Discontinuum Mechanics Using Finite and Discrete Elements - WITrress ‘WIT Press publishes eadingboks in Sc nd Tey. "Vision webs fr new ed caren Is of ties ‘em ipres om WiTetibrary Making the latest research accesible the WIT elctoniclibrary leurs papers presented at Wessex Institte of Technology’s prestigious intmational conferences. “Toaccess the ibrary and view abstracts fee of charge please visit wiv Wwitpress com os TR AME ee, ssa c ca ca Discontinuum Mechanics Using Finite and Discrete Elements S. Mohammadi University of Tehran, Iran WIT PRESS southampton, noston we0 rn u fn Ly cs Ga 9 € a ‘Author: S. Mohammadi 3 Deparment of Ci Engiering ; nies of ean in| en Fan en “= ° Farifav Engineering Research Company Teva ran Published by WIT Press Asurst Lage, Ashurst, Southampton, SO80 TAA, UK Tel: 44 (0) 238029 3223; Fax: 44 (0)238 029 2853 ‘E-Mail: witpressawitpress.com ‘tpi witpress.com For USA, Canada and Mexico Computational Mechanics Ine 25 Bridge Stet, Bilevia, MA 1821, USA Tel: 978 667 S841; Fax: 978 667 7582 E-Mail: infouso@witpress com tipo. itpess.com ied Livery Cataloging ablation Dat ’ Ctalogue record for this books avallble ‘om the Beis Library ISBN: 1-85312-959.3, Library of Congress Catalog Card Number: 2002116429 [No responsibil is assumed by the Publisher, the Elitors and Authors for any injury andlor damage o persons or property a8 a matter of products liability, negligence oF otherwise, or flo any use or operation of any methods, products, instrotions or ideas ‘contained in the material herein WIT Press 2003, Printed in Gret Brin by Athenaeum Press, Gateshead, _Allighs reserved. No pat of ths publication may be eproduced, stored ina erie sytem rise, without the prior written permission ofthe Publshe oni. 0 Chapter 2 transmitted in any frm orby any means, electonic, mechanical photocopying, Contents Chapter 1 Introduction 14 Discontinuum Mechanics, Why? . 1.2. Alternative Approaches 1.3 Scope of the Book . . 1.4 Notations 21 Geomechanical Applications 24.1 Rock Blasting... . 2.2 Mining, Rock Burst. 2.41.8 Failure of Cut Embankment 2.14 DDA Slope Stability Analysis 2.1.5 Shear Band Slope Stability 2.1.6 Falling Objects 24.7 Porous Media 7. ©. 22. Granular Materials . 22.1 Granular Flow in Silos 2.2.2 Superquadric Elements 23. Impact Analysis (Progressive Fracturins) 2.3.1 Penetration of a Missile Discontinuum Mechanics ~ A Review a 2 .2 18 “4 16 16 ra 18 a a25 28.2 Metal Cutting 23.8 Demolition of Structures, 234 Composites 23.5 Masonry Structures. 236 Repair Modelings . Particulate Plow 24.1 Shockwave Analy Othor Applications . . Chapter 3 Constraint Enforeing Methods 34 32 33 Sa 35 Introduction Definition of a Constraint. Constraint Enforcement 3.3.1 Impeneteability Constraint 3.3.2 Penalty Method 33.3 Lagrangian Multiplier SL Pertuabed bayeanglan Method 3.5 Augmented Lagrangian Method Variational Forms... The Penalty Method 3.5.1 Contact Instal lity Chapter 4 Nonlinear FE Discretization 4a 42 Introduction 5 Large Deformation Kinematics 424 The Motion 4.22 Deformation Gradient. 4.23 Polar Decomposition 4a 4 6 a) 32 34 wu 42.4 Generalized Strain Tensors 42.5. Rate of Deformation 426 Stress Measures... + 42.7 Objective Stress Rates. « 42.8 Multiplicative Deeompesition 43. Initial/Bowndary Value Problem. 438. Kauiibrivm Bquation 43:2 Variational Form 44 Finite Element Disretization « . 4.5, Explicit Time Integration « 45. Central Difference Method 45:2 Lumped Mass Matrix 45. Stablo Time Step «++ « Chapter 5 Normal Contact 5. Introd bd Mydrocorde Mormulatlon oes ees Algorithms 52.2 DYNA 2D Formulation . . 5.3 DYNA Penalty Formulation . .. . 5.3.1 Search For the Master Segment . 5:22 Finding the Contact Point 833° Contact Bore. s— - 5A 2D Node to Edge Penalty Contact 541 Contict Geometry... 542. Global Search Algorithm. 549. Local Search Algorithm 88-8 8.3.3.8 <9” 0 zaChapter 6 61 62 63 64 65 66 5.44 Normal Contact Interaction . . 5.4.5 Consistent Node to Exige Contact alg to Hy Contact. : 4 Frictional Contact Introduction Frictional Contact - The Basies . .. . a 6.21 Frictional Contact Model 62.2 Basic Kinematics 6.2.8 Plasticity Theory of Friction 2D Frictional Contact Models Without Hardening. 6.3.1 Residual Force Vector 8.2 Stishwess Contrib stent Linewzation . 38 Gon SD Frictional Conthet- Models Without Hardening . 64.1 Preliminaries 64.2 Plasticity Theory of Friction 6.4.3 Numerical Integration 6.4.4 FE Diseretization D Frictional Contact Models With Tietional Hardening. Miers 65. PhysicdAspects of Fictional Phenomena 65.2 Numerical Simulation of 83D Work Hardening Frictional Contact 65.3 Numerical Integration Scheme... 654 Consistent Tangent Modulus. . Node to Face Fictional Contact Models 65.1 2D Node to Bdge Model 66.2. SD Node to Quadrilateral Face Model ui <5 118 an 121 121 121 12 126 = 135 187 138 2 7 ur ur 150 152 155 185 150 163 164 168 166 170 Chapter 7 ma 72 738 Chapter 8 a1 82 83 84 85 Discontinuum Contact Mechanics Introduction TAA Historie! Development 7.1.2 Applications 73 Specifications . Contact Detection 7241 Contact Geometry 72.2 Global Search Algorithms . 723° Buller Zone. 724 Contact (Field) Zone... - 7.2.5 Binary Tree Structures . 7.2.6 Spurn Devompesition. . . 72.7 Alternating Digital Tree (ADT) Object Representation . 731 Circular Disks... 782 Disk Clusters . 7.3.3 Ellipse Shaped Particles 7.34 Superquadric Objects . . 785 Geometric Models for Permeable § Remeshing Techniques Introdvetion ‘Mesh Dependency Fracte Mechanics Strain Softening Models . Damage Mechanics . .. . . 8.5.1 Higher Modes of Fracture 181 +181 245, - 287ca oa eo a ca cs 56 Anisotropic Material Models 86.1 Tsai-Wu Model 86.2 Hoffman Anisotropic Yield Criterion... 87 Localization and Crack Direction . . 827.1 Loss of Uniqueness 87.2 Discontinuous Bifurcation . 88 Remeshing Algorithm Appendix A UTDem code AA Introduction A2 Data File Structure AS Supple Data File Preface Lam always obliged to a person who has ight ve « single word? ‘The finite clement method has become « meture power approach, for ‘numerical simulation of various engineering and industrial applications, With repid development ofthe computing fies, the analy of largo: scale ten! dscontinnu problems has now become realty. ‘A month after I returned to Tran, Twas asked to arrange for 8 new state of the art course for the PhD. students studying structural mechanics. My answer was ‘Computational Contact Mechanics. After three years, a comprehensive book has no became available, covering the discrete clement. methodology and principlox of discontinuury, mechanics. Tt is mainly prepared aa a postgeadunte textbook and fF) research institutions undergoing research on general contact mechanics fand discontinuum probleme, Here, T should mention my studenta in the Univetnty of liran, whom frequent questi Kol aw bo prepare HR: Arle wilh, Lops, hapa enone ‘The book can be classified into three parts. The first part is dedicated to the basic concepts of constraint definitions atv! various methods for constraint enforcement: aa well as & short «ivenssion on the finite element discretization and necessary finite deformation theory as the bases of the ‘combined finite/discrete olament method. ‘The second part i devoted to the contict interction and detection procedures. Physical and experimental observations, normal and tangential contact formulations, »theocetical considerations and computational algorthins will be discuss tor bes :ronnlar, low and deformable fnite/diserete element modeling. ‘The final part describes the progressive fracturing phenomena. Stability and convergence discussions, an anisotropic material model,ea ¢3 3 2: a eS € ea Ioealization difficulties, remeshing teclmiques and their asocinted computational algorithune will be discussed, A limited version of the UTdem discrete element code has been inched in the accompanying CD. It is primarily aimed at giving the vader an overall insight of how complex discrete element. software performs and how various control parameters may affect the progressive cracking and granular flow simulations. A sample manual for preparitig ‘the input date file has been included as an appendix I would like to sratefully apprecinte my colleague Mr. A. Najafi Amin for his outstanding ‘work on pre and postprocessing modulus of the UTdem. ‘The present textbook is a reault of infinite’ research works for many years all over the world, I have tried to appropriately acknowledge the Ichievements of corresponding researchers within the text, relevant Figures, tables and formulae, Tam much indebted to their ontstanding. cee Hwatha, om ony alioebeoning, in vulfevuttyarkaendenging Cen be acenely eegrttest First of all, I would like to express my sincere gratitude to Prof. D.R.J. Owen who was the first to-open my eyes beyond the continuum harriers, introducing ne to new challenging frontiers of the computational reionce. Also to Dr. AL. Crook, my first discrete element tencher and to Prof. D, Perie and the late Prof. E. Hinton for valuable discussions. To this end, I have used and bullt 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 T got involved with this new subject; I gratefully acknowledge and appresiate them, 1 wl het extent ay ts eat Dr. SS. Yat nd tote Farinkav Buginesring Research Company for the software support. The technical support of the Computer Centre, Department of Civil Dnginooring of University of Tehran for facilitating the preparation of the manuscript is also neknowledged. Finally, amy special thanks to my family for their never ending support, Uhave spent many hours preparing this text; hours that had to be spont for my wife and little ‘sogol’. Without their understanding and Jove this book may not ever have been written S)’Mohammadi Tehran To: Mansoureh & Sogoles Cc ea Chapter 1 Introduction 1.1 Discontinuum Mechanics, Why? ‘A sand hourglos has for centuries bean a shu ple ronutyen Hi mnwlng tool as well as a relatively accurate reliable measure fur scientific expert iments, Nevertheless, until very recently, even nt the age of supercomputers fand efficient numerical methods such aa the finite element method, it has rot been possible to accurately model this simple phenomenon. A highly varying geometry, and interacting objects (sauls) were the main obstacles to any efficient approach (see Figure 1.1). ‘An interesting set of problems which have recently utteacted spectal st tention, includes the general behaviour of granular materials. In this class of problems, large number of interacting bodies, usually simple rigid elements, are interacting in a domain which will govern (he: geseral response of the ‘medium through these individual interactions. ‘ie best example, may be the filling or emptying a silo with/from granular materials as depicted in Figure 1.2 {1 Figure 1.1: Sand flow in @ typical hourglass. walia co es co 3 Discontinuum Mechanics Figure 1.2; Filling aslo with granular material [1] ‘The behaviour of any individual element is usually governed by the clas- seal rigid body dynanvies, although a general finite element formulation may bie ania Uo convider deformable response for then, However, to include the interuction of the bodies into the analysis, conventional continuum methods ‘ean not be effectively used (if at all). Additionally, more complex phenom- ena such as dynamic filling/emptying, arching behaviour, shockwaves due to sudden closure, thermal effects, and lateral seismic behaviour were almost untouchable. This lack of success was not only limited to that simple case; almost anywhere in the Industry and academic world, several applications could have been found that analysts ceased to be able to accurately simulate. One of the major deficiencies was in the field of new advanced materials bicing subjected to dynamic and hazardous loadings. Gradually replacing onventional materials, advanced materials such as new metallic alloys, composite laminates, ceramics and plastics are now widely used in many applications involving dynamie loading such as ma- chinery, pressure vessels, defence structures, vehicles, sports equipment and, notably acrospace structures. Industries such as the automotive, and recre- ational industries have also been placing increased reliance on high per. formance materials. ‘The reason 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 characteristls, high fatigue strength, corrosion resistance and versatility to meet spetife requirements, Chapter 1, Introduction = ~~ 3 Fragmentation Materia) factre Delamination Figure 1.3: Progressive fracturing and fragmentation in typical composite specimen subjected to impact loading. Even with the significant advances that ast Iwing mole with these new ‘materials, there is still a great deal unknown about their mechanical chart: teristics. These unknowns can be attributed to tle large number of different ‘materials being used, to the contelbution of micronschavicnl Interactions of differnt components to the overall behaviour and the snelety of falhure ‘mvecaisins involved, 1¢ is also well known (ht the presouce of Inhomos geneities plays an important role in the behaviour of these materials, One of the major problems that dominates the design procedure for structural applications isthe strength of material to transverse impact which ‘may cause substantial internal damage. ‘These impact damages are often ‘embedded inside the material, hardly detectable by tie naked eye [2, 3}. AC- cordingly, detection of impact damage i critically important in maintaining the integrity of the structures jn service. By examining the numerous con- tributions to this area of research (4, 5, 6, 7, 8), it is evident that Impact loading represents a highly éomplex phenomenon composer! of several inter- acting processes. Figure 1.3 represents the progressive fracturing and fragmentation phe- nomena in a typical composite speeimen subjected to impact loading. ‘Th schematic representation, is perhaps only related to the failure observed in high velocity impact. For low velocity impact, however, while it is une likely that extensive fragmentation will be observed, material fracture and delamination will be the likely modes of fallure that exist. Tt has been shown that internal eracking of a structure subjected to low or high velocity impact loadings, are progressive phenomena which may rapidly propagate throughout the component, ‘This ight result in the ereco c c Ce) € 4 Discontinuum Mechanics Bh resi tes on Figure 1.4: Composite specimen subjected to impact loading. ‘tion of new totally separated parts, which interact with their surrou reyions, Conse ing ently, a powerful scheme is required to be able to monitor Uw fracturing process and to effectively model both individual and inter- section behaviour. Recent developments of discrete element methods have prepared the ground for a new approach to modelling this behaviour based ‘on the concepts of the diseontinuum mechanics. ‘he tralitional approach to the simulation of stress distributions in ar- bitrazy shaped components under possible nonlinear geometric and material conditions is by fine element techniques. However, the traditional nite el tment method (FEM) is rooted in the concepts of continuum mechanics and js not suited to general fracture propagation problems since it necessitates that discontinuities be propagated along the predefined element boundaries. "The corresponding elasticity and fracture mechanies concepts are applica- bie only in situations dealing with single crack or a low-fractured area without any fragmentation (9. Ia contrast, the discrete element method (DEM) is specifically designed to solve problems that exhibit strong dis- ‘outiities in materi] and geometric behaviour (10). ‘The diserete element ‘nthod iealizes the whole medium Into an assemblage of individual bod- ies, which in addition to their own deformable response, interact with esch ‘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 fnite element