Extended finite element and meshfree methods 1st edition timon rabczuk - Own the ebook now with all

Page 1


Extendedfiniteelementandmeshfreemethods1st EditionTimonRabczuk

https://ebookmass.com/product/extended-finite-element-andmeshfree-methods-1st-edition-timon-rabczuk/

Instant digital products (PDF, ePub, MOBI) ready for you

Download now and discover formats that fit your needs...

Partition of Unity Methods: The Extended Finite Element Method St¿Phane Bordas

https://ebookmass.com/product/partition-of-unity-methods-the-extendedfinite-element-method-stphane-bordas/

ebookmass.com

Computational Methods for Fracture in Porous Media: Isogeometric and Extended Finite Element Methods 1st Edition Renã© De Borst

https://ebookmass.com/product/computational-methods-for-fracture-inporous-media-isogeometric-and-extended-finite-element-methods-1stedition-rena-de-borst/ ebookmass.com

Finite Element Method: Physics and Solution Methods Sinan Muftu

https://ebookmass.com/product/finite-element-method-physics-andsolution-methods-sinan-muftu/ ebookmass.com

Why We Fly Kimberly Jones

https://ebookmass.com/product/why-we-fly-kimberly-jones/

ebookmass.com

Jewish Martyrdom in Antiquity: From the Books of Maccabees to the Babylonian Talmud Yair Furstenberg

https://ebookmass.com/product/jewish-martyrdom-in-antiquity-from-thebooks-of-maccabees-to-the-babylonian-talmud-yair-furstenberg/

ebookmass.com

A History of Genomics across Species, Communities and Projects Miguel García-Sancho

https://ebookmass.com/product/a-history-of-genomics-across-speciescommunities-and-projects-miguel-garcia-sancho/

ebookmass.com

LOGAN Punished: A Dark Mafia Romance (New York Mafia Vengeance Book 2) Iff

https://ebookmass.com/product/logan-punished-a-dark-mafia-romance-newyork-mafia-vengeance-book-2-iff/

ebookmass.com

The great conversation: a historical introduction to philosophy Eighth Edition Melchert

https://ebookmass.com/product/the-great-conversation-a-historicalintroduction-to-philosophy-eighth-edition-melchert/

ebookmass.com

Gandhi and the Psychology of Nonviolence, Volume 2: Applications across Psychological Science 1st ed. Edition V. K. Kool

https://ebookmass.com/product/gandhi-and-the-psychology-ofnonviolence-volume-2-applications-across-psychological-science-1st-ededition-v-k-kool/

ebookmass.com

https://ebookmass.com/product/modernism-and-the-meaning-of-corporatepersons-lisa-siraganian/

ebookmass.com

EXTENDEDFINITE ELEMENTAND MESHFREEMETHODS

EXTENDEDFINITE ELEMENTAND MESHFREEMETHODS

TIMONRABCZUK

BauhausUniversitätWeimar

Weimar,Germany

JEONG-HOONSONG

UniversityofColoradoatBoulder

Boulder,CO,UnitedStatesofAmerica

XIAOYINGZHUANG

TongjiUniversity

Shanghai,China

LeibnizUniversitätHannover

Hannover,Germany

COSMINANITESCU

BauhausUniversitätWeimar

Weimar,Germany

AcademicPressisanimprintofElsevier 125LondonWall,LondonEC2Y5AS,UnitedKingdom 525BStreet,Suite1650,SanDiego,CA92101,UnitedStates 50HampshireStreet,5thFloor,Cambridge,MA02139,UnitedStates TheBoulevard,LangfordLane,Kidlington,OxfordOX51GB,UnitedKingdom Copyright©2020ElsevierInc.Allrightsreserved.

MATLAB® isatrademarkofTheMathWorks,Inc.andisusedwithpermission. TheMathWorksdoesnotwarranttheaccuracyofthetextorexercisesinthisbook. Thisbook’suseordiscussionofMATLAB® softwareorrelatedproductsdoesnotconstitute endorsementorsponsorshipbyTheMathWorksofaparticularpedagogicalapproachorparticular useoftheMATLAB® software.

Nopartofthispublicationmaybereproducedortransmittedinanyformorbyanymeans, electronicormechanical,includingphotocopying,recording,oranyinformationstorageand retrievalsystem,withoutpermissioninwritingfromthepublisher.Detailsonhowtoseek permission,furtherinformationaboutthePublisher’spermissionspoliciesandourarrangements withorganizationssuchastheCopyrightClearanceCenterandtheCopyrightLicensingAgency, canbefoundatourwebsite: www.elsevier.com/permissions.

