potential '11 can be defined as. a cylinder with radlus
isotropic frictional contact:
\J!(f)
,,
= 111t11
11/t11
linear slip
surface
e) Nonlinear friction law.
for
(6.192)
'
and the slip direction is then defined as the outward normal to the slip
potential '1!:
Figure 6.8: Slip criteria and associate/non-associated slip rules (72).
162
1
p
Ut
{J'iI!
= -"(8ft
(6.193)
Box 6.3 - Constitutive model for the plasticity theory of friction

with nonlinear evolution [72].
Additive decomposition of displacement rate
or,
f= -At
(6.194)
= ~ +f
= (1 -
n n) 'l.t
Elastic prediction for contact forces
where t is the unit tangential vector,
in
= Dn e = Dn Yn
= Dt e
Dn = -an(n n)
it
(6.195)
To obtain the continuum. tangent modulus Dep,
consistency condition:
i'
Dt =--at(l-nn)
is solved from the
Slip criterion
<P(f ,w) = llftll
f)p) .
( oft
ft+
( f)p ) .
fJfn
fn+
(f)p)
8w
w=O
(6.196)
Slip potential and non-associated slip rule
\J!(f) = llftll
itf = -'Yt
and then substituted into the constitutive law, defining the following linearized forra:
(6.197)
where
Dep
+ (! n,w)llJnll
8.:V -~
- 8Tt
- llJ1ll
t-
Hardening law
w= - f t itf = i'll f t 11
is the non-symmetric continuum tangent modulus [72]
Loading/ unloading conditions

cp~o,
-Ot
[ ~ll 11:1(!!.i!.)l
t
(3 "
Dw
'Y2:0,
i'4>=0
t t) .
(6.198)
sliding direction on the contact surface. The last two remaining terma,
indicate contribution of hardenittg/Sbftening phenomena to the stick and
slip stiffness matrices in the sliding direction, respectively.
Box 6.3 summarizes the constitutive model for the pla...,ticity theory of
friction with nonlinear evolution [72].
with
f3=1+(fn,w)llf
Ot
11 2 (
"
8
8w
(6.199)
The first term of nev indicates the stiffness in the normal direction,
while the second term, denotes the adhesion stiffness perpendicular to the
6.5.3
Numerical Integration Scheme
Among severa} available numerical integration schemes [105, 102, 106], the
implicit backward Euler method is a.dopted in this section. Similar pro~
dures may be followed for alternative numerical integration schemes. Box
Discontinuum Mecluwics
1()4
6.4 summarizes the numerical integration procedure for nonlinear hardening

frictional problems based on the work by Hashimoto et al. [72}:
6.5.4
Consistent Tangent Modulus
In order to obtain quadratic convergence, the consistet tangent modulus

has to be employed. The 111ethodology given by [72] is followed. Beginning
with the incremental resp(lse function,
(6.200)
Box 6.4 - Nonlinear hardening frictional scheme [72 .

l. Update configuration for the iteration i.
1.1)
x}2 1 = x;~> + Au}2 1
1.2)
1.3)
,(/~~H 1 ,w;), ~
n,j+l
= f n,j + Dn AUj+l
and
anf.. n
2. Elastic predictor phase
2, 1) Elastic trial tangential force

trial
t,j+l
= Jt,j + D t ''-\Uj+l
A
2.2) Check tht:l stick/slip condition
IF
ci>trial
Hl -_ llftrial
t,J+l
or
SET
11 - (/(i)
n,j+l' W_ )11/ 11,.J 11 l
~O THEN
{ft,w,, (811f.. 11)' (~) }j+l = {}}':tr'
EXIT
il!
ELSE
LJ.
3. Slip corrector phase
According to Equation (6.201), df can be derived
8f(Jj, Wj, AUj+i)
df ]+1 = -~--------. au
3.1) Solve the consistency equations for A/
+ ( a+ ) trial A/ + ( 8<t>) trial A = O
t.rial
fiTe
J+l
(6.202)
j+l
8w j+l
with Aft = l"YDt tJ+1
6w = }~f 1 11/ n,;+1 llA/
resulting in
df ;+t = Dnd>u+1
A"'(=
+ D,dt.u;+1 .
11/CriG&
11 trial ll / n,;+111
C.11 -i+l
ac+,.triG&(.?fu.)tric.111/
. 112
,..;+1 l1W 1+1
n,3+1
3.2) Evaluate new stress state, slip surface and i,
et1t+1db7 - a,t.7 ( :;, ) d>f1,;+1

(6.203)
A
t
t,j+l - '-\"'(C<t
l trial
Wj+l'::::: Wj + llft,J+lllArj+l
Jt,j+l
=i
(f~!j+2t Wj+i) and ~

3.3) Check the stick/slip condition
The following equations are solved for dA/

(6.~04)
<l>j+l
--
llft.;+111- (f~!j+l' WJ+i)llf n,j+lll .
IF ;+1 >O THEN
(6.205)._.
SET
ep
{li w,, ( 811fnll), (~) };::z = {};+1,

1
GOTO 3.1
Finally, a linearized rate form is obtained,
f ;+t =De Uj+l
ENDIF
(6.206)
i;.
,::;;; . >
l
f(f i' Wj, AU;+t) = fj +A/ n,j+l + Aft,j+l
186.
Chapter 6. Frietional Contnct
ENDIF
166
nep=
e
167
11
-an(n n) - t.llf !;.{U (I - n n - t t)
timestep j+ I
t,;+111
11
timestep j
n
(6.207)
Ht.
.
On
(-4-)
llf
811Jn11 . .
__ ,,.,,,"'"'
------
11] (t n)
!'j
whern D~P is the consistent tangent modulus, and

v
L_j
= 1+ (H1ll/n,j+ill
Ot
6.6
6.6.1
_b/y) (aw)
.
llfn,;+111
(6.208)
Figure 6.9: 2D node to edge frictional contact
j+l
8 -
X1) . t
= Yt
+ (eJ+1 - eJ)l;
(6.213)
.~'
k
1
K=Kn+Kt
+1
=
O!n
.
r({}gn.)T
8.'9n + gT !_ (Yn)]
8u
8u
nau 8u
(6.214)
(6.215)
L ,
(6.209)
)]
K -. t .[(8gt)T
-8gt + 9t -8 (Yt
8u
8u
8u -.
aut -
Yt,j+l
(6.212)
and for the stiffness matrix,
Kn
T
1 (X
= O:n9n
_,., g
t
l int
t
- v.t t.au ,
Based on the work by Crook [52], the necessary formula.e for a two dimensional nade to straight edge frictional contact problem are provided in this
section, though the general p.umerical formulation has previously been discussed in detail. Figure 6.9 shows a typical node to edge contact problem
in a two dimensional space.
ec =
8g '
2D Node to Edge Model
From the geometric~ description of the problem, at time step j
{}g
= a g __!!,
f int
n
n "au'
Node to Face Frictional Contact Models
'''
;e
(6.210)
(6.216)
Using the same methodology as described in Section 5.4.5 for evaluating

the derivative of the gap function, will lead to,
Elastic Tangential Contact

Beginning with elastic tangential contact, the internal f~rce vector can be
defined as:
(6.211)
lj {T + r9nN m }
f tint = t9t y
(6.217)
(6.218)
i'
169
Chapter 6. Frictiona.1 Contact
Discontlnuum Mecha.nlcs
l8
>
,,
where
langlllllal
lilllplltlol
pe. 1,
/Ol'OI
T; = [t
- (1-e)t
(6.219)
-etJ
11uvd111um
1hear1trus
T~ =[O
N! = [n
(6.220)
t]
-t
-~n]
- (1-e)n
(6.221)
Figure 6.10: Coulomb friction law.
N~ =(O
(6.222)
-n n)
(6.223)
;~ = sign(ft)
""
(6.230)
The stress update procedure may then be.followed as (see Fig1_1re 6.U)~
D
B = {
(N,N~ +NmN;)-(T,T~ +TmT;)- 2~"(TmN~ +Nm'r~)}

.
<eJ+i - eiW
(6.231)
I + O!tU.9t
A
t,j+l = j
f trial
(6.232)
!:l.gt =
(6.224)
When 9n = O, the npdal forces are defined by distributing the local

tangential force using the shape functions of the master segment and transformiug to t.he global system using t.
~:~J+'1 =
111:1+'1 ll - fn ::; O
t,J+l = fn,;+1 sign(J,t,'/;'.~)

f:l."( =
Coulomb Frictional Contact
~t [111:1+'1 ll -
fn,j+l]
!:l.gf = A-;sign(ft,J+l)
The classical assumption of Coulomb friction, as stated in Figure 6.10, is
that the body is at rest until the maximum shear force is reached. However,
in reality, bodies exhibit sorne relative movement within the contact area
prior to sliding.
slip rule:
slip potential:
(6.226)
l/tl -
(6.227)
,fn
8\1!
1.:f = !:l."( 8f t
'I!(ft)
(6.236) ..
(6.225)
!t = atgr
~ =
slip condition:
(6.235)
(6.237)
gf + gf
elastic region:
'(6.234)
The consistent tangent mat~ix, which is non-symmetric, may then be

defined by:
The rnain equations can be summarized as:

9t =
(6.233)
= 11! tll = [! t f t1;
(6.238)
resulting in:
(6.228)
1
(6.229)
(6.239)
170
171
f,
a) Geometrical description.
Figure 6.11: Stress update procedure for Coulomb friction law.
where
C={Ts+ 9Nm}N';
(6.240)
b} Basic vectors.
3
6.6.2
3D Node to Quadrilateral Face Model
x,1
)-11,
x,
ln this section, a three dimensional node (slave) to four noded que.drllateral

face (master) frictional contact model is presented. This is basically based
the material provided by Crisfield (107].
011
Frictionless Contact
e) Initial and current
Geometrical description ti}he model is depicted in Figure 6.12a. In this

model, S. represents the ~ave node and element 1234 is the. quadrilateral
finite element. e is the point immediately below son the master surface.
It is irnportant to note that the contact surface, in general, does not lie in
aplane.
configurations.
It is not a trivial matter, as it is in a 2D case, to find the (e, 17) values

associated to point C with r(e, 11) position vector.
Using the standard shape functions Ni(e, 77), associated with node i, the
position vector r may be expressed by the nodal position, W,
.
Figure 6.12: Node to quadrilateral face frictional <ontact. ,

.
Discontlnuum Mecha.nlcs
172
Chapter 6. Frictiona.1 Contact
173'
(6.241)
where the shape functions Ni, are well defined:
(6.248)
where J is the jacobian matrix at

using t4e variational mathematics,
(e, 77).
Back to Equation (6.241), and
Ni= i(l +e)(1+11)

(6.249)
N2 = l{l - e)(1 + r)
N3 =
N4
(6.242)
Hi - eH1 - ,,,)
= i{l + e)(1 -
where,
77)
(6.250)
and
U
T = ( 'UT , 'UT , U2T , U3T , U.1' )

8
(6.251)
(6.243)
o
o
(6.252)
l {(1 + r), -(1 + r), -(1 -
Nf = (t/.)
r), (1 - r)}
NT=
7J
(ii-)
= l {(1 +e>, (1 - e), -(1 - e), -(1 +en
NfTJ=
(,N)
8(8rj
then using (6.245),

(6.244)
(6.253)
i {1, -1, 1, -1}
where
aml the vector connecting s to its estimated projection onto the quadrilateral surface, e, g,
(6.245)
(6.254)
and for the remaining variables:
Once the correct value of (e, i]) is known, the vector g will be orthogonal
to the two vectors r~ and r,, which lie on the master surface (See Figure
6.12b), i.e. g 11
{6.255)
By defining new variables ae anda,,, we may derive an equation to define

(e,11):
(6.256)
. aE=rEg=
or
ae
{6.~~6)
';.~~ .~\:
a,, = r~[Hu- A~)
w~~:.1?i~~t"::,ic:f:~{
+ gT.H"u + (gT re 11 )8~
174
175
(6.267)
(6.259)
so 8HT and n are to be determined. For determining n we have,
(6.260)
n =
In practice, from the contact detection, we have an ..estimate for positi?n

of C(~, 77) which we may wish to improve. By fixing the nodal coordinates,
8u = O, and adopting a Taylor series:
fgjj
(6.268)
so,
(6.269)
(Ao1dl
+ DJI{, =O
wit/i D =-ATA+ (gT re,) [
~ ~]
(6.261)
and from (6.248) and (6.253), we get,
which can be used in a Newton-Raphson scheme to find an improvement ~

and therefore, to update the initial estimates for (e, 77).
By finding thc variation of
9ni
(6.270)
the residual (internal) force vector is
whcre
calculated:
(6.262)
(6.263)
[ Ge ]
G,,,
'
G~ = H~n
G.,,=H.,,n
(6.271)
Therefore,
(6.272)
g,. =
(;n g)
70
(6.264)
g = n [Hu-A6(]
where nis the normal vector to the surface at the contact point. Knowing
ATn=O,
= {H - An- 1YT]u
with
(6.274)
(6.265)
and following the definition of the residual force vector, re
= ~,
resulta
(6.273)
Therefore, n can now be determined:
in
(6.275)
(6.266)
The consistent tangent matrix is then defined by finding the deriva.tive

of ,,.e with rnspect to u,
.
'
- and to calculate HT,
176
177
(6.285)
(6.276)
where
(6.277)
(6.286)
Finally, by using (6.277) and (6.275) in (6.267), the ~qnsistent tangent

matrix is defined:
, , ,....
K e =K1e+K2n
e
The residual force vector, is then defined by:
(6.278)
(6.287)
where
(6.288)
(6.279)
e= [
t
K~n =
~ [HT H - HT nnT H
+ HT AD- 1 AT H + BnHT AD- 1G
ff1
{ ff2
= Ot9tl
= Ot9t2
(6.289)
and fi.nally, for the consistent tangent matrix,
+gnGTD- 1 ATH + g~GTn- 1 a]

'
ft ]
ff2
'
(6.280)
(6.290)
with
Sticking Friction
Referring to Figures 6.12b,c, which depict the initial and current configurations, the tangential gap vector, Yt, can be defined by:
(6.291)
(6.281)
(6.292)
where E= [e 1 , e2, e 3 ] is the orthogonal triad based on the contact point,
where
(6.282)
(6.293)
(6.283)
(6.2&4)
i
and X= L:, Ni(eo, r0 )Xi is the initial configuration, while x

is the current configuration of the contact point.
L.J
=E Ni(~, TJ)Xi
We will also require derivative of Yt with respect to u, or:
W = ATre.(F)T +Frf.A+(<FATr~,+(F, 0 )Tx~,) [
~ ~]
(6.294)
and
178
F"' = [
~ ~
179
.,...-jc,trial
K,-
{6.295)
] /"'
(6.296)
11J~ 1 trial11
(6.306)
Therefore, the interna! force vt:t;tor is defined by:
BT/cHl
r e_
-
(6.307)
g = Bu
(6.308)
(6.297)
Q= [
~f]'
(6.298)
(6.309)
G~ 11
=n
7'H
(6.299)
~r1
and the non-symmetric modular matrix, Ct, as

(6.300)
!
(6.310)
(6.301)
with,
Sliding Friction
(6.311)
Finally, the nurnerical formulation is extended to incorporate sliding .frictional behaviour. Assuming a Coulomb type of sliding friction,
<P
where
UDT = (ffi,f{2 ).
For (j
= 11/fll + f~
(6.302)
+ l)th stress update iteration,
(l _
O:t:l.'Y ) c,trial
llf~'trialll
= K./c,trial
t
(6.304)
(6.305)
Far non-associated slip rule, /:l.gP is normal to the cylinder
Constant, hence, ~Y~+l = /:l.gP,trial. Frorri =O, we get,
(6.312)
The tangent stiffness matrix, finally, is defined by:

(6.313)
(6.303)
e .
J t,J+l
f n,j+l = O:n9n
llnll
L_,
,----!
Chapter 7
Discontinuum Contact
Mechanics
7.1
Introduction
The pioneering work by Cundall and his colleagues, who completed the original work by Goodman in 1968 (108} on jointed rocks, marks the beginning
of modelling of discontinuum media [109}. They developed an algorithm for
modelling the behaviour of jointed rigid rocks, soon termed as the Distinct
Element Method.
By advancing the capabilities of the finite element method, and increasing power of computing facilities, fully defo~mable blocks replaced the original rigid bodies, with the new Discrete Element Method terminology.
Nowadays, the discrete element method has reached t.o a.u ever increasing
popularity for modelling ali potential discontinuum media. Nevertheless, it
is mainly used for two classes of problems:
Granular ftow: where a ll'ge number of simple elements ( usually rigid)

are interacting with each other and with the smTotmding bounda.ries
(rigid or deformable). Granular How in silos and the slope stability
analysis are the most attractable types of problems in this class.
Progressive fracture: where a continuum is subjected toan extremely

high condition such as explosive loading or high velocity impact, caus- ;
ing extensive cracking and possibly fragmentation. The behaviour of

the model is continuously cha.nging towar<l the discontinuity a.nd the
original geometry of the body is changing by the extension of cracking.
181
182
Chapter 7. Discontinuum Contact Mechanics
7.1.2
The essential point is that the finite element method is rooted in the
concepts of continuum mechanics, thus not suited to general fracture propagation and fragmentation problems. The firiite element method may only
effectively deal with a single crack ora low fractured area without any fragmentation, whereas the discrete element method is specifically desigried to
.solve problems that exhibit strong material and geometrical discontinuities.
'
;,
Applications .
The discrete element method has found extensive attraction within various
fields of the industrial and scientific applications. The main categories are:
l. Geomechanical applications
........ ~
- J ointed rock mass [l09]- Rock blasting problems (18, 113].

- Underground structures (114, 115].
Before dealing with the main issues, a quick review of historical qevel:
opments and present industrial/scientific applications is provided.
7 .1.1
183
- FaHing rocks [30).
Historical Development
- Simulation of sands (116].

As mentioned earlier, the original development of the discrete element method
2. Porous media (32]
may be attributed to the work by Cundall in 1971. In the following, a brief

nvinw of t.110 111ain hiHl:orical developme11f.H of thu 1nothod is providod:
3. Granular nm.teritilH
- Disk and sphere elements [117, 33, 32].

- Ellipsoidal elements [118].
1968 : Analysis of jointed rocks by Goodman [108].
- Superquadric elements [32].

1971 : Analysis of jointed rocks by Cundall [109].
4. Industry
1!)88 : Fully deformable discrete elements included (Ghaboussi (110]).
- 3D printing
5. Impact analysis (progressive fracturing)
1990 : Beginning of large scale simulations.
- Reinforced concrete slabs

1995 : Combined finite/discrete element method for fracture simulatiou of brittle media (10].
- Metal cutting [34].
1995 : Coupling discrete elements with fluid or gas ftow (111].
- Composites [39].
[119].
- Shock/thermal initiation of explmorn; (36].
1996 : Parallel and object oriented computing [112, 24).
6. Motion planning in robotics (120].
1996 : Modelling granular flow in silos [1].
7. Computer Graphics .
- Animation [121, 122, 123].

1998 : Metal cutting using adaptivity techniques [34].
- Virtual reality [124].
1998 : Impact analysis of anisotropic three dimensional composite

shells [39].
7 .1.3
1999: Damage investigation and repair modelling of masonry structures/bridges.
L 1
I t should be noted that for each ca.se, earlier less sophisticated modela
can also be found in the literature and the mentioned years show the time
of major advancements of the method.
--~-------------ft - - - -
Speciflcations
Table 7.1 briefiy provides compaxative information about the principie speclifications of continuum (finite element) analysis and discontinuum (discrete
element) analysis.
-!
~'."; :.,
:.':;
i:
:~~.
t{'.j,
Principal components of a general discrete element simulation can b

classified into the following five groups:
184
185
l. Object representation
Table 7.1: Cornparing continuum and discontinuum modellings.
2. Contact detection
continuum
- only finite elements
- explicit definition of contact surfaces
geomctry
3. Contact interaction
4. Physics of objects
discontinuum
- finite elements (FE) and rigid"'bodies (RB)
- automatic evaluation of contact surfaces
5. Visualization
This chapter mainly deals with the contact detection procedures closely
coupled with object representation. The contact interaction algorithms have
already been discussed in previous four chapters. Physical behaviour of
objects may essentially aff~ct the material modelling (plasticity) within the
finite element procedure, and may also cause cracking/fracturing which will
be dealt with in the next cha.pter. Finally, visualization is a necessary tool
for a successful discrete element analysis. Changing geometry and mesh,
creation of new bodies, and checking the interaction stresses/forces require
a powerful visualization tool, without which, the huge bulk of result data
may not ever be verified and discussed. However, it is a mu.tter of co111n:tl:\r
graphics and CAD /CAE systems and is out of scope of this book.
continuum
- simple algorithm
global search
discontinuum
- must be efficient for large numbers of bodies
- must be efficient for dense/loose packs
. ; _ must be applicable to FE a.nd RB
.....
continuum
- no9e to node contact search
- simple surface node to element edge contact
local search
contact
interaction
discontinuum
- complex algorithm (arbitrary shapecl bodies)
- different algorithms for different types of RB
7.2
continuum
- interface elements
- n<?de/node, node/face and face/face contacta
- penalty and various Lagrangian formulationa
In this section, the contact detection procedures are briefly revi~wecJ. and
their main advantage/disadvantage points are discussed [125). Then, the
alternating digital tree, as one of the fastest geometric intersection sea.rch
algorithms, are explained in detail and its application to general contact
detection problema will be reviewed by providing samplc probl~ms.
discontinuum
- node/node, node/face and face/fac,e contacta
- penalty formulation for FE bodies
- closed form solutions. for rigid elements
- energy conserving/dissipating algorithms
The problem of detecting the interacting bodies in multi-body analyses, also known as the geometric.. intersection search, has become a major
computational research topic ..
Assume there is a system of N interacting hodies; ali may happen to

come into contact with any other body. A naivo coutact. dutection method
requires the checking for contact between each body and every other body
within the system. Figure 7.1 shows how such a simple approach requires
a checking link between each ( target) body and the remaining ( contactor)
bodies.
- - continuum
- explicit or implicit dynamic solution
solution
discontinuum
- only explicit dynamic solution
1---
continuum
- simple deformed geometry visualization
visualization
discontinuum
- advancecl graphical postprocessing
- continuously changing geometries
Contact Detection
The number of operations required to detect ali contacti; between N

bodies will then be proportional to
1
1
N-1
N-2
(7.1)
186
2
3
4
2
-3
4
0
D
[[]
187
In multibody analysis, however, the above method obviously becomes

extremely expensive. Several other algorithms have been proposed to im- .
prove the detection procedure. In the best case, the computational efforts,
has been reduced to a factor of
Nlog2 (N)
(7.2)
The existing detection methods have so far laid in between the two
extremes .
The alternating digital tree (ADT) algorithm, which developed initially
to solve the problem of mesh generation, reduces the number of operations ,
required to deterniine the contacts between bodies by creation of short lista
of potential contactors for each target body. Figure 7.2 shows a sample part
of the created short list for a set of N contacting bodies. In this case, a ,
direct checking is undertaken for the number of relevant bodics of a ta.rget,
and the procedure is repeated for other target objects.
Figure 7.1: All to all check, the simplest contact detection procedure.
7.2.l
Contact Geometry
Depending on the type of modelling, two types of discrete elements may be

defined:
l. Rigid bodies (Figure 7.3)
Figure 7.3: Simple rigid discrcte elemeuts .
2. Deformable finite elements (meshed polygons) (Figure 7.4)

Figure 7.2: Short lists of contactors.
The contact geometry is then either computed from the input definition
of rigid bodies, e.g. a circular disk is defined by a centre point and radius, or
188
189
Chapter 7. Dlscontln1111m Contact Mechanlcs
Dlscontlnuum Mechanlcs
F.
.E
.,,, ...
___/
a 1
Figure 7.4: Deformable discrete elements (meshed polygons).