methods. ‘Consider a test specimen subjected to an impact loading as depicted in Figure 1.4. Barly material cracks are likely to appear near the position of applied impact load. As the analysis advances, two separate regions can be distinguished. ‘The fist one is a highly fractured region, and the second ‘one is the remainder of the body which presumably contains no fracture atterus, ‘The fracture region is usually formed in the vieinity of impact loscling and may comprise further separate parts Chapter 1. Introduction 5 Figure 1.5: Discrete element modelling of a composite plate, It should be emphasized that drawing such a conclusion does not neces- sarily require 6 fracture analysis. A relatively fast elastoplastic analysis can be performed by simple sell or solid elements to determine the different regen necoriing, Lv sono offoctve atrews oF sttnin eritevin, ‘the predicted fractured delaminated regions may then be exinined in te Inter stages of the analysis through the combined finite/discrete element algorithm. ‘The regions’ boundaries should be further extended if the cracking reaches to the boundaries of the fractured region (12, 13) Figure 1.5 shows @ typical section of the above micutioned specimen (here, a quacter of a composite plate). In a combined FE/DE method, the possible fractured region is modelled using a discrete clement mesh and the remainder of the speetmen is modelled by a stasidard finite element ‘mesh, Its also possible to modal the whole structure with discrete elements, In this ease, the possibility of delamination is investieatel throughout the structure. A combined mesh enables us to prevent unnecessary contact tection and interaction calculations which comprise « major part of the ‘analysis time, It is worth noting that even by modelling the whole structure with discrete elements, we are still using a combined finite/ discrete element approach, owing to the fact that" elements are used for modelling the 4 deformable behaviour of individual discrete elements. For the particular case of composites, each ply or a group of similar pilies is modelled by one discrete element. acl: discrete elesnent will be discretized by a Gnite element mesh and might have noulinear material properties or geometric nonlinearities (large deformations). ‘The interlami- nar behaviour of disercte clemente ie governed by bonding laws, including contact and friction interactions for the post delamination phase.eo © c ‘ Discontinuum Mechanics Interactions between finite elements (not those which are used for DEM discretization) and discrete elements are modelled by transition interfaces, ‘A transition interface is defined asa bonding interface with very high bond- ing strengths which prevent debonding under all stress conditions. Al interfaces, firstly, are monitored against the delamination criterion, Once two layers are delaminated, the corresponding interface will still be capable of further contact and friction interaction. However, there will be ho re-bonding after delamination. One important aspect of this type of modelling, which disting it Tron vthor contact baeed delamsiantion algorithvus (14, 46}, is that Ie does not require any predefined interface element. Being free from the restrictions of interface elements, provides major advantages. Firstly, there |s no need for the nodes on different layers to match each other, which cesses the way that data are prepared. This is essential in defining the transition interfaces. Secondly, in progressive cracking, particularly material fracturing, we may endl up with new nodes, edges, and boundaries that could destroy the compatibility required for these interface elements, 1.2 Alternative Approaches ‘There are alternative approaches developed and implemented for special applications over the past few decades competing the faite element based methods, Some of them can be regarded as special solutions of very specific Dries, while a others hm vlad for m wi ta of pl cenbennn Flaven thy: Hone of Hote con crmpete with the generality of the Jinite clane aiethod. ‘The boundary element method (BEM) isan important technique inthe computational solution of a number of physical oF engineering problems. ‘The boundary clement method has the important distinction that the {xetization only applies to the boundary of the domain of interest. Hence the computational advantages of the BEM over other methods can be con- siderable, Table 1-1 compares some of the main characteristics of the finite clement and boundary element methods as presented by Hunter (16) A combination of the finite element method and the boundary element Incthod would enliunce the general performance by utilizing the distin- {guished features of bot is for simulating various engineering prob- lems. This kind of modefftg, for instance, may be efficiently used in mod- clling the behaviour of tunnel lining and its surrounding rock medium sub- jected to seismic waves. As an example, Figure 1.6 illustrates a simple crack analysis studied by Kim|17] and the results achieved by a combined finite/boundary element method. t f Chapter 1. Introduetion ‘Table 1.1: Comparison of FEM and 15/M characterinties. e6ory FEM Mesh generation | entire domain roel A Tist te boundany Solution. entire domatn | ica we din potat Approximation “| ~"yoverming PDE] _voiiary coi 2 ma Tully populated Matrices spe =| usualy symmetric now-syinmetric: Tatograls | easier to evaluste | far harder to evaluate Nonlinearity —| widely acceptable | only Tinear problems Tmplementation easy, ‘uch snore difuTe Ey fat yee y eM qomon | | beth ey Ep) = Evexp (Bx) (we 8)8 Discontinuum Mechanics e 1.3. Scope of the Book As mentioned earlier, this text is prepared as a textbook for introducing the mathematical and computational concepts of the contact mechanics which are used in increasing industrial and academic applications of the combined fiaite/diserete element method. The goal has been set to make some contribution in discussing new ideas in a less clearly understood area, and to highlight the main problems that are met when dealing with highly complex phenomena of discontinuum mechanics. Chapter 2 provides some representative results and data on various ap- plications that involve the general concepts of the contact mechanics and related topics. Chapter 3, reviews the available methods for enforcing a constraint re- lation onto a set of simultaneous equations. A detailed study of the penalty ‘method provides basie information for use in other chapters. In Chapter 4, the main general computational aspects of the analysis will be discussed which includes a compreliensive discussion on large deformation theory, as is usually the case in the practical simulation of impact loading cf structures. Then the basics of the finite element discretization will be ‘explained. A short review of the explicit algorithm as the time integration scheme adopted will close the chapter, Chapter 5 is devoted to providing historical and state of the art algo- rithms for normal contact problems. The fundamental aspects of classical theory of normal contact will be discussed and practical ambiguities will be adresse, ‘The relatively complicated theory of frictional contact will be explained Jn Chapter 6. Tt includes discussions on various frietional contact phe- nomena and provides computational procedures for updating material state les for hardening/softening conditions and in general non-assoclated formulation, Chapter 7s dedicated to diseontinuum approaches of contact mechanics. Te 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 will also be explained which are Closely associated to contact detection procedures. ‘Chapter 8 summarizes the remeshing techniques and principles of crack initiation and propagation phenomena and discussions on the problem of size effect in fracture analysis of solids, softening and damage mechanics methodologies ns well as introducing a three dimensional anisotropic ma- terial model which consists of an otress return algorithm and deriving the consistent tangent modulus. Chapter 1. Introduction 9 Finally, Appendix A is devoted to describing the cata file structure for a 2D finite discrete element code called UTdem. This is not primarily aimed 15 a professional general purpose finite/dlserete clement code, but to provide the user with an overall insight of how a discrete elenient code may be used for modelling progressive fracture analysis of structures is well 98 granular flow simulations, 1.4 Notations In this section, the adopted veetor/tensor notations new briefly explained. ‘These notations are used throughout the book unless itis explicitly stated otherwise, 1. 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; sch as m, t, 4. 4. Seoond order tensors are represented ws bold face upper case Italle letters, such as D, N, A. A fow Greck boldmath letters, however, are used for stresses (¢) and strains (¢). 5. The transformation ofa vector by a second order tensor into another ‘vector is represented by simple Juxtapasiion, teon or t=Tn 6. The transformation ofa second order tensor into another second order tensor by a fourth order tensor ia repress by simple juxtaposition, o=Ee or H=DN 7. The inner produet of two vectors represented! by & 8. The cros (vector) product of two vectors is represented by a X, axb 9. The tensor product of two vectors is a msi dined a A=xoy As = 20s 10. The inner product of two second rank tetsors is represented by + A: B=tr(a7s)0 Chapter 2 Discontinuum Mechanics - A Review To attain a realistic overview of the extent @ contort bist algorithm ean be sed for analysing various academic, engineerin incl insustrial problems, ‘a quick review of potential applications are provided. It is not intended, primarily, to compare the results with available cata in the literature, as it is usual in academic papers, but to illustrate to the reader the applicebility of the method to different applications that may be analysed by the use Of the computational discontinaum mechantes. I is also aimed at sparking new ideas for further research and future challenges in this subjeet. ‘This chapter reviews the following engineering applications, amongst ‘many others, which are currently being researclied in many research insti- tutions thronghout the world, ‘+ Geomechanical applications © Granular materials + Impact analysis (progressive fracturing) «Particulate flow # Computer graphies umeee ee eee Discontinuun Mechanics 2.1 Geomechanical Applications 2.1.1 Rock Blasting Nock blasting is an interesting area for application of discrete clement ‘nethod. Ross shown in thi section are taken from (28), nse on ue ing simpli sold rock» detonation gu hternction models Figure 2.1: A 2D bench blasting simulation (18) Bruch bhesting is a very frequent rock blasting operation, In this exam- ple, 20 problem is modelled. The initial geometrical configuration and progressive fractute patterns in successive times from the initial blast are Atepicted in Figure 2.1 In order to climinate reflection of waves, some artifical damping has Introciwced close to the fixed bottom and left edges, thus modelling them a8 non reflecting infinite boundaries. Other edges are free surfaces. Figure 2.2 illustrates the fragmentation patterns in a multiple bench blasting proliem. In this problem, only the bottom edge is assumed to eee ‘Chapter 2. Discoutinwun: Mechanics ~ A Review 18 Figure 2.2: Fragmentation patterns in @ multiple bench blasting simular tion (18). be an infinite boundary modelled by a non-reflcting boundary condition. These results may be combined with reaulte obtained for bench in order to-estimate fragmentation pattern more accurately. Optimal fragmentation ‘can then be obtained by estimating the correct spucing, which fy one of the important geometry paraineters in a practical beuch blasting problem, 2.1.2 Mining Rock Bursts [Rock burst is an energy plienomenon sdccompan i ch filure in the form of brittle frecture induced by mining. Energy changes wceut during the course of mining activities and a part of the ener stored in the rock mass surrounding the excavation. ‘The ultimate stress state is created by the action of the pre-mining and mining induced stress. ‘The vlianges brought about by mining merely trigger latent seismic cxcuts that slerive mainly from the strain energy produced by geological liferences in the state of stress ‘The rock burst is @ sudden manifestation of the release of strain energy stored in the rock mass, The possible sources for the liberated energy are:“4 Discontinaum Mechanics Chapter 2. Discontinuum Mechanies - A Review 1b Figure 2.5: Experimental failure eesults: and numerical simulation of a sim- | | ple rock bursting problem (20) " 1. Stove strain energy in the surrounding mass of an opening. | us 2. Change in tlie potential energy of the rock mass. : 5. Minor slippage along rock contacts. 4. Gas extrusion ‘The rock bursting process can well be simulated by the combined fi- 2 a sac et ig 9 Beef ny op after the mining rock bumps, this field has remained a very important and active part of mining eyineering. Figure 2.4: Failure of eut embankment (24) 3 £3 Figure 2.3 illustrates some experimental and numerical simulation re- sults reported by Hazzard et al, [20] based on (21, 22, 23). 2.1.3. Failure of Cut Embankment In this example, a wall of 2600 oqui-size spheres betweeti two vertical plates ‘was modelled by O'Connor [24]. Gravity was first applied for 1000 iterations at a time step of one milliseconds. Att = 1s, the left boundary was suddenly L removed, The ensuing collapse configurations are shown in Figure 24. The simulation wt run for a total of 3000 iterations [24)16 Discontinuum Mechanics 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 developed by Shi [25] to model the behaviour of diseontinuum media. The DDA is based on dy- namic equilibrium, which considers the kinematics of motion of individual Dlocks [26]. In DDA, each discontinuity (joint) may be assigned » differ- ent frletion angle, cohesion and tensile strength. ‘The blocks are essumed to be elastic deformable bodies contacting to neighbour blocks through the penalty enforcement of the impenetrability constraint. The DA may be assumed as a special case of the more general finite/discrete element tech- nique. Figure 2.5 illustrates the results of slope stability analysis for various discontinuity patterns and friction angles based on the discontinuous defor- ‘mation method as reported by MacLaughlin et al. (26) 2.1.5 Shear Band Slope Stability In contrast to the DDA method, a shear band slope stability analysis may be performed by using a fully deformable nonlinear finite element simu- lauoa. An adaptive remeshing scheme has to be employed to avoid excess distortions of the finite elements clase to the highly deformed shearing band. (Chapter 2. Discontinuum Mechanics - A Review W Figure 2.6: Remeshing process and the 45° shear band development in a ‘tensile plate undergoing large lateral necking phenomenon. Figure 2.7: Two shear band slope instability problens (28, 29} ‘The concept of shear band deformation can be best. 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, Figure 2.7 illustrates two-different examples of slope Instability simu- lations performed by Stead et al. (28) and Cramer et al. {29} in two and three dimensions, respectively. In the 2D ease, a h-adaptivo finite element ‘method has been adopted, whereas in the 3D example, only large deforma- tion theory has been considered. 2.1.6 Falling Objects Figure 2.8 shows the application ofthe discrete vlcamnent approach to a full three dimensional simulation involving randomly shaped grains fuling under {gravity into a box. lwo sides of the box on which some particles have come to rest have been removed for rendering purposes (30,18 Discontinuum Mechanics Figure 28 Discrete element modelling of falling objects [24). 