Thisbookandtheindividualcontributionscontainedinitareprotectedundercopyrightbythe Publisher(otherthanasmaybenotedherein).

Notices

Knowledgeandbestpracticeinthisfieldareconstantlychanging.Asnewresearchandexperience broadenourunderstanding,changesinresearchmethods,professionalpractices,ormedical treatmentmaybecomenecessary.

Practitionersandresearchersmustalwaysrelyontheirownexperienceandknowledgein evaluatingandusinganyinformation,methods,compounds,orexperimentsdescribedherein.In usingsuchinformationormethodstheyshouldbemindfuloftheirownsafetyandthesafetyof others,includingpartiesforwhomtheyhaveaprofessionalresponsibility.

Tothefullestextentofthelaw,neitherthePublishernortheauthors,contributors,oreditors, assumeanyliabilityforanyinjuryand/ordamagetopersonsorpropertyasamatterofproducts liability,negligenceorotherwise,orfromanyuseoroperationofanymethods,products, instructions,orideascontainedinthematerialherein.

LibraryofCongressCataloging-in-PublicationData

AcatalogrecordforthisbookisavailablefromtheLibraryofCongress

BritishLibraryCataloguing-in-PublicationData

AcataloguerecordforthisbookisavailablefromtheBritishLibrary

ISBN:978-0-12-814106-9

ForinformationonallAcademicPresspublications visitourwebsiteat https://www.elsevier.com/books-and-journals