Figure 7 .6: A system of rigid bodies (disks) moving in a domain.
a.utomatically evaluated for deformable finite element bodies by evaluating
ali exterior edges / facets and grouping them for each discrete element, as
depicted in Figure 7.5.
DE
1
cient for both loose and tight packs of elements. A single approach might not
achieve all the mentioned goals, and different approaches may be adopted
for different applications.
Contact Geometry
1 3
3 2
2 1
5 6
6 7
7 8
8 5
1 2
6
2 3
4
2 4
5
5 6
2
In this section, different search algorithms wiU be briefiy reviewed and

some of the important associated techniqus will be addressed.
Direct Checking
Assume there is a system of N interacting bodies; all may happen to come
into contact with any other body (Figure 7.6).
The simplest approach for detecting the potential contact8, will check for
contact between each body and every other body within the i;ystmn. Figure
7 .1 shows how such a simple approach requires a checking link between each
(target) body and the remaining (cont~tor) bodies.
Figure 7.5: Definition of contact geometries for discreta elements.
7.2.2
.AB stated before, for each body, N - 1 potential contacts have to be

assessed. Therefore, the total number of operations (assessments) required
to detect all contacts between N bodies will be proportional to
Global Search Algorithms
A general global search algorithm must be efficient in dealing with a large

number of bodies, suitable for both rigid and deformable bodies, and effi-
N-1
N--.
2
(7.3)
190
---
F.
,,..---l. . .
' , . 1/
1 ',
\ /
\
\'
......
.""
'
'w""
..... __ _,,_,
191
' '}'-. // -,,..7'>.--,

/
-....._ Ar- // '
'l
/
\.\
\)..--- I B
/
'
/~
~
/\
,._-'1
\/ G
,,,,---1. . . .
y , He1
/\
/\
9-\-~v~~-
Figure 7.7: Defining a potential contact zone for a system of moving bodies
( disks ).
Direct Evidence Search
In this method, the mtinber of contact node searches for ea.ch target ~ur
face is reduced by forming a reduced set of nearby contact nodes, thereby
reducing the total number of contact searches. This reduced set (short list)
has to be periodically updated. Figure 7. 7 shows a simple way of defining
a buffer zone for creating a reduced set of ,nearby nodes.
Figure 7.8: Moving disks within a buffer zone.
Therefore, the first step in such a search algorithm is to define the buffer
zone for each target, and then finding the disks iuside the specified zone
(Figure 7.8).
By knowing the velocities at each time step, we may approximately
evaluate the distance in which a disk can travel in next n time steps. Accordingly, the need for a new glooal iearch for fiuding nodes inside each
new zone can be decide<l from the leaving/entering nuy nodPs from/into the
zone.
If the max:imum velocity of an object in consecutive time steps is Vmaz,

then as a safe estimation of the mnimum number of time steps for that
objcct outside the buffer zone of a target to read1 tho tn.rget will be:
Afterward, only the associated reduced set will be searched for potential
contacts with any node.
Almost all of the developed search algorithms differ mainly on how to
create such a short list in a faster systematic way for specific or general
applications.
7.2.3
Buffer Zone
As indicated in Figure 7.7, a buffer zone may be defined for ea.ch target. In
L
cai;;t~ of rnoving disks, the buffer zone may be defined by a circle with radius
n,,.
(7.4)
Therefore, no contact is expected to happen before N t::.t time steps and
the global search to locate all contactor objects within a target buffer zoe
is performed every N At time steps.
The sarne concept is also used to enable direct evidence checking in association with meshed polygons and more complex object shapes. Different
rules for forming a buffer zone can be utilized depending ou the object shape
nml type of application:
193
Discontinuum. Mechanics
192
r--
..,,
l~-r--------,.-"":1
/Buffer ZOne
,,
LL
1
1
1
1
1
1
1
.~
1'
11
1
.,
'
"'
-r--------ll..1 ',
1
1
1
1
1
1
1
',
'
'
'
'
',
'
'
',
'
'
,---------------.....-'
1
',
',
'
'
~'l
}'
~---~-------------------../~'
,c.":'-1--------~---------,
1
11
lt1
'1
1
1
1
1
1
Figure 7.9: Definition of the buffer zone; uniform expanslon of the mesh
boundary (left), and a rectangular box (right).
11,,
+-~------.,1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
~---------,-------1
-h,
I~...:---+- - - - - - - --!:-----..:---..:-..:---..:-..:-..:-J-'
Figure 7.11: Forming the buffer zone based on each segment geometry.
In order to minimize the total time spent on up<lc.i.Ling Lht: shcrt Hst
of contactors, the size of the buffer zone should be defined with respect to
the actual size of the bodies under consideration. Sorne reports proposed an
optima! size equal to the average size of ali bodies, although its applicabllity
to general meshed polygon systems ls in doubt.
Figure 7 .10: Effeet of contact probability on the shape of buffer zone.
If the relative motion of the bodies is small, a small buffer zone will be
most efficient. The faster the IJ\Otion of elements, the larger wll be the
necessary size of buffer zone.
l. Unifo;rm expansion of a faceted discrete element formed from de-
formable finite elements (Figure 7.9)

2. A rectangular box (Figure 7,9)
Flying Through
3. Buffer zone formed to account for direction of motion (Figure 7.10)
There exists an analytically derivable critical time step in a stable explicit

time integration scheme limiting the maximum time step increment of the
dynamic analysis of a system. Unfortunately, this time step increment may
still be too large to assure the detection of all contacts during the analysis.
The velocities of bodies under consideration can be high enough to allow
the bodies to fly through one another (Figure 7.12), unless the time step is
further reduced.
.
4. To avoid a large number of node to edge local searches, a segment
based buffer zone can also be defined (Figure 7.11)

In a general discrete element analysis, in order to decide when to update
the list of nearby contactors again, the displacements must be monitored
and accumulated. When the largest accumulated displacement exceeds the
dcfinnd size of t.he buffer zone, the complete conta.ct detection procedure
mu8t be performed again.
j___
Based on the dimension of the buffer zone, db, t.he tinw stcp increment
194
Clmpter 7. Discontinu um Cont&ct Mechanlcs
195
Bufferzone
contactor
flying
through
the target
t'~' tf':-~
................
I
"",
}
I .,
' , ___ ...

\
n
'
\
.\
\\
>
///
Figure 7.12: Flying of a contactor object through the buffer zone of a target.
Figure 7.13: Ambiguities of the way point D penetrates BAC during an

impact.
lJ
must satisfy,
(7.5)
where
7 .2.4
Vma:t
is the maximum velocity observed in the system.
Contact {Field) Zone
As a consequence of satisfying the time step reduction (7.5), it is possible

that two bodies are found to be penetrating into each other's buffer zone
prior to their actual contact. Nevertheless, the accuracy of the numerical
solution for multibody contact problems can not necessarily be guaranteed.
contact (field) zone
For example, consider Figure 7.13, where after the impact, point D has
penetrated the body B AC. However, by only monitoring the displacements,
the program can not simply decide whether the point D has penetrated
though the segment B A or CA. The !fme problem may arise near the
corners, where sveral paths seem valid for the penetration of the point lnto
the body.
1
A contact zone (field zone) is now introduced (Figure 7.14) in order to

gain further control over the accuracy of the numerical solution,
Taking into account both the buffer and field zones, Equation (7.q) is
modified [1],
Figure 7.14: Definition of a contact (field) zone.
196
l::i..t <
-
Chapter 7. Discontinuum Contact Mecha.nics
d2 +d ( 1- d7
d2)
-;
b
V~
(7.6)
where d is the current distance between bodies, d is the field (contact) zone
thickness and v~ is the relative normal velocity between interacting bodies.
The condition 7.6 ensures a gradual change of,tfie time step increment
during contact, leaving a :QJ.inimum number of time steps for treating a
general impulse/contact problem. As a result, the accuracy of the solution
is also preserved.
..;""'
iR
(j
Binary 'free Structures
A binary tree structure is a specific method of sorting data that allows new
data to be ea.sily added (inserted) or removed (deleted). Binary trees a.re
one of the most important non-sequential types of data structures [126]. At
each node, the information, stored consista of data and two pointers known
s the left and right links that point to further data. Each added link can
either be equal to zero or equal to the position in memory where another
node of the tree is placed.
1 -------- _,
rcm ~
rn:;i
kwlD
7.2.5
------------------
l.@~1,,
~ ~
----lrle/2
Figure 7.15: A simple binary tree data structure.
Compared to a linear sequential arra.y, the binary tree structure requires

only two extra storage locations per item left and right links, and provides
a much greater degree of flexibility.
The first node in the tree is known as the root node. From one node of
the tree it is possible to point a.t most two other nodes, while for ettCh node
(except the root), there is one arid only one link :pointing at it. i A node
without any pointer to' other nodes is called a terminal node.
Figure 7.15 shows a typical binary tree structure with three levels of
information and six nodes. The pointers on each node refer to the memory
location for the left and right links, respectively. For example, pointer Ls
refers to the memory location that holds the set of D data, i.e. left link to
the B set.
1
'
L..)
If a node at level i points to a node at level i+ 1, then the no.de at level i is

the father and the other one is the son. No informatlon at a particular node
(except the root) can be retrieved unless the root is exa.mined in advance.
Figure 7 .16 shows how the data can be stored for defining a simple binary
tree.
Each no de of a binary tree can be assigned a binary number. Assume
each left branch is recognized by a O and each right branch by a 1 digit.
Starting from the root, the furthermost left digit of a specified node number
,,,..,
. \
""' ......
1111
,,..
rlght
notk
m3
,4
x-4
-B
x.
"' "-,;;-\
"'s"',,........ . . ,,61........
' ',
... ....,
\
'"'e
-o
~- .. B
\
'F
,,,
Xc
'
Xo
7.,
"
'
x,
Figure 7.16: Storing the data for a binary tree structur~.
le
198
Chapter 7. Discontinu um Con ta.et Mecbanlcs
199
----------------~ level O

(011)
------- level 3
(110)
a) Degenerated tree
b) Well balanced tree
Figure 7.18: Degenerated and well balanced binary trees.

Flgure 7.17: Enumerating a bin~y tree components with binary numbers.
is rcad and the branch corresponding to that digit is followed to a node of

the first level. At the node of the first level, the next digit of the particular
node number is read and the corresponding branch is followed to the node of
level 2. This is continued until the last node is reached. Figure 7.17 depicts
the same procedure for enumerating the node 1 by a binary number.
An optimized ordering procedure or node inaertion can be developed

to consider the possibility of balancing a tree struct me by adopting a new
order of insertion. Such an optimized tree structure may be found extremely
effi.cient if a binary tree for geometric intersection search has to be rebuilt
and searched through relatively often.
Information RetrievaJ/Tree Traversa!

Creation of a Binary Tree
The fi.rst step in creation of a binary tree structure is selecting a root nade.
Adding new data items to the binary tree depends on definition of a criterion for choosing between the left or the right branch for insertion. Every
insertion then starts by checking this criterion at the root node. and then
traversing the tree until an empty place is found.
The criterion for insertion of data iteras and traversa! of the binary tree
is in fact a measure of relative spatial position of two nodes of the binary
tree.
The order of object (body) insertion determines the final shape of the
binary tree structure. The shape of the binary tree substantially influences
the computational cost of the global contact searching and the insertion
of new data items. Low insertion and search costs are obtained from well
bala.nced trees (Figure 7.18b), whereas poor performances are expected from
hi~hly tlcgenerated binary trees (Figure 7.18a).
A nade in the tree can only be visited or examined if all its ancestors are
visited first. However, an efficient systematic retrieva.l of data requires that
each node is only visited once. This can be achieved with a recursive scheme
of the so called stay left rule,
l. Retrieve data for the node at iev;i i.
2. If the left link exists visit the pointed node at level i
+ 1.
3. Else if the right link exista visit the pointed node at level i
+ 1.
4. Else return to level i - l.
Figure 7.19 illustrates this simple rule for a typical tree. The order of
traversa! is: A-+ C -+ D -+ E-+ F-+ B -+ G --+ H. Tlw procedure can
be summarized as Box 7.1.
200
Clmpter 7. Discontinuurn Contact Mechanlcs
201
Programming languages such as C++ and F90 support automatic recursive algorithms, thus a relatively easy implementation of the traversing
technique ca.11 be performed. However, the more common F77 language does
not provide this capability and morti complkatcd implementation, such as
the stack technique is required.
ortkr ofretrieval: .4-C-D-E-F-B-G-H
Figure 7.19: Data retrieval in a binary tree.

Box 7.1 - 'Iraversing procedure for a binary tree (Figure 7.19).
In a so ca.lled stack technique the items are inserted onc by one, and then
extracted in the reverse order one by one. During the insertion procesa, the
value of the right link is stored in order to enable the subsequent traversal of
the right subtree. In fact, a list containing the address of all right subtrees
encountered along the way which are yet to be traversed must be kept and
continuously updated [126] 'j
The traversa! continues until a zero left link is encountered. Then, the
last right link inserted in the stack list is retrieved. The subtree rooted a.t
this address is traversed and the corresponding right link is removed from
the stack.
Box 7.2 - A traversa! algorithm
\uu;~
on the stack technique [126].
l. Visit the root A
2 .. Take the left child (C) and traverse its subtree

- Visit C
- Take the left child (D) and traverse its subtree
- Visit D; Skip (no left and right children)
- Take the right. child (E) ~nd traverse its subtree
- Visit E
- Skip (no left branch)
- Take the right child (F) and traverse its subtree
- Visit F; Skip (no left and right children)
3. Take the right child (B) and traverse its subtree
l. root_address
= address of the root node
2. size =O
2.1 - visit the node stored at root_address
2.2 - IF rightJink :/:- OTHEN

-size
= size +1
- stack (size) = rightJink

2.3 - ENDIF
2.4 - IF leftJink =/= O THEN

- root-addres.5
leftJink
- GOTO 2.1
2.5 - ELSE
- Visit B
- Take the left child (G) and traverse its subtree
- Visit G
- IF size =/= OTHEN

- root.address = stack(size)
- Skip (no left branch)
- size = size - 1
- Take the right child (H) and traverse its subtree
- GOTO 2.1
- Visit H; Skip (no le.ft and right children)

- Skip (no left and right children)
- ENDIF
2.6 - ENDIF
2.7 - IF size =O END
202
Discontlnuwn Mech8f1lcs
203
7. 2.6
A simple non-recursive implementation of the traversal algorithm based

on the stack technique is provided in Box 7.2 (126). With the help of a stack
technique, any recursive algorithm can be implemented in programming
languages such as F77 which do not support automatic recursive routines.
An efficient domain decomposition scheme is an essential part of a contact

detection algorithm, because it directly affects and controls the subsequent
steps of the search procedure.
The last remaining point about the binary tree structure is to define
a methodology for removing a node. If the undesired -TIOde. is a terminal
node, it is sufficient to change to zero the corresponding link of its father.
Therefore, the occupied memory will be available for new node insertion.
Space cell decomposition methods, in general, comprise finding the suitable shape of a cell, space subdivision into cells, and cell enurneration.
In the following, various methods of space decomposition are .briefly
reviewed.
In the case of an intermediate node, the unwanted node is replaced by

a terminal node chosen from among its descendants. This is performed
without moving any nodes from their physical memory positions and only a
il
i
Space Decomposition
Coordinate Based Space Decomposition
simple modification of the left and right Jinks of the father of the unwanted
node will be sufficient. Figure 7.20 illustrates the deletion process for a
simple binary tree. The deleted node C has been replaced by one of itJs
descendants terminal node H.
In this method, each coordinate axis is divided into finite intervals and ~a.ch
finite interval is given a number. Ea.ch cell is defined as a set of points that
have coordinates that belong to the same interval (Figure 7.21).
If a particular application demands frequent deletion/insertion processes, a memory rnanagernent system is necessary for the efficient implementation
of tree str~ctures so that new nodes can be placed in the memory space
released by the deletion of previous nodes. With programming languages
featuring dynamic allocati9n, this is automatically provided, F77, however,
<loes not offer this capability, and the problem should be handled by using a
linked list structure to record all the available memory spaces. For further
details see [126].
Each cell can be enumerated by two (for 20) or three (for :m) numbers.
Instead, a single cell number may also be constructed using a binary number
structure.
(6,7)
8
Cis removed
7
6
C '1
-......_
(7,2)
l_
Figure 7.20: Deletion process in a binary tree.
Figure 7.21: Coordinate based space decomposition in 2D.
Discontinuum Mecha.nics
204
205
Cha.pter 7. Discontinuum Conta,ct Mechanics
Suppose the problem space with n bodies is divided into m cells. On

average, there will be ( '!!i) bodies in each cell. To check for contacts in one
cell, (~)(~ -1}/2 check.s are necessary, and to check all cells n('!!i -1)/2.
An advantage has occurred because of the implicit assumption that bodies in one cell can not interact with bodies in another cell. However, in
above, the necessary checks for assigning bodies to parti~ular cells are neglected. In addition, because bodies may span the grid ~between cells and
a body may move from one cell to another with slightest motion, frequent
numerically expensive cell updates are necessary.
o
2
? 8 9 10
Space Bisection Algorithm

This method provides an,ppropriate ~~proa.ch as the base for the node
insertion criterion ih construction of a digital tree. In this method, the
space is systematically divided into two parts at ea.ch level. Figures 7.22
and 7.23 show this systematic bisectioning of a 2D space.
11 11 JJ 14
10
11
13
11
Figure 7.23: Space bisection procedure.
To better explain the way the tree is constructed, assume the problem
stated in Figure 7.24, which consists of eleven disks moving in a domain.
To form the tree an optional order of. points should be assumed, e.g.
B - t C -+ D -+ E ~ F -+ G -+ H -+ 1 -+ J ~ K. If the root
(here A} or the order of the points is changed, another tree structure will
be formed.
-t
1
1
- - ' - - - t
---------(b ------ ~--'
---1--------<b----
. .-tr--'.! o_J!______ _
t
'
1
1
---+:: -
1
1
'
levelO
level 1
leve/ 2
level 3
level 4
- - farther levels

1 .G
K
J.
Figure 7.24: Space bisection"'lg6rithm for a typical problem.

Figure 7.22: Space bisection algorithm.
n
LJ
DiscontlziUum Mechanics
206
Chapter 7. Discontinuum Conta.ct Mechanics
207
"
--------------~-------'
.,---------'----"-----
..
"
Figure 7.26: Target search within a space bisectiou procedure.
--------=,------'-----------.
.9.f '
.E
2-
'
:~,)
--~-
" e
--------I. _ _
J_-'----------------~
'
2 -- ------ -A --:---+--1
C
1 " e
Figure 7.25 shows a step by step procedure of constructing a tree associated to binary space decomposition of problem stated in Figure 7.24.
The main advantage of the geometric ~earch based on the binary tree is
that the area associated with descendants of a no<le A.t. l'lny leveJ ii:; always
contained within the area associated with "the node [52]. For example in
Figure 7.23, the area associated with nodes 3 and 4 are contained within
the area associated with node 1 (and so node O). Therefore, in a contact
detection analysis, if a target does not intersect. with tlw area. associated
with an object, the object and ali its descendants will not be considered as
potential contactors for the target; i.e. they will not be included in the target
short list of contactors. Figure 7.26 illustrates this key feature. Objects A
and B and ali their descendants can not be a candidate for contact with the
target (as it is clear!y observed from the figure) .
Hextree Decomposition
Je
_____J__
Figure 7 .25: Space bisection algorithm for a typical problem.
Another extremely eflicient and systematic decomposition approach is the

hextree decomposition procedure. In this method, sets of he:wgonalaggregates are used to decompose the space. The first level of a hextree is formed
by taking a single hexagon and its six neighbours (Figure 7 .27).
A second level aggregate is defined by taking a first level aggregate and
its six first level neighbours. Figures 7.28 and 7.29 show the second and
208
209
third level
aggregate 1
leve/3
- - level2
- level 1
Fip;urn 7.27: First lovel aggregate for a hextree space decomposition [127).
lJ
secondlevel---....._
aggregate 1
level2
Figure 7 .29: Third level aggregate for a hextree space decomposition (127). .
- - levell
23
third levels of aggregates, respectively. This process continues until the

required level of representation is reached (127].
;
The address of any particuiar hexagon is composed of a series of digits.
The lowest order digit identifies which of the seven cells (O - 6) in the first
level aggregate is being addressed. The second digit identifies the first level
cell number within the second level aggregate, and so on.
04
4
Figure 7.28: Second level .~egate for a hextree space decomposition [127].
To realize the level of detall attainable by this method, it is estimated

that a 20 level system can represent the whole surface of the earth within a
lOcm resolution.
This system of representing spatial data has particular advantages. Firstly,
all addressing involves integer arithmetic. Secondly, only those a.reas where
objects exist need be represented in the tree. Both thc::sc feH-Lm'es ma..'ke
retrieval and manipulation of spatial data extremely efficient.
210
211
Chapter 7. Discontinuum Canta.et Mechanics
Adaptive: Grid
In this method of space decomposition an adaptive grid of cells is used. In
this method, a variable fine cell positioning system is used to define the
space which is occupied by the objects. Elsewhere, coarse grid of cells will
be used. Figure 7.30 illustrates the way an adaptive grid is constructed for
a 2D problem.
'~ ,
1
Y1
-"
L.
....
.... '
~-
____....,.______--4________.....,___________
--
'' -e'
~
"'
...
,.
Figure 7.31: Spatial heapsort approach.
i~
be sorted, and the second array (rank a.rray) is uscd to rnnk the order llf
the objects when they are srted.
Figure 7.30,: Adaptive grid for 2D space decomposition.
Spatial Heapsort Technique
A hoap is gen~rally defined as a trcc with the property that a.11 of the nodes
below a given nade contain a key value that is less than that of the parent
[24].
Figure 7.31 illustrates the heapsort algorithm for a collection of 20 objects. The lower bound extents of the object boundaries are projected onto
the. X and Y roces. To perform the spatial. heapsort, .two arrays of integer
fields are require<l for each dimension of the problem doma.in. The first
array (id array) of each pair is used to store the set of object identifiers to
The search procedure consists of two steps:
1. At the first step, an object from the id array is selecte< 1 (called pivot),
which is the lower bound search index. Then using a binary search, the
sorted list is traversed to identify overlapping objects in this direction

by checking if the largest ordinate of the pivot object parallel to the
search axis lies beyond the smallest ordinul-t~ of objcctH lying ahead
of it. The process is stopped at the index location of the object that
<loes not have a lower bound extent leas than the uppcr bound extent
of the pivot, i.e. the upper-bound index.
2. The same procedure is followed for the remaiuiug axes to identify both
an upper-bound anda lower-bound index for the pivot.
212
Chapter 7. Discontinuum Contact Mecha.nics
213
Body Based Cells
In a geometric intersection problem, the goal is to find the objects which

overlap a certain subregion of the space being under consideration.
Another alternative is to associate the search cell with an element, so that

each element has a cell surrounding it.. This is as if each element has its
ow,n radar which scans out to a given distance. The element is only aware
of other elements within its sean radius, saving them in the so caUed short
list of the neighbours (Figure 7.32).
,
An alternating digital tree {ADT) algorithm is one of the most eflicient