2.1.7 Porous Media : ‘Awother interesting application of the contact mechanics is the geometric tiodelling of porous media. Modelling materials such as permeable soll 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 of the bulk material. The model, fou then be used as the input model for numerical experimente such as fuid flow simulations (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 east can be guaranteed to eouuecl aexus5 the houndaries, ensuring a permeable model. In order to obtain valid void Chapter 2. Discontiauum Mechanics - A Review 19 Figure 2.9: Porous objects with holes (31) Figure 2.10: 3D particulate packing in a porous medium (24).123 3 co 3 5 20 Discontinuum Mechanics { geometries, the modelling process must first generate a matrix that is in ‘mechanical equilibrium. Only then can we take a cast of a Yoid space that is geometrically valid and exists as a stable mechanical system [32]. Porous solids such as models illustrated in Figure 2.9, ean be used for simulating particulate flow in porous medi, as shown in Figure 2.10. 2.2. Granular Materials 1 Granular Flow in Silos Silos represent a vital part of the industrial infrastructure. Failure of a slo ‘often causes great economic loses either by wasting the ensiled materials, delaying production lines or disrupting transportation plans. In this ex- sumple, the prediction of pressure nnd flow in sllos has been investigatod ‘ulizng the discrete element method. Silo and granular material are both ‘modelled in this approsch as depicted in Figure 2.1 (88, 32). The results of typical conducted analyses may be used to guide the silo design procedures by pointing out any upanticipated loading conditions and pressure distri- butions which might allse during operation, as well as phenomena such as arching, diferent filling/emptying regimes, seismic loading, ete rere 2.11: Discrete element modelling of granular flow In a typical silo (33 Chapter 2. Discontinuum Mechanics - A Review a 2.2.2 Superquadric Elements Superquadtrics are « fonlly of parametelefunctinns, reveutly proposed for wv in mukibody dynamics analysis [92}. ‘Their vetnition nd interesting characteristics will be discussed later in this book. Here, atypical aplica- tion using superquadre elements is lustrated. Figure 2:12 depicts & depo- sition simulation using this kind of object representation technique within & discrete element analysis. As itis clearly seen, various shapes of the object fan be obtained while sinilae formulation i wt forall of thom ‘“ POV VERY OS Figure 2.12: Diserete element modelling of atypical deposition problem [3 2.3 Impact Analysis (Progressive Fracturing) Progressive fracturing and fragmentation analysis of impact or explosive loadings of structures have benefited substantially from the development cof the discrete element method. In the followin, some of the interesting, ‘applications are briefly reviewed 2.3.1 Penetration of a Missile Structural design of a shelter, armoured military eqwipmcut nnd safety mes- sures for bullet-proof vests inay force a designer/analyst to cheek for im- penetrability response of & structure subjected to high velocity object. ‘A complete analysis of an object penetrating a structure and developing,2 Discontinuum Mechanics extensive damage i fini lise ele thas only become possible by the use of combined techniques. Figure 2.13, illustrates how the erack patterns are propagated within a typical ceramic plate as a bullet penetrates the plate in different time steps. igure 2.13 Progressive fracturing in a structure Impacted by a high veloc- ity let, 2.3.2 Metal Cutting Metallic alloys age particularly susceptible to adiabatic shear localization 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 due to the high rate of energy generation (34). ‘The present example simulates machining of « TH6AL4V titanium alloy sample. Adopting an error estimator based on the rate of fracture indicators, has resulted in the evolution of the mesh refinement process as illustrated in Figure 2.14, Plastic deformation first takes place near the free surface of the workpicee. ‘The advancing tool causes the development of the shear bani, At huter stages, friction between the chip and tool eauses an increase of plastic deformation near the tool/chip interface which may give rise to 8 secondary shear zone, Figure 2.15 Mlustrates another example of simulatioa of « Llanking pro cess, In this example, an adaptive remeshing procedure has allowed for the Chapter 2, Discontinuumn Mechanics - A Review 2 Figure 2.14 Shear band development and chip creation int « metal eutiing application [34]. igure 2.16: High epeed metal eutting prablow (25)« m4 Discontinuum Mechanics simulation of the whole cutting process from the frst contact until the onset of total separation [35]. 2.3.3 Demolition of Structures Demolition of structures is among the group of problems that we roquire a realistic assessment of fractured regions and to what extent the scattered ‘materials are spread; to ensure the safaty of neighboring structures. This {kind of problem always exhibits extensive fracturing and frogmentation. Figure 2.16 gives an example of a tall conercte chimney tower. A mass ‘of specific explosive is detoneted within a tunnel in the base of the tower. ‘The problem was simulated by Munjiza (18). At frst, the basis ofthe tower Js 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 ofthe tower (18, 36). 2.3.4 Composites” Composite laminates are now widely used in many applications involving ‘dynamic loadings notably in the aerospace industry. One of the major problems that affects the design and performance of composite materials for structural applications is thelr vulnerability to transverse impact which ‘may cause substantia internal damage of the component due to matrix cracking, fibre failure Wid delamination, There is, however, agreement that ‘the most dominant causes of damege during impact are matrix cracking coupled strongly with complex mode delamination mechanisms (37, 38] Deleminsted sublaminates ae also particularly susceptible to dynamic load buckling when subjected to impact loading. It 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 inerease the risk of buckling (ee Figure 2.17), AAs another example, the results of 3D delamination analysis of « com- posite plate subjected to impact loading are presented. Figure 2.18 ilus- trates the combined finite/diserete element modelling [40]. ‘The composite ply pattern is set to (90, 04,90q; 0p, 00a). The impact loading is simulated by a trinagular load with a variable peak force. Figure 2.19 illustrates the debonding patterns at different layer interfaces for two different stages of the loading. Delamination patterns are clearly developing from the central region of the plate, ie. the impacted zone, towards the edges of the plate. These igures depict only the DE part of the ‘whole mesh, Chapter 2. Disoontinaum Mechanies - A Review 25 Liek Figure 2.16: Demolition of a chimney tower subjected vo explosive load. ing [126 Discontinuum Mechanics Figure 2.17: Different failure modes in composites subjected to impact load- ing (30). Mesh ee ie Figure 2.18: Combined FE/DE mesh of the composite plate. It should be noted that these results were schieved without considering material facture analysis, and only the bonding fracture was activated. In a practical test, however, the illustrated large deformation will cert involve extensive material fracture. 2.3.5 Masonry Structures ‘Several interesting implementations of the discrete element method have bbeen proposed for predicting the bahaviour of masonry structures [1, 41, 42, 43), Nevertheless, the predictive modelling of the nonlinear behaviour of mimonry structures remains a challenge, due ta theie sembdiserete and composite nature. Chapter 2. Discontinmm Mechanics - A Review 7 Interface between layers 4 and 5. 1) T = 0.00006 we. by 100012 aoe, Figure 2.19: Delamination pattems at layer interfaces at different time steps [39].ca «2 n a c c 2 Discontinuum Mechanics Figure 2.20: Discrete element modelling of masonry arch failure (1. Chapter 2, Discontinuum Mechanics - A Review 29 yur 2.20 ilastentew Us Koll problon of the fuiline of « two KBAR bridge studied and simulated by Petrinie [1]. ‘The problem was treated a ‘s combined finite/discrete element problem with the masonry blocks being represented by deformable discrete cements in frictional contact, the fill ma- terial by spherical discrete elements and the foundation region was modelled fas a Mohr Coulomb material by conventional finite elements. The failure mode brought about by settlement of the central pier ancl the traversing of ‘an abnormal two point load Is indicated. ‘As « real simulation, Figure 2.21 shows a 60-year old railroad two span. masonry bridge and the finite/discrete element model. ‘The bridge was incrementally loaded in place until severe cracking and large bridge key deformation were observed as reported by Marefut et al. (4). ‘A combined finite discrete element simulation was performed to simulate the failure behaviour of the structure. Cracking patterns similar to the test observations were predicted according to Figure 2.22 (45 Figure 2.21; A two spen masonry bridge and the finite element model [44].30 Discontinuum Mechanics Figure 2.22: Crack propagation pattems at different times (44). 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. Anexperimental test was performed to assess the performance of the ma- somry bridge until the collapse of the structure, Figure 2.23. A numerical simulation was performed by Klerck [43] based on a combined fnite/discrete clement technique, as depicted in Figure 2.24. The failure modes are inter- ‘stingly 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/disorete element method is capable of modelling the whole structure (old masonry structure and ‘new anchor elements), taking into account the bonding phenomena that tny exist between 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 Chapter 2. Discontiauum Mechanics - A Review Figure 2.25; Strengthening the masonry bidder hy anchoring. aL2 Discontiauum Mechanies Figure 2.26: Finite element simulation of anchor pull out test [43] 2.4 Particulate Flow 2.4.1 Shockwave Analysis A wall of 2600 equi-size spheres is constructed by placing the spheres be- ‘owoen two vertical plates. Gravity was frst applied for 1000 iterations at «& time step of one millisecond. A stream of slightly denser particles of the same size are dropped under gravity from above the wall. Oa impact, the streain 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 thy waves still originating at the source of diaturbance (24). 8 Chapter 2. Discontinium Mechanics - A Review ay Figure 2.27: Shockwave propagation in particulate fluid [24],3 eo ow ea 4 Discontinuum Mechanics 2.5 Other Applications ‘The discrete element methodology is increasingly being utilized for a wide variety of new industrial applications. Grouting, shoterete modelling, sup- porting systems, shot pening process (47), hot forging problems, print- ing and applications involving smart materials are amongst the engineering problems while animation, virtual reality [48] and artificial intelligence tech- nique are among various computer science applications that have benefited from the development of the contact mechanics and the discrete element nnethodology. An important area of research is the development of efficent parallel processing finite/diserete element algorithms for modelling of large seale diseontinuum problems [49} Chapter 3 Constraint Enforcing Methods 3.1 Introduction Many different methods have bon developed for enforcing a constraint com dition on the governing equation ofa well established physical behaviour. In this chapter, the following four methods for enforcement of constraints within a finite element analysis are reviewed + Penalty method «Lagrange multiplier method + Perturbed Lagrangian method «+ Avgmented Lagrangian method ‘The final part of the chapter describes the penalty method 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 (siusle point constraint) of a relationship between two or more freedoms (multipoint constraint). 35np r cc me ca c 36 Discontinuum Mechanics Contact Constraint Figure 3.1: Impenetrabilty constraint for two contacting bodies. rare a repre gia imponstraity constraint Baten two con- tacting bodies. ‘This constfaint defines the necessary conditions to prevent the bodies from penetrating each other. ‘The mathematical description of a constraint equation may be written in the form, cu=Q (1) where C is « matrix of constraints, u is the vector of freedom and Q is a vector of constants. @ in many eases may become a null 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, ie. A, B and C, are to be constrained to stay in a straight line, eg. for symmetry reasons. ‘Che constraint equations may then be formed as: Chapter 3. Constraint Enforcing Methods a Bh 4—% Figure 32: Nodes A, Band Co stay in straight lie, u-up=0 {S80 ea) Rearranging it into the standard format (3.1) G3 aq[2)-[) as ‘areal co) |s the matrix of constraints for this particular problem, Example 2 Figure 3.3 depicts a model combined of a continuum msi and a structural imesh, To define the connection between a structural element to continuum clements, ic. to link the rotational freedom of the structural element, 4a in Figure 3.4, to the displacement freedoms 11, us and uy of the continuum. ‘model, the rotational equilibrium about node 2 can be considered: lu thus — G2 = 0 (35) ‘Therefore, the constraint equation ean be fornied38 Discontinuum Mechanies Rigid Rotational Link Continuum Element 7 Figure 3.3: A combined continuum/structural model. I—> uy, My : 2 3—> Us, igure 3.4: Rotational equilibrium about node 2 should be satisfied. Chapter 3. Constraint Enforcing Methods 39 nodes 2 and 3 are just in contact “Figure 3.5: A simple two bar model. be (a1 -n[8]-0 0) 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 direction. A 0.1 unidirectional rightward displacement is applied to node 1 of the left bar. Bach bar behaves as a linear spring, 50 w —10 Kak [ 1 | on ‘The assembled system of equations will be oe uy = OL fA | F| - o 0 10 =10 us a. G8) 0 0-0 w}fueo} ba since the equations are uncoupled, the results will be: {er (a)E 40 Discontinuum Mechanics 01 me 1 e—————_e2 J: 4 Figure 3.6: Uncoupled solution for the two bar problem. Figure 3.6 shows the deformed shapes ofthe bars for this analysis, which cloarly shows overlapping the elements ‘To avoid this, the following constraint equation should be enforced: w-m20 (10) ‘We will later use this simple example to verify the methods adopted as constraint enforcing methods. 3.3. 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 briefly reviewed and compared. One approach, which has been widely used by many researchers, is the concept of minimization of the total potential energy for deriving the nec- ‘sary equations. The total potential energy of n linenr elastic system sub- Jected to static loading and consisting of two discrete bodies, : and ay be written es (Figure 3.7) Bein [ — wdan-[ ue? (31) n - In, +0, O25 fay4% ith Using @ standard discretization procedure based on appropriate trial functions Chapter 3. Constraint Enforcing Methods a Figure 3.7: A system consisted of two interacting bodies, Ulu) = Au? Kau" R (22) 2 where w is the nodal displacement vector, 1 Is the system stiffness matrix and Fis the force veetor. Without additional constraint, equation, bodies %; and M do not interact and the system is uncoupled. In an alternative approach, imposing the constraint equation is per~ formed by means of variational principles [50]. In this approach, we ean introduce the constraint by forming a constraint functional, 1°, in addle tion to the functional of the system [61] f= Mu) +0"(C,u,..) (3.13) where f1(u) is the functional of the unconstrained system. The variational form of (818) will then be oft = anon" (3.14)I c a oe (as o 2 Discontinuum Mechanics which is valid providing IT = 0 and the constraint is imposed, simultane ‘ously. Further development of the formulation requires a clear definition of the constraint equation, C, and the constraint functional, TI", in equations (213), (314). 