Publisher: MatthewDeans

AcquisitionEditor: BrianGuerin

EditorialProjectManager: IsabellaC.Silva

ProductionProjectManager: SuryaNarayanan Jayachandran

Designer: MarkRogers

TypesetbyVTeX

Preface xiii Nomenclaturexix

1.Introduction1

1.1. Partitionofunitymethods1

1.2. Movingboundaryproblems6

1.3. Fracturemechanics8

1.4. Levelsetmethods10

1.4.1. Implicitinterfaceandsigneddistancefunctions11

1.4.2. Discretizationofthelevelset12

1.4.3. Capturingmotioninterface12

1.4.4. Levelsetsfor3Dfracturemodeling14 References15

2.Weakformsandgoverningequations19

2.1. Strongformforpuremechanicalproblems19

2.1.1. Onedimensionalmodelproblem19

2.1.2. Modelprobleminhigherdimensions20

2.1.3. TotalLagrangianformulation21

2.1.4. UpdatedLagrangianformulation22

2.2. Fromthestrongformtotheweakform24

2.2.1. Weakformfortheone-dimensionalmodelproblem24

2.2.2. WeakformforthetotalLagrangianformulation26

2.3. Variationalformulation27

3.Extendedfiniteelementmethod29

3.1. Formulationandconcepts29

3.1.1. StandardXFEM29

3.1.2. Hansbo-HansboXFEM34

3.2. Blending,integrationandsolvers36

3.2.1. Blending36

3.2.2. Isoparametric2DquadrilateralXFEMelementforlinearelasticity40

3.2.3. Shapefunctions41

3.2.4. TheB-operator42

3.2.5. Theelementstiffnessmatrix44

3.2.6. Integration46

3.3. XFEMforstatic/quasi-staticfracturemodelingin2Dand3D50

3.3.1. XFEMapproximationforcracks50

3.3.2. Discreteequations54

3.3.3. Crackbranchingandcrackjunction57

3.3.4. Crackopeningandcrackclosure59

3.4. XFEMfordynamicfracturemodelingin2Dand3D60

3.4.1. Diagonalizedmassmatrix60

3.4.2. Limitations64

3.5. Smoothedextendedfiniteelementmethod65

3.5.1. IntroductiontoSFEM67

3.5.2. EnrichmentinSXFEMandselectionofenrichednodes70

3.5.3. Displacement-,strainfieldapproximationanddiscreteequations72

3.5.4. Numericalintegration75

3.6. XFEMforcoupledproblems77

3.6.1. Hydro-mechanicalproblems77

3.6.2. Thermo-mechanicalproblems89

3.6.3. Piezoelectricmaterials92

3.6.4. Flexoelectricity100

3.7. XFEMforinverseanalysisandtopologyoptimization105

3.7.1. Inverseproblem105

3.7.2. Optimizationproblems115

3.7.3. Mathematicalformofastructuraloptimizationproblem116

3.7.4. Solidisotropicmaterialwithpenalization(SIMP)117

3.7.5. Levelsetbasedoptimization118

3.7.6. Nanoelasticity118

3.7.7. Nanopiezoelectricity130

3.8. Conditioningandsolutionofill-conditionedsystems146 References147

4.Phantomnodemethod153

4.1. Formulationandconcepts153

4.2. Acracktipelementforthephantomnodemethods154

4.2.1. Three-nodetriangularelement154

4.2.2. Four-nodequadrilateralelement157

4.3. Multiplecrackmodeling158 References159

5.Extendedmeshfreemethods161

5.1. Introductiontomeshfreemethods161

5.1.1. Basicapproximation161

5.1.2. Completenessandconservation162

5.1.3. Consistency,stabilityandconvergence164

5.1.4. Continuity165

5.1.5. Partitionofunity165

5.1.6. Kernelfunctions167

5.2. Somespecificmethods171

5.2.1. Approximationofthedisplacementfield171

5.2.2. Spatialintegration177

5.2.3. Essentialboundaryconditions185

5.2.4. Comparisonofdifferentmethods186

5.3. Numericalinstabilities190

5.3.1. Instabilityduetorankdeficiency192

5.3.2. Tensileinstability193

5.3.3. Attemptstoremoveinstabilities193

5.3.4. Materialinstabilityinmeshfreemethods194

5.4. Fracturemodelinginmeshfreemethods209

5.4.1. Thevisibilitymethod209

5.4.2. Thediffractionmethod212

5.4.3. Thetransparencymethod215

5.4.4. The“seethrough”and“continuousline”method217

5.5. Theconceptofenrichment217

5.5.1. Intrinsicenrichment219

5.5.2. Extrinsicenrichment222

5.6. (Extrinsically)enrichedlocalPUmeshfreemethods225

5.6.1. Enrichedmethodswithcracktipenrichment226

5.6.2. Enrichedmethodswithoutcracktipenrichment230

5.6.3. Crackbranchingandcrackjunction236

5.7. Extendedlocalmaximumentropy(XLME)238

5.7.1. LocalMaximumEntropy(LME)approximants239

5.7.2. Numericalintegration243

5.7.3. Conditionnumber245

5.8. Crackingparticlemethods245

5.8.1. Theenrichedcrackingparticlesmethod246

5.8.2. Applicationstolargedeformations250

5.8.3. Thecrackingparticlesmethodwithoutenrichment250

5.8.4. Crackingrulesforcrackingparticlemethods251

5.9. Comparisonofdifferentmethods253

5.9.1. ThemodeIcrackproblem253

5.9.2. Themixedmodeproblem260

5.10. ExtensionstomodeIIkinematics263