_.available search algorithms offering the following specifications,
A natural extension of the digital tree search technique.
Efticient solution of the geometric searching.
Offers the possibility of inserting/removing points.
Optimally searching for the points co!ltained inside a given region .
.;-"'
I
I\
<> \""'"'.
Applicable to 11ny number of dimensions.
'
'-...-/
....,....,
..... .....,
..........
o
/
--'
/'
/
~'7'
body
based
cells
'O'
,_/
\
Allowing treatment of any geometrical object in a R 11 space as a polnt

in a R 2n space; thereby allowing the solution of geometric intersection
problems.
When the distribution of polnts/elements to bo trcated is reasonably
tiniform, the necessary cost to solve the geometric searching and intersection
problema is found to be proportiorial to nlog(n) where nis the number of
itema.
Here, the methodology developed by Petrinic [1] for contact detection
problema based on the original work by Bonet et al. [126] for ~vancing
.front mesh generation technique is followed~
Body Location Mapping

Figure 7.32: Space decomposition by the body based cell approach.
7.2.7
Alternating Digital Tree (ADT)
The problem of developing an efficient. search algorithm basically arises in

two classes of problems within the general computational mechanics sul>ject;
the geometric searching and the geometric intersection problem.
A geometric searching method is a searching algorithm for determining

which members of a set of m points in an n dimensional space lie inside a
prescribed space subregion. The algorithm may also be extended to handle
finite size objects as well as points. The computational expense is proportional to log( n).
The first step in a global search algorithm is to represent a general shaped

body by- a. standard simple botinding box. Each body cn.n h1: drcumscribed
with a bounding box whose edges are parallel to the global coordinate system axes as depicted in Figure 7.33. This bounding box object will be used
in ali global search procedures.
To comply with previously defined point based space hiS<~ction algorithm,
a body i from the Rn space can be represented by a point in the R 2n space
via an invertible mapping L : Rn.... --+ R 2n that combines the coordinate
limits of the body's bounding box according to:
...., . -...,,
[xli,min1 ... , Xni,mint Xli,ma:r:t ... , Xni,max ] T
(7.7)
is a unique location of an object in the R 2n space. Figure 7.34 shows the

mapping procedure for a relatively simple one dimensional case (n = 2).
Xi
214
215
Creation of the Binary tree _
Basic specifications of binary trees have been previously rliscussed. The data
items stored at the nodes of a binary tree, whkh is cwated with respect to
the geoIJletric intersection search, must contain iuformntion on geometric
properties of the bodies under consideration.
By performing the location mapping (7.7) for all bodies from the Rn into
the R 2n space, the creation of a binary tree data structure for a geometric
intersection problem can start. It begins by selecting one body from a given
set as the root and continues by inserting new nodes.
The node insertion criterion for the ADT is based 011 decomposition
of the space region assigned to the body at a particular node by a space
bisection algorithm. The bisection is performed with respect to the body
location coordinate component which is used for checking the relative position during the tree traversa} and in turn is associated with the binary tree
level containing the traversed node'[l]. By defining
L_J
_ [, 1
.
n
.
1
. n
Yi,mint ... , Yi,min1 Yi,maxi ... , Yi,ma:c
Yi -
]1'
(7.8)
Figure 7.33: Definition of a boun<ling box for a 30 body [1].

as the coordinate limits of the bounding hypercube region {limite of the,
bisection level corresponding to body i), the following condition must be
satisfied for a body i with its location in the R 2 n space defined by :e,
Uf,min ~ X. ~ Y,ma:c
f or
j = 1, 2, ... , 2n
(7.9)
defining a simple though efficient geometric search criterion. Yi should ~ '.; ,

also stored in order to complete the d~ta tem stored at the root node.
Figure 7.35 shows the tree cpnstruction procedure for a typical one di-:
mensional problem. The ord~r ofinsertion is: A --t B --~ C-+ D-+ E-+ F .
'-....- ... ...
~-"'
..
,.
bodyi
Intersection Search
~
1'
x,..,,,,
...
Figure 7.34: Mapping a lD segment to a point in R 2 space (1).
The recursive nature of the space bisection process for creating the binary
tree data structure provides an appropriate basis for nn efficient automatic.
geometric intersection algorithm
In additibn to the information of the mtt.pped points in the R 2n space,
each tree node contains the information of its as1mciated bounding h,y-pc.rcube region. A hypercube stored at any node of the biuary tree supersedes
all bounding hypercubes stored at the descending nodes, providing a key
feature to the fast geometric intersection search algorithm.
216
217
'~
.....
x',._
- ...
x'
l~IO
.'
40
JS
AS S)
JO ..___
(J~ 4)
CIF
IS
_..__...__.....__._..__....__...__.-........_ XI
IS
:JO :IS
tU
..
10
P( 4,11
10
CJIB
10
l'IW
:4
:JS
level 1 .
Jllil
30 JS 41
o
I
10
IS
21 JJ JI
--
JS 40 X!.,,.
AD
cCJ
level O
Y,";.r.
level 1
L.J
BD
levd2
levelO
~
q.1
.iL
D[J
cm
D&'.]
cm
LI
,......,
!~h.._
level o
______________.
kvell
[~
o
lnll2
levelO
AD
EO
level O
levell
'b
""'' 1
~,.,,..
1...,.:1k+-______.._______...
,.,
Figure 7.35: Creation of an alternating digital tree data structure and ddn~
ing the coordinate limits of the bounding hypercube region.
218
A hyper rectangular region (ai, bi) in the R 2n space, representing the

search region associated with the body i, can be defined as [1],
ai. -
.
-
n
1
n
Yr,min1 ... , Yr,min' xi,min' ... , xi,min
[ 1
]T
(7.10)
bi
where the subscript
[xi,ma:z:, ... , Xt,ma.a:l Yr,ma.:z:l ... , Yr,maa: JT

1
n+l
2n
.....
denotes the root node.

"
In other words, the locatio11; mappi~g projections of all objects inside

the bounding box around the body i will be placed inside this region.
'
(7.11)
JI
'
'
.1
"
1l
"
'
,.
"
,_
H
n "
n "
JO
Jll ,__ __ ,___ r--1--
--1--+--r--i--r--r--i,-0-1-
J4 --- -- -- -- --
---+--+---t----+--+--r--
..
:14
(7.12)
- x'
"
JI
40
i---.-- -r--- -- --r--r--r----r--r---r--r--1

1 ~-
l--f--lo~--4--+--+--+---t--+--+--+---+--D-Lin --
;
j == 1, 2, ... , 2n
e-
Furthermore, the bounding boxes of bodies i and k are found to be overlapping if the mapped location of a body k satisfies the following condition
f or
lf+ITTH
-l
1
-e
IJ
b=11=~~1
-~--+---+--t--r--r--;----i--1-. -~
:.1:1 ,__,__ .....
r---:
:.10
,.
.....
o.~
-.... .
-1
.. -- . ...
~o
11 ,__,__ ~- - - --4--H-:J.+--+--+--+--+-
This condition is checked while traversing the binary tree in order to generate the short lists of potential contactors k for each target body i [1], leading
to creation of short lists of all potential contact couples.
For a selected target body f, the binary' tree is traversed until the targets'
search region intersects with the bounding hypercube of a body k ata visited
node. If the body k satisfies the condition (7.12), it becomes a contactar
to the target body i and is added to the targets short list of contactprs.
Otherwise, the whole branch of the body k descendants can be exclu:ded
from frther search [1].
The process of creating short lists of contactors for each target body
from a given one dimensional set of objects is illustrated in Figure 7.36.
It is worth noting that a further reduction in the number of required operations to accomplish the intersection search may be achieved by comparing
ouly those coordinate components which are associated with the binary tree
bisection level at which the traversed node is. inserted. For example, according to Figure 7.36 only the first coordinate components of nodes A, E, F, D,
N and Pare required to be compared with the first coordina.te components
of a and b search region, while the second coordina.te componerits.of nodes ...
n, C, I, G, K, L, M and O will be compared with the second coordinate
components of a and b search region.
16
,___,f----+- - .. . it ~
14 ---- .
1:1 1--f--+--+- --10 ,_ .. ..
..'
, 1 l~ -tl
--+---+---+--_,___+--.--..-
...
... .
6L-J....-1.-J"--1.-Ji---.,j......L......i.--i.--i.-i.--i.-'--1--'---'--'-~_.~xmm
'
l
'
11 1 10 1:1 u 11 11 11 u " 26 28 Jo n 34
tree
level
search
component
220
Chapter 7. Discontinuum Contact Mecbanics
221
X"""'
'
JI
36
Accurate Local Contact Search
-----------b.
(2)
-f2....i2-,~~ -- - ~~
... -
----
--r--
--1--+--+--t--+-- - - -
-~
The global search method based on the ADT approach uses a bounding box
search to a.void numerous difficulties that me.y u.riHe iu dealing with complex
geometi:ies of real objects. As a result, the resulting short lists of potentlal
contacta have to be re-examined within an accurate local contact search
procedure to determine the intersecting objects. The contact interaction
algorithms may then be utilized to define the surface tractions arising from
contacts between intersecting bodies.
-- ---::ir
u1---l-+-t--~-1--1----t--1-+-41-+-+--t--+tM:H-'-f
31 -+-+-+--l--+--+--+--1---1-+---+--t--+--+--t--1
JO
i--r---
-+--+--+-'-+----t-+-t--l-f-t-IJ-t-+--f--1---+--t
u1--r~-+--1--t--t--+--+-al+l-+----+--t-+--t---ir-t
u1--1-1-+-+-+--t--+--+-+-i-lt-t-+-t--t---t--1
This accurate local contact search, also called local contact resolution
phase, requires a detailed.geometric description of the contacting couples in
order to accurately deterfuine the bodies really intersect.
n ----1--~-+-~-+--+--+----+-+-+--1--<1-+-+--1
/ll--l-IC-Q-1--+-+-l-i~~IH-+-i-lt-+-+-t--t---t--I
00
r=(O,U) ce-+..............-+'""'!l!l!-J-+--+--+-~f-li-t-t--r--r-i
10 - - - ... -- -
'
11~-- --~} - +O.,__..__ -
- - - --- ---+--+--+--+-1--1
---
Rigid discrete elements have simple shapes with closed form surface definitions. Therefore, an accurate local contact search can be performed based .i}~
on ma.thema.tical intersection solution of potential contact couples.
111--1-l-+-+-+--l---l--+-+-~-+-+-+--t---t--1
~-t--+--<f--t-+--t----t
6.__._.__.,_.__.__.__._......_..._......._._..._.__.__._...__l.~x
'
'
10 IJ "
16
11 "
'1
u " "
JI J:l
In contrast, if the surfaces of the bodies are approximated using meshed : \::.
polygons/polyhedrons as a result of the finite element discretiza.tion, the
geometric intersection search procedure can be subsequently rearranged to
use the edges/facets rather than the whole bodies in order to simplify the
accurate local contact search.
..
Finally, from advanced programming point of view:
This approach allows object orientd programming techniques to be

1:1sed in accordance with the objects' hiernrchy used for finite elements
[1).
The conta.ct detection procedures, rely heavily on indire?t addressing. As a result, it is virtually impossible to vectorize them to any
significant degree.
'
Nevertheless, they offer possibilities for parallel compute11-3 as work on
disjoint subtrees can, in principie, be carried out independently.
7.3
Figme 7 .36: Geometric intersection search.
Object Representation
In this chapter the main classes of object representation methods are discussed. It includes circular disks, ellipse shaped disks, and the general
superquadric forms. Additionally, the meshed polygon systems, which frequently encountered in general finlte element c.ontact. analyses are also
among the object representation techniques. They usually provide speclfic
222
Cba.pter 7. Discontinuum Contact Mech,anics
223
A relatively simple computational sequence fer disk element analysis can

be summarized according to the following:
problems within a conta.ct detection or interaction procedure which have

already been discussed and will not be addressed a.gain.
For all contact pairs, follow the force-displacement law:
7. 3 .1
Circular Disks
Circular disks/ spheres a.rflprobably the most frequently used type of element in modelling of granular ftow by the discrete element metbod. They
have been used as rigid body objects interacting ea.ch other in a granular
fiow simulation. Both penalty and continuum mechanics based methods
have been used for contact interaction formulation.
From the object representation point of view, they consist of the simplest
forros for two and three dimensional mbclelling. Their geometric representation includes the coordinates of the centre and the magnitude of radius.
The motion of particles can be readily calculated from the equations of rigid
'
body dynamics.
Figure 7.37 shows the geometrical description of a system of two interacting 2D disks.
Rele.tive velocities (i = 1, 2)
Xi= (xia -
ni =
.Xini ,
xib) - (OaRa + bRb)ti

ii = .Xiti
Relative displacements
.6.n = n.6.t,
At = i.6.t
Contact force incremente
AFn = OnAn,
Total forces at time i
+ IJ.F11 ,
F~ = F~- 1
Check for slip
Ft
= min(Ft, Fntan</> +e)
For all particles, used the equations of motion

Calculate moment
Ma=L:FaRa,
Assume constant force and moment from ti-t to ti+t
Accolorntion
.. j_~
x, -
'
Velocity
,f+! -- Xi.j-! + .u~

Atd
(ji+i =(ji-!+ .6.tijJ
--
Assume constant velocities from ti to tH 1

Displacements
= x! + .6.txi+i
'
'
oJ+i = Oi + IJ.tOi+i
a;1+i
Figure 7.37: Geometry of two interacting disks.
At the end of sequence, the time is incremented and the whole procedure
is repeated. A more sophisticated approach is presented here to clarify the
main specifications of a disk ha.sed discrete element technique as described.
by Petrinic [1].
Dlscontlnuum Mechanics
224
Chapter 7. Discontinuum Canta.et Mechanics
225
~.
Figure 7.38: A model for normal contact force between two circular disks.
Normal Contact Force

Figure 7.39: Normal contact between two circular disks.
Although the size of the overlap is small compared to the radii of the disks,
only the contact zone is considered to be deformable. The contact force is
assumed to be proportional to the overlap size of the two disks in contact
and their relative velocity in the normal direction (model described in Figure
7.38),
K1
_
-
l-111
G1
(7.15)
(7.13)
where Pn = Pn(u) is the spring force, Cn is the viscous damping coefficient,

v:;, is the relative velocity in the direction normal to the contacting surfaces.
For small overlaps, the nonlinear spring behaves linearly which can be
expressed as
'
The spring force Pn is defined using the elasticity solution for two disks
in contact,
(7.16)
where Kn is the spring stiffness,

(7.14)
--
K _
n -
where R is the radius of disk, G is the shear modulus, v is the poisson's

ratio, b is half the width of the surface of contact (defined in Figure. 7.39)
and,
dPnl
dun
(7.17)
Un=O
The viscous damping coefficient is represented by a chosen percentage

of the critica! damping for a collision of two disks,
226
227
(7.18)
(7.19)
w here Mi is the mass of the disk i.
Tangential Contact Force

Geometrical idealization causes disks to be less resistant to rolling than the
actual round shaped bodies they represent. Therefore, in order to model
the formation of phenomena such aA arching in granular ftow using circular
disks, an additional part of tangential component of contact force between
disks has to be employed. Here, a so called rolling reslstance is applied by
means of a viscous damping force.
Sliding Friction
Figure 7.4~: Kinematics of two contacting circular disks.
,,
~.-. component of contact force is represented by the

This part of the tangential
model shown in Figure 7.40.
{7.21)
r1
e,
LJ
where is the coefficient of friction, Fn is the normal component of the

COntact force,
is the re}ative tangentia} di:;p}aceUWllt.H of the.contactmg
disks, Ct is the viscous damping coefficient a.11d v[ iH t.he relative tangentlal
velocity.
F,,
ur
The relative tangential di~placement between the two disks is obtained

from the solution of the globn.1 cquations of motion,
K,
(7.22)
Figure 7.40: A model for sliding friction force between two circular disks.
The actual cxpression for sliding friction is obtained following study of

the sliding contact for locomotive driving wheels,
The relative tangential velocity is determined from the disks kinematics

(Figure 7.41 ),
(7.23)
(7.20)
L
where Pt
= Pt ( n1.) is the spring force defined as
and the duration of contact is defined by:
228
(7.24)
where Un is the size of normal. overlap, tl.t is the length of the time step
during which the contact occurred and v is the disk velocity vector, while
229
Rolling Frlction
Consider the situation where the disk is set to roll on a rough horlzontl
plane (Figure 7.43),
(7.25)
is the relative normal velocity.
Figure 7 .42 illustrates the relation between the relative tangential displacement and thefrictional spring force.
Figure 7.43: Rolling disk on a rough horiwntP.! imrfo.ce.

The sliding friction cannot provide any resistance to the movement of the
disk rolling on a rough surface since there is no relative tangential velocity at
the contact point (v[ =O). Therefore, an additional term for the tangential
component of the contact force is also required; called the rolling frlctlon,,
..._-----------------Ur
n
(7.29)
Figure 7.42: Relation between the relative tangential displacement and the
frictional spring force.
where
(7.30)
The viscous damping coefficient Ct is represented by a chosen percentage

of the critical damping
is a chosen percentage of the critica! damping (7.27), a.nd
(7.26)
v[c = (v1 - v2) t

Ct=2
(7.27)
(7.31)
is the relative tangential velocity of the centroids of the disks in conta.ct.

The rolling friction force should also satisfy the following condition
with Kt as the spring stiffness
: (7.28)
--
Also, if the rolling friction obtained from (7.29) results in
(7.32)
230
11,'t. 1- ,1r 1 '_.. Ji'u
Chapter 7. Discontinu um Contact Mechanics
231
(7.~3)
it should be re-calculated in order to give priority to sliding
Fr = sign(v[c) lFn -
IFtll
(7.34)
Condition (7.32) is often sa.tisfled automatically since the critica! damping (7.27) depends on the sliding friction stiffness (7.30), which decreases
when approaching the maximum allowed friction force Fn (Figure 7.43).
This is why the rolling part of the tangential component of the contact force
is chosen to be applied in a form of damping. It ensures good cooperation
of sliding and rolling friction.
l
1
Applying the force at the centroids of the disks also implies adding a
resisting moment in the direction opposite to the direction of rolling as
Figure 7.45: Unknown variables of a basic disk cluster.
(7.35)
1
7.3.2
Disk Clusters
(7.36)
where T is a displacement function of disk i
Disk clusters have been introduced as a new generalization of circular disk

objects [128]. They are formed ftom a group 'of virtually connected rigid
circular disk elements and crea.te a wide v,ariety of irregular particle shapes.They provide higher resistance to the rolling movement of particles and
therefore, do not require any artificial rolling damping etc. as required by
standard circular disks.
=[1
O -(y-yo) ]
O 1 (x-: xo)
(7.37)
Applying the principie of mnimum potential energy, the following system of simultaneous equations is derived for a syst.em consh;t.ing of n disks,
Figure 7.44 illustrates five simple basic disk clusters. They can be utilized to generate various particle shapes.
KD=F
(7.38)
where. T is a displacement function of disk i
Ku
K21
K12
K22
Figur.,.44: Basic disk cluster elements.
Kn1
Displacement vector U = (u, vf of a point (x, y) of a disk can be

defined as a function of rigid displacements and rotation of its centre
=
(u, v0 , 'Y)T
U?
Kn2
..
K2n
Kin
...
Kn.n
l[ l[ l
D2
D1
F2
F1
Dn
Fn
.
..
.
..
(7.39)
where KiJ is a 3 x 3 submatrix. Equation (7.39) can be solved by one oi .

the available equation solvers. For further details of the formulation ~
references [128, 129).
232
Disk clusters are assumed to be rigid bodies similar to circular dislcs,

except for their contact (spring) bounda.ries. Therefore, there is only one
unknown per cluster, U of the centre of cluster (see Figure 7.45).
Displacement vector of any point of a cluster can be defined using a
displacement function similar to T of Equation (7.37) where (x 0 , y0 ) is the
centre of the cluster. In fact, there is only one set of unknowns for each
cluster and the size of the problem is not cha.nged if the clusters are replaced
by equivalent number of circular diflks. The formulatin for inertia. and
damping terms are very similar and the a.rea and angular moment of inertia
of a circular disk are replaced by correspondirig values of a cluster about its
centre, respectively.
!l
1
LJ
An important aspect of clusters in c~mparison to ellipse elements is their"

contad detection procedure.. For clusters, the same procedure as circular
<liAkR cn.n ho follwecl. Ali conto.r.t.R l\re Msumed to occur e.long tho center
Hnes. No contact contr~ts required for disks within a cluster.
Disk clusters are expected to perform well in problema that rolling resistance has a major part in overall behaviour of the pa.rticle/object. For
example, Figure 7.46 illustrates the problem solved by Thomas et al. (128].
An individual circular disk will a.lwa.ys roll, whereas a cluster may slide,
roll or remain unchanged depending on the slope angle, tangential friction
coefficient and the cluster shape.
Cha.pter 7. Discontinuum Contact Mechanics
7 .3.3
233
Ellipse Shaped Particles
With more widespread use of disk and sphere based numerical codea, and
the recent development of sphere based constitutive models for granular
assemblages, it is particularly important to assess the degree to which these
models are applicable for real non-spherical materials.
One common problem when using disk and sphere ha.sed discrete element
modelling of soil is the low aggregate friction angle inherent in these systems,
regardless of the angle of inter pa.rticle sliding friction, which is used.
Particle shape has the 1largest effect on mechanical behaviour, with reported increases I\ peak interna! friction angle up to 10 for systems consisting of angular particles compa.red with round particles.
With the realization that disk based discrete element model has serious
deficiencies when used for modelling real granular materials, it has recently
become popular to use the clHpse as th~ barde parl.ldt1 r>lmpo. The elllpee
shape has the advantage of having a unique and c0ntin11rn1s 011tward normal
and no singularities along its surface.
Solution for ellipse-ellipse contact detection requires solution of fourth
degree algebraic equations, which can be done analytically rnther than with
iterative procedures. For these objects, normal c:ontnct forces actlng eccentrically on a particle can generate applied moments w hich potentlally
inhibit particle rotation. As a result, this shape is well suited ph)i'sically
and numerically to modelling granular soilS, powders and grains. '
Contact Decomposition
.............. ~
.........
Figure 7.47 indicates the nomenClature for two ellipses in contact. Points
A and B, which can be usecA as a measure of the total normal overlap
(penetration) between the two objects, a.re determined from the current
ellipse-ellipse intersection algorithm.
1
1
To 88Sess the relative importance of rolling and sliding mechanisms of

deformation within the granular assemble.ge, the contact deformation is separated into portions due to individual particle rotation and particle translation.
Figure 7.46: A basic cluster may slide, roll or remain atable on a rough
slope.
For particle i, .the vector from the centroid in the direction of the presumed point of contact is:
1.
rci =:te - X1
(7.40)
234
235
perpendicular to the contact normal to yield the incrementals norm8.1 and

tangential contact velocities: .
dvcn
= dvnn = (dvc n)n

= (dv 1tn1 + dvcyn'2)n
(7.46)
= dvtt = (dv 0 t)t

= (dvut1 + dvcyt2)t
(7.47).
ellipsej
dvct
noting that
(7.48)
where nis the unit .outwards normal at the contact for ellipse i and t is
the unit vector perpendicular to the normal, defined clockwise positive tO
particle i.
Figure 7.47: Contact between two ellipses.
(7.41)
The velocity of the contact on particle i dueto rotation and translation

of i is:
Ata given instant, the individual terma in (7.46, 7.47) may be separated
lnto the incremental net contact deformation, nct normal contact deformation, or net tangential contact deformation due to particle translation or .
particle rotation.
The contribution of rotations of particle i and j to the net tangentlal.