3.3.1 Impenetrability Constraint ‘This constraint defines the necessary conditions required for preventing the contacting bodies from entering into each others boundaries. In this section only a frictionless contact is considered. Figure 38 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 time 1 iy ooN time t+dt Se w © Figure 3.6: Potential penetration of two contacting bodies. ‘The necessary constraint equation may be written as o=le n>0 0 onf=rinr? (3.18) ‘where P® is the common contact surface of two bodies, z! and «? are the defonned positions of bodies 1 and 2, respectively, n is the normal to body 2 and 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 equa tions are simplified g=Cu (3.16) Chapter 3. Constraint Enforcing Methods 43 ign(n)[t. — 1] (7) vy (218) cau ° (3.19) 1k i important to completely clarify the munrical importance of the g function. In a general contact problem, if the coustraint equation Cu = @ is exactly satisfied, we will have Cu @Q = 0. However, in almost any ‘numerical analysis of « contact problem there will be some penetration ‘hich could be interpreted as the g function g=Cu-Q (320) ‘There are several methods for enforcing a constraint condition to a sys tem and solving the resulting coupled set of equations. Here, four methods which havo boon frequently used In various contact auinlywe are coweetbeds ‘+ Penalty Method ‘+ Lagrangian Multiplier Method « Perturbed Lagrangian Method + Augmented Lagrangian Method 3.3.2 Penalty Method ‘The Penslty method was probably the first approach adopted for’ con- straint enforcing method. It was developed by Hallquist and his colleagues in Lawrence Livermore National Laboratory during the late seventis for ‘modelling impact/contact problems. ‘To obtain the necessary equations, and comparing to the first term in Equation (3.12), the total potential energy for « constrained problem can be written as Tw) =1(w) + 3aPag @ayi I a4 Discontinuum Mechanics ‘where @ is a normal contect stiffness, called penalty number, end in general is a diagonal matrix of penalty terms for each degree of freedom. g is the normal gep vector and for g = 0 the constraints are fully satisfied; A(u) = Nu). Minimization of the total potential energy will result to sti = [Bou + oF asu] (322) (3.23) ‘To maintain equilibrium, ff should be equal to 0. ‘The first term on the right hand side of (3.23) is the well known stiffness equation ’ am on Ku- (3.24) Fa 7 Ku-k (3.24) and for the second term, we have a=Cu-Q (3.25) (328) Fe 67088 = (Cu-QiFaC=CTaCu-CT0Q (827) ‘Therefore, the modified st ness equation will be [+07 ac]u=R+O7Q (3.28) ‘The term, CT aC should be added to the system stiffness matrix to incorporate the impenetrability constraint stiffness Chapter 8. Constraint Enforcing Methods 45 Remarks on Penalty Method ‘The main features of this method are: ‘Enforcement of constraints requires no extra equations. ‘The constraints are only satisfied in an approximate manner and the correct range of penalty numbers have to be chosen. If is too low, the constraints are poorly satisfied, while if is tov large, the stifiness matrix becomes poorly conditioned (the difference hecween in and out of diagonal terms becomes very high). As an initial estimate for a OSB 0 one (3.81) where g is the gap function, «1 and 2 are the deformed configuration of ‘body 1 and 2, respectively, n is the normal to the body at the contact, surface, and I, is the contact domain, P= 1? nT? ‘Therefore, the varintioual form of the constrnint exustion (3.74) may be explicitly expressed ns awe =f aégiu)io (3.82) hn. owe [| ag = Suda (3.83) Equation (3.83) may be.re-written in terms of the contact foree vector oy a) | owen = Fm Fd (3.84) Attention is now focused on single boundary node in contact to fore imulate the residual contribution of contact coustraint, x. The component {orm of the virtual work of the contact forces associated to the contact node is then given by (57) Pym out (3.85) 5" = fe" bau = SE"co ca [on 3 (am) O Discontiauum Mechanics 11 Baler tne! “Eyelet oh Prion ae Figure 8.12: Normal and tangential gaps, where k = n,t and i= 2,y, and uf is the the é-component of displacement vweetor at noe s, g ~ (qy8s) is the relative motion (gep) vector in normal tnd tangential directions, respectively, and fis the contact force vector ‘over the contact area AS, preme eoee[G ae]. em ‘where ais the penalty term matrix, which ean vary for normal and tan- ential gaps anc even between single contact nodes. ‘The corresponding recovered residual force is then evaluated as (87) ‘The partial derivative part of equation (3.87) defines the direction and \lstribution of normal and tangential contact forces. The calculated contact force has then to be distributed to the target and contactor nodes. ‘The potential normal and tangential gaps for each contacting couple are evaluated by monitoring the coordinates of contacting couple nodes in each time step. Then by projecting the coordinates in the current and previous time steps to a reference configuration, the possible gaps are calculated (Figure 3.12), Note that for a debonding problem, both penetration and telease normal gaps (positive and negative) are valid. 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 reason can be attributed to ‘he foct thet the inpeetrability condition (g = ) 1s only approximately ~ Chapter 3. Constraint Enforcing Methods a satisfied by this method. This is clearly observed from Equation (3.21) where «non-zero term is added to the total potetial energy’ of the system. Without any additional treatment, the penalty method will cause the ays tem to gun nergy artificially, although sometimes this extra energy i 8 cornpensation forthe loss of deformation eneray dive to contact penetration. In a central difference one dimensional contact analysis based on the penalty method, the contact force may be defines! as, en Be ee (0a where g is the normal penetration, One may expect the ceuteal diference scheme to be stable when applied with Kehoe ca (a0) However, numerical tests show that in some cases considerable energy ia 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 (ball) on to-a rigid wall as depicted in Figure 3.13. 1) The all escaping ot ofthe wall, 6) The Bll penerating int the wall Figure 3.13: A material point entering and leaving a riyid wall at successive time steps (58). At a time step when the material point is about to enter the wall bound aries, for some time ty < At no contact force exists to resist penetration.8 Discontinuum Mechanics ‘The resulting penetration generates contact force and stiffness for subse ‘quent time steps. When the material is leaving the wall boundaries, the con- tact force continues to be pushing out the material point from the boundary for some time tg < At, although there is no penetration any longer. Con- sequently, each time the ball enters and leaves the wall boundary, it gains some additional energy (58). Figure 3.14“illustrates simple contact problem in which a material point of mass m = 0.1kg is moving between two parallel walls with an Initial velocity of w = 10m/s. The walls are on the distance ! = 7.5m from each other and the penalty coefficient a = Le +06. Figure 3.14: A material point moving between parallel walls. Figure 3.15 shows the velocity of the material point after it has passed through the starting position m times. The results are obtained by unmod= ified central difference time integration scheme (3.88)(58]. It is clearly seen that as the At is increased, the velocity of the material point is also rapidly increased, Le. the system becomes faster and faster, representing the system gaining extra energy during the time steps. Larger ‘values of Af leads to larger values of t; < AC¢ for which larger energy values are added to the system. ‘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 Alifferont selena which have been tinod to stabilize the contact analysis, ace brielly reviewed. Chapter 3. Constraint Enforcing Methods 63 Figure 3.16: Velocity of the material point after It passes m times from its original position. A Modified Central Difference Scheme | I this method, at euch Lime step the contact fore div to the penetration | | ia calculated from the modified version of (3.88)/10) ° if gt <0 and gf <0 if of > Oand gt <0 (3.90) if of < and gt >0 if gf > Vand git >0 where v is @ user defined constant and a negative value of g* means no penetration Por K = LAr? < 05 and v < 1.0, numerical ests have shown Feabiey and at th soo tina the damping tne cried odReation is negligible (0.64 Discontinuum Mechanics ‘A Nonlinear Contact Force It is sometimes convenient to assume that the contact foree is not linearly proportional to the value of the penetration g. A natural way to define ‘nonlinear g — f relation is by assuming that the value of penetration is limited to a maximum, ealled field zone. Penetrations larger than the size of the field zone are usually ignored. For the exact definition and further details of the field zone, see Section 7.2.4 ‘The nonlinear contact force can then be written in terms of the g and the field zone, (3.91) In order to obtain a stable central difference scheme, at time f, the contact force f is calculated as Sm a(t noses (a2) ‘The stability of the scheme is controlled by parameter 0 < @ <1; proper @ has to be chosen in order to avoid instability due to nonlinear contact force. 0 = 0.6 proved to give a stable scheme in a variety of the discrete clement applications (58). Relative Velocity Deffuping ‘A simple damping scheme ls provided by applying an additional damping force according to the relative velocity {v}, Fo" = 0g + lv) (3.93) where ¢ is the contact damping couficlent, For normal contact ang + eld] (3.94) with (0.98) s/t { aig = Ema /AO? Chapter 8. Constraint Enforcing Methods 65 ‘where mys = min(m,,m,) isthe minimum mass of the contacting bodtes 4 tnd j. €,¢ are input parameters controlling the penalty term, usually taken 0.08 Itis proved that the central difference tine inteyration scheme is stable for any value of At if the following condition is satisfied e}o}-yMeN, (4.19) it 1 e=)> glnA conn (4.20) i 4.2.4 Generalized Strain Tensors ‘A goneral formulation for the Lagrangian and Fulerinn stenin tensors way be defined by generalization of equations (4.10, 4.11) and (4.19, 4.20), (21) (4.22) and BO => ap- yen; (4.28) : em = Ligaen) YRa- a" mom (4.24)4 309 esc a Ea ” Discontlauum Mechanics which implies 9 2 REO RE (4.25) In the particular case n —+ 0, this provides the Logarithmic strain ten- BOS inNON =U (428) = ladmnsGny = nV (427) 4.2.5 Rate of Deformation ‘The velocity gradient matrix is defined as, 1-229 oy (4.28) On “Therefore, the rate of deformation gradient may be expressed as ou ox (4.29) 1=FFO (4.30) ‘The rate of deformation matzix (also called the velocity strain, or stretch- ing) is defined according to the velocity gradient matrix, l, 1 iP) geet) (431) a= en (432) where d gives the rate of extension per unit current length of a line element having a current direction defined by m. The velocity strain measures the Chapter 4. Nonlinear FE Discretication % current rate of deformation, but gives no information about the total defor- mation of the continuum. Furthermore, its integral in time for a material point does not yield a well defined, path dependent teusor so that infor- ration regarding phenomena such as total stretching (or 1waximum strain) are not available, Therefore, to obtain @ 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 awed (439) vere 1 Ty w= lar) aa) aod ants te(ou-!—u-) RP w= RR" 4}R(U--U-W) Fl (435) "The second term vanishes in rigid body motion for which d = 0. 4.2.6 Stress Measures ‘The Cauchy stress, which is also called the true stress, is defined as a spatial ‘tensor, @%, which provides the necessary and suificient condition for satis- faction of the momentum balanee law. This syunctric tensor relates the ‘normal vector n to the traction vector t t(n)=on (4.36) ‘and satisfies the general equation of dynamic motion diva + fot = pi (437) where n is the unit normal to the boundary of the current configuration; pis the density at the current configuration, @ is tho acceleration of the. ‘body and f° is the body force measured per unit volume of the currentmca 8 w Es 76 Discontinuum Mechanics ‘configuration. The variational form of equation (4.37) leads to the principal of virlual work (Hor more detail see section 4.3) ‘The Internal viral work nay then be expressed as, (4.38) Pairs such at o 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 internal virtual work in terms of an integral cover a volume that is known and can incrementally decompose the stresses ‘and strains in an effective manner (64). Sometimes itis easier to deal with the reference configuration than with the current configuration. The Kirchhoff stress is defined per unit area of the reference configuration reJo (4.39) J=deF (4.40) Which is work conjugate to the rate of deformation tensor with respect to the initial volume, sweet = fr saav (at) Another stress measure isthe frst Piola-Kirchhoff stress which is defined at the reference configuration P=JoF" (4.42) ‘and is work conjugate to F. The second Piola-Kirchhoff stress tensor, an- other frequently used stress measure, is defined as, S=JFtoF* (4.43) and is conjugate to the Abbterial strain rate tensor, E, with respect to the Initial voluue, Flually, the logasithmale strain measure (In), defined in (4.26), is work conjugate to the rotated Kirchhoff stress, R”7 Re (65). Chapter 4, Nonlinear FE Discretization nm 4.2.7 Objective Stress Rates ‘An important aspect of the measure of stress is the vljectivity (or frame jnvariance) ofits rates, An objeetive rate must transform according to the law for a second order tensor when the spatial reference frame is rotated. ‘This is important because constitutive equations are often expressed in rate form, so ifthe rates are not objective, the material behaviour will depend fn the choice of the spatial coordinate system, which is obviously physically uunzealistie and must be avoided, Although the Cauchy stress tensor is an objective tensor, ito rate doce fy the objectivity condition. Therefore, for a general constitutive ‘material model we are bound to use other stress rates. Only the most popular ones are briefly reviewed here. ‘The Truesdell Stress Rate is constructed directly from the terins required, for the objectivity of Cauchy stress rate, lo-al + (ube (add) ‘The Green-Naghadl stress rate is defined by ignoring the stretch compo- nent of F in the Truesdell stress rate (66), 6 +oRR"- RRs (4.45) Ifthe un-symmetric part, RAT, is approximated by the spin tensor ww 0° =6-ow-wo (4.48) the Joumann stress rate is obtained. ‘This tensor has two important prop- erties: : (0% : 0°) = 20° : (oP yrunted (47) a a (09)! = RTO R (448) sand if 2° is a deviatoric tensor, then (2%)"** is also deviatoric. Any frame invasiant stress and strain measures may be used in the con- striction of a constitutive equation. One of the most popular rale form is to express the Jaumann rate in terms of the velocity strain.E3 ee ee oo ers as & 8 Discontiauum Mechanics ‘d (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 @, 2°, d or other state variables, Note that it ean not be a function of & or d because they ‘are not femme invariot ‘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 ‘necount, for the effect of the rigid body rotation on the components of © wha it bs expresses in a fixed coordinate system. ‘This is not necessary If the constitutive law is expressed in a corotational framework, e=ca (4.59) If the corotational coordinate is updated by the rotation of the material based on the polar decomposition theorem, the rotation of the principle strain rates, then & corresponds to the Green-Naghai stress rate [67]. Note that such corotational formulations are much more suitable for anisotropic tmotetials than Jaumann or Truesdell rate formulations. When the latter ‘re used with anisotrople 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], (451) where T is the material rate of the rotated Cauchy stress tensor and D is the rotated rate of deformation, t-RToR (4.52) & pentar ‘Then, the Cauchy stress tensor can be evaluated from the expression, nes = Rass (e+ ["at) Rn (459) Chapter 4. Nonlinear FE Discretization np 4.2.8 Multiplicative Decomposition ‘The additive decomposition ofthe infinitesimal strain tensor into elastic and Inelastic parts is basic to virtually all the subsequently developed theory of plasticity, +e (4.54) For strains that ara too large to he treated as infinitesimal, the decom= position still works, at least for longitudinal strain, when this is taken as the logarithinie strain, Jn order to formulate a theory of plasticity for large de> forination, we have to establish the appropriate locomposition for arbitrary states of deformation. ‘An additive decomposition of the logarithmic strain is equivalent to a ‘multiplicative decompesition of the deformation gradient F, P=FF, (4.88) 5 ‘which was frst explicitly formulated by Kroner J}, anc! further developed by Loe [TO], Nemat-Nasser [71] and others on tlie bass of consideration of the behaviour of crystals. ‘The reasoning behind this, is comprehensively presented, among others, by Lubliner (61). ‘The transformations involved in the decomposition are shown schemst- ically in Figure 4.2. ‘The unloaded element (2, + dé) is usually said to be in an intermediate configuration. ‘The velocity gradient then becomes (eke raee (4.58) where l= PFz) (4.57) b= PF, (4.58) and finally the Bulerian rate of deformation is d= de+(FelpFs')* + 4.59)3 3 ca a 80 Discontinuum Mechanics oa Figure 4.2: Multiplicative decomposition of the deformation gradient. ‘Box 4.1 summarizes thé stress integration procedure for a large defor- mation plasticity. \ 4.3. Initial/Boundary Value Problem 4.3.1 Equilibrium Equation Consider a body, B, occupying @ region 9 with a boundary T subject to body forces f° tliroughout its domain ©. Here, the boundary is assumed to consist of @ part with prescribed displacement ug, P,, and a part with prescribed traction force f2""/, T, (Figure 43). ‘The boundary conditions ‘may then be described as m= fe" on Ty s Am (4.60) where o represents the Cauchy stress tensor and n. represents the unit otter normal along Py For this body to be in a state of static equilibrium, the following condi- tion must be satisfied i gta [pte =0 (461) Chapter 4. Nonlinear FE Discretization a Box 4.1 - Stress integration procedure for a lane leforuuation plagticity. + Compute current deformation gradient Pest Vu © Elastic predictor phase (Frey = BM (PR) = Re (Ue ‘Til clastic stress is evlunted | + Chock for ying IR (ansnyiriet nym so + Blastics update Set (o)"tt = ((o)r2y rIsE « Plastic corrector phase Perform smal etraln stress update ont on ‘¢ Update intermediate configuration Fee (PME pao (erty sit ENDIF and for a state of dynamic equilibrium, f Sodas i petray— | pa (482) Applying the divergence theorem to the first term in the above equation ‘and using equation (4.60), the following is obtained ii folio | onda = [vad (4.63) Ir, ‘Thus, the equilibrium condition for the body B becomes fue + 9) do = viedo (4.64)€ a3 1 c 2 Discontinuum Mechanics [As this condition holds for any arbitrarily chosen subdomains, the inte- ‘grands should be equal pointwise, dive + 52%! = pit (4.65) which represents the dynamic equilibrium condition at a point within the body. It is also called the strong form of the linear momentum balance (or translational equilibrium) written in spatial description. 4.8.2. Variational Form ere, 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 applied pointwise, but rather in an average sense, by a weighted residual scheme, [ (owe ‘where the weighting functions w satisfy the natural boundary condition over Ty, ive. w=0. Applying Green's theorem and integrating by parts, results in the following identity sir [ (se! an) -ode=o (6) ~ fs Vado + fe fot ade ~ fr pitudy (487) + Jpen-uda + fy, (71! ~on) sda =0 whieh is simplified to Joos te [ inc = [pede f gS nda (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 6, 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 fougg in Zienkiowice etal, (51). Chapter 4, Nonlinear FE Diseretization 83 Figure 4.3: Description of the boundary value problem. In addition, tis assumed that a part of boundary, P=. may be in contact with another body (Figure 4.3) according to the contact. boundary condi tions (72, 56] on ont. if ge >0 Snag me eee (4.09) where gu is the gap between the bodies. By denoting Vie (Su: duy=0 on Ty.) (4.70) the space of admissible variations, the variational (oreak) form of the dy- namic initial/boundary value problem may be expressed as {73 74] Wee fu) + M (Buy) = We*(Gu) + (5u) (ATHY where wit Gua) = I as) o(wido (472) Miia) = [Seni amyca a Discontinuum Mechanics werou) = ustedes i bw fda (474) ween gu) = / S60) Fa (475) denote, respectively, the virtual work of internal forces, the inertia forces contribution, the virtual work of external forces and the virtual work of contact forces. Here ois the Cauchy stress tensor, ¢ is the strain tensor, uw is the displacement vector, while g represents the contact gap vector. Observe that in the present formulation the contact terms correspond to a penalty formulation of contact interaction, ‘To incorporate appropriate stress and strain measures in » finite strain analysis, the virtual torms may be expressed in terms of an arbitrary virtual velocity from the current position of the body. Using the same analogy as before, and concentrating on the virtual work of the internal forces, one may write . wit fo: bado (4.76) which 6d is the virtual rate of deformation Alternative expressions may be given for other conjugate stress and strain measures, wine Jon 7 édav (am where 7 = Jar is the Kirchhoff stress tensor and is the initial (reference) volume of the body, and we = [Pi ska (478) ° wis a fs: sbav (479) for the fist and second Piola-Kirchhoff stress tensors, respectively. Chapter 4. Nonlinear FE Discretization 6 4.4 Finite Element Discretization ‘A stondard methodology according to Zienkiewice [61] and Hughes [75] is followed ‘The rte forms desived in section 4.3 may be exploited with ap- propriate linearization to generate incremental forius. Lincarizntion of functional W(x,m) at the known configuration x = X eat be obtained through the dizectionsl derivative formula DwvGanyu= Z] ween (490) lo in the dretion usresultng in the linear part ofthe functional Yn) LW m)] = Wn) + DX n}ewe (481) Equilibrium, of course, must be satisfied for every configuration x Woonf=0 vnev (4.82) te residual force may theu be descrbed as (se (1.71), ie f a vemnde— vet — vee — (483) In where WV" is the virtual work of the contact forces and W/"*" is the virtual work of external loading wet [prints [rine 49 ‘Now, consider the space V* C V as finite dimensional spproximation to described in (4.70). Furthermore, let subspace ”" be generated through spatial finite element discretization in the current confiiration Beco.) with xing fYxiny) =0 if az | (488) where Ne a the number of finite elements. So tnt over the typical element , +(e) one has the interpolation86 Discontinuum Mechanics lx Dele me (4.86) whore Noy is the number of nodes of each individual finite element and N,(2) is 0 stantard shape funetion over euch sulspaco, that transposos the continuum problem to a finite element problem. We may now represent the discrete gradient operator in the following form vee ag = La Batle (aan) Be=V'"Ng (4.88) "Therefore, the discrete version of (4.88), (4.84) supplies the element residual force vector fen fot se (489) where $2" = {2%} = £ coy Bile} (4.90) a Fe (Gah fg Nelanides [o.Nalsmtde (490) ‘ve 0 calle! eloment internal and external force vectors, respectively. The contbation ofthe coutat fare i conieeddirstly rough the moc. ated contact nodes, Fa" = AR” (eg) (4.92) in which « penalty based algorithm is adopted to evaluate contact forces. ‘AGM a thécotact arn asocated to ach contact node, and ete palty coefficient. ‘The global residual vector is then obtained, Chapter 4, Nonlinear FE Discretization ar (4.93) ‘where Now is the number of nodes in contact. ‘The finite element discretization of the inertia term in Uhe boundary value equation (4.71) results in the consistent mass matrix M.- [| pN.Nutn (494) 3) ‘The effects of the inertia term is directly considered in the explicit time integration scheme to evaluate the current displacement, velocity and 86 , Mi (625) ‘where nesn is the number of contributing slave nodes, Mm is the mass at rode m, Mj, is the contribution of mass of the slave node # to master node ‘m, and superscripts * and * denote post (after) and pre (before) impact finnutilies, wespectively. of and vt are the normal velocity of master node ‘i uu lave node s, respectively. ‘Pho tatgentlal component of the velocity ‘yector remains unchanged during the momentum balance. 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 masta gent, base on the approach given by TOODY code, With reference to Figure 6.3, the contribution of pressure to master segiuent m from the slave element s, pis defined by: pit dons ge a ee (6.28) whereof the effective stress in slave clement s perpendicular to master segment, ae (627) where ff and 3+ are the internal normal forces in slave elements s and 8 + L respectively I the inenuality (tm.a41 “tm & lm) is satisfied, 73? becomes the master sogitent pressure, otherwise the contribution to the pressure of the next oweelapping slave element is computed accordingly, Chapter 5. Continuum Based Normal Contact Algorithms 108 ates daet oy om ‘and ifthe inequality (rm,o42 ‘tm 2 fm) is not satisfied, » is incremented by ‘one and the procedure continues until the inequality is satisfied. ‘The total pressure for master segment m is then computed, i (6.29), ‘where nes isthe tgtal number of contributing slave segments, The pressure. is then distributed to the master nodes m and m +1. Master Surface Accolorntion Using the TENSOR algorithm, the normal interface force at master node 1m is defined by: asp = Silt = Mal + AMn Mute (6.20) ‘where a; is the coriolis acceleration, FR" = Mm internal force in master element m. ‘The norma! rode m is then computed: and J? is the normal eration ofthe master Fr -age (6.1) ‘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 DYNA, the slave velocities and accelerations tuormal to the master line are reset, not imposing the slave to remain exactly fon the master surface. For the slave node in contact with the master sogment m, the tangential, velocity and scoeleration, vf and af, are computed from the slave element lobal velocity and acceleration vectors, and die normal compoments are interpolated from the relevant master nodes,104 Discontinuum Mechanics uy = (L= Gaor + Gong? (6.32) (1G am + cam? (5.33) ‘Then, the new velocities are the combination of the tangential compo- nents from the slave side and normal components from the master side, ‘while the new accelerations can be computed from the dynamic equations of relative mation. 5.3. DYNA Penalty Formulation ‘The frst practical penalty based finite element contact analyse was do eloped by the DYNA team in the eighties. Tt 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 symme- try ofthe appronch, eliminating any bias in choosing the slave and master surfaces [54 A general interface treatment using penalty method may be outlined as follows: ‘+ For each slave node ¢, locate the closest master node m, and corre: 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 6.7 shows a typical contact problem in which four master segments can harbour the slave node s, and node m is the closest master node. If m and 9 do not coinelde, ¢ can usually Le shown to lie in o master segment: via the following tests: Chapter 5. Continuum Based Normal Contact Algoritiins 15 Se Figure 5.7: Node m is the closest master node to the slave node 8. (rms t) (Png X Fst) > 0 co) (rma X tae (ta X Fmat) > 0 (5.35) where fia is the position vector along the cominon exlge of the master segments i and {1 and outward from node m, t, is the projection of the 9 vector, connecting points m and 3, t=9-G-nyn (636) Pmi Xmas Frm Pmt en ‘neo the eliding constraints keep # close but not necessarily on the master surface, and since # may lie near or even on the intersection of two master106 Discontinuum Mechanics sogments, the inequalities of (5.34) and (5.35) may be inconclusive, failing to be positive or giving more than one positive result [bd]. In this case, » is assumed to lie along the intersection which yields the maximum value for ‘the quantity Stat, in, (5.38) 5.3.2 Finding the Contact Point ‘A contact point is defined as the closest point on master segment to the slave node s. Ifa quadrilateral finite element discretization is used, each point, on the master surface segment i, can be represented by the position veetor r (see Figure 5.8), r= falGsn)in + falEsm)ia + Jal mbia (5:39) Figure 58: Finding the closest master point ¢ to the slave node s. Chapter 5. Continuum Based Normal Contact Algorithms 107 with, Aen) = Nal? (6.40) Ns(6.n) = F066) +n) (a) where NV, are the standatd shape functions, and 2°” is the nodal coordiiate of the j** node in the i'* direction. The normal to the surface is non-zero, nat ca z #0 (642) ‘The contact point coordinates (f,7e) must satisfy the following condi- tons: Se Gon) re =r] = 8 (6.43) Fem): Ira ~ elon =0 (a4) 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 force 's applied between the slave node and its contact point, with a magnitude proportional to the amount of penetration g, g= 1: [ry (Gon (645) -agn if g<0 (6.46) whore J; is the contact force scting on the slave node s. An equal and ‘opposite force is distributed over the master segment nodes,Ee ao 108 Discontinuum Mechanics Fins =—NlGene)fs tf 9 <0 (6.47) In DYNA, the penalty atifiess factor, ois given in terme of the bulk nous, K, voluuie V, and the face aren A of the master element, where is a scale factor for the interface stiffness, usually taken as 0.10. Larger values reported to cause instabilities unless the time step size is reduced. 5.4 2D Node to Edge Penalty Contact In this section, a simple and logical approsch for two dimensional node to edge normal contact based on the penalty method is discussed. The ap- prowch was originally developed in 1980 by Cirea[63}, and later documented by Crook (52) In this approach, the contact surfaces are explicitly defined. Two sur- {faces are then linked to form a slideline or a contact pair (or contact couple). ‘A sitnple global search algorithm will be followed by an accurate local search ‘method to identify all contacting couples. A penalty based formulation ts ‘adopted for contact interagpion and evaluation of contact forces and forming _ tthe 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 e1, then point 2 ‘will penetrate edge ez. Therefore, we need to search for both surfaces at ‘each slideline Figure 5.9: ‘Pwo surface search fur u sideline, Chapter 5, Continuum Based Normal Contact Algorithins 109 5.4.1 Contact Geometry Figure 5.10 shows a typical slideline between two contacting bodies, defining ‘he contacting nodes and segments. segment m Figure 6.10: Definition of a contact geometry. ‘The contact surfaces are defined by a list of element edges (segments) stored in order from the start ofthe surface. The order is defined by insisting, the he outward normal i coord with righthand srew rue (Pigus ). (normal) t (tangential) Figure 5.11: Definition of normal and tangential voctors. adh seginent nodes of each surface will be checker peuctration of the sogments of the opposite surface.uo Discontinuum Mechanics 5.4.2 Global Search Algorithm "The first step is to find the closest node m on surface 2 (master) to node ‘son surface 1 (slave) (soe Figure 5.12). This is achieved by evaluating the distance to each node, and finding the minimum value: lon =rnin ia) = 2B], maine. 4D) where ‘2, ls the numberof nodes on surface 2 ‘The global search is numerically very expensive, and i unless itis proved to be necessary. Fortunately, for many applications such fas 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 lst. segment m7 \ ‘closest node to $ Figure 5.12: Finding the closest node to node s. 5.4.3 Local Search Algorithm Chapter 5. Continuum Based Normal Contact Algorithms un segment m segment m1 m ml wT ™ Figure 5.13: Finding the segment in potential contact with node 8. Pma'tm 20 (8.50) then there is a potential contact with segment m, and if Tme tat 20 1) there will be a potential contact with segment m —1, Figure 5.13 depicts the above mentioned cases based on the sign of the inner product of the position and tangential vectors, : Ifonly one potential contact is encountered for point + we use the stan= dard case of contact with single segment, otherwise the special case of cone tact near a corner has to be used. 5.4.4 Normal Contact Interaction ‘To compute the normal contact force for preventing the penetration of & contact node into the associated eontact segment (target), a standard step, by step procedure is followed. ‘The ist step is to check for penetration. ‘The normal vector is defined by: neesxt (6.52) and the penetration distance is computed n= Fimo (6.53) and the penetration occurs if gy <0.33 ss u2 Discontinuum Mechanics Figure 5.14: Local coordinate system and projected contact point. ‘The second step is to locate the contact point. By defining a local natural coordinate aystem € € (0,1), and a local segment numbering system according to Figuee 5.14, the local contact point is defined a Hler-a)-t (654) ‘The last step is to compute the contact force and its associated stiff ness, The model is constructed by inserting the penalty spring between the contactor node of body 1 and the target segment of body 2, as depicted in Figure 5.15. equivalent spring system ~S~__(activated after penetration) igure 5.2 Equivalent penalty linear spring. Chapter 5. Continuum Based Normal Contact Algorithis 3 ‘Tho contact coustenlut will Involve the contetor nose saul the two nodes (1 and 2) on the target segment, leading to: Kote = f° (5.55) where We = [thy t,t] (556) ‘The magnitude of the contact force in the local system is defined by: f= 09% (6.57) where a is the normal penalty coefficient. In terms of global variables, f° becomes: f= amNs (6.58) n Nas | -NilG)n (6.59) —Na(E)n where Ny = 1—€ and Nz = € are the target sezment shape functions. Figure 5.16 shows the geometric representation of the contact forces. Tt is worth noting that in any case, the angular momentum of the body should not be changed. ‘The contact stiffness is then defined by: K.=aN,NT (6.60) ‘The penalty stiffness matrix, defined in (5.60) is not consistent; ie. it does not contain the second order terms. Therefore, the described formule ton does not converge quadratically, or at all in some circumstances. Contact Near a Node A node to edge based contact problem which happens near to a comer node, of in more general form of multinode «tacts near the comers,es 1 c &3 € c aa Discontinuum Mechanics i | i 7 Se t Figure 5.16: Distribution of the contact forces, Figure 5.17; Oscillatory behaviour around the corners. pose numerical ambiguities as well as numerical instability in particular ‘conditions. For instance, there are always cecllations between two corner nodes in iterations. Actually, we may have oscillations even without having any real penetration, Figuce 5.17 shows a typical situation where during the analysis, pint 3 may seem to be penetrated through its opposite segment, causing the contact forces to be generated to return the node to an unpenetrated postion. At the same time another force with opposite direction is acting: ‘bn node 1 to push it out of the region occupied by body 3 near the corner polut 2. This causes oscillation of the systems around the corner points ‘There exist remedies to treat this set of problems which may frequently happen in progressive fracturing phenomena, for which several small parts of s fractured and fragmented body may interact with each other simulta- neously. One simple remedy is to slightly enlarge the end segments, so & node to edge contact can be performed without eny ambiguities. However, Chapter 6. Continium Based Normal Contact Algorithius 15 performing a costly this is not a practical way in general problems, accurate local search is unavoidable. 5.4.5 Consistent Node to Edge Contact It has alroady boon pointesl ot that the almplo penalty wetted will Hud to f stiffness matzix which is not in a consistent foru. By vsing te prinelple of minimization of the total potential energy of the system, including the Contact constraint, a consistent form for the siffiucss matrix can be derived. 'As described earlier in Chapter 3, the equilibrium condition can be ex: pressed as: (6.61) Considering the contribution from the contact constraints alone, the tangent matrix is obtained by further differentiation of (5.61) with respect to the nodal variables, Therefore, the force vector and the tangent stifiness matrix for a single element are defined by: foot (6.62) o| (2) + eetde | (658) ‘The first teem in (5.63) corresponds to the stiffness matrix (5.60). Differentiation of the gap function ‘To evaluate Equations (5.62, 5.63), derivative of hiv gap function g has to bbe computed. Here, the same methodology as described by Crook [82] 18 followed. Referring to Figure 6.18, the gap function is defined as: gam (a ~ 2) (564) "The first variation of (5.64) is: 6g = 1- (Sty — 5t) + 5m > (a5 ~ 23) (6.65)c C3 €2 3 16 Discontinuum Mechanics Figure 5.18: Defining the gap function. Itis worth noting that we did not consider the change of n (67) during ~ the contact in previous approach. én can be expressed as: neext (5.66) fn =e xt (6.67) and for the tangential vector t, tahoe) or) and it =} (6uy — dun) - 46l(a — 21) (6.69) =} 1) ~ jp 8lla2 — 2 7 ‘To.evaluate di, we have : P= (@2~ a) - (a2 - 21) (6.70) sm Hes aifbun 60) am) it = j(bus— Su) ~ 75(@2-21)(6ua ~ Sun)(aa— as) (6.72) Chapter 5. Continuum Based Normal Contact Algorithins ur Using (a-bJe = (¢@ b)a, will lead to: it = N bus — bus) — J lle — 21) @ (22 ~ 21)}(6ua— Hun) (679) and finaly st = FU te Gur ~ bu) (674) Substitutions of (6.74) in (6.67) and then in (5.65) gives, Fe, — 21)(t © n)(5ue — Sus) (6.75) Sg =n (buy —6u1) Noting that = $( ay a1) ty (6.78) becomes dg =m: (Bu ~ (1~ &)6u1 ~ &:52) (6.76) on, ig = m- (54g ~ Ni (Ge)Bun ~ NalGe)6u2) (e771) ‘Therefore, an expression identical to (5.58) is derived for evaluation of the contact force vector: =n [ =NilGojn | (6.78) —NalEelr ‘The first term in the consistent tangent: matrix (5.63) becomes: sux. = ofa (32)” 3 a0 om) SuK hu = eviul(buy—Nidus — Nodus):(n@n) “(bug ~ Niduy — Nba] oy18 Discontinuum Mechanics By using the definition of N, shape function, the linear part of the stiffiaess anatrincithen defined by’ ines oN,NT a1) which is identical to (6.60). "To derive an expression for the nonlinear stiffness torm, the second va ation of (6.75) is derived, which will eventually result in the following con- stent fore: TNT NgTT+2NuNE)] —(6.82)( _ 0 we-| "| 6s 0 Tm =| -t | (5.84) ‘ t n [is ee) Mt Note that for zero 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 nake of convergence we need this term, 5.5 Edge to Edge Contact Figures5.19 illustrates a simple contact problem defined in terms of edge/edge contact couples. Although it is not as popular as the node to edge contact, it has been used by many research groupe working on general erack proper ‘ation problams by discrete clement technique. Utilizing the Boussinesq solution for the concentrated forve acting ou an ‘elastic half space in combination with the Winkler elastic foundation theory Chapter 5. Continuum Based Normal Contact Algoritivus ne Figure 5.19: An edge to edge contact. 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 lnlf space. Petrinic [1] has adopted and successfully implemented this methodology to analyse various solid/granular Interaction problems. According to [t] the ‘vertical displacement of 1 general surface point is efines! by tun(2) ~ tn(O) = (ha + ha) (8,2) (6.86) ‘where py is the intensity of the normal surface pressure at the other end of the contact surface, bis half the width of the coutact surface and $042) =P nw) —Z nto (2 -9F) were) eat—a) aN hale ‘ee (5.88). nate120 Discontinuum Mechanics Figure 5.20: Description of an edge to edge contact. where G represents the shear modulus and v is the Poisson's ratio. ‘The normal contact stiffness can also be derived ak e : 1 Tate area) | tnd ally the normal contac fore ean be evalsted Fe" = oust) 9) Pad = KO tae 69) bag = KS ots Chapter 6 Frictional Contact 6.1 Introduction In this chapter, the main features of computational frictional contact al- gorithms are discussed. ‘The chapter is divided into five parts. The first section, explains the basics of frictional contact problems including the kine- matics and principles of the plasticity theory of friction. ‘The second 8e0- tion, provides a detailed formulation for two dimensional frictional models ‘without frictional hardening. ‘The third section generalizes the approsch to three dimensional models and the fourth section will introduce frictional hardening within the 3D models. Finally, a 2D node to edge model and a 3D node to quadrilateral face model for Coulomb frictional contact without hardening are presented. 6.2 Frictional Contact - The Basics In this section, some of the fundamental aspects of a frictional contact. be- haviour ere discussed and some Of the principal issues regarding the math- ‘ematical formulation of the computational methods are aciiressod. 6.2.1 Frictional Contact Model Because of microscopic nature of frictional contuct, including the mlero- scopic geometrical conditions shown in Figure 6.1, friction can not be easily modelled adequately in the macroscopic numerical analysis of many appli- cations auch as metal forming processes. waca a ca kd a fame} 12, Discontinuum Mechanics Figure 6.1: Microscopic nature of frictional contact. ‘The frictional sliding is simulated by assuming mathematical relation- ships between contact stresses and other related state variables, which de- scribe the contact conditions. These relationships can be detived from the tmicromechanical models of fretion and then incorporated into an appropr- ate constitutive theory (84). At frst, only a quasi static type of friction is considered, which occurs, for exazaple, when two metalic surfaces are slowly pressed together and relatively displaced in a quasi static equilibrium (85). ‘The most conspicuous mechanisms contributing to this type of friction plastic deformation of asperities. ‘+ formation of elastoplastic junctions on the contact surface. + possible fracture of junctions. + strong coupling between normal and tangential plastic deformations. «dissipative mechanisms caused by the junction deformation. ‘Thermal coupling aspects of friction, which recently received consider- able attention in the computational literature, will not be considered. Thelr inclusion, however, is possible without any major modification in the con- text of the general thermo-mechanical coupling strategy. 6.2.2 Basic Kinematics In this section, basic kinematics and geometric assumptions required for lesiving the necessary équations governing a frictional contact Lehaviour fare explained, Figure 6.2 shows a typical metal forming problem in two Chapter 6, Frictional Contact 133 Master Master Inidat Current Configurasion Figure 6.2: Contact of a rigid master and a deformable slave solid. configurations. In the reference of Initial configuration, the objects ore yet to make any contact, while the master and slave objects are in contact in the ccurtent configuration, In metal forming application, the tool is regarded fs @ rigid master body and the workploce is treated os n deformable slave body. It should be noted that, however, in « general contact elgorithea, the role of master and slave bodies are interchangeable and the 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 configuration, showing both relative normal and tangential displacements. In this figure: nn; outward normal on the master surface t+ tangent orientation € + dizeetional coordinate along the minster surface Jy + normal gap 9, + tangential gap where € is measured from a fixed point on the master, and gy and gy F@ ‘wo independent Kinematic variables. “At the current configuration, the position of slave node , 5, is defined as=Xstus (6.1)124 Discontinuum Mechanics Pigure 6.3: Relative normal and tangential displacements of two contacting, bodies. whore ay and Xs denote the current and initial positions of slave node rrrr—es—E configuration. Point M geometrically represents the closest projection of $ onto the ‘master surface Ty lew —@s|0 THEN ser {fom (aii) (#8) } coro 31 ENDIF ENDIF Hea166 Discontiauum Mechanics DpH ~on(non)—abhiaetr—non—t08) o, [BEE] eon) m (6.207) are DP isthe consent tangent moduli, and ‘Hs fnseall or «- (toilfsel ar) (38), , Masel 6208) 6.6 Node to Face Frictional Contact Models 6.6.1 2D Node to Edge Model Based on the work by Crook (52), the necessary formulae for a two dimen- ‘Sonal node to stzaight edge frictional contact problem are provided in this fection, though the general pumerical formulation has previously been dis- cussed in detail. Figure 6.9 shows 8 we ypleal node to edge contact problem in a two dimensional space. From the geometrical description of the problem, at time step j +1 & e=te-a)-t (6.200) hy (6.210) gegen = 9+ (Ban Elastic Tangential Contact Beginning with clastic tangential contact, the internal foree veetor can be defined as: oe ft (2) Chapter 6, Frietional Contact 167 timestep j+1 peo Figure 6.9: 2D node to edge frictional contact nga 88, fa = Onde (on) man, fe= aun (6213) and for the stiffness matrix, (6.214) in = On (( (6.218) enon (3B (ox ‘Using the same methodology as described in Section 5.4.5 for evaluating, the derivative of the gap function, wil lead to, si aunt {64 8a} oan Kwa as nee (6218)3 E ca 108 Discontinuum Mechanics where =|t -G-@t -@] (6.219) Th=[0 -t t) J (6.220) In -G-én —én (6.221) NE=(0 -n n] (6.222) (6.223) Ba (N+ NnNT)~ (CR + TaD) ~ HATaNR + ate)} (6.224) When gq = 0, the nodal forces are defined by distributing the local tangential force using the shape functions of the master segment and trans- forming to the global gystem using f. Coulomb Frictional Contact ‘The classical assumption of Coulomb friction, as stated in Figure 6.10, is tthat the body is at rest until the maximum shear force is reached. However, in reality, bodies exhibit some relative movement within the contact ares prior to sliding. ‘The main equations can be summarized as: anette (6.225) elastic region: fi = axgf (6.226) sip condition: © =|fil ~ fn (6.227) (6.228) (6.229) Chapter 6. Frictional Contact 169 upcenso ioe Te Figure 6.10: Coulomb friction law. ov ay samt) (6.230) ‘The stress update procedure may then be fllowed as (see Figure 6.11): doe= (G1 - " (6231) SGA fy todo. (6.232) PH = MAGAM — Hf <0 (6233) Jose = Mfg S0(F558) (6234) ay = © [GSU - ufos] (6238) Add = Sista hegsi) (6.236) ‘The consistent tangent matrix, which is non-symmetric, may then be defined by: a te (eas ia Ken (6238) resulting in: Khe (6.239)n oo ca ea J Cc £3 £3 Ea 170 Discontinuum Mechanics Mae Mfas 0 where (6.240) 6.6.2 3D Node to Quadrilateral Face Model In this section, « three dimensional node (slave) to four noded quadrilateral face (master) frictional contact model is presented. This is basically based ‘on the material provided by Crisficld [107]. Frictionless Contact Geometrical description af the model is depicted in Figure 6.122. In this ‘model, $ represents the Slave node and element 1234 is the quadrilateral finite clement. Cis the point immediately below S on the master surface. It is important to note that the contact surface, in general, does not lie in a plane. {tis not a trivial matter, as itis in a 2D case, to find the (E,7) values aswociated to point C with r(6,n) position vector. Using the standard shape functions (é,7), associated with node s, the position vector r may be expressed by the nodal positions, a, Chapter 6, Frictional Contact = - am 9) Geometrical description, ) Basic weetors. . €) Initial and current configurations. Figure 6.12: Node to quadrilateral face frictional contact,s 7 Ce ee | a E 1m 8 iscontinuum Mechanios rn) = ME nas (6.241) where the shape functions Nj, are well defined: M=40+90+n) M=Ht-O0tn) (6.242) Ns=ja-90-) Me=}0+90-) and NNT = (Na, NayNay Ma) (6.243) Nb= GH = HG+n-C +m. -0- 0), 0-m) nt= @M) {(14+8,0-8,--9.- +O} (6-244) N= GMD = rand the vector connecting to its estimated projection onto the quadrilat- eral surface, C, 9, 9 = te—r(En) = Be News (6.245) ‘Ones the correct value of (€,9) is known, the vector g will be orthogonal to the two vectors re and Ty which lie on the master surface (See Figure 6.128), Le. gy. By defining new variables ag and aq, we may derive an equation to define n} r ac=rto= (3) 0 = (646) (6.247) (Chapter 6. Frictional Contact 173 arn (% batre ral = Jo (6.248) where J is the jacobian matrix at (€n)- Back to Equation (6.241), and using the variational mathematics, br = AKA SO NSU, (6.249) where, at = (af 27,29, 23,27) (6:250) wT = (uf uf uf uu) (6.251) = 0) (6.282) then using (6245), bu — ABE (6.253) where H = {0,-NiL,~Nal, -Nol,-Nal] (6254) nd forthe remaining varibles: sre = itebu+ (Sew) (6288) rg = Hybu t+ (Nets) 8 (6.256) bag = rP (Hou — Abe] + 9TH edu + (97 req)5n (6.257), 3 Say = 1 [Su ~ ABE) ‘eH + (97 ren )5E pe where 2h 05 alasca co aes (a ea 65 € e9 a4 Discontiawum Mechanics He = (0, Ned Nal. Nel, Nel] (6.259) Hy = (0,Noil, Nya, Nool Nos] (6.260) In practice, from the contact detection, we have an,gstimate for position of C(E.n) which we may wish to improve. By fixing the nodal coordinates, Su =0, and adopting a Taylor series: (Aua)" + D5E=0 with D -aTasorre) |} a] (6261) which can be used in a Newton-Raphson scheme to find an improvement 5§ land therefore, to update the initial estimates for (6,7). By finding the varintion of gq, the residual (internal) force vector 1s | calculated | | waao (6.262) nbon = 9769 (6.263) tyr bin (2s) ig =n" |Hbu— ABE) (6.264) where isthe normal vestor tothe surface atthe contact pont Knowing A™n=0, Hu = bu Hn (6.265) 50 and following the definition ofthe residual force vector, = i rules po = fSHTn = Onin (6.266) ‘The consistent tangent matrix is then defiuel by finding the derivative Chapter 6. Frictional Contact 15 Gr? = andguH™n + OngnSH™ + Ongn ET ine (6.267) so 6H and én are to be determined. For determining dr we hu ‘ase isi (628) in lt = nn" lig (6.269) and from (6.248) and (6.253), we ee, [ATH + onGbu+ DIE =0 (6270) where « S o-[&]. (eras amy ‘Therefore, 6¢= DOYTSu (6.272) 6g =|H - AD“ Y" Su (6.273) with ; YT =-ATH- 9G (6.274) ‘Therefore, bn can now be determined: : dan? Hbu (6.275) H+ AD“ATH + 9967]] 6u— bao in “and to calculate 6H”,176 Discontinuum Mechanics SH n= ~5EH Tn ~ inn (6.276) SH™n =-GTE=C™D NATH + Glu (6.277) Finally, by using (6.277) and (6.275) in (6.267), the ggnsistent tangent matrix is defined: = Kp=Ki+Ke (0278) where a Ki =0,Han™H (6.279) Kot = [HTH HT ant + HTADATH + 99H ADC +gn67D~'ATH + G7 DG] (6280) Sticking Friction Referring to Figures 6.12b,¢, which depict the initial and current configura tions, the tangential gap vector, g,, ean be defined by: of = (ga. 912] = (ex€2]"(@ ~ X) (6.281) where E = [e1,€2,€a) is the orthogonal triad based on the contact poi as (6.282) Tel eg=n (6.283) eax ey (6.284) sand X= Na (Go, 0) Xs the intial configuration, while w = J) Na(E,n) 4 is the current configuration of the contact point. ‘We will also require derivative of g, with respect to u, oF: Chapter 6. Frictional Contact, fq, = Aedé = AgD™ YT 5u where As = ened" ze20) ‘The residual force vector, is then defined by: ears ‘and finally, for the consistent tangent matrix, K.=Ki+K? with K2=0,¥DATA,D“'YT =YDUWDY"+ Kat (K)” where ’ Pak, -FTZAYD" ATE +Q- Ferhy Ht — W = Are FF 4 eB + (BaP rey + (Pa) 24n) | $ and , im (6.285) (6.286) (6.287) (6.288) (6.289) (6.290) (6.291) (6.292)a wie Ee = | = " w 118 Discontinuum Mechanics — (6.295) Fan Af® (6.296) Paslevelfi (6297) oO Qc = HIF. o-[% }. (Gcake: ber Gey =n" Hey (6.299) Hey = (0,Nenl, Neval, Noga Nened] (6.300) B= ff Het TH, (6.301) Sliding Friction Finally, the mumerical formulation is extended to incorporate sliding fric- tional behaviour, Assuming aCoulomb type of sliding friction, b= Nfl tafe (6.302) where (J2)" = (ffi Sf)» For (+1)! stress update iteration, Figar = Fig t+ ule — Bagh) (6.303) Fig = (: - itty) eens (oan rage = Ing + ADn (6.305) For nonassociated slip rule, Ag? is normal to the cylinder ft. = Constant, hence, Ag}, = SgP"l, From & = 0, we get, Chapter 6. Frictional Contact —nfgiriet wer ‘Therefore, the internal force vector is defined by: r= BTS) fg = Bou and the non-symmetric modular matrix, Ce, 08 Faced with, “spit sity Faa=nrf Fase = Onda ‘The tangent stiffness matrix, finally is defined by: K=BICB+K? + KE 179 (6.308) (6.307) (6.308) (6.309) (6.310) (e311) (6.312) (6.313)ey Chapter 7 Discontinuum Contact Mechanics : 7.1 Introduction ‘The pioneering work by Cundall and his colleagues, who completed the orig inal 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 Distinet Blement Method. By advancing the capabilities of the finite element method, and inerens- ing power of computing facilities, fully deformable blocks replaced the orig inal rigid bodies, with the new Discrete Element Method terminology. Nowadays, the discrete element method has renched to wn over increasing, popularity for modelling all potential discontinuin media, Nevertheless, 1 {is mainly used for two classes of problems: «© Granular flow: where large numberof simple elements (usually rigid) ‘are interacting with each other and with the surrounding boundaries (tigid or deformable), Granular flow in silos and he slope stability ‘analysis are the most attractable types of problems in this class. + Progressive fracture: where a continuum is subjected to an extremely ‘high condition such as explosive loading or high velocity impact, cous- ing extensive cracking and possibly fragmentation. The behaviour of ‘the model is continuously changing toward the discontinuity and the ‘original geomet of the body is changing by the extension of eracking. 181 wha12 Discontinuum Mechanics ‘The essential point is that the finite element method is rooted in the concepts of continuum mechanies, thus not suited to general fracture prop- ‘gation and fragmentation problems. The finite element method may only effctively desl with a single crack or a low fractured area without any frag- ‘mentation, whereas the diserete element method is specifically designed to solve problems that exhibit strong material and geometrical discontinuities. Before dealing with the main issues, © quick review of historical devel- opments and present industral/sientifc applications is provided. 7.1.1 Historical Development [As mentioned earlier, the original development of the discrete element method may be attributed to the work by Cundall in 1971. In the following, a brief review of the win historical developments of the method ig provided: ‘© 1968 : Analysis of jointed rocks by Goodman {108}. #1071 : Analysis of jointed rocks by Cundall (109). + 1088 : Bully deformable discrete elements included (Ghaboussi [110)). ‘© 1090 : Beginning of large scale simulations. ‘+ 10995 : Combined finite/diserete element method for fracture simula- tion of brittle nein (10). + 1005 : Coupling disrete elements with uid or gs fow [123]. + 1006 : Parallel and object oriented computing 112, 24) ‘+ 1906 : Modklling granular low in slo 1) + 1008 : Metal cutting using adaptivity techniques [34]. ‘+ 1998 + Impact analysis of anisotropic three dimensional composite shells [89] ‘+ 1099: Damage investigation and repair modelling of masonry struc- tuures/bridges Te should be noted that for each case, earlier less sophisticated models ‘an also be found in the literature and the mentioned years show the time of major advancements of the method. a Chapter 7. Discontinuum Contact Mechanics 183 7.1.2 Applications ‘The diserete element method has found extensive attraction within various ficlds of the industrial and sceatifie applications. The main categories are: 1. Geomechanical applications = Jointed rock mass [109]- - Rock blasting problems [18, 11}. + Underground structures [114,115] + Falling rocks (30. ~ Simulation of sands (116). Porous media [82] Granular materials = Disk and sphere elements (117, 33, 3: Bilipsoal elements (118 ~ Superquadric elements (32. 4. Industry ~ 8D printing 5. Impact analysis (progressive fracturing) = Reinforced concrete slabs (119) = Metal cutting (34), - Shock/thermal initiation of explosions [36 = Composites [39]. 6. Motion planning in robotics {120} 7. Computer Graphics + Animation (121, 122, 123). + Virtual reality (124), 7.1.3 Specifications ‘Table 7.1 briefly provides comparative information about the principle speo- ifleations of continuum (finite element) analysis and discontinuum (discrete clement) analysi ‘ Princtpal components of » geueral discrete clement simulation an be classified into the following five groups:I ca es ca co c=] 1st Discontinuum Mechanics ‘Table 7.1: Comparing continuum and discontinuum modellings. geometry continuum only finite elements = explicit definition of contact surfaces «+ discontinuum = finite elements (FE) and rigid Bodies (RB) + automatic evaluation of contact surfaces slobal search + simple algorithm ‘+ discontinuum += must be efficent for large numbers of bodies, + must be efficent for dense/loose packs = must be applicable to FE and RB local search * continuum node to node contact search + simple surface node to element edge contact + disontinuum + complex algorithm (arbitrary shaped bodies) + ferent algorithms for diferent type of RB contact Interaction ‘continuum Interface elements + node/node, node/ace and face/face contacts + penalty and various Lagrangian formulations + discontinuum = node/node, node/face and face/face contacts = penalty formulation for FE bodies ~ loced form solutions for rig elements + energy conserving, disipating algorithms solution > continuum + explicit of implicit dynamic solution + iscontinuum ~ only explicit dynamic solution visualization "> continuum ~ simple deformed geometry visualization © discontinuum + advanced graphical postprocessing + continuously changing geometries ‘Chapter 7. Discontinuum Contact Mechanics 185 1, Object representation 2. Contact detection 3. Contact interaction 4, Physics of objects 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 affgct the material modelling (plasticity) within the finite element procedure, and may also cause cracking/fracturing which will be dealt with in the next chapter. Finally, visualization is a necessary tool for a successful discrete element analysis. Changing geometry and mesh, ‘reation of new bodies, and checking the interaction stresses/forces require 2 powerful visualization tool, without which, the huge bulk of result date may not ever be verified and discussed. However, itis a matter of computer graphics and CAD/CAE systems and is out of scope of this book. 7.2 Contact Detection In this section, the contact detection procedures are briefly reviewed and ther main ecvantago/disedvantage points are disuse (125 ‘Then, the alternating digital tee, as one ofthe fastest goomtric intersection search algorithms, ate explained in detail and its applicetion to general contact detection problems will bo reviewed by providing sample problems. ‘The problem of detecting the interacting bodies in multi-body analy- seo, also known as the geometri itartection Search, has become & malor ‘computational research top. Assume there is system of 1 interesting bovis; all may happen to come into contact with any other body. A nalve cuitact detection method equies the checking for contact between each body and every other body ‘within the system, Figure 7.1 shows how such a simple approach requires 2 checking link between each (target) body and the remaining (contactor) bodies. ‘The numberof operations required to detect all contacts between bode wll then be proportional to ne (rayi a Figure 7.2: Discontinuum Mechanics Short lists of contactors. Chapter 7, Discontinuum Contact Mechanics 187 {In multibody analysis, however, the above method obviously becomes extremely expensive. Several other algorithms have been proposed to im- pprove the detection procedure. In the best case, the computational efforts hhas been reduced to a factor of Nlog,(N) (7a) "The existing detection methods have so far laid in between the to extremes, ‘The alternating digital tree (ADT) algorithm, which developed initially to solve the problem of mesh generation, reduces the number of operations required to deterntine the contacts between bodies by creation of short lists ‘of potential contactors for each target body. Figure 7.2 shows « sample part of the created short list for a set of NV contacting bodies. In this case, a direct checking is undertaken for the number of relevant bodies of a target, land the procedure is repeated for other target objects. 7.2.1 Contact Geometry Depending on the type of modelling, two types of discrete elements may be defined: 1. Rigid bodies (Figure 7.3) Ow re Figure 7.3: Simple rigid diseret ements, 2. Deformable finite elements (meshed polygons) (Figure 7.4) ‘The contact geometry is then either computed from the input definition ofigid bodies, e.g. acireular disk is defined by a centre point and radius, oFn [ae 2 63 ta a & ime Ea 188 Discontinuum Mechanics Figure 7.4: Deformable discrete elements (meshed polygons). automatically eveluated for deformable finite element bodies by evalusting all exterior edgee/facets and grouping them for each discrete element, as depicted in Figure 7.5. - Contact Geometry TY]i ad aa “ wee 24 « [ope A 67 ' 78 7 88 ssa} . » Popes 23 4 a : 562 Figure 7.5: Definition of contact geometries for discrete elements. 7.2.2 Global Search Algorithms A general global search algorithm must be etticient in dealing with a large number of bodies, suitable for both rigid and deformable bodies, and eff- Chapter 7. Discontinuum Contact Mechanics 189 Figure 7.6: A system of rigid bodies (disks) moving in a domain. cent for both loose and tight packs of elements. A single approach might not fachieve all the mentioned goals, and different approaches may be adopted for different applications. In this section, different search algorithms will be briefly reviewed and some of the important associated techniques will be addressed. Direct Checking system of 1V interacting bodies; all may happen to come any other body (Figure 7.6). “‘Thesimplest approsch for detecting the potential contacts, will check 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. [As stated before, for each body, N ~ 1 potential contacts have to be assessed, Therefore, the total number of operations (assessinents) required to detect all contacts between NV bodies will be proportional to (73)190 Discontinuum Mechanics Figure 7.7: Defining a potential contact zone for a system of moving bodies (disks), Direct Evidence Search In this method, the number of contact node searches for each target sur face is reduced by forming a reduced set of nearby contact nodes, thereby reducing the total number of contact searches, This reduced set (shortlist) has to be periodically updated. Figure 7.7 shows a simple way of defining 1 buffer zone for creating a reduced set of nearby nodes. ‘Afterward, only the assoctated 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 AAs indicated in Figure 7.7, a buffer zone may be defined for each target. In ‘caw: of moving disks, the Buffer zone may be defined by a clrele with radius hy, Chapter 7. Discontinuum Contact Mechanics 191 Figure 7.8: Moving disks within a buffer zone. ‘Therefore, the first step in such a search algoritiun is to define the buffer zone for each target, and then finding the disks inside the specified zone (Pigure 7.8). By knowing the velocities at each time step, we may approximately ‘evaluate the distance in which a disk can travel in next 7 time steps. Ac- cordingly, the need for a new glo6al search for finding nodes inside each new zone.can be decided from the leaving/enteriny any nove from/into the If the maximum velocity of an object in consecutive time steps is Umass ‘then as a safe estimation of the minimum number of time steps for that object outside the buffer zone of a target to reac the target. will be: Nat (ra) * Teed ‘Therefore, no contact is expected to happen before Nar time steps and the global search to locate all contactor objects within a target buffer zone is performed every Na time steps ‘The seme concept is also used to enable direct evidence checking in as- sociation with meshed polygons and mare complex object shapes. Different rules for forming « buffer zone ean be utilized depending on the object shape sand type of epplication:2 3 192 Discontinuum Mechanics Buffer Zone Buffer Zone La Figure 7.9: Definition of the buffer zone; uniform expansion of the mesh boundary (left), and a rectangular box (right). Figure 7.10: Effect of contact probability on the shape of buffer zone. 1. Uniform expansion of a faceted discrete clement formed from de- formable finite elements (Figure 7.9) 2, A rectangular box (Figure 7.8) 53, Bufler zone formed to account for direction of motion (Figure 710) 4. To avoid a large number of node to edge local searches, # segment Dased bulfer 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 twa size of the buffer zone, the complete contact detection provedure ust be performed agai Chapter 7. Discontinuum Contact Mechanics 193 on Figure 7.11: Forming the buffer zone based on each segment geometry. In order to minimize the total time spent on updating tie short Ret of contactors, the size ofthe butler zone should be defined with respect to the actual size ofthe bodies under consideration. Some reports proposed an ‘optimal size equal to the average sizeof all bodies, although its applicability to general meshed polygon systems isin doubt. IEF the relative motion of the bodies is small, «small buffer zone will be most efficient. ‘The faster the motion of elements, the larger will be the necessary size of buffer zone. Flying Through ‘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. ‘Based on the dimension of the buffer zone, di, te time step increment3 ca et 3 194 Discontinuum Mechanics Pigure 7.12: Flying of a contactor object through the buffer zone of a target. must satisfy, ars (78) where tinge is the maximum velocity observed in the system. 7.2.4 Contact (Field) Zone ‘As consequence of satisfying the time step reduction (7.5), itis 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 Eblution for multibody contact problems can not necessarily be guaranteed. For example, consider Figure 7.13, where after the impact, point D has penetrated the body BAC. However, by only monitoring the displacements, the program can not simply decide whether the point D has penetrated though the segment BA or CA, The same problem may arise near the comers, where several paths seem valid for the penetration of the point into the body. [A contact zone (field zone) is now introduced (Figure 7.14) in order to gain further control over the accurecy of the numerical solution, ‘Taking into account both the buffer and field zones, Equation (7.5) 18 modified (1), Chapter 7, Discontinuwm Contact Mechanies 198 Fue 7.18 Ambiguities of the wy pint D penetrates BAC during an impact. ‘contact (field) one Figure 7.14: Definition of a contact (eld) zoneea y 196 Discontinuum Mechanics Chapter 7. Discontinuum Contact Mechanics @ 4. fte0-§) es) ‘where d's the current distance between bodies, dy isthe field (contact) zone thickness and vf is the relative normal velocity between interacting bodies. ‘The condition 7.6 ensures a gradual change of thé time step increment during contact, leaving & minimum number of time steps for treating {prneral impulse/contact problem. As a result, the accuracy of the solution Is also preserved. 7.2.5 Binary Tree Structures ‘A binary tree structure is # specific method of sorting data that allows new date to be easily added (inserted) or removed (deleted). Binary trees sre ‘one of the most important non-eequential types of data structures (126). At ‘each node, the information stored consists of data and two pointers known Figure 7.15: A simple binary tree data structure, fs the left and right links that point to further data, Bach added link can tither be equal to zero of equal to the position in memory where another node of the tree is placed. Compared to « linear sequential array, 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 fist node in the tree is known as the root node, From one node of the tree it is possible to point at most two other nodes, while for each node {except the root), there is one and only one link pointing at it. 'A node Gittoue any poner to other nodes i clled terminal node. ee Figure 715 shows a typical binary tree structure with thee levels of : ar Inforatinn and six nodes. ‘The posters on each node refer tothe memory ARS ‘Jocation for the left and right links, respectively. For example, pointer Lp . refers to the memory location that holds the set of D data, 1.e. left link to - . ihe Bet nok anode at level points toa node a level +1, then the node a evel fis RELE OS the father an the other on isthe eon, No information a a particular nie (encapt the rot) can be retrieved unles th rot is examined in advance. {fur 716 shows how the data can be stored fr defining a simple binary ik Sch ie ins was Figure a binary tre structure. tre. ‘Bach node of a binary tree can be assigned a binary number. Assume coach left brunch is recognized by a O and each right branch hy a 1 digit. ‘Starting from the root, the furthermost left digit of eepecified node number198 Discontinuum Mechanies anny aa Figure 7.17: Enumerating a binary tree components with binary numbers. is read and the branch corresponding to that digit is followed to a node of the first level. At the node of the frst 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 J by a binary number. Creation of a Binary Tree "The first step in creation of a binary tree structure is selecting a root node. ‘Adding new data items to the binary tree depends on definition of a crite- rion for choosing between the let or the right branch for insertion. Every insertion then starts by checking this eriterion at the root node and then traversing the tree until an empty place is found. ‘The criterion for insertion of date items and traversal of the binary tree isin fact a measure of relative spatial position of two nodes of the binary tive, ‘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 nearching and the insertion of new data items. Low insertion and search costs are obtained from well Dalanced trees (Figure 7.18b), whereas poor performances are expeclet! {rom highly degenerated binary trees (Figure 7.189). Chapter 7. Discontinuum Contact Mechanics 199 ) ) (>) (®) ® © O© O 1b) Well balanced tree 1) Degenerated tree Figure 7.18: Degenerated and well balanced binary trees, An optimized ordering procedure for node iusertion can be developed to consider the possibilty of balancing » tree structure by adopting @ new order of insertion. Such an optimized tree structure may be found extremely ficient if binary tree for geometric Intersection search has to be rebuilt ‘and searched through relatively often. Information Retrieval/Tree Traversal ‘A node in the tre can only be visited or examined if al its ancestors are Visited frst. However, an eficient eystematic retrieval of data requires that cach node is only visited once. Thisan be achieved with rocusive scheme ofthe socalled stay lft rule 1, Retrieve data for the node at Tevel i, 2. Ifthe left link exists visit the pointed node at level +1 3. Else ifthe right ik exista visit the pointed node at Lith 4. Else return to level i 1. Figure 7.19 illustrates this simple rule for a typical tree. ‘The order of, traversal ss A+ C—» D+ BF B+ Gi. Tho procedure can, bbe summarized as Box 7-1.200 Discontinuum Mechanics Figure 7.10: Data retrieval in « binary tree. Box 7.1 - Traversing procedure for a binary tree (Figure 7.19). 1, Visit the root A 2, Take the left child (C) and traverse its subtree = Visit © = Take the left child (D) and traverse its subtree - Visit Dj Skip (no left and right children) - Take the right child (E) and traverse its subtree - Visit EB = 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 + Visit B = Take the left child (G) and traverse its subtree = Visit G = Skip (no let branch) = Take the right child (H) and traveree its subtree = Visit H; Skip (no left and right children) = Skip (no left and right children) Chapter 7. Discontinuum Contact Mechanics 201 Programming languages such as C++ and F90 support automatic re- cursive algorithms, thus a relatively eagy implementation of the traversing technique can be performed. However, the more common F77 language does rot provide this capability and more eomplicatce implementation, such a8 the stack technique is required. In aso called stack technique the items are inserted one by one, and then ‘extracted in the reverse order one by one. During the insertion process, the ‘value of the right link is stored in order to enable the subsequent traversal of the right eubtree. In fact, a ist containing the adldress ofall right subtrees ‘encountered along the way which are yet to be traversed must be kept and continuously updated (126) , ‘The traversal cortinues until a zero left link is encountered. ‘Then, the last right link inserted in the stack list ia retrieved. The subtree rooted at this address is traversed and the corresponding right link is removed from the stack. Box 7.2 - A traversal algorithm based on the stack technique {126). 1, root_address = address of the root node 2. alae = 0 2.1 - visit the node stored at root address 22 - IF right.tink # 0 THEN + siae = size 41 + stack(size) = right-tink 23- ENDIF 24. IF leftlink # 0 THEN = root address Ieft-link - GOTO 21 25 - ELSE < IF size # 0 THEN «= root.addreas = stack(size) + size = size 1 = GOTO 21 - ENDIF 2.6 - ENDIF 2. -IF size = 0END202 Discontinuum Mechanics [A simple non-recursive implementation of the traversal algorithm based ‘on the stack technique is provided in Box 7.2 [125]. 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. ‘The last remaining point about the binary tree structure is to define ‘a methodology for removing a node. If the undesired mode is @ terminal node, itis sufficient to change to zero the corresponding link of its father. Therefore, the occupied memory will be available for new node insertion, In the caso 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 end only & simple modification of the left and right links of the father of the unwanted rode 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 its descendants terminal node H. Ifa particular application demands frequent deletion insertion processes, ‘a memory management system is necessary for the efficient implementation of tree structures so that new nodes can be placed in the memory space released by the deletion of previous nodes, With programming languages featuring dynamic allocation, this is automatically provided. F77, however, does not offer this eapabilty, and the problem should be handled by using linked list structure to record all the available memory spaces. For further details see (126) Figure 7.20: Deletion prociss iu a Vinay twee. Chapter 7. Discontinuum Contact Mechanics 203 7.2.6 Space Decomposition 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. ‘Space cell decomposition methods, in general, comprise finding the suit- able shape of a cell, space subdivision into cells, and cell enumeration. In the following, various methods of space decomposition are briefly reviewed. Coordinate Based Space Decomposition In this method, each coordinate axis is divided into finite intervals and each finite interval is given a number. Each cell is defined as a sct of points that hhave coordinates that belong to the same interval (Figure 7.21). ach cell can be enumerated by two (for 2D) or three (for 3D) numbers, Instead, a single cell number may also be constructed using» binary number structure. 7) 0? GLO iD (7.2) Figure 7.21: Coordinate based space decomposition in 2D.3 03 = 3 3 208 Discontinaum Mechanics Suppose the problem space with n bodies is divided into m cells. On average, there will be (2) bodies in each cll. To check for contacts in one cell, (%)(-% —1)/2 checks are necessary, and to check all cells m(3 — 1)/2. [An advantage has occurred because ofthe implict assumption that bod- jes in one cell ean not interact with bodies in another cell. However, in ‘above, the necessary checks for assigning bodies to particular calls are ne- losted. In addition, because bodies may span the grid between cells and ' body may move from one cell to another with slightest motion, frequent numerically expensive cell updates aro necensary. Space Bisection Algorithm ‘This method provides aifappropriate approach as the base for the node Insertion criterion in construction of a digital tree. In this method, the space is systematically divided into two parts at each level. Figures 7.22 ‘and 723 show this systematic bisetioning of a 2D space. ‘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. ASB >C4D+E+FIG+H +145 9K. i the root (here A) or the order of the points is changed, another tree structure will bbe formed, —— loeio level 1 mene level ——~ further levels Chapter 7. Discontinuum Contact Mechanics Figure 7.23: Space bisection procedure. Figure 7.24: Space bisection'nlgorithm for a typical problem,<= 3 3 @ Figure 7.25: Space bisection algorithm for a typical problem. Chapter 7. Discontinuum Contact Mechanics 207 Figure 7.26: Target search within a space bisection procedure, Figure 7.25 shows a step by step procedure of constructing a tree asso- ciated to binary space decomposition of problem stated in Figure 7.24. ‘The main edvantage of the geometric aearch based on the binary tree is that the area associated with descendants of a node at nny level i 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 90 node 0), ‘Therefore, in a contact dotection analysis, if a target does not intersect with the area associated ‘with an object, the object and all its descendants will not be considered as Potential contactors forthe target Le. they will not be included in the target shortlist of contactors. Figure 7.36 illustrates this key feature. Objects A ‘and B and all their descendanté can not be a candidate for contact with the target C (as it is clearly observed from the figure). Hextree Decomposition, Another extremely efficient and systematic decomposition approach is the hextree decomposition procedure. In this method, sets of hexagonal aggre- gates 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 sic first level neighbours. Figures 7.28 and 7.29 show the second and