5.10.1. Enrichingintheshearbandplane263

5.10.2. EnforcingmodeII-kinematicswiththepenaltymethod265

5.11. Discretesystemofequationsforpuremechanicalproblems265

5.11.1. Methodswithoutenrichment265

5.11.2. Enrichedmethods267

5.11.3. Extensiontodynamics270

5.12. Spatialintegration283

5.13. Timeintegration286

5.13.1. Explicit-implicittimeintegration286

5.13.2. Explicittimeintegration,criticaltimestepandmasslumping287

5.13.3. Crackpropagationintime304 References306

6.Extendedisogeometricanalysis315

6.1. Formulationandconcepts315

6.1.1. B-splinesandNURBS315

6.1.2. Bézierextraction317

6.2. HierarchicalrefinementwithPHT-splines320

6.2.1. PHT-splinespace321

6.2.2. Computingthecontrolpoints323

6.3. Analysisusingsplines324

6.3.1. Galerkinmethod325

6.3.2. Linearelasticity327

6.4. Numericalexamples329

6.4.1. Infiniteplatewithcircularhole329

6.4.2. Openspanner330

6.4.3. Pinchedcylinder331

6.4.4. Hollowsphere332

6.5. Adaptiveanalysis333

6.5.1. Determiningthesuperconvergentpointlocations333

6.5.2. Superconvergentpatchrecovery337

6.5.3. Markingalgorithm340

6.6. Multi-patchformulationsforcomplexgeometry341

6.7. XIGAforinterfaceproblems341

6.7.1. Governingandweakformequations342

6.7.2. Enrichedbasisfunctionsselection345

6.7.3. Enrichmentfunctions347

6.7.4. GrevilleAbscissae348

6.7.5. Repeatingmiddleneighborknots349

6.7.6. Inversemapping350

6.7.7. Curvefitting351

6.7.8. Intersectionpoints353

6.7.9. Triangularintegration354 References355

7.Fractureinplatesandshells359

7.1. FracturesinshellandplatesusingXFEM359

7.1.1. Weakform359

7.1.2. ImplementationbasedontheQ4element361

7.1.3. Shearlocking362

7.1.4. Curvaturestrainsmoothing363

7.1.5. Extendedfiniteelementmethodforsheardeformableplates365

7.1.6. Smoothedextendedfiniteelementmethod367

7.1.7. Integration368

7.2. Fracturesinshellandplatesusingthephantomnodemethod370

7.2.1. PhantomnodemethodfortheBelytschko-Tsayshellelement370

7.2.2. Phantomnodemethodbasedonthethree-nodeisotropic triangularMITCshellelement378

7.3. Extendedmeshfreemethodsforfractureinshells392

7.3.1. Shellmodel393

7.3.2. Continuumconstitutivemodels396

7.3.3. Crackmodel397

7.4. Animmersedparticlemethodforfluid-structureinteraction402

7.5. XIGAmodelsforplatesandshells408

7.5.1. Kinematicsoftheshell408

7.5.2. Weakform410

7.5.3. Discretizationofthedisplacementfieldandenrichment412

7.5.4. Discretesystemofequations419

7.5.5. Edgecrackedplatesundertensionorshear422

7.5.6. Pressurizedcylinderwithanaxialcrack428

References432

8.Fracturecriteriaandcracktrackingprocedures437

8.1. Fracturecriteria437

8.2. Crackingcriteria437

8.2.1. CriteriainLEFM437

8.2.2. Globalenergycriteria440

8.2.3. Rankinecriterion440

8.2.4. Lossofmaterialstabilitycondition441

8.2.5. Rank-one-stabilitycriterion443

8.2.6. Determiningthecrackorientation444

8.2.7. Computationofthecracklength444

8.3. Cracksurfacerepresentationandtrackingthecrackpath445

8.3.1. Thelevelsetmethodtotracethecrackpath447

8.3.2. Trackingthecrackpathin3D451

8.3.3. Adaptivecrackpropagationtechnique462

8.3.4. Comments464 References466

9.Multiscalemethodsforfracture471

9.1. ExtendedBridgingDomainMethod472

9.1.1. Concurrentcouplingoftwomodelsatdifferentlengthscales474

9.1.2. Consistencyofmaterialproperties479

9.2. Extendedbridgingscalemethod479

9.2.1. Consistencyofmaterialproperties481

9.2.2. Upscalinganddownscaling483

9.3. Multiscaleaggregatingdiscontinuity(MAD)method491

9.3.1. Overviewofthemethod491

9.3.2. Coarsegrainingmethod494

9.3.3. Micro-macrolinkage500

9.4. Crackopeninginunitcellswiththehourglassmode503

9.5. Stabilityofthemacromaterial504

9.6. Implementation507

9.7. Numericalexamples508

9.7.1. 3Dmodelingofcracksinananocomposite508

9.7.2. Hierarchicalmultiscaleexample508

9.7.3. Semi-concurrentFE-FEcouplingexample510

9.7.4. ConcurrentFE-XFEMcouplingexample512

9.7.5. MD-XFEMcouplingexample513 References516

10.Ashortoverviewofalternativesforfracture521

10.1. Numericalmanifoldmethod(finitecovermethod)521

10.1.1. Thecoverapproximation522

10.1.2. Theleastsquare-basedphysicalcoverfunctions523

10.1.3. Theimpositionofboundaryconditions524

10.1.4. Fracturemodeling524

10.1.5. Geometricandmaterialnonlinearanalysis527

10.2. Peridynamicsanddual-horizonperidynamics528