contact deformation is:
dv~t
(7.42)
(7.43)
[(-BiTci sin a)+ (83Tcj sin OJ)] t2

- [(B1rci cosa.) ...: ( -Bjrcj cos nJ)] t l
(7.49)
while the contribution to the .net normal contact deforma.tion is,
The relative velocity of the contact between i and j is then
dv~n =
[(-Birci sin a)+ (B;rcj sin Oj )] n1

(-fJj'f'cj cosaj)) n2
+ [(eirci cosa) -
(7.50)
(7.44)
The contribution of translation to the net tangential cont.act deformation

is:
dvc
= . [(vxi - fJirci sino:i) - (Vxj - 83rc3 sina3)] i

+ [(vy + Oirci sinai) - (vyj + Ojrcj sina3)]j
(7.45)
where the terms with vare attributed to individual particle translation, and
O to particle rotation. This relative velocity may be resolved parallel and
(7.51)
(7.52)
236
237
Here, the geometric potential idea is extended to 3D. Two contacting

ellipsoid can be defined as
Numerical tests have shown that the particle rotation accounts for twice
as much contact motion for round particles as does particle translations
[118].
Ei =
3D Ellipsoid Elements
Three dimensional ellipsoid elements have not been adopted until very re- .
cently due to their difficult contact detection procedure.
E2
Algorithm described in Section 7.3.3 is based on the assumption that

the contact point of two ellipse shaped particles is the midpoint of the line
that joins the intersection points of the partlcles [118]. Unfortunately this
can not be easily extended to three dimensional problems.
A stable detection algorithm has been proposed by NG (117} utilizing the

idea of geometric potential. According to this method, potential J = f(x, y)
represents a class of conforma! ellipses with different geometric values. As
illustrated in Figure 1.48, there is a point on ellipse 1 (point 1) that has the
lowest potential with respect to ellipse 2. The contact occurs if this point
lies inside ellipse 2. The same procedure is followed for point 2 of ellipse
2. The final contact point is assumed to be the midpoint between points 1
and 2. Instead of minimizing the geometric potential, Ting (118} adopted
an alternate approach by~ fi.nding the position of the intersection points of
two contacting ellipses althe solution of a set of two quadratic simultaneous
equations.
a1x
2
+ biy 2 + c1z2 + 2fiyz + 2g1zx
+2p1x + 2q1y + 2r1z +di= O
+ 2h1xy
a2x 2 + b2y2 + c2z 2 + 2f2'yz + 2g2zx
+ 2h2xy
+2p2x + 2Q2Y + 2r2z + "2 =O
~(7.53)
(7.54)
where "' bi, ... , ri, di are constants.

The. Lagrange multiplier method is aclopted to find mnimum of E2 with
the constraint E1 = O
(7.55)
with
""
~ = O -+
2a2x + 2g2z + 2~y + 2p2

+A(2a1.x + 2g1Z + 2h1y + 2p1) =O
(7.56)
= O -+
2b2y +2hz + 2h2x + 2q2

+..$2b1Y + 2ftz + 2h1:c + 2q1) =O
(7.57)
Pj! =
O -+
2c2z + 2f2y + 2g2x + 2r2
+;\(2c1.t + 2fty + 2g1x + 2r1) =O
(7.58)
simplified to
and the constraint is

Figure 7.48: Mnimum geometric potential for an ellipse.
+ b1y 2 + c1z2 + 2ftyz + 2g1zx + 2h1xy

+2p 1x + 2q1Y + 2r1z +di= O
a1x
(7.60)
238
239
Chapter 7. Discontinuum Conta.ct Mechanics
(7.63)
(7.64)
8/t(z1 tVl ,.i1)
_ __..8....
z __ = O
8h(:i:hu'1
(7.65)
,.i1 >
- - - - - - =0
(7.66)
with
,-1
:
LJ
Figure 7.49: Common normal concept.
'(7.67)
Equations (7.59, 7.60) pro.vide four equations for determining four unknowns (x, y, z, ..$. The remaining part of the algorithm is similar to 20
approach.
L.\i
( 8f(x~:' z1) )' + ( 8f1(w:;:''' z1) ) ' + ( /Jf (x~;i.1) )'
One important deficiency of the mnimum potential approach is that it

rnay produce unparallel norm~ls at points 1 and 2 of two ellipses. Consequently, the osculating plane can only be approximately defined. The idea
of c9mmon normal concept has been developed to overcome this problem
[l~O].
(7.68)
L.\i =
Figure 7.49 illustrates a 2D representation of this approach. The method

s aimed to determine positions of points A and B so that normals to ellipses
at these points hccome parallel to the line AB, i.e. the common contact
normal.
Points A and B satisfy the general equation of their corresponding ellipse
function,
(7.61)
(7.62)
The common normal concept will lead to the following conditions:
8f2(x~~1,z1) + (8h(xi:i.1)
f+ (~M:~:1,z1))2
(7.69)
Equations (7.61-7.69) llave to be solved nunwrically.
7.3.4
Superquadric Objects
Superquadrics (superquads) are a generalization of mathematical functions

known as quadric surfaces. The extension comes about by raising the exponents of the variable terms to values other than 2. They are a family of
parametric functions, introduced in mid 60's and later proposed for use in
multibody dynamic analysis by Willlams (131]. It is estimated that about
80 percent of ali manufactured components can be represented by boolean
combinations of the superquadric forma.
240
241
n.
LJ
Figure 7.51: Superquadric elements [24].
D
n
lJ
where the parameters A and 'Y define the strengU1 and the degree of decay
of the isopotential field :F surrounding the potential source I.
Figure 7.50: Superquadric elements (24].
To generate even more complex objects, severa! localized isopotential

functions can be coupled:
From the family of possible superquadric functions, the best known is

the super ellipsoid (24]: C
M(x,y,z) =
(7.70)
where (xo, yo, zo) is the origin of the function, (ai, a2, aa) are the dimensions of the superquadric semi-major axes extents, and a and {3 are the
roundnes&-squareness. parame~ers of fnction in. two .perp~9~~war. 4.irections, respectively.

<
:F(x, y, z)
= -ye-AI(x,y,z)
(7.71)
(7.72).
which results in aggregate geometries M.

The ~upercfufidtic ftlnctioff'l;r\iid'etflt'Wlth of information about tht:.~~~~?jt;~~~,":"
geometry of the object. For example, it is straightforward to calculate the
surface normal, the surface curvatura, the volume and various moments of
volumes. These properties can be extremeiy diffi1...ult to cakulate using other
representations based on surface patches such cubic splines or cells.
Figures 7.50 and 7.51 illustratevarious objects that can be represented

by a superquadric function.
To manipulate the surface definition, new control parameters can be
applied to fine tune the locality and intensity of the function,
:.ri
-1
..... &:,
Equation I( x, y, z) provides a measure of distance of the point (x, y, z)

from the surface of superquadric. This property provides an extremely
useful check, sometimes called an inside-outide check, on whether a point
lies inside or outside the surface,
242
< O : The point lies inside
> O : The point Hes outside
= O : The point lies exactly on the surface
(7.73)
This check is exactly the one needed when determining if one body is1 in
contact with another in. a contact detection procedure.
7 .3.5
Geometric Models for Permeable Solids
As an interesting application of the contact detection/interaction processes

and the object representation phenomena, geometric modelling of permeable
solids, based on the work by O'Connor [24} is briefly reviewed.
n
lJ
Figure 7. 52 shows a typical porous medium where some set of the cavities
in the bulk material have formed channels through the region.
Chapter 7. Discontinuum Contact.Mechanics
243'
which form networks that cross the boundarie8 of the bulk material. This
is achieved by taking a solid cast of the pore spaces in a densely packed
assemblage of particles.
The process can be visualized as injecting wax into the pore spaces and
dissolving the solid particles in some manner. The wax which rema.ins is a
cast of the original pore space. Dueto the conncctivity of tho initial particle
assemblage, the pore spaces of the wax cast can be guaranteed to connect
across the boundaries, ensuring a permeablemodel [24).
In order to obtain valid void geometries, the modelling process mllst first
generate a matrix that is in mechanical equilibrium. Only then we can take
a wax cast of a void spacE\ that is geometrically valid and exists as a stable
mechanical system.
The numerical model, therefore, must support multibody contact mechanics. To model more complex media, the superquadric representation tech..
niques can be used to describe the particle geometry and the mecha.nisms
by which they interact.
a-. , .:.- '" . ~
Figure 7. 53 ill ustrate. a sifupte portS mdel generation [31]. Flrst

of solid bodies or particles placed and packed into a container using some
deposition mechanism. The inter-body contact mechanisms will form the
structure of the packed material. For the material to be stable, any objecta
must be supported underneath by a network of contacts rna.ching down tO .
the bottom of container. The entire volume of the container is treated as
a solid material and then the particles are removed. This is simply an
inversion of the space, interchanging solid forvoid and vice-versa.
Figure 7.52: A typical section of a porous medium.

Figure 7.53: A simple porous model generation [31}.
LJ
While classical methods propase algorithms for mechanical or computational mo<lelling of void spaces and their mechanical behaviour within a
geometrical void-less model, the present approach has been designed to create a geometric modelling of permeable solid. The model, can then be used
as the input model for numerical experiments such as fluid flow simulations.
Geornetric modelling of permeable materials requires the creation of a
geometric representation of the solid matrix enclosing the interstitial spaces
Mathcmatically, the aggregation can be represente<l iu the form of Equation (7.72)

n
A(x, y, z)
= L :Fi
i=l
(7.74)
244
Discontinuum Mecbanlcs
..-
Chapter 8
Crack Propagation and

Remeshing Techniques
8.1
o
Figure 7.54: Typical porous models [24}.
where :Fi is the mathematical. representation of the objects in the form

of Equation (7. 71). Similarly, the container is modelled as a single object;
C. Therefore, the geometry of the pore.space, 'P, can be defined by subtracting the aggregate expresion for part'icles .A from the expression for the
container,
P(x,y, z)
= C -.A
(7.75)
Figure 7.54 shows typical porous blocks created by this procedure (24].
Finally, the process of generating a porous model can be summarized
.
into three distinct stages:
l. Object representation. Generating the geometric description of the
particles with support for contact detection.

LJ
2. Compaction. crelfng a dense assembly with contact networks to

support the structure.
Introduction
Creation and propagation of eraclcs are frequently encountered in many engineering and industrial problema. Forming a crack is followed by releas.ing
energy and redistributing the forces which ca.used the initiation of the crack.
If this procedure happens immediately after occurrence of a crack~ lt wlll
lead to inaccurate solutions, and more lmportautly, to rnimlts that strongly
depend on the size of the finite elements used in the analysis (14].
To remedy this, strain softening plasticity procedures have been pro- '
posed. In concept, normal hardening plasticity procedures can be applled
to softening materials, by simply introduclng a Boft.ening mther than a hard
ening equivalent stress - equivalent plastic strain relationship. In practlce,
however, there are many difficulties; a umber of which are still unresolved.
The most important one is the mesh dependency of the finite element analysis, in which by continually r~fining the mesh, we can make the crack
propagate at lower and lower loads (107].
Having recognized the serious limitation of stress-based softening failure
criteria, we could move to methods directly involving fracture mechanics.
An alternative procedure, is to introduce a softening stress-strain relation,;_;,
ship which is related to the fracture energy. In the following, both methods
for considering the load transfer in cra.cked regions will be reviewed.
In this chapter, only some of the main computational issues are addressed. Firstly, the phenomenon of mesh dependency is <liRc.nssed. It has
3. Solid space casting. Calculation of the solid complement of the void

space.
245
246
bcen an important rea.son for development of fracture mechanics which will

be briefly explained next. Strain softening method, as a reliable alternative
will be described and the associated damage mechanics techniques Will be
discussed. A simple logical remeshing technique for geometrical mod~lling
of cracks will then be introduced.
8.2
Mesh Dependency
Bazant, among others, has studied the problem of mesh dependency and
has explained many associated diffi.culties that there exist with stress based
failure criteria for purely brittle and quasibrittle materials such as concrete
f132).
Consider a simple tensile plate, discretized by a constant stress triangular

finite element mesh as depicted in Figure 8.1. In this figure, a notch or ftaw
has been shown by the she.ded element. Assuming that the maximum stress
is now in element A, once A has cracked, a brittle model would simply
reduce the sti:ffness of this element to zero.
The concept does also apply to other types of tinitt~ nlements, i.e. Gauss
point cracking in higher order elements. lt is worth noting that there is a
class of problems that does not seiverely su:ffer from these difficulties. In some
analyses, integration points through the depth of beam or shell elements may
be used. In these circumstances, even by degrading the properties to zero
and while there would be a stress concentration in a full 3D or 20 ana.lysis,
no such concentration occurs in relation to the adopted strain assumption
that 'plape sections remain plane'.
8.3
Fracture Mechanics
A classical paper published by Inglis in 1913 {134] ma.rks the fil'.st theoretical description of stress concentration at the tip of a crack. He obtained
the elastic solution for stresses at the vertex of an ellipsoidal cavity in an
infinite solid. He then noticed that as the ellipse approaches a line crack,
the stress at the vertex of the ellipse tends to infinity.
Figure 8.1: Finite element modelling of a tensile ple.te.
247
The extensive need for methods which quantify the effects of the presence of
cracks on material performance has led to the evolution and development of
the theory of fracture mechanics. With various types of cracks being always
present in all structures, to sorne degree, a fundamental requirement for a
fracture mechanics theory is some means of asseHsing Uw sta.bility of such
cracks. In this respect, the most significant advance has been the introduCtion of the stress intensity fa~tor. as a single parameter for categorizilig the.
onset of crack propagation {133).
'--'
Chapter 8. Crack Propagation and Remeshing Techniques
However, because of the fiaw, there is a stress concentration for a more

refined mesh, the load to reach the critica! stress in the first element ( equivalen t to element A) would be reduced. This problem not only a.pplies to
the first cracking, but also applies to the crack propagation. With a refined
mesh, the load for the floo; cracking (after A) in the current most highly
stressed element (namely B) will also be reduced. As aresult, pointed out by
Bazant, by continually refining the mesh, we can make the crack propagate
through the plate at lower and lower loads.
Griffith [135, 136] re-analysing the se.me problem concluded that ln p~

ence of a crack, the stress value can not be used as a criterion of fa.ilure slnce
the stress at the tip of a sharp crd in an elastic continuum is infinite n
matter );\ow small the applied load is (132].
Therefore, he proposed a~ energy criterion of failure, which serves as the
basis of the classica.l linear elastic fracture mechanics (LEFM) and its
eral form of elastic fracture mechanics {EFM). Acf!ording to this criterlon, .,;,,
the crack will propagate if the energy ava.ilable to extend the crack by a unlt, }!i(
surface area equa.ls the energy required. to do so; representing the energy -~~ 1~:\
that must be supplied to break the bonds in the material microstructure.
sen ;_'.(.';: .
The rather academic work of Grlffith, was further <kveloped by US Navy

ship fracture experiments during world war 11, and followed by the discovery
of J-integral. by llice [137] which was th~ missing part of relating the enerl)'
release re.te to the stress and stra.in ficlds close ta the crack tip for any elastlC
material; enabling fracture analysiS for more general nonlinear behaviou....,
by a logical tool.
,
248
Discontlnuum Mecha.nics
The next major achievement was the introduction of the stress intensity
factor K by Irwin in 1957 [138) as a single parameterfor categorizing the
onset of crack propagation and relating the intensity of stresses close to the
crack tip to the energy release rate [133].
.
Categorizing the types of cracking, there exist three pure modes of fracture as depicted in Figure 8.2; Mode /, which is a pure opening mode Mode
JI which represents in-plane shearing mode, and Mode 111 for an~iplane
shearing mode.
Chapter 8. Crack Propagation and
8.4
Remeshing Tedmiques
249
Strain Soft'ening Models
An alternative procedure, pioneered by Hilleborg et al., is to introduce a

softening stress-strain relationship which is relate<l to tl1e fraeture energy.
In this way, fracture mechanics is indirectly introduced (107].
The main concept, borrowed from fracture mechanics, is the assumption that the fracture energy release G, is a material property (fracture
toughness) rather than a local stress-strain curve. The implementation
of the G = cte concept, leads to the importaut conclusion that the local strain-softening law depends on a fracturing zone with characterization
length, le, depending on the finite element mesh. As a result, the fracture
energy concept leads to a non-local format of the equivalent softening relation which is fundanwntally different from the loen.! conHt.it.11t.ivc forma.ts of
strain-softening plasticity and continuous damage mechanics l141 }.
The simple model of Figure 8.3 is used to illustrate these effects. A
softening stress-strain relationship of 8.3a governs the behaviour of ea.ch
constant stress (and strain) eleien of the finite element model (Figure
8.3b).
Model
Modell
Modelll
Fgure 8.2: Graphical r~presentatio.n of three fracture modes.
Mode I has been extensively used in different applications. Adopting

a single mode, will significantly reduce the theoretical and computationa.l
efforts, though its physical applicability has to be closely verified. In a general approach, however, a complex mi.Xed mode fracture, including modes
l, JI, and even lll may be simultaneously activated. A simple criterion tbat
governs crack growth could be presented by a linear interpolation of the
energy release. rates of aH three modes which has proved sufficiently accurate in recent finite element based simulation of delamination and material
fracture dueto low-velocity impact [139, 8].
It should be noted that, since ~his method requires a re-analysis of the
whole model in each step, it is not suitable for an explicit dynamic analysis of
multlple crack problema, and certainly not applica.ble in hlghly progre8sive
fracturing and fragmentation processes [140, 9].
A displacement controlled solution, based on the longitudinal displacement A is adopted (Figure 8.3b). At the maximum load P, the stress in
the element would each have reached the point B in Figure 8.3a, i.e. Ut
Following this stage, a possible equilibrium path in the load/defiection
space is given by the falling curve 1 in Figure 8.3c. For this path, it is
888umed that each of the elements follows the Hoftening Ht.rcss-strain llne
BC.
On the other hand, the falling line 2 in Figure 8.3c ca.u be obtained if
three of the elements softened dGwn BC in Figure 8.3a, while one of them
unloaded elastically down BA.
In relation to the adopted model, the only stable path would be pa.th 4,
for which only one element softened while the others all unloaded elastically. :;}::1,,~
If the mesh was refined, this. ata.ble path would be a line doubling' back;
directly on the elastic loading line.
'
. ~;~' ~
~,.
We consider the latter patb as eta.ble, because if the strength of each

element W8S fractionally perturbed, the model structure would reach its
ma.x:imum load when the weakest of the elements first reached' its t:lt value,
and beyond this level, only the weakest element would soften. Clearly, the
situation in Figure 8.3c involves a form of material bifurcation. For thJs:
sltua.tlon, the curve 4 would be uaoclated wit.h oue 11og1tt.ive plvot ln the
tangent stiffness matrix (associated with the snap-back behaviour} whlle In
'
250
Discontlnuum Mecbanlcs
251
Chapter 8. Crack Propagation and Remeshing Tedmiques
CJ
Fracture lndicator
<l f : Frnrturr EntrllY
a,
l'
Figure 8.4: Fracture energy softening modcl.

a) uniaxial strain softening model.
moving from curves 3 to 2 to 1 we would on each occasion add a. further
negativa pivot for having passed material bifurcn.tion points.
Elements
.
p
' b) simple
bar.
While with a material imperfection, there would strictly bP. only one attainable equilibrium curve, and so resulting in a mesh independent solutio.t:!,
with finite steps, a finite element solution procedure can easily converge on
to a higher unstable state. It is true that these problems are at their most
severa for constant stress states such as depicted in Figure 8.3, nevertheless, equivalent problems can plague finite element solutions even in other
circumstances with a varying stress gradient.
''
One model that provides a simple approach to localization zone simulation is the Rankine softening plasticity model and in this work a bilinear
local soaenlng model is adopted (10] (See Figure 8.4).
The fracture energy relea.se, is de.fined as the integral of the area under
the softening branch of the ~tress-strain curve
e) structural response.
Figure 8.3: Stress localization and alternative equilibrium states.
G =
[~ft (fu -
where ft is the tensile strength and

ultimate strains respectively.
fu
Et)] le
(8.1)
and Et are the tensile fracture and
The introduction of the characteristic length, le, is a result of expressing

the fracture energy of a smeared crack model by a discrete crack model .
(Figure 8.5). The fracture energy G for a discrete crack of width w can be
expressed as
/,
-11
252
253
Ohapter 8. Crack Propagatlon and Remeshlng Techniques
finite element width. Hence, as a close approximation, it may be defined

based on the area A, or the volume of the fractured element, V, (52)
1
1
for 2D
(8.6}
le=
-fwl-
vi
/or 3D
The softening modulus is then defined as
EP =/lle
2G
(8.7)
The position of the stress point on the softening branch, or the value
of the fracture indicator (Figure 8.4), could be used as a measure, being
compared to a predefined maximum value, to quantify the level of material
dama.ge for different regions.
Figure 8.5: Discrete and smeared crack models.
1
1
GJ =
Undw
8.5
(8.2)
while for a smeared cr& model, w is smeared across the width le
UJ = / Undecr
(8.3)
Damage Mechanics
The concept of damage as a load bearing area reduction was first introduced
by Kachanov in 1958 [142]. Since then, the terminology damage mechanica
has been used in many different _ways with the common theme of reductio[l
in the secant stiffness. In the following, a brief outline of the !'~lativelY; '
simple scalar damage is discussed.
Referring once again
to Figure 8.3, lf poiut
F coindded with the orlgln
at point A, a so called elastic dama.ge model is fonued. Otherwise,
where ecr is the equivalent crack strain smeared over le
the .
unloading line DF would define an elastoplastic damage model.

For an elastic damage niodel (Figure 8.6),
u = (1 - d)Eoe = (1 - d)a
(8.8)
Therefore, {8.3), (8.2) with _the help of (8.4) gives,
G =le g
which is equivalent to (8.1) for the adopted strain softening model. In

general, le is contained within one element, however the stress state that
causes the crack in the discrete crack model is inclined to the local axis of
the finite element, and therefore, the characteristic length will not be the
(8.5)
which applies with loading phase (e= fmoa:), ur unloading phase (e< frncl#)
where fma:i: is the maximum strain experienced so far. Eo is the initial
elastic stiffness and d is the scalar damage para.meter,
(8.9)
- 1
254
Cbapter 8. Crack Propagation and Remeshing Techniques
255 .
..
Differentiating Equation (8.8),
a
&
a,
= (1 -
d)Eof. - dEoe
(8.14)
with
Q
Et.
. dd.
d =-f=---
de
a-1 e2
(B.15)
a.e,
(8.16)
Figure 8.6: Simple tensile damage relationship.

Using (8.11) to evaluate ~;, resulta in,
Equation (8.9) can be written as,
u
d
= --.
(t - :!)
a-1
e
~ (t -
a-1
Tt)
r
f or aet > E > Et
= (1 - d)E0 e -
--~ -Et E
a-le
1
1-a:
oe = - - E oe = E tE
(8.17)
(8.10)
u.
wherE.'. a~ auxiliary parameter T has been introduced which can be related

to the energy of the undamaged material: \
= (1 -
d)EoE. - -Q- Tt~OE

oE = Et.
a -1 r 2 2
(S.18)
where Et is the tangent modulus. A stress updn.t:o n.lgorithm bttsod on the

described damage model is given in Box 8.1.
(8.U)
A damage function f can also be introduced,
More or less the same procedure applies for multidimensional scalar

damage, as illustrated in Box 8.~. In this algorithm, no precise form of
the damage relationship between d and T has been defiued. Clea.rly, it
should be related to experimental data. For instance, an exponential form
for the tensile behaviour of concrete, can be defined .as
- -
f =E - max{Eold, Et}~ O
(8.12)
f = r - max{ Toldi rt} :5 O
(8.19)
with,
E< Et
{
E>
Et
~ O:Et
no damage; f < O
material damages
aged material; d = O
(8.13)
where the constant A is chosen based on the fracture energy. Clearly, Equations (8.10) and (8.19) can not be directly used with algorithm given in Box
8.2, because the complete material would then degrade in this tensile soft..;
ening manner. A simple solution can be to define two different damage'.
parameters for the tensile and compressive behaviour.
256
Box 8.1 - Stress update algorithm based on lD scalar dama2e.
= (1 -
7"=
~E
= 'T - maz(Told, Tt)
f =O
= 'T ,
d = a~l (1
(!) + (!I)
-1:")
G1
=d
dold
Higher Modes of Fracture
The simplest mbced mode of fracture is a linear combination of the separate modes,
IF/~OTH~
'Told
8.5.1
There have been enormous contributions to the literature reporting on the

effects of different fracture modes on the crack growth due to various loading
conditions (6, 143, 144, 145, 146, 147]. For example, in delamination analysis
of composites, both fracture energy releeses related to the first mode, G,
and to the second mode, G m contribute considerably to the growth of the
delamination, indicating a mixed mode crack growth.
dozd)Eoe
Et = (1 - d)Eo
e
= (1 -
257
Chapter 8. Crack Propagation and Remeshing Techniq ues
(8.20)
= 1
G 11
oc
d)Eor:.
Et= (1-d)Eo -
--L
(i;.
-E0
2 Eo = a- 1 ~
T V"2
1-a
1
where Gis the energy release rate, and

is the fracture toughness of the
material. The general form of the mixed model, may then be expressed
1
as [148]
:!,
ENDIF
Box 8.2 - Multidimensional scalar damage stress update alp:orithm.
= ( 1 - dozd)DoE = (1 - dotd)iT
O'
Dt = (1 - d)Do
. T
= (~e 7'DoE) t
= r - max( Told, Tt)
IF
j~O
Told
U'
G 11
The linear model is achieved by a = 2, and

mixed model.
= (I - D]
~oE =
T,
= 1
(8.21)
= 4 leads to the qua.dra.tlc
_!!_p]
Eo
l+K
[1 -
THEN
(8.22)
where it and denote stress and strain (or traction and relative displacement) with opening mode 1 and shearing mode 1I, re:;pectively,
d = d~r,rt)
dold
G1
A damage mechanics model [107] is required to establish the evolution

of the stress stra.in relationship. For the purpose of implementation in a
finite element code, a damage model proposed by Mi et al. [14} is aclopted,
which can be written as
T = ( !o-T DQ" 1) !
f
(!)' + (!)'
= d
(8.23)
= (1 - d)DoE
Dt = (1 - d)Do - ~ dd~;Tt) (Doe)(DoE)T
= (1 .
d)Do _ l
'T
dd(-r,Tt) T
d-r
E=
ENDIF
and Eo is a diagonal stiffness matrix,
E
-E
(8.24)
258
Eo = Diag
[~:]
in equation
(8.27)
Attempts to develop a strength theory which is invariant with respect

to the coordinate system, )was initiated by Ashkenazi [153] and Goldenblat
(154}. Subsequent efforts in this direction led to the development of the
. Tsai Wu crlteriorlin 1971 [155].
and the matrix F takes the form of
where ft, ev.. and t are illustrated in Figure 8.4. The scalar
(8.22) is defined as
(8.26)
K
Damage is assumed to occur for K. > O, and the material is considered

not to take any load when the Dii terms exceed unit.
It has been analytically shown that the linear mixed mode formu~ation
is satisfied if
i
lJ
tJ
tJ[
(8.28)
f.11 = pl
which implies a proportional straining. it has been proposed that a be left

as a user parameter, which can take non-integer values between 2 and 4 to
obtain an interaction relationship lying between a linear and a quadra.tic
mixed mode (148].
8.6
259
the mecha.nical response of single crystals and metals. Hill (1950)r'[l49] generalized von Mises' formulation to include anisotropy. A variation of maximum stress theory for unidirectional composites wa.s proposed by Stowell
and Liu [150] in which the failure stress of the fibres was taken as the
limiting .strength of the lamina in the fibre direction, while the limiting
transverse and shear stresses are those of Lhe rr.atrix. In 1965, Hill's theory
was adopted by Azzi and Tsai [151) as a strength criterion for composites.
Hoffman {152], in 1967, altered Hill's criterion to provide for unequal tensile and compressive strengths by introducing linear terrns in the functiona.l
form.
(8.25)
F=Diag[~]
fu - t
Chapter 8. Crack Propagation and Remeshing T'clmiques
Anisotropic Material Models
Anisotropy may be seen in various applications due to physical/geometrlcal

prope~ties of the mate~ial/b.ody. ?omposite materials are the b.est example
for th1s type of behav1our m wh1ch a layered configuration ca.uses difrerent behaviours in and through the layers directions. Another source of
anisotropy is the creation of cracks in which different behaviours are expected along and normal to the crack direction.
Severa! of the anisotropic strength or failure theories are extensions of
isotropic yield criteria. Much of the early workon the strength of anisotropic
materials was conducted independently in the areas of wood technology.a.nd
Development of strength theories is not concluded here, and every year,

a few new theories are proposed. Most of these theories are modifications of
the classical theories. For example, Whitworth [156] has proposed a failure
criterion which is based on the Tsai-Hill theory, however unlike the majority
of other failure theories, takes in~ aecount the influence of the elastic properties on failure. Despite considerable effort that has been devoted to the
formulation of anisotropic strength theories and to their correlation with
test data, no single analysis has emerged which is fully adequate. In other
words, the strength theories have been either insufficient or over complicated, which restricts their use. The analysis by Tsai-Hill and Ho:ffman are
relatively easy to use but are va.lid only under special orthotropy. This ham"."
pers their direct application to the classical analysis of complete lamina.tes,
though there is no restriction in their use in a finite element analysis. The
Tsai-Wu criterion is invariant with respect to the roces of the reference, but
use of this analysis is inconvenienced by the need to evaluate coefficients
under biaxial states of stress.
8.6.1
Tsai-Wu Model
In an effort to more adequately predict experimental results, and inftuenced

by the work of Soviet scientists, Tsai-Wu (1971) proposed a failure criterion
in stress space of the form [155]
(8.29)
where F and Fi3 are the strength parameters. The linear term in O'' talces .
into account interna! stresses which can describe the difference between
positive and negative stress induced failures. The quadratic terms CT'CTJ
260
define an cllipsoid in the stress space. The main features of this criterion
are as follows:
261
are
deflned
where the projection matrix P, and the project.iou vector p
based on the nine material yield strengths ami a. normalized yield strength
; (see Schellekens et al. (159]),
It is a scalar equation and is automatically invariant. Therefore, it is

valid for all coordinate systems.
Since the strength -components a.re expressed in tensora, their transformation relations are well established. So we can readily rotate the
material roces_ once we are attempting to study the off-axis or transformed properties. Most other existing criteria are limited to specially orthotropic materials. These -criteria can only be applied by
transforming the externa! stresses to the material axes. Rotation of
the material roces can not be done because the transformation of the
strength criteria are not known."
Interactions among ali stress -componente are independent material

properties. Therefore, the off-diagonal components of 1':1 are related - "to the interactioJ of two stress components in the strength criterlon.
As a result, sim~ uniaxial or pure shear tests will not be sufficient.
This is _the most inconvenient aspect of this theory, specially while
realizing that relatively small changes in F,; can significantly affect
the predicted strength [157). Lack of accurate material parameters
usually leads to use of simplified versions of this crlterion.
(8.31)
2 (0:31
0:12)
-20:12
P=
-2as1
-20:12
-20:31
-20:23
2 (0:23
0:12)
-2a23
2 (0:31
o
o
o
o
o
+ 0:23)
o
o
o
60:44
o
o
o
o
o
6a55
where
au
= -2
(1
(uc _
11T)
O'UCO'U T
The criterion predicts the imminence of failure but provides no information 'l;l.bout the failure mode.
The study by Gregory [158] has confirmed the discrepancies found by
other researchers regarding this criterion. The inability of this theory and
other similar criteria to predict the correct direction of crack growth is not
d ue to an inaccurate finite element analysis, but due to limitations in the
theories themselves.
o
o
(8.34)
(8.35)
0:33
-2
(1
(liase - iiaar)
-
(8.36)
0'33C0'33T
(8.37)
8.6.2
Hoffman. Anisotropic Yield Criterion
j2
0:55
According to the Hoffman criterion, a. geometrical yield surface is con-
= 32
(8.38)
31
Htructed from three tensile yield strengths ~"T' three compresslve yield
strengths
as:
aiiC,
and three shear yield strengths iis;, i
1= j.
It may be defined
(8.39)
(8.30)
(8.40)
262
Chapter 8. Crack Propagation a.nd Remeshing Techniques
263
{8.41)
1
(8.42)
Only one shear strength is required for implementation in thin shell
elements, and the size of the projection matrix is reduced to 4 (160].
The Tsai..:Hill yield criterion is .a special ca.se of this criterion, in which
the projection vector Poc is reduced to a null vector. The projection matrix
P a is determined by (8.32) under the restriction that O'iiT = O'c.
Integration of the Rate Equation

The key issue of computational pla.sticity modela is often the integration of
the fiow rule in a finite time step. For this purpose, the backward Euler
method coupled with the Newton-Raphson iterative scheme is used; which
has been found to be generally stable and efficient (161, 162].
Figure 8. 7: Stress return algorithm.
For the jth iteration of the ith load step, the additivity postulate is
adopted,
Then, the final stress at the end of the iteration is solved from this
relatiou
(8.43)
(8.47)
Applying the linear ela.stic stress-strai~ telationship and the a.ssociated

flow rule,
(8.48)
which can be re-written in terma of the D matrix
(8.44)
; (8.49)
where A>.J is the plastic multipUer. The derivative of ~ (8 ..30) with respect
to O'j is the flow vector a,
'
8~
a= = P at.Tj + P
8u;
A graphical interpretation of this algorithm is given in Figure 8.7 (163].

Linearization of q> with respect to A.X gives,
(8.45)
(8.50)
Re-writing equation (8.44) with the aid of Cui-1 - Ef~ 1 =O, will eventually result in
(8.46)
_,.-:.~
Hence, the derivative of with respect to AAJ is now needed. For hardening or softening materials, ~is also a function of the hardening/softeni11..g
parameter, K. Therefore,
264
(8.51)
The first part of the first term is the ftow vector a, and equation (8.49)
is used for the second part
Ohapter 8. Crack Propagat1on and Remeshing Techniques
-D(I + A>..;DP0 ) - 1 [P0 D(I + A>..;D)- P 0

( e:: 1 + Ae; - A>..;p0 ) + Pa]
,,,
ffii
(8.59)
(8.60)
1
-p(I + A>.;DPa)- [D(I + A>..;DPar P 0

(e~:l + Ae; "'""A>..;p0 ) + Pa]
(8.52)
(8.61)
To evaluate the second term, an isotropic hardening la.w is adopted[164]
Box 8.3 summarizes the stress return algorithm for the Hoffman yield
surface with isotropic hardening/softening.
(8.53)
Consistent Ta11ger1t Matrix
(8.54)
and from equation (8.30)
Although there is no direct need for ev~.luation of a consistent tangent ni&:

trix within an explicit dynamic analysis, this matrix is derived here for its
potential use in the crack propagation algorithm.
The total strain at the end of i~ration j is given by
8<JJ
-{}jj = -2cr
(8.55)
(8.62)
(8.56)
With the relations for incremental elaatic nnd plast.ic strnins, the streeestrain relationship can be written as
(8.57)
Therefore, from (8.51), one gets
(8.63)
Taking the time derivative of(S.63) results in
(8.58)
{}2<JJ
e= Cu3 +A>.-D'j
3
'
80'J
. 84!
+ ..\3
(8.64).
OCT j
or in terms of the D matrix,
where A = ( aT a) 1 and H is the hardening/softening slope.

I t should be noted that for isotropic materials, D (I + A>..; D P 0 ) - and
P commute in (8.52) and thus the formula.tion becomes much simpler
1
(8.65) .
266
Chapter 8. Crack Propagation and Remashing Teclmiques
267
The last term in {8.65) v~ishes for infinitesimal increments, which results in the classical elastoplastic tangent stiffness matrix. For finite loading
increments, however, the contribution of this term becomes significant. Rearranging {8.65) with the aid of
=a and ~ = P co results in
Box 8.3 - Stress return algorithm for Hoffman surface.
IJ:
Compute trial stress {Elastic prediction phase)

atr
= u i-1
+ D.6.e
.. -
Compute yield function for utr

tr
IF
~ tr
u!1 - 2
sO
(8.30)
(8.67)
{Elastic update)
To evaluate ~, the consistency condition is considered

1
Gj :::: <Ttr
(8.68)
(Elastoplastic updte)
ELSE
lterative loop for Plastic correction phase
n
LJ
where Bis the second term in {8.51),
Compute stress at point of return (e)

O'c
[J + .6.>..DPa]- 1 [atr - .6.,\Dp0 )
{8.4~}
'.
. {8.69)
Compute current yield stress and softening modtilus

H, ac(K) = + e:~ 1 H
Compute current yield function
<Pe
= u~J -
IF 1~c
(8.66)
{c) 2
By premultiplying (8.66) by aT and then substituting into (8.69),

then be solved for
>. may
{8.30)
/ {c) 1~ TOLER
2
(8.70)
<Tj =<Te
ELSE
and when substituting back into (8.66) ylelds,
a=/!;= Pauc +Pa
A= (aTa) i, e~}=
8CT
lM.>..
E {i,~;>.A
from {8.52) or ( 8.61)
FIR$T = aT
\k+l -
where the consistent tangent niatrix, Dce, is givcu hy
lfx
SECOND = 2fi,H
u.Aj
(8.71)
(8.53)
\/e.
~/\i
Next iteration
[A'+ ~aTPaJfJJ] (8.58)

(>e
- FIRST-SEdNb
Dct
=Q
aaTQ ]
[1- B.+aTQa
(8.72)
{850}
.
8. 7
Localization and Crack Direction
ENDIF
ENDIF
The presence of discontinuities in the form of cracks or shear bands pushes

the continuum mechanics framework to
the limits of its validity. Therefore, an indication of failure which somehow quantifies the level of discontinuity is of paramount importance for loclized failure analysis. In thls
near
Discontlnuum Mechahics
268
269
section, the conditions for localized failure are examined on the grounds of
bifurca.tion analysis, following the methodology reported by Pearce [164).
Beginning with the classical elastoplasticity, the rate of stress tensor can
be described by the rate of conjugate strain tensor,
(a)
(b)
(8.73)
w here D
is the material tangent operator. For general non-associated plastic
ftow,
el b i<:>.. a . Del
D = net - D . . 'OI
H +ti: Del: b
(e)
{d)
(8.74)
1
1
111'
where a = ~ is the gradient pf the loading surface, bis the plastic fiow
direction and H is the hardening/softening modulus. In general, D will
not be symmetric, Dijkl :/:- Dklij, although m.inor symmetry is assumed:
Di;kl
D;ikl
!.........- ONIMof
1
llltlp-bock
1
1
bdavlolll'
8
Di;lk
Plastic loading is characterized by a positive value of the plastic multiplier ~:

a: Del:
A = _H_+_a_:_D_e..,...,-:-b
(8.75)
The numerator of (8.75) can be regarded as a. scalar product of an elastically predicted trial stress rate and the normal vector to the loading surface. Therefore, the numerator is always positive, and according to (8.75)
the dominator must also be positive,
H
+a
: Del : b > O
(8.76)
Figure 8.8: Idealized uniaxial stress-strain cuni'es. a,b) general behaviour;

c,d) total brittle behaviour [164).
~
E"= El( .
E+H
(8.78)
!\
Noting that (EP-+ -oo) for (H--. -E).

Recalling Equation (8.7), the softening modulus E" and so Ha.re functions of the fracture toughness G and the characteristic length le which
in turn depends on the element.size. Therefore, Equation (8.77) defines a
constraint on the maximum element size that can be used within a fracture
energy based softening approach.
or
8.7.1
H >-a: Del: b
l_j
introducing a lower bound for the modulus H. In the limit, H

represents totally brittle behaviour.
Loss of Uniqueness
(8.77)
= -a: Dei: b
Figure 8.8 illustrates the same concept for an idealized uniaxial problem,
where the elastoplastic modulus is defined by:
The losa of uniqueness of the stress rate, also called the onset of material
branching or simply the limit point is associated with a null stress rate
tensor,
il'=D:=O
(8.79)
Discontlnuum Mecbanlcs
'270
Equation (8.79) is characterized by a singularity of the tangent modulus

D. Denoting the strain. rate tensor corresponding to the zero eigenvalue of
the singular D by
ecr, we have
D:
ve
ecr =O
(8.80)
The stationary value of ;, , can be related to the elastic strain rate using
'
(8.81)
(8.82)
Chapter 8. Crack Propagatlon and Remeshlng Techniques
271
Equation (8.86} is true only for zero critical hardening/softening modulus corresponding to perfect plasticity (H = O). In other words, when
H = O, the tangent modulus D ~~()!11-eB singular, and for any strai.ning
in the critica! eigendirection equivalent to b the stress rate remams a. ni:ill
tensor. The same concept is valid in associated plasticity, except for the critica! eigendirection which becomes proportional to the normal to the loading
surface a.
It should be noted that H =O represents a necessary but not sufficient
condition for the loss of uniqueness. Under certain conditions such as kinematic constraints it is possible that the critica! eigendirection will not be
activated.
To verify the stability condition, severa! postulates ha.ve been propase(!.
According to the Drucker's stability condition, the external work in
of a cycle of application and removal of the externa! loading, must not be
negative. In Hill's postulate, an equivalent stability condition is defined as:
case
(8.83)
(8.88).
De is positive definite, thus
(8.84)
(8.85)
Therefore, the direction associated to singular D or the critica! eigendirection is proportional to the plastic fiow direction b. From Equations (8.85,
8.80, R74),
De - De : b a : De) : b = O
H+a:De:b
and the critica! values of H may be obtained for which iT : = O. For

associated plasticity (b = a), the critica! hardening/softening modulus for
which stability is lost is zero, Her= O, whereas for non-associated flow, the
loss of stability may occur even with a po~itiv0 value of H. This conclusion
implies that negative second order work can be experienced befare the lose
of uniqueness (164]
O< H <Her
(8.89)
By substituting Equation (8.73) in (8.89), tho onsd of material lnstability can be defined:
(8.86)
E:D:e=O
(8.90)
or in indicia! forro (164],

The un-symmetric part of the tangent operator D can be eliminated due
to the quadratic form of Equation {8.90)
--
(8.87)
D"m:
=O
(8.91)
where Dsym is the symmetric part of the tangent modulus D.
(8.92)
Dlscontlnuum Meche.nics
272
Equation (8.91) states that the eigenvalues of the tangent operator are
bound by those ofits symmetric counterpart. Thus, the loss of material
stability occurs when the symmetrized tangent operator is no longer positive
definite, characterized by at least one zero or negative eigenvalue [164]. For
a symmetric tangent operator, the instability indicator coincides the loss of
positive definiteness of D.
8. 7.2
Chapter 8. Crack Propagation a.nd Remeshing Techniques
Strong Discontinuity
In a strong discontinuity, a jump in the displacement rate is also admitted.
Therefore, preservation of the displacement rate jump, (), along the plane
is required. The directional deriva.tive Qf ((u) in a tangential direction ze
to the plane must vanish:
D())t = (Vu) mt
Discontinuous Bifurcation
The occurrence of loalization can be considered as a. bifurca.tion problem,

whereby the local constitutive description becomes unstable, allowing the
existence of a kinematically acceptable alternative strain field in the form
of a highly deformed material zone. ,., .
In the classical cortlin of localization, a jump (discontinuity) in the
spatial gra<lient of the displacement ratea is assumed in the form of a shear
band. Two types of discontinuity are defined based on the level of displacement jumps:
(Vu) = mn
(e)= 4(mn+nm)
while the
= mn
__
...-
(8.97)
1 (
=2
2m1n1
m2n1 + m1n2
m3n1 + m1 n3
m1n2+ m2n1
2m.2n2
m3n2 +m2n3
(8.98)
Plane Stress Case
(8.93)
To further investigate the problem, a plane stress case, w hcre the disconti
nuity plane normal is aligned with the x 2 axis is considered. Therefore, the
discontinuity normal vector n becomes
u must be continuous along the plane,

(u) =o
(8.96)
or in component form,
()
(Vu)
(8.95)
The jump in the corresponding strain rate is <lefined as the symmet.a:!~

part of (Vit)
Weak Discontinuity
According to the Max:well's compatibility condition, only the normal derivative of u across the plane of discontinuity can be discontinuous
=o
and the kinematic restriction resulted from the Max:well's compatibility condition
Weak discontinuity: where a jump in the displacement rate gradiente

is allowed.
Strong discontinuity.: where a jump in the displacement ra.te gradiente
and the displacement ratea may occur.
273 . t,
(8.94)
where n denotes the orientation of the discontinuity plane (normal), and

m describes the form of the discontinuity (failure modes), i.e. the relative
movement of the material either side of the imminent discontinuity due to
the introduction of the disconthwity. Mathematically, m is the norm~ directional derivative of llitll across the discontinuity. ()denotes the difference
in the quantity on one side of the discontinuity plane to the quantity on the
othm side.
. .
n = (0, l)T
(8.99)
and Equation (8.98) reduces to
(e)=! ( o
2
m1
m1 )
2m2
(8.100}
.,
274
275
Localization Condition
For a strain jump problem, the corresponding discontinuous stress rate eld
can be expressed as,
n
(&)
mlln
= D:
(e)
(8.104)
spUttnode
e+-s-=F-o
a-s-=F-o
i/-u=o
;,+.,;,.,.o
In contrast, the normal traction rate 'n must be continuous due to

equilibrium considerations across the discontinuity plane
(&n) b n -(O-)= n D : () =O
mln
{8.105)
Substitution of (8.97) in (8.105) leads to,
sh1armo
(8.106)
Figure 8. 9: Weak and strong discontinuities for splitting and shear failure
modes [164].
or,
(0-n) =(nDn)m=
Two limiting cases for the vector m are parallel and. perpendicular to
the vector nas illustrated in Figure 8.9,
parahel to n
perpendicular to
m = (0, l)T
= (l,O)T
(8.101}
m.Ln
l_J
() = (
()
= !2 (
oO m2
o )
O
m1
m1 )
A non-trivial solution of the classical localization condition (8.107) can

be derived if the determinant of the localization tensor Q becomes zero,
det Q
= det (n
D n) = O
(8.108)
Q is often referred to as the chara.cteristic tensor or acoustic, following the

analogy to the acoustic wave pro:pagation problem in solids.
Equation (8.100) can then be evaluated for each case,
mlln
(~.107)
(8.102)
As opposed to the conditin for general bifurcation, the discontinuous

bifurcation condition (8.108) depends not only on the state if the material
(D}, but also on the orientat.ion of the discontinuity plane n (164].
The closed form solution of (8.108) may only be derived for two dimensional problema and a simple iterative approach is req11ired to solve this
equation for n.
(8.103)
where mlln represents the splitting localization mode, and m .L n represents the shearing localization mode (see Figure 8.9). A variety of mixed Iocalization mode can be potentially initiated in between these two extremes.
Localization Failure lndicatot. _

To achieve a measure of localize.tion fallure, an indicator may be
based on the definition of acoustic tensor Q,
dened
276
277
F"'
e
Figure 8.1 O: Localiza.Uon failure indicator for different discontinuity orientations.
q.
Fzoc =
detQ
det Qe
(8.109)
where qe .is the acou'stic tensor based on. an elastic tangent operator De,
and :fioc is a localization failure indicator. For :Fioc = 1 the response is
fully elastic, while ':Fioc =O represents the 1 imminent localization. In practice, for different discontinuity directions, the localization failure indicator ..
is evaluated. Figure 8.10 illustrates a typical curve obtained fr 'different
orientation angles. The maximum failure indicator will identify the discontinuity direction.
8.8
Remeshing Algorithm
Material fracture may result in the creation of new discreta bodies which
are in contact by friction interaction with neighbouring bodies. Therefore,
remeshing techniqu~s have to be developed to maintain compatibiiity conditions in newly fractured regions. Here, a simple, though efficient approach
is introduced which can be used for both 2D and 3D applications.
The failure indicator and the crack direction for ea.ch individual element are evaluated from the material model routines. A weighted averaging
scheme is then used to evaluate both the failure indicator and the average
Figure 8.11: Weighted averaging of the failure indicator and the crack direction for a. failed point.
crack direction of each node. Figure 8.11 illustrates this scheme for t.) t-r1 dimensional problem. The same a.pproach may be used for three dimensional
problems with an appropriate weighting. factor for the third dimension.
The next step is to geometrically represent the crack and perform the
necessary split, separation and thet remeshing processes. Figure 8.12 representa the two dimensional remeshing algorithm which comprises four steps:
splitting the element, separating the failed nodes, creating new remeshing
nades, and dividing uncracked elements to enforce compatibility a.t new
nades.
Adopting this local remeshing algorithm will provide a relatively finer
mesh in the fractured region and prevents the distortion of the elements in
this region, improving the finite element approximation of the analysis.
Alternative remeshing techniques include adaptive remeshing algorithme
(as a.n example see Figure 2.6 which illustrates an adaptive shear band sim
ulation). They are a class of e:ffective techniques for improving the finite
element approximation which can be applied to almost any proolem. Unfortunately, they usually require a. large a.mount of computational time and
278
n
LJ
crackdirection~ ..,.
....::::-
node separation
Appendix A
~7
crack
~'"
. :;7
.~ing
UTdem Discrete Element

Code
--
newnode
Figure 8.12: Remeshing technique for ~odelling a crack ata failed point.
A.1
memory every time a remeshing update is undertaken. Due to the continuous changing geometry and behaviour of discontinuum problems, severa!
remeshing updates are required, causing massive computational expenses.
Therefore, using an adaptive approach should be carefully studied and then
Le adopted.
Introduction
In this appendix chapter, the data file structure of a general combinad

finite/discrete element code is explained. This limited version f UTdem
program, gives an overall insight to the user how a combined finite/discrete
element software may be used for modelling progressive fracture analysis of
structures as well as granular ftow simulations.
First, necessary blocks of data are explained and a sample data file is
provided. Several samples have been inch~ded in accompanying CD. For
a more comprehensive data file description manual or accessing the source
files and full unlimited version of the program, please contact the author.
A.2
Data File Structur:e
In this section a brief mamutl for preparlng a U'l'< lnm dnt.a filo iH provlded.
The preprocessor creates this da.ta file based on the information giwn by
the user, and it is not expected to prepare this file by hand. Bowever, in
certain circumstances the user may wish to change parts of the data file for
a new run.
The model is defined as a set of different groups. Each group of elements
may have different element types and material properties. Each block of
data begins by a { and ends by a }. The position of each block is trivial
279
Appendix A. UTdem Discrete Element Code
within the data file. The srune holds for the positlon of each subblock withln
any block. In the following data file description, capital words represent the
beginning of a new block/subblock of data while lower.case words represent
numbers defining the dimension of the corresponding array. Words in capital
letters must remain unchanged. Any irrelevant block may be discarded from
the data file.
4. ELEMENTS : one block for each group
280
The main parts of the data file are:

l. UTdem DATA FILE
'rhis is an interna! control 'Iine for the program and must be kept
exactly the same.
2. GROUPDATA
ngrcode =O
NUMBERS : block specifies the numl?ers associated to each group
ngroups : total number of groups
ACTIVITY: defines the active/inactive state of ea.ch group
(1= active, 2= inactive)

FRACTURING : defines the possibility of fracturing for
each group
(1= no fracture, 2= fracture)
DEFORMABILITY: defines the deformability/rigid state of
each group
(1= deformable, 2= rigid)
3. ELEMENTTYPES : one block for each group

igroup : group number
TYPE : element type number
4= 3 nodedtlane stress
5= 4 noded plane stress

CONNECTIVITY : node numbers of each element
'l
nnodes : number of nodes of each element

ngelements: number of elements of group.
NUMBERS : element numbers
5. THICKNESS : one block for each group

.
THICKNESS : thickness at each node of eiement
ngelements: nmber of elements of group
nnodes : 'number of nodes of each element
ANGLEFLAG: material angle at each node of element
(O=isotopic, l=angle w.r.t. global X, 2=angle w.r.t. local x)
6. NODES
nncode =O
NODES: node numbers
npoints : total number of points
COORDINATES: coordinates of each node
ndim = 2
7. MATERIAL : one block for each material data set
imaterial : material number

ELASTICFLAGS : setting material elastic flags
nef = 4
ELASTICPROPERTIES : defining material elastic properties
nep = 15
'
Ea;, Ey, Ez, V~"' Vyz, Vzll:I Gxy' Gyz1 Gzx1 ax, ay, z, p,u.,P
6= 3 noded plane strain
NSTATES: number of sta.te variables
7= 4 noded plane strain
SURFACEFLAG : surface type flag
8= 3 noded axisymmetric
SURFACEPROPERTIES: surface properties
9= 4 noded axisymmetric
MATERIAL : material number
EOPTIONS: group options
nopts = 50
SURFACE : surface contact specification numbers
nsurfp = 2
281
nsp = 1
PLASTICFLAGS : defining plestic flags

O , plastic type number
npf= 2
PLASTICPROPERTIES : defining plastic properties
npp
=2
282
8. DISCRETEPROPERTIES: discrete element general properties
13. GLOADS : global loads. one set for each load case
DAMPING : contact damping coeffi.cient
igload : load case number
FIELD : size of field zone
POINTLOAD : point loadj_ng_
ZONE : size of buffer zone
ndim
MINSIZE : mnimum size of finite elements
CONTACT : contact type

9. DEBOUNDARJES: discrete element surface assignment
ndecode
283
=2
npsets : number of point load sets

PPOINTERS : pointers of point load sets to corresponding nades
npnodes: number of points subjected to POINTLOAD
nsetl = 2
=O
DEBOUNDARY: discrete element boundary segment lines

nnodseg: number of nodes on each DE segmenta
ndeseg : number of DE segments
DEPOINTERS: pointers to associated DE surface number

10. SURFACE: one block for each surface contact property set
isurface : surface number
DISPLACEMENT : prescribed displacement

ndim = 2
ndsets : number of prescribed displacement sets
DPOINTERS : pointers of prescribed displacement sets
to corresponding nodes
ndnodes : number of points subjected to DISPLACEMENT
nsetl
TYPE: contact type
=2
11
[I
VELOCITY: prescribed velocity
1= impenetrability contact
nvsets : number of prescribe<:} velocity
3::::: bouding law
11
Hd.8
ndim = 2
VPOINTERS : pointers of prescribed velocity sets to
corresponding nodes
nvnodes : number of points subjected to VELOCITY
7= delaminatio'n
PROPERTIES :
ncsp = 4
1
FRICTIONTYPE : ftiction type number

FRICTIONPROPERTIES: friction properties
ncsffp : n umber of friction properties

11. LOADINGS: general loading information
nlcode =O
LOADCASES : load case numbers
nlcases : number of load cases
ACTIVITY : active/inactive flag for each load case
(1= inactive, 2= active)
nsetl = 2
ACCELERATION : prescribed acceleration
na.sets : number of prescribed acceleration sets
ndim
=2
APOINTERS : pointers of pres<;ribed acceleration sets

nanodes : number of points subjected to ACCELERATION
nsetl
=2
14. CONSTRAINTS : defines the constraint information

nccode =O
12. LOAD CURVES : load curve definition. one set for ea.ch load case
DISPLACEMENT : displacement constraints
iload : load case number
ndim = 2
ntpoints : num~f of points defining the load curve
ndsets : number of disp!acement constraint sets
TIME : time points

LOADFACTOR: loading factor at corresponding time point
DPOINTERS : pointers of displacement l:unstrnint sets

rl
11
11
f:
284
STOPTIME: termination time
ndpoints : number of polnts

nsetl
=2
OUTPUTFREQUENCY : time step frequency for output post pro- '

cessing
OUTPUTTIME: time frequency for output postprocessing
VELOCITY : velocity. constraints

ndim
=2
FOUTPUT : time step frequency for text output
nvsets : number of velocity constraint sets
TOUTPUT : time frequency for text output

MESSAGESTEPS : time step frequency for screen message
VPOINTERS: pointers of velocity constraint sets to

corresponding nodes
nvpoints : number of points
nsetl = 2
A.3
ACCELERlrJON: acceleration constraints

ndim = 2
APOINTERS : pointers of acceleration constraint sets

napoints : number of points
ACTIVITY { ngroups .
nsetl = 2
1 1
}
15. HISTORY : history variable output
FRACTURING { ngroups
ihistory : history set number
1 1
}
NODES : node numbers
DEFORMABILITY { ngroups
nhnodes : number of nodes for history output
1 1
}
nvalues : number of requested history varia.ble outputs

GLOBALVALUES: requested general glObal history variables
ELEMEN11'YPES { igroup
TYPE {
4
ngvalues : number.of requested history global variable outputs

STEPS : number of time steps for a new history output
MATERIAL {
1
16. CONTROL : general control specifications

}
TITLE: title of the control block
EOPTIONS { nopts
INITIALTIMESTEP : initial time step
F F F F F F
F F F F F F
F F F F F F
}
SURFACE { nsurfp
1 1
.; TIMESTEPCHANGE: maximum allowed time step chan~e

MINTIMESTEP : Minimum time step
MAXTIMESTEP : Maximum time step
CRJTICALFACTOR : Critica! time step
L.J
Sample
Data
File
.
1
UTdem DATA FILE

GROUPDATA { ngrcode
NUMBERS { ngroups
1 2
'""
nasets : number of acceleration constraint sets
NVAL UES : requ~sted history variables
285
Appendix A. UTdem Discreta Element Code
F
F
F
SOLU'fION : always 1
NUMBEROFTIMESTEPS : maximum number of time steps
ELEMENTTYPES { igroup
F F F F F F
F F F F F F
F F ""F -
F
F
F
F
F
F
F
F
F
F
F
F
Discontinuum Mecbanics
286
287
L_;
TYPE {
}
}
THICKNESS { igroup
THICKNESS { ngelements nnode
MATERIAL {
10*1
2
}
ANGLEFLAG {
EOPTIONS { nopts
F
F
F
F
F
F
F
F
F
F
F F
F
F
}
F F F F
F F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F
F F F
F
F
F
NODES { nncode
NODES { npoints
591
592
601
602
593
603
'
10
594
604
695
605
696
606
597
598
599
600
SURFACE { nsurfp
2
}
ELEMENTS { igroup
CONNECTIVITY { nnodes ngelements
COORDINATES { ndim npoints
601
108
605
-50
-48.7772
3.79429
600
107
NUMBERS { ngelements
1
2
3
4
MATERIAL { imaterial
991 . 992
993
994
995
~96
997
998
10
999 1000
ELASTICFLAGS { nef
o1oo
}
ELASTICPROPERTIES { nep
1000000 O O O O O O O O O OO 1000
ELEMENTS {igroup
CONNECTIVITY { nnodes
141
ngelements
NSTATES {
12
315
222
521
147
SURFACEFLAG {
212
}
}
NUMBERS { ngelements
SURFACEPROPERTIES { nsp
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
0.2
}
}
'
THICKNESS { igroup
THICKNESS { ngelements nnode
3000*1
MATERIAL { imaterial
ELASTICFLAGS { nef
o1 oo
ANGLEFLAG {
ELASTICPROPERTIES { nep
oo
288
10e+5 O O O O O O O O O O O 1000
Appendix A. UTdem Discreta Element Code
oo
SURFACE { isurf ace

TYPE {
PLASTICFLAGS { nef
o8
PROPERTIES { ncsp
-PLASTICPROPERTIES { nep
1000 50
FRICTIONTYPE {
1e+006
NSTATES {
O -10e-10
12
FRICTIONPROPERTIES {.ncsffp
SURFACEFLAG {
0.2
}
1
}
SURFACEPROPERTIES { nsp
0.2
SURFACE { isurface
TYPE {
1
}
}
DISCRETEPROPERTIES {
DAMPING {
PROPERTIES { ncsp
1e+006
O -10e~10
FIELD {
LOADINGS { nlcode
LOADCASES { nlcases
0.2
ZONE {
ACTIVITY{ nlcases
1
}
MINSIZE {
0.2
LOADCURVES { iload
TIME { ntpoints
CONTACT {
o.o
1
}
}
}
LOADFACTOR { ntpoints
DEBOUNDARIES { ndecode
DEBOUNDARY { nnodseg ndeseg
45
1
1.0
44
24
GLOADS { igload
POINTLOAD { ndim npsets
-10.1
DEPOINTERS { ndeseg
2
}
}
18.2
PPOINTERS { npnodes nset1

8
12
13
289
r
L_.;
290
;___)
""
\1
-10.1
18.2
12
1
13
1
14
1
124
2
135
2
VELOCITY { ndim
-0.02 o
0.02 o
2
1
209 210
1
1
1
1
DPOINTERS { nd.nodes nset1

8
2
VPOINTERS { nvpoints nset1
DISPLACEMENT { ndim ndsets
210
2
1
406 407 409

411 '64 56
410
408
403
404
10
1
104
105
106
107
207
8
2
208
}
HISTORY { ihistory
NODES { nhnodes
156
ACCELERATION { ndim nasets
157 45
413 412
NVALUES { nvalues
"x_DISP"
APOINTERS { nanodes nset1
23
1
GLOBALVALUES { ngvalues
"E_KIN" 11 E_EL 11
CONSTRAINTS { nccode
DISPLACEMENT { ndim ndsets
STEPS
o
1
}
{
1
1
CONTROL {
TITLE {
DPOINTERS { ndpoints nset1

1
11
10
Sampledata 11
196
197
198
199
200
201
202
203
204
206
1
2
3
1
1
195
205
L...
}
INITIALTIMESTEP {
0.1
}
TIMESTEPCHANGE {
100
}
MINTIMESTEP {
VELOCITY { ndim nvsets

1
.'-,
20~
APOINTERS { n~point~ nset1
414
1
1
207
1 o
o1
-0.2
107
ACCELERATION { ndim nasets
106
VPOINTERS { nvnodes nsetl

209
104 106
j
l
l
nvsets
291
Discontinuwn Mechanlcs
- --
292
Dlscontinuum Mechanics
MAXTIMESTEP
____;
1
}
References
CRITICALFACTOR {
0.9
}
SOLUTION {
1
} NUMBEROFTIMESTEPS
[1] Petrinic, N., Aspects of Discrete Element ModeUing Involving Facetto-Facet Contact Detection and Jnteraction. Ph.D. thesis, Department
of Civil Engineering, University of Wales Swansea, UK, 1996.
20
}
(2] Grady, J. & Sun\ C., Dynamic delamination crack propagation in a -
STOPTIME {
graphite/~poxy
100
}
FRESULTS {
[3] Roh, Y. & Chang, F.K., Effect ofimpact damage on lambwave prop&gation in laminated composites. Dynamic Response and Behamourof
Composites, eds. C. Sulk B.,,.Sankar & Y. Rajapakse, ASME AD-Vol.
46, pp. 127-138, 1995.
50
}
OUTPUTFREQUENCY {
50
""
MESSAGESTEPS
1
TOUTPUT {
o
}
lamina.te. Oomposite Mnterials: Fatigue and Fractu~
ASTM STP 907, ed. H. Hahn, 1986.
(4] Abrate, S., Impact resista.nce of composite materials - a review. Applied Mechanics, 44, pp. 155-190, 1991.
(5) Agarwal, B. & Broutma.n, L., Analysis and Performance of Fiber

Composites. John Wiley and Sons, Inc., 1990.
[6} Kinloch, A., Wang, Y., Williams, J. & Yayla, P., The mixed-mode delamination of fiber composite materials. Composte Scence and Technology, 47, pp. 225-237, 1993.
[7J Matthews, F. & Rawlings, R., Oomposite Materials: Engineering and

Science. Chapman a.nd Hall, 1994.
[SJ Razi, H. & Kobayashi, -.A., Dela.mination in cross-ply laminated composite subjected to .low-velocity impact. AIAA Journal, 31(8), pp.
1498-1502, 1993.
[9] Ortiz, M., Finite element analysis of impact darnage and ballistic penetration. Fourth US National Congress on G'omputational Mechanics,
USNCCM IV, ed. M. Shephard, pp. 10-13, 1997. San Francisco, USA., .
{lOJ Munjiza, A., Owen, D. & Bicanlc, N., A combined finite-discrete e~

ment method in transient dynamics of fracturing sods. Engintiering
Computations, 12, pp. 145-174, 1995.
293
..
- . . :~-~it,1y~--- '
W4
lteferences
{22] Chandler, N., Dixon, D., Gray, M., Hara, K., Courant, A. & Tillerson,
(11] Bicanic, N., Munjiza, A., Owen, D. & Petrinic, N., From continua to
?isc?~tinua. - a ~ombined finite element / discrete element modelling
m c1v1l engmeermg. Developments in Computational Techniques for
Structural Engineering, ed. B. Topping, Civil-Comp Press pp. 1-13
1995.
'
'
J., The tunnel sealing. experiment: An in situ demonstration of technologies for vault sealing. 19th Annual Conference of the Canadian Nuclear Society, pp. 4o-44, 1998. Toronto, Canada.
[23] Hanks, T. & Kanamori, H., A moment magnitude scale. Journal of

Geophysical Research, 84, pp. 2348-2350, 1979.
(12] Mohammadi, S., Foroozan-sepehr, S. & Asadollahi, A., Contact based
(24] O'Connor, R., A Distribut~ ]Jiscrete Element Modelling Environment
delamination and fracture analysis of composites. Thin Walled Structures, 40, pp. 595-609, 2002.
(
- Algorithms, Jmplementation and Applications. Ph.D. thesis, Department of Civil and Environmental Engineering, MIT., 1996.
[13] Forouzan-sepehr, S. & Mohammadi, S., A contact based method for

[25] Shi, G., Discontir.uous de!ormation analysis, a new nurr:erica! model
for the stqtics and dynamics f>/ block systems. Ph.D. thes1s, Umversity
of California, Berkeley, 1988.
3D delamination analysis of composites subjected to impact loading..

Computational Mechanics: New Frontiers for the New Millenium, eds.
S. Va~iapan & N. Khalili, PP: .?91-696, 2001. Sydney, Australia.
[14] Mi,
y,
[26] MacLaughlin, M., Sitar, N., Doolin, D. & Abbot, T., Investigation of
Crisfield, M. & Davie~, G., Failure modelling in composite

Computational Mechanics in UK - 5th ACME Conference
Crisfield, pp. 36-39, 1997. London, UK.
'
slope stability kinematics using discontinuous deformation analysis.

Jntemational Joumal o/ Rock Mechanics f3 Mining Sciences, 4, PP
1-10, 2001.
struc~ures.
ed.
M.
i !
[15] Zhang, X., Hitchings, D. & Davies, G., Modelling composite failure
[27] Lak, M., Adaptive Shear Band.Slope Stability Analysis. Master's the. sis, Department of Civil Engineeri~g, University of Tehran, 2002.
and delamination during low velocity impact. Computational Mechanics in UK - 5th ACME Conference, ed. M. Crisfield, pp. 44-47, 1997.
London, UK.
[28] Stead, D., Eberhardt, E., Coggan, J. & Benko, B., Advanced numerical techniques in rock slope stability analysis - applications and
limitations. Landslides - causes, impacts and countermeasures, PP
615-624, 2001. Davos, Switzerland.
[16] Hunter, P., FEM/BEM notes. Lecture Notes, Department of Engi.

neering Sciences, The University of Auckland, 2002.
1
[29] Cramer, H., Findeiss, R., Steinl, G. & ~u~1derlich: W~, An
[17] Kirn, J., Mixed mode fracture 'analysis via coupling the finite ele-
P;., ~
proach to the adaptive finite element analys1s m associated and nonassociated plasticity considering localization phenom:ena. Oomputer
Methods in Applied Mechanics and Engineering, 176, pp. 187-202,
ment method and the symmetric galerkin method. Presentation Notes,

Department of Civil and Environmental Engineering, Functionally
Graded Materials Group, 2001.
1999.
[18J M11njiza, A., Discr'Cte Elements in Tmnsient Dynamics o/ Practured

Media. Ph.D. thesis, Department of Civil Engineering, University of
Wales Swansea, 1993.
[30] Masuya, H. & Kajikawa, Y., Numerical analysis of the colllsion b&.
[19] Malagaris, G., Finite/discrete elements in blasting and mining applications. Master's thesis, Un~versity of Wales Swansea, 1994.
[31] Williams, J. & O'Connor, R., aeometric modelling of porous ~,.
[20] Hazzard, J., Young, R., Pettit, W.
tween a falling rock and a cushion by distinct element method. Oo"t.

puter Methods and Advances in Geomechanics, pp. 493-498, 1991;;.~&~;-,
' ','
&
[32]
[21.J Martin, C., Strength o/ Massive Lac du Bonnet granite around un-
.:':
'.1 ~--~
using the discrete element method. Symposium on the Mechanica .Of

Materials with Discontinuities and Heterogeneities, eds. A. Misr&; ~t
C. Chang, ASME Joint Applied Mechanics and Materials Corm
'
1995. UCLA-,usA.
~ ~~'.<~
Collins, D., Examining unstable shear fracture using dynamic numerical modela ~d IJlicroscopic
monitoring. Technical report, University of Liverpool,':2002.
derground openings. Ph.D. thesis, University of Manitoba, Winnipeg,

1990.
.
-!
A
Williams, J. & Rege, N., Coherent structures in deforming gran: J" w~

materials. International Jouma.l o/ Mechanics of Cohesive-.FHctionol, _
Materials, 1996.
296
Discontinuum Mechanics'
[33) Petrinic, N., Owen, D., Munjiza, A. & Bicanic, N., Rolli~g resistance
of disks in contact. Extended Abstracts for the 3rd ACME UK Conference, pp. 105-110, 1995. Oxford, UK.
[34] Vaz-Jr., M., Comm.tational approaches to simulation of metal cutting
process. Ph.D. thesis, Department of Civil Engineering, University of
Wales Swansea, 1998.
[35] Adibi-asl, R., Adaptive Simulation of Machining Procesa. Master's

thesis, Department of Mechanical Engineering, University of Tehran
2001.
'
[36} Schnurr, N ., Thermal initiation caused by fragment impact on cased
297
References
[45] Ostadhossein, H., Combined Finite/Discrete Element Modelling of

M asonry Structures. Master's thesis, Department of Civil Engineering, University of Tehran, 2001.
[46] Rockfield Software Ltd., ELFEN helps save our historie structures.
Newsletter 4, 200L
[47] Mak, K.W., Some aspects o/ combined discrete/finite element simulation of shot peening problems. Master's thesis, University of Wales
Swansea, 1998.
[48] Connel, M., Virtual reality simulation for finite/discrete element problema. Ph.D. thesis, University of Wales Swansea, 1998.
1
Computational Techniques for Contact, Impact, Penetration and Perforation of Solids, eds: L. Schwer, N. Salamon & W. Liu,
ASME AMD-Vol. 103, 1989.
(49] Macedo, J., ln'begrated approach to the parallel processing of explicit

finite/discrete problema. Ph~D. thesis, University of Wales Swansea,
[37J Mohammadi, S., Owen, D. & Peric, D., A combined finite/discrete
(50] Brunetto, M. & Baria, G., A variational approach to discrete element
e~plosives.
1997.
element algorithm for delamination a.nalysis of composites. Finite Elements in Analyss and Design, 28, pp. 321-3~6, 1.998.
modelling in rock mechanics. 1st US Oonference on Discrete Element

Methods, eds. G. Mustoe, M. Henriksen& H. Huttelmaier, CSMPress,
1989.
.
[38] Forouzan-sepehr, S., Analysis of lnitiation and Propagation o/ Interlarninar Cracks in Composite Shells Subjected to lmpact Loading. Master 's thesis, Department of Civil Engineering, University of Tehran,
[51} Zienkiewicz, O. & Taylor, R., The Finite Element Method. McGraw
2001.
[39] Mohammadi, S., Combined Finite/Discrete Element Analysis of Impact Loading of Composite Shells. Ph.D. thesis, Department of Civil
Engineering, University of1Wales Swatlsea, 1998.
[40) Worswick, M., Strazincky, P. & Majeed, O., Dynamic fracture of fiber
reinforced composite coupons. Dynamic Response and Behaviour of
Composites, eds. C. Sun, B. Sankar & Y. Rajapakse, ASME AD-Vol.
'1G, pp. W-'11, 1n!J5.
[41] Price, D., Discrete element modelling of masonry walls. Master's thesis, University of Wales
~ansea,
1997.
[42] Yu, J., A contact interaction framework for numerical simulation of

rnulti-body problems and aspects of damage and fracture for brittle
rnaterials. Ph.D. thesis, University of Wales Swansea, 1999.
Hill, 4th edition, 1994.
(52] Crook, A., Combined finite/discrete element-method. Lecture Notes,

Department of Civil Engineering, University of Wales Swansea, 1996.
[53] Belytschko, T. & Neal, M., Contact-lmpact by the. pinball algorithm with penalty, projectlon, and lagrangian methods. Oomputational Techniques /or Oontact, Impact, Penetration and Per/oration
of Solids, eds. L. Schwer, N. Salamon & W. Liu, ASME AMD-VOL.

103, 1989.
[54] Hallquist, J., Goudreau, G.
& Benson,
D., Sliding interfaces with
contact-impact in large-scale lagrangian computations. Computer

Methods in Applied Mechanics and Engineering, 51, 1985.
[55) Cundall, P. & Hart, R., Numerical modelling of discontinua. 1st US
Conference on Discrete Element Methods, eds. G. Mustoe, M. Henriksen & H. Huttelmaier, CSM Press, 1989.
[56] Peric, D. & Owen, D., Computational model for 3-D contact problems
[43] Klerck, P., Fnite element modelling of discrete fracture in quasi-brittle

materials. Ph.D. thesis, University of Wales Swa.nsea, 2000.
with friction based on the penalty method. Intemational J oumal For

Numerical Methods in Engineering, 35, pp. 1289-1309, 1992.
[44] Marefat, M., Ghahremani-Gargari, E. & Naderi, R., Full scale test of
[57] Schonauer, M., Rodic, T. & Owen, D., Numerical modelling

ther
momechanical processes related to bulk forming operntions. Joumal
De Physique IV, 3, pp. 1199-1209, 1993.
a 60-yea.r old ma.si;-coucrete arch bridge. World Congress bn Railway

Research, pp. 114-118, 2001. Koln, Germany.
of
1
1
~.:
;
!'.
;.!':'.
!
1
;
~
11
'
298
[58] Munjiza, A., Bicanic, N., Owen, D. & Ren, Z., The central difference
time integration scheme in contact-impact problems. 4th Jntemaonal
Conference on Nonlinear Engineering Computations - NEC-91, eds.
N. Bicanic, P. Marovic, D. Owen, V. Jovic& A. Mihanovic, pp. 569575, 1991.
[59] Haghshenas, S., 3D Progressive Fracture Analysis of Composite Shells.

Master's thesis, Department of' Civil Engineering, University of
Tehran, 2002.
[60] Yu, J., Report on discreta element 3D contact analysis algorithm implementation. Interna! report, Rockfield Software Ltd., Swansea,
UK, 1996.
[61] Lubliner, J., Plasticity Theory. Macmillan Publishing Company, 1990.

- ........
[G2] Bonet, .J. & Wood, R., Nonlinear connuum mechanics for finlte element analysis .. Lecture notes, University of Wales Swansea, 1995.
[63] Belytschko, T., An overview of semidiscretization and time integration procedures. Computational Methods for Transient Analysis, eds.
T. Belytschko & T. Hughes, Elsevier Science Publishers, chapter 1,
pp. 1-65, 1983.
[64] Bathe, K., Finite Eleme4.Procedures. Prentice Hall, 1996.
[65] Peric, D., Owen, D. & Honnor, M., Simulation of thin sheet metal
forming processes employing a thin shell element. VDI BERIOHTE,
894, pp. 569-586, 1991.
(06] Green, A. & Naghdi, P., A general theory of an elastic-plastic continuum. Archive for Rational Mechanics Analysis, 18, pp. 251-281,
1965.
(67] Belytschko, T. & Leviathan, l., Physical stabilization oNhe 4-node
shell element with one point quadrature. Computer Methods in Applied Mechanics and Engineering, 113, pp. 321-350, 1994.
[68] Nagtegaal, J., On the implementation of inelastic constitutive equations with special reference to large deformation problema. Computer
Methods in Applied Mechanics and Engineering, 33, pp. 469-484,
1982.
(69] Peric, D., Owen, D. & Honnor, M., A model for finite strain elastoplasticity ha.sed on logarithmic strains: Computational issues. Computer Methods in Appled Mechanics and Engineering, 94, pp. 35-61,
1992.
[70] Lee, E., ElaHtic-plostic deforma.tion at finite strains. Joumal o/ Applied Mechanics - ASME, 36, pp. 1-6, 1969.
299
References
[71] Nemat-Nasser, S., On finite defor~ation elasto-plasticity. International Joumal of Solids and Structures, 18(10), pp. 857-872, 1982.
[72} Hashimoto, K., Neto, E.S., Peric, D. & Owen, D., A study on dynamic
frictional behaviour of coated steel sheets: experiments, forrnulation
and finite element simulations. Technical report, U niversity College. of
Swansea, 1993.
(73] Mohammadi, S., Owen, D. & Peric, D., Delamination analysis of composites by discrete element method. Computational Plasticity, COMPLAS V, eds. D. Owen, E. Onate & E. Hinton, pp. 1206-1213, 1997.
Barcelona, Spain.
'l'
'
[74] Mohammadi, S", Owen, D. & Peric, D., Discontinuum approach for
damage analysis of composites. Computational M echanics in UK - 5th
ACME Con/erence, ed. M. Crisfield, pp. 40-44, 1997. London, UK.
[75] Hughes, T., The Finite Element Method. Prentice-Hall International,
Inc., 1987.
[76] Kole, J., Figge, M. & Raedt, H.D., Unconditionally atable algorithms
to solve the time dependent maxwell equation. Phys Rev E, 64, pp.
66-75, 1999.
[77] Cook, R., Malkus, D. & Plesha, M., Goncepts and Applications of
Finite Element Analysis. John Wiley & Sons, 3rd edition, 1Yts9.
(78] Hinton, E., Rock, T. & Zienkiewicz, O.,. A note on mass lumping and
related processes in the finite element method. Earthrwike Engineer
ing and Structural Dynamics, 4, pp. 245-27U, 1!>76.
[79] Bicanic, N. & Maca, J., An improved mass lumping technique by
rnatching element spectral characteristics. Technical report, University College of Swansea, 1991.,
[80) Wilkins, M., Calculation of elastic-plastic flow. Methods in computational physics, ed. B. Alder, Academic Press, volume Vol. 3, pp.
211-263, 1964.
[81] Maenchen, G. & Sack, S., The TENSOR code. Methods in computational physics, ed. B. Alder, Academic Press, volume Vol. 3, pp.
211-263, 1964.
[82] Bertholf, L. & Benzley, S., TOODY II, a cornputer program for two
dimensional wave propagation. Sc-rr-68-41, Sandia N ational Laboratories, USA, 1968.
[83] Circa, A., Determining the normal stlffness of joint

32-41, 1980.
'
fu(~H.
ACME, pp.
300
References
301
[84) Schonauer, M., Unified numerical analysia of cold and hot metal forrning processes. Ph.D. thesis, University of Wales Swansea., 1993.
(98) Wrlggers, P., On consistent tangent matrice8 for frictlonal contacfi

problema. Proceedings o/ NUMETA '81, p. C15, 1987. Swansea., UK.
[85} Oden, J. & Martins, J., Modela a.nd computa.tional methods for dy-
[99) Peric, D. & Owen, D., A model for 3-D contact problema with frlction. Proceedings o/ the European Conference on New Advances in
Computational Structural Mechanics, 1991. Giens, France.
namic friction phenomena. Computer Methods in Applied Mechanics

and Engineering, 52, pp. 527-634, 1985.
(86] Curnier, A., A theory of friction .. lnternational Journal of Solids and

Structures, 20, pp. 637-647, 1984.
[87] Fredriksson, B., Finite element solution of surfa.ce nonllnearities in
structural mechanics with special emphasis to contact a.nd fracture
mechanics problems. Computers & Structures, 6, pp. 281-290, 1976.
._.,
(88] Michalowski, R. & Mroz, Z., Aasociated and non-associa.ted sliding

rules in contact friction problem&~ Archive Mechanics, 30, pp. 259276, 1978.
'
[89] Kalker, J., Mathematica.l modela of friction for contact problema in

elasticity. Wear, 113, pp. 61-77, 1986.
(100) Simo, J. & Taylor, R., Conaistent tangent operators for rate indepen-.
dent elastoplasticity. Computer Methods in Applied Mechanics antl
Engineeri,ng, 48, pp. 101-118, 1985.
[101) Simo, J. & Hughes, T., General return mapping algorithms for rate
independent pla.stici~y. Constitutive Laws for Engineering Materials:
Theory and Applications, ed. C.D. et al., pp. 221-231, 1987.
(102) Mitchell, G. & Owen, D., Numerical solutions for elastic-plastic pro~
lems. Engineering Compu.tations, 5, pp. 277-284, 1988.
(103) Marsden, J. & Hughes, T., Mathematical F'oundations of Elasticity.

Prentice-Hall, Englewood Cliffs, New Jersey, 1983.
(90] Back, N., Burdekin, M. & Cowley, A., Review of the research on fixed
and sliding joints. Proceedings o/ the 19th Intemational MTDR Conference, eds. s. Tobias & F. Koenigsberger, pp. 87-97,: 1973. London,
UK.
[91] Connoly, R. & Thornley, R., Deterclining the normal stiffness of joint
faces. ACME, pp. 67-6, 1967.
[92] Wriggers, P., Van, T.V. & Stein,
[104] Suh, N. & Sin, H.C., GenesiS of friction. Wear, 69, pp. 91-114, 1981.
[105] Ortiz, M. & Simo, J., An a.nalysis of a new class of integration algorithms for elastoplastic conatitutive relations. Intemational Joumal
jor Numerical Methods in Engineering, 23, pp. 353--366, 1986.
(106] Crisfield, M., Non-linear Finite Element Analysis of Solids and Struc-.
E.,
Finite element formulation of

large deformation impact-contact p~oblems with friction .. Comuters &
Structures, 37, pp. 319-331, 1990.
(93} Kragelsky, I., Dobychin, M. & Kombalov, V., Priction and Wear Calculation Methods. Pergamon Press, Oxford, 1982.
[94] Avitzur, B. & Zhu, Y., A friction model base<i on the upper bound
approach to the ridge and sublayer deformations - update. Proceedings
oj the 19th North American Metalworking Res. Conference, 1-IESL
Publication Edition, pp. 103-109, 1985. Berkeley, USA.
[95] Kobayashi, S., Oh, S. & Altan, T., Metal Forming and the Finite
Element Method. Oxford University Press, NewYork, 1989.
!
nonline~
[96] Wriggers, P. & Simo, J., A note on tangent stiffnesa for fully
contact problems. Communications in Applied Numerical Methods, 1,
pp. 199-203, 1985.
[97] Parish, H., A consistent ta.ngent stiffness ma.trix for three dimensional
nonlinear contact analysis. Intemational Joumal for Numerical Methods in Engineering, 28, pp. 1803-1812, 1989.
tures. John Wiley & Sona Ltd., 199i.
[107) Crisfield, M., Non-linear Finite Element Analysis of Solids and Structures. Volume 2: Advanced Topics. John Wiley & Sons Ltd., 1997.
[108] Goodman, R., Taylor, R. & Brekke, T., A model for mechanics of
jointed rock. Joumal of Solid Mechanics and Foundation, ASCE,
SM3, p. 94, 1968.
(109] Cundall, P., A computer model for slmulat.ing progrn:;::ve, large scale
movements in blocky rock system~ Proceedings of Intemational Syrn,-:
positt.m on Rock Structures, 1971. Na.ncy, France.
[110] Ghaboussi, J., Fully deformable discrete element analysis using a finite
element approach. Computefi"s nd Geotechnics, 5, pp. -175-195, 1988.
[111] Foster, N. & Metaxas, D., Visuallzation of dynamic fluid simula.tions:
waves, splashing, vorticity, boundaries, buoyancy. Engineering. Computations, 12, pp. 109-124, 1995.
[112] Hustrulid, A. & Hall, B., Parallel implementation of the discrete element method. Technical report, Colorado School of Mines, 1996.
302
(113] Xian, L., Bicanic, N., Owen, D. & Munjiza, A., Rock blasting simulation by rigid body dynamic analysis and rigid-brittle fracturing
model. 4th Intemational Conference on Nonlinear Engineering Computations - NEC-91, eds. N. Bicanic, P. Marovic, D. Owen, V. Jovic
& A. Mihanovic, Pineridge Press, pp. 587-597, 1991.
References
303
[125] Wensel, O., Search algorithm for detecting geometric overlapping in a

discrete element context. Master's thesis, University ofWales Swansea,
1992.
[114] Hakuno, M. & Yamamoto, T., Simulation analysis of dynamic non-
[126] Bonet, J. & Peraire, J., An alternating digital tree (ADT) algorithm
far 3d geometric searching and intersection problems. International
Journalfor Numerical Methods in Engineering, 31, pp. 1-17, 1991.
linear behaviour of underground. structures by the extended distinct

element method. Computer Met~ods and Advances in Geomechanics,
pp. 481-486, 1991.
[127) Pande, G., Beer, G. & Williams, J., Numerical Methods in Rock Mechanics. John Wiley and Sons Ltd., 1990.
[115] Dutton, A. & Meek, J., Distinct element modelling of longwall gate
road roof support. Computer Methods and Advances in Geomechanics,
pp. 1757-1764, 1991.
.. "'
[128] Thomas, P. & Bray, J., Capturing nonspherical shape of granular

media with disk cluster~. J oumal o/ geotechnical and geoenvironmental
engineering, 125(3), pp. 16~178, 1999.
[116] Sawada, S., Pradhan, T. & Iwas.ki, Y., Simple shear simulations of
sand using d.e.m. Computer M ethods and Advances in Geomechanics,
pp. 511-516, 1991.
{129] Holst, J., Rotter, J., Ooi, J. & Rong,'G., Numerical modelling of silo
filling, ii: discrete element analysis. Journal o/ engineering mechanics,
125(1), pp. 104-110, 1999..
[117] Lin, X. & Ng, T., Contact detection algorithms for three-dimensional
(130) Lin, X. & Ng, T., A three-dimensional discrete element model using
arrays of ellipsoids. Geotechnique, 47(2), pp. 319-329, 1995.
ellipsoids in discrete element modelling. Intemational Journal for Numerical and Analytical Methods in Geomechanics - Short Communication, 19, pp. 653-659, 1995.
[118] Ting, J., Meachum, L. & Rowell, J., Effect of particle shape on the
strength and deformati&" mechanisrns of ellipse-sha.ped granular assemblages. Engineering Computations, 12, pp. 99-108, 1995.
[119) Morikawa, H. & Sawamatp, Y., Local, fracture analysis of a reinforced

concrete slab by discrete element method. Proceedings o/ the .Second
lntemational Conference on Discrete Element Modelling (DEM), eds.
J. Williams & G. Mustoe, 1-IESL Publication Edition, pp. 275-286,
1993. MIT, USA.
[120) Canny, J., The complexity o/ robot motion. planning. fh.D. thesis,
MIT, ACM Doctoral Dissertation Award, 1987.
[121] Moore, M. & Wilhelms, J., Collision detection and response for computer animation. ACM Comput~r Graphics, 22(4), pp. 289-297, 1988.
...
[122] Herzen, B., Barr, A. & Zatz, H., Geometric collision for time dependent parametric surfaces. ACM SIGGRAPH 90, 24, pp. 39-48, 1990.
{131] Williams, J., Contact analysis of large numbers of interacting bodies using discrete modal methods for simulating material failure on
the microscopic scale. Engineering Computations, 5(3), pp. 150-161,
1988.
[132] Bazant, Z. & Planas, J., .Fracture an4 Size l!J/Ject in Concrete .nd
Other Quasibrittle Materials. CRC Press, 1998.
[133) Owen, D. & Fawkes, A., Engineering Fracture Mechanics. Pineridge
Press, 1983.
(134] Inglis, C., Stresses in a plate due to the presence of cracks and sharp
cornera. Journal o/ Institutio'h o/ Naval Architects, 55, pp. 219-~41,
1913.
[135] Griffith, A., The phenomena of rupture and fiow in solids. Joumal o/
Philos Royal Society Architects, 221, pp. 163--197, 1921.
(136) Griffith, A., The theory of rupture. The first intemational conference
o/ applied mechanics, pp. 55-63, 1924.
[123] Vatti, B., A generic solution to polygon clipping. Communications of

the ACM, 35(7), pp. 56-63, 1992.
[137] Rice, J., A path independent integral and the approximate analysis
of strain concentration by notches and cracks. Joumal o/ Applied Mechanics - ASME, 35, pp. 379-386, 1968.
[124] Pentland, A., Computational complexity versus simulated environments. ACM SIGGRAPH Computer Graphics, 24(2), pp. 185-192,
1991.
.
[138] Irwin, C., Analysis of stresses and strains near the end of a crack
traversing a plate. Joumal of Applied Mechanics - ASME, 24, pp.
361-364, 1957.
304
305
[139] Liu, S., Quasi-impact damage initiation and growth of thick-section
[152] Hoffman, O., The brittle strength of orthotropic materials. Joumal of

Composite Materials, 1, pp. 200-206, 1967.
L___;
and toughened composite materials. Intemational Journal of Solids

and Structures, 31, pp. 3079-3098, 1994.
[153] Ashkenazi, E.,. On the problem of strength anisotropy of construction

materials. Soviet Physics - Technical Physics, 4(3), pp. 333-338, 1959.
[140] Camacho, G. & Ortiz, M., Adaptive lagrangian modelling of ballistic

penetration of metallic targets. Computer M ethods in Applied Mechanics and Engineering, 142, pp. 269-301, 1997.
English translation.
(154] Goldenblat, l. & Kopnov, V., Strength of glass-reinforced plastics in

the complex stress state. Polymer Mechanics, 1(2}, pp. 54-59, 1965.
[141] Hinton, E., Numerical Methoda and Software for Dynamic Analysis
of Plates and Shells. Pineridge Press, 1987.
[155) Tsai, S. & Wu, E., A general theory of strength for anisotropic materials. Composite Materials, 5, 1971.
[142] Kachanov, M., Time of rupture process under creep conditions. Izv
Akad Nauk SSR, 8, pp. 26-31, 1958. In Russian.
[156] Whitworth, H. & Yin, 18.W., A strength theory for failure of composite laminates. Failure Analysis: Techniques and Applications, ed.
J. Dickson, 1991.
[143] Banks-Sills, L. & Schur, D., On the influence of crack plane orientation
in fatigue crack propagation and catBBtrophic failure. Fracture Mechanics: Perspectives and Directions, ASTM STP 1020, eds. R. Wei
& R. Gangloff, pp. 497-513, 1989.
[157) Rowlands, R., Strength (failure) theories and their experimental correlation. Handbook o/ Composites, Vol. 9 - Failure Mechanics of Composites, eds. G. Sih &? A. Skudra, Elsevier Science Publishers B.V.,
chapter2, pp. 71-125, 1985.
[144] Russell, A. & Street, K., On the development of a mixed mode tensileshear interlaminar fracture criterion for graphite-epoxy composites.
Modelling Problems in Crack Tip Mechanics, ed. J. Pindera, Martin
Nijhoff Publishers, 1983.
[145] Wang, A., Fracture analysis of interlaminar cracking. Interlaminar
(158] Gregory, M. & Herakovich, C., Predicting crack growth directlon in

unidirectional composites. Composite Materials, 20{1), pp. 67-85,
1986.
Response of Composite Materials, ed. N. Pagano, Elsevier Science

Publishers B.V., chapter 2, pp. 69-149, 1989.
[159) Schellekens, J. & de Borst, R., The use of Hoffman yield criterion
[146] Whitney, J., Experimental characterization of delamination fracture.

Interlaminar Response o/Composite\ Materials, ed. N. Pagano, Elsevier Science Publishers B.V., chapter 4, pp. 161-250, 1989.
[147) O'Brien, T., Analysis of local delaminations and their influence on
j
'
[149] Hill, R., The Mathema1cal Theory

Press: London, 1950.
o/ Plasticity.
Oxford University
(150] Stowell, E. & Liu, T., On the. mechanical behaviour of fiber-reinforced
Strnctures, 37(6), pp. 1087-1096, 1990:
[160) Koh, C., Owen, D. & Peric, D., Explicit dynamic analysis of elasto-
'j
plBBtic laminated composite shells: implementation of non-iterative

stress update schemes for the Hoffman yield critedon. Computational
Mechanics, 16, pp. 307-314, 1995.
[161j Peric, D., On a claBs of coqstitutive equations in viscoplasticity: Formulation and computational issues. lnternational Joumal o/ Numerical Methods in Engineering, 36, pp. 1365-1393, 1993.
composite laminate behaviour. Delamination and Debonding of Materials, ASTM STP 876, ed. W. Johnson, pp. 282-297, 1985.
[148] Mi, Y. & Crisfield, M., Analytical derivation of load/displacement

relationship for the DCB and MMB and proof of the FEA formulation. Ic-aero report 97-02 issn 0308-7247, Aeronautics Department,
Imperial College of Science, Technology and Medicine, 1996.
in finite element analyses of anisotroplc composites. Computers and
[162] Simo, J. & Taylor, R., A return mapping algorithm for plane stress
elastoplasticity. Intemational Journal of Numerical Methods in Engineering, 22, pp. 649-670, 1986~
[163] Schellekens, J., Computational Strate,gies for Composite Structures.
Ph.D. thesis, Delft University of Technology, 1992.
crystalline materials. Joumal o/ Mechanical Physics of Solida, 9, pp.
242-260, 1961.
[151] Azzi, V. & Tsai, S., Anisotropic strength of composites. Experimental

Mechanics, 5(9), pp. 283-288, 1965.
[164] Pearce, C., Computational Plasticity in Concrete Failure Mechanics.

Ph.D. thesis, Department of Civil Engineering, University of Wales
Swansea, 1996.
307
lndex
DYNA, 94, 99, 103, 104, 107, 108
Index
acoustic, 275, 276
adaptive, 182, 210, 277, 278
ADT, 187, 212, 213, 215
alternating digital tree, 185, 187,
212, 213, 217
anisotropic, 8, 78, 133, 182, 258260
associated flow rule, 134, 262
associated slip rule, 134, 160, 161
augmented Lagrangian, 35, 43, 5254,57, 94
'-_J
...
~\
BEM, 6, 7
binary tree, 196-200, 202, 207, 215,
218
bisection, 204-207, 215
boundary element method, 6
bounding box, 213, 218
buffer zone, 190-194, 282 ., ,
constraint, 8, 35-37, 40-54, 5660, 82, 93, 98, 105, 113,

115, 126, 129, 137, 147,
. 150, 160, 237, 269, 271,
283, 284
contact damping, 35, 64, 282
contact detection, 5, 8, 174, 185,
186, 192, 203, 207, 222,
232,
233, 236, 2~2, 244
,,
contact instability, 60
Coulomb, 29, 121, 130, 131, 146,
155, 168-170, 178
crack, 3-6, 8, 22, 24, 26, p8, 182,
185, 245-248, 251, 252,
257, 258, 260, 265, 267'
276-278
damage, 3, .8, 22, 24, 182, 246,
249, 253-257
damping, 2, 12, 62-65, 88-91, 225,
228-230, 232
~nnA, 16
defending node method, 65
DEM,4, 6
direct checking, 187, 189
direct evidence, 190, 191
discontinuity, 181, 272, 273, 275,
276
..
discontinuum, 1, 4, 8, 11, 16, 181,
183, 184, 278
discrete element, 4-6, 9, 12, 20,
21, 181-183, 192, 222
discretization, 6, 40, 48, 50, 69,
85,87,106,147, 152, 153,
221
distinct element method, 181
circular disk, 221, 222, 224-226,

230, 232
combined finite/discrete eleinent,
5, 8, 14, 22, 29, 30, 182,
279
composite, 2-5, 24, 26, 182, 183,
257-259
consistent tangent, 8, 94, 87-89,
108, 113, 115, 117, 118,
135, 142, 146, 151, 152,
164, 169, 174, 176, 177,
265, 267
constitutive, 77, 78, 122, 135, 139,
142, 146, 148-150, 153,
160, 162, 163, 233, 249,
272
306
EFM, 247
elastoplastic, 5, 122, 126, 130, 148,
150, 153, 159, 253, 266268
ellipse, 183, 221, 232-238, 240, 247,
260
Euler, 70, 71, 73, 79, 143, 144,
150, 163, 262
')'
failure, 3, 13-15, 275-277

fracture, 3-5, 8, 22,' 26, 182, 245,
247' 249, 251, 253, 255,
257' 269,. 279
fragmentation, 3, 4, 12, 13, 21, 24,
26, 181, 182, 248
frictional conta.ct, 8, 29,.121, 122,
126, 135, 142, .147, 148,
150, 153, 155, 159, 160,
166-168, 170, 171
frictionless, 42, 128, 170
geometric intersection, 185, 199,
212, 213, 215, 220, 221
geometric search, 207, 212, 213,
215
'
global search, 108, 110, 184, 188,

191, 213, 221
granular, 1, 2, 9, 11, 20, 119, 181183, 233, 222, 233, 226, t
279
impenetrability, 16, 21, 36, 42, 44,

58-60, 65, 93, 129, ~60,
282
instability, 17, 60, 62, 64, 90, 108,
114, 271, 272
interface, 6, 22, 24, 27, 49, 93;
126, 129, 134, 159, 184
1
ti
Lagrange, 35,47-54, 57,237

Lagrangian, 35, 43, 47, 50-54, 57,
70, 71, 73, 94, 184
LEFM, 247
left link, 196, 199, 201
linearization, 85, 134, 138, 142,
144, 263
local search, 108, 110, 115, 184,
192
localization, 22, 250, 251, 267, 272,
274-276
master, 93, 95-108, 110, 123-126,
135, 136, 139, 140, 147,
168, 170, 172
mesh dependency, 245, 246
mixed mode, 248, 257, 258
Newton, 144, 174, 262
non-associated fiow, 271
non-associated formulation, 8
non-associu.ted plastic flow, 268
non-associated slip rule, 134, 135;
149, 150, 160, 161, 163,
178
non-consistent, 108
hardening, 8, 121, 126, 132, 135,

147-149, 155, 156, 159,
163-165,245,263-265,268,
271
heapsort, 210, 211
HEMP, 95
hexagon, 207, 209
hextree, 207-209
Hoffnian,259, 260, 265, 266
hydrocode, 94, 95, 99, 104
hypercube, 215, 217, 218
object oriented, 182, 221

object representation, 8, 21, 185,
221, 222, 242, 244
parallel processing, 34
particulate fiow, 11, 32
. permeable, 18, 242, 243
perturbed Lagrangian method, 35,
43,50, 51, 57
porous, 18-20, 183, 242-244
spring, 39, 45, 46, 112, 224-226,

228, 232
progressive, 3, 6, 9, 11, 12, 21, 22,

114, 183, 248
propagate, 3,4,22,90, 245-247
propagation, 4, 8, 30, 33, 245, 275
stability, 16, 35, 60, 63, 64, 90,

181, 247, 271, 272
stack, 201, 202
state variable, 8, 78, 122, 148, 159,
quadratic, 113, 134, 139, 151, 164,

236, 257~259, 271
Rankine, 251
recursive algorithm, 202
refinement, 22
remeshing, 8, 16, 17, 22, 245, U6,
276-278
,,
retrieval, 199, 200, 209
right link, 196, 199, 201, 202
rigid, 1, 2, 61, 75, 78, 98, 123, 125,
126, 131, 139, 147, 156,
181, 184, 187-189, 221,
222, 230, 232, 280
rolling, 158, 226, 229, 230, 232,
233
root, 196, 198, 200, 201, 204, 215,
218
281
stick, 126, 127-130, 133, 142; 143,
159, 160, 163, 165,176

stiffness, 41, 44-46, 53, 54, 104,
108, 112, 113, 115, 118,
120, 128, 134, 138, 139,
142, 151, 153, 154, 162,
" .., .
shear, 16, 17, 22, 23, 120, 248,

257, 259, 260, 262, 267,
272, 274
short list, 186, 187, 190, 193, 207,
212, 218
sil~
1, 2,
2~
181, 182
slave, 93, 95-108, 110, 123-125,

135-139, 142, 147, 160,
170
slideline, 93-97, 108, 109
sliding, 93, 105, 122, 129, 133, 155157, 163, 168, 178, 226,
229, 230, 233
slip, 14, 125-127, 130, 132-135,
139, 142, 143, 145, 146,
148-152,159-161,163,165,
168, 178, 223
softening, 8, 126, 132, 133, 148,
155, 156, 163, 245, 246,
249-253, 255, 263-266, 268,.
269, 271
163, 167, 179, 225, 228,

230, 246, 249, 253, 257,
267
strong discontinuity, 4, 272-274

subtree, 200, 201, 221
supercomputer, 1
superquad,21, 183,221,239-241,
243
TENSOR, 95, 98, 103
TOODY, 95, 96, 98, 102
transition interface, 6
translation, 82, 233-236
traversal, 198, 199, 201, 202, 215
traverse, 200, 201, 211, 215, 218
traversing, 29, 198, 200, 201, 218
trial function, 40 .
uncoupled, 39-41, 89
uncracked, 277
undamaged, 254
uniqueness, 269, 271
UTdem, 9, 279, 280, 285
variation, 41, 56, 57, 59, 76, 82,
83,87,115, 118,125,129,
137-139, 142, 153, 173, 1
174,259
virtual reality, 34
viscous damping, 92, 224-228
visualization, 184, 185
weak discontinuity, 272
WITPRESS
Structuras Under Shock
and lmpact VII
Editors: N. IONES, University o/
Liverpool, UK, C.A. BREBBIA, Wessex
lnstitute o/ Technology, UK, and
A.M. RAJENDRAN, US Army Research
Laboratory, USA
The shock and impact behaviour of structures
is an extremely challenging area. It is
therefore important to ..recognize and utilize
the contributions emerging from theoretical,
numerical and experimental studies on
structures, as well as investigations into
material properties under dynamic loading
conditions.
Reflecting the broad range of research work
currently being carried out in academic
institutions and industry, this book contains
the papers presented at the Seventh
Intemational Conference on Structures Under
Shock and Impact. The topics highlighted
are: Protection of Structures from Blast
Loads; Surface Structures Under Explosive
Shock; Missile Penetration and Bxplosion;
Interaction Between Analytical and
Experimental Research:
Structural
Crashworthiness and New Software for
Shock and Impact; Bnergy Absorbing
Systems; Seismic Engineering Applications;
Material Characterization; Material Response
to High Rate Loading; Behaviour of Concrete
Structural Components; Behaviour of Steel
and Composite Structures; Fluid Solid
Interaction; and Analysis of Unbound
Domains.
Editor: M.H. AUABADI, Queen Mary

College, University of London, UK
Bringing together recent research on
fracture of rocks, this volume covers a wide
range of subjects, including hydraulic
fracturing, blasting and fragmentation,
transport problems and creep.
Ali of the chapters have been conttibuted
by leading scientists in the field and foous
on modelling and analysis techniques.
Topics covered inc!ude: Boundary Blement
Analysis for Rock Fracture;
The
Displacement Discontinuity MethOd for the
Analysis of Rock Structures - a Fracture
Mechanics Approach; Computationally
Efficient Models for the Growth of Large
Fracture Systems; Transport of Colloids in
Saturated Fractures.
Series:
Advances in Fracture Mechanics, Vol 5
ISBN: 1-85312-542-3 1999

HSS.OO/US$248.00/232.SO
440pp
We are now able to supply yuu with details of new

WIT Prcss titles via
E-Mail. To subscribe to this free service, or for

infonnation on any of our titles, please contaot the
Marketing Department, WIT Press, Ashurst Lodge,
Ashurst, Southampton, S040 7AA, UK
Tel: +44 (O) 238 029 3223
Fax: +44 (O) 238 029 2853
Series:
Structures and Materials, Vol 11
ISBN: 1853129119 2002

U89 .OO/US$294.00/283.50
Fracture of Rock
E-mail: marketing@witpress.com
612pp
L._
Advances in Dynamics '

and lmpact Mechanics
Editors: C.A. BREBBIA,
We~sex
Institute
of Technology, UK and G. NURICK,
Thermomechanical
Fatigue and Fractura
Computational Methods in Movine Boundaries VII

Computational ModeUing of Fre.s
Contact Mechanics VI
Editor: M.H. AUABADJ, Queen Mary

College, University of London, UK
&litor: C.A. BREBBIA, Wessex lnstitute of

Technology, UK
Thermal and mechanical fatigue leads to

initiation and growth. of cracks in many
structures such as pressure vessels, nuclear
reactors, high temperatute engines,
electronic packaging and denW materials.
This book brings together up-to-date
contributions from leading specialists.
Contents:
Thermomechanical Crack
Growth using Boundary Elements; Finite
Blement Modelling of Thermomechanical
Fracture - Theory and Case Studies;
Prediction of Crack lnitiation and Growth
under Thermal Fatigue; Predicting Crack
Growth Along Polymer Interfaces Due to
Water. Attack and Thernial Fatigue; The
Saturation of Thermomechanical Fatigue
Damage in Brittle Materials; Plasticity
lnduced Heating in the Fracture and Cutting
of Metals; Tensile Ductility and Fracture of
Aluminum-SiC Composites under Changing
Temperature Conditions; Bffects of Residual
Stress on Thermal Fatigue .Phenomena of
Aluminium Die Casting Dies.
Modern engineering design has led to the

realization of the importance of contact
problems in many fields of technology.
lncluding discussions of mechanical models,
numerical
aspects,
experimental
measurements and engineering applications,
this volume features the probeedings of the
Sixth Internatibnal Conference on
Computational Methods and Experimental
Measurements in Contact Mechanics.
Particular emphasis is placed on the
appli<~ation of advanced theories, while the
contributors have also been encouraged to
critically review existing ideas and to explore
new rese:arch methods.
The. paprs presented are. grouped into three
different sections: Computational Metbods;
Experimental Results and Computational
Mechanics; and Fracture, Fatigue and Wear.
University of Cape Town, South Africa

This book contains invited chapters by
internationally renown researchers in
<lynamics and impaCt mechanics .and covers
a wide range of topics including both
experimental and theoretical studies.
Inspired by and dedicated to Professor
Norman Jones of the University of
Livcrpool, UK, the volume acknowl*dBes
his outstanding contribution to the field.
Contents: Tribute to Profes sor Norman
Jones; On the Dynamic Collapse of
Circular and Square Tubes Under Axial
Impact; On Analysis of Axisymmetric
Crushing of Tubes and Frusta; Crushing
Strength of Ship Structures; On the Side
lmpact of Hollow Composite Rail Body
Shells; Multibody Systems Dynamics
Methodologies for Crashworthiness; On the
Numerical Simulatioq of the Progressive
Collapse of Polymer.Composite Sandwich
Beams Under Impact Loading; .Simplified
Analysis of Square Plates Under Bxplosion
Loading;
Ballistic Penetration and
Perforation of Steel Plates: An Experimental
and Numerical lnvestigation; Dynamic
Behaviour of Large-Scale Structures Due to
a Low Speed lmpact; Dynamic Response
and Failure of Free-Free Beams Under
lmpact or Impulsive Loading; Failure
Behaviour of Prestressed Concrete Beams
Under High-Speed and lmpact Loadings.
Series:
. lmpact and Damage on Structures, Vol 1
ISBN: 1-85312-928-3
2002
!107 .OO/US$165.00/f.l 60.50
304pp
Series:
Advances in Fracture Mechanics, Vol 7
ISBN: 1-85312-549-0 2002
U 15.00/US$173.00/172.50
Series:
Computational and Experimental
Methods, Vol 8
ISBN: 1-85312-963-1
2003
!89.00/US$142.00/133.SO
232pp
344pp
Find us at
~
Ali prices correcl al time of going to press hui
subject lo change.
WIT Press books are available through your
booksel/er or direcl from the publisher.
Save 10% when you order from our

encrypted ordering service on the web
using your credit card.
and Moving Boundary Problema

Editors: A.A. MAMMOLJ, University of
New Mexico, USA and C.A. BREBBIA,
Wesse:X lnstitute o.f Tethnofogy, UK
Containing papers presented at th,e Seventh
Intemational Conference on Compubttional
Modelling of Free and Moving Boundary
Problems, this book demonstrates the rapid
growth of work in the are a by scientists and
engineers in both academia and industry.
The intcrdisciplinary nature of the problema
is shown in thc diversity of the pbysical
models and numerical methods descrlbed and
the papers featured consider such areas as:
Phase Interface Phenomena; Instabilities;
Chemically Reacting Flow; Flow Through
Porous Media;
Sedimentation and
Infiltration; Cavitational Flow; Free Surface
Flow; Soli<l-Liquid Phase Change; Uquid
Vapour Phase Change; Fracture and Contact
Problems; Inverse Problems; Finite and
Boundary Element Methods; Spectral
Methods~ Seawater Intrusion; Latent Heat
Storage; Casting, Welding, Solidification
Processing; and Optimisation.
Series:
l:omputational and Experimental Methods.
Vol JO
ISBN: 1-85312-987-9
2003
apx 300pp
apx '9UO/US$153.00/161.12
WITPress
Ashu.nt Lodge, Ashurst, Southampton,

S040 7AA, UK.
Tel: 44 (O) 238 029 3223
Fax: 44 (O) 238 029 2853
E-Mali: witpress@witpress.com
,.

Discontinuum Mechanics Using Finite and Discrete Elements

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

Discontinuum Mechanics Using Finite and Discrete Elements

Uploaded by

Copyright:

Available Formats

'I

WIT Press publishes \eaing hooks in Scic111.:1.~ am\ lh\Jnology.

.... ....: . : ~ '

Department of Civil Engineering

Fariflav Engineering Research Company

Discontinuum Mechanics, Why? .

Scope of the Book

For USA, Canada and Mexico

Computational Mechanics lnc

Discontinuum Mechanics - A Ueview

Failure of Cut Embankni.ent .

DDA Slope Stability Analysis

Shear Band Slope Stability

/\ Catalogue record for this book is av ai la ble

WIT Press 2003.

Granular Flow in Silos .

Printed in Great Britain by Athenaeum Press, Gateshead.

Impact Analysis (Progressive Fracturiug)

Generalized Strain Tensors

Objective Stress Rates ..

Constraint. Enforcing Methods

3.2 Definition of a Constraint

l 'cq1.111hnd l111~;r1u1~l1111 rvlc1t.l111d

Augmented Lagrangian Method.

The Penalty Method

Finite Element Discretization

Explicit ,Time Integration . .

Central Difference Method.

Luroped Mass Matrix

Stable Time Step .

t.. ~ llydrntuth H.1111111latluH

DYNA Penalty Forroulation ..

Search For the Master Segroent

Finding the Contact Point .

Large Defonnation Kinematics

2D Node to Edge Penalty Contact

Local Search Algorithm .

Normal Contact Interaction . . . .

Com;istent Node to Edge Contact . .

Frictional Contact - The Basics

Discontinuum Contact Mechanics

7.2 Contact Detection

Frictional Contact Model

Global Search Algorithms

Plasticity Theory of Friction

7.2.4 .Contact (Field) Zone .

2D Frictional Contact Models Without Hardening

Residual Force Vector

Binary Tree Structures .

Alternating Digital Tree (AD'I') .

3D J:i!ictionaJ Canta.et Models Without Hardening

Pla.':it.icity Theory of Friction

6.5 3D Frictional Contact Modela With

Ellipse Shaped 'Particle~

Geometric Models for Permeah!P Soli1 Is

Physica9A.spects of Frictional Phenomena .

Numerical Simulation of 3D Work Hardening

Numerical Integration Scheme

Consistent Tangent Modulus .

Node to Face J:irictional Contact Models

Strain Softening Models

2D No'de to Edge Model . . . . .

3D Nade to Quadrilateral Face Model

Higher Morles of Fracture