10.2.1. Dual-horizonperidynamics531

10.2.2. Thedualpropertyofdual-horizon539

10.2.3. Wavepropagationin1Dhomogeneousbar543

10.2.4. Numericalexamples544

10.3. Phasefieldmodels562

10.3.1. Concepts563

10.3.2. Governingequations568

10.3.3. Discretization569

10.3.4. Solutionschemes572

10.3.5. Implementations573 References575

11.Implementationdetails581

11.1. Computerimplementationofenrichedmethods581

11.1.1. Pre-processing582

11.1.2. Processing585

11.1.3. Post-processing589

11.2. Numericalexamples590

11.2.1. Crackpropagationangle591

11.2.2. Hydro-mechanicalmodelwithcentercracks591

11.2.3. Hydro-mechanicalmodelwithedgecrack592 References597

Part1.Appendices

A.Derivationofshapederivativeforthenanoelasticityproblem601

B.Derivationoftheadjointproblemforthenanopiezoelectricity problem603 Index

Preface

Theobjectiveofthisbookistoprovideanoverviewandthetheoretical/computationalbackgroundofpartition-of-unitybasedcomputational methods,theirimplementationandapplications.Thefocusisonextended finiteelementandmeshfreemethodsandtheirapplicationwithfocuson modelingmaterialfailure.Itisassumedthatthereadersarealreadyfamiliar withfiniteelementmethodsorsimilarcomputationalapproachesincluding theirimplementation.Thecontentofthisbookiswrittenfromanengineeringpointofview.Itexplainsconceptsandformulationsandprovides detailsontheimplementationthroughsimpleMatlab® codes.Weprovideclassicalbenchmarkproblemsforwhichstate-of-the-artcomputational methodsaretestedatandpresentsomeinterestingnumericalexamplesto demonstratethepowerandperformanceoftheoutlinedmethods.The bookhoweverdoesnotcontainmathematicalproofsconcerningforinstancetheconvergenceoftheabovemethods.Convergenceplotsarejust shownnumericallyforspecificexamples.Thoughsomeofthemethodsare implementedincommercialsoftwaresuchasABAQUS,thebookdoesnot provideadescriptionontheuseofthesefunctionswithinsuchcommercial codes.

Thebookisaimedforstudentsandresearcherswhoareinterestedin learningandimplementingpartition-of-unitymethod,especiallyextended finiteelementandmeshfreemethods.Itiswellsuitedforstudentsandpostdoctoralfellowstostartresearchinthisdirectionandwhoareinterestedin methoddevelopmentortheapplicationofdescribedmethodstochallengingproblemsinengineeringandmaterialsscience.Itisalsoofinterestto readerswhoareinterestedinstate-of-the-artcomputationalmethodsfor linearandnonlinearfractureandchoosinganadequatemethodfortheir problemofinterest.Thecontentofthisbookistooextensivetobecoveredinasinglecoursethoughpartsofitcouldbethebasisfora1-semester courseonmeshfreemethodsorextendedfiniteelements.

Chapter 1 providesanintroductiontocomputationalchallengeswhich occurinproblemswithmovingboundariessuchasfracture,fluidmechanics,fluid-structureinteraction,inverseproblemsoroptimization.Italso presentsthelevelsetmethodwhichiscommonlycombinedwithpartitionof-unitymethodsforthoseproblems.Chapter 2 summarizesthegoverning equationsforpurelymechanicalproblemsforapplicationsinstaticsand

dynamicsasmostofthemethodsarepresentedinsuchasetting.They areprovidedinstrongandweakformincludingaTotalLagrangianand updatedLagrangiandescriptionofmotion.Chapter 3 isfocusedontheextendedfiniteelementmethod(XFEM)andvariationsorimprovementsof it.Withinthischapter,challengesrelatedtoso-calledblending,numerical integration,enrichmentandsolutionproceduresarediscussedandpotential solutionsarederived.TheimplementationofXFEMforstaticanddynamic fractureproblemsisdescribedindetailandtherepresentationofcomplex featuressuchascracknucleation,crackbranchingandcrackcoalescenceis discussed.WealsoproposeavariationoftheclassicalXFEMforfracture,i.e. thesmoothedextendedfiniteelementmethod(SXFEM).SXFEMavoids theintegrationofthesingularityincaseofasymptoticcracktipenrichments.Itfacilitatesthesubtriangulationcommonlyemployedincracked elements,andinheritscertainsuperiorpropertiesofthesmoothedfinite elementmethodincludinglesssensitivitytomeshdistortionanditshigh accuracyfortriangularelements.Wesubsequentlypresenttheextendedfiniteelementformulationsforcoupledfractureproblemsincludingdifferent enrichmentstrategies,implementationdetailsandpotentialchallengesrelatedforinstancetotheill-conditioningofthefinalsystemofequationsto besolved.Thermo-mechanical,hydro-mechanicalandelectro-mechanical aredescribed;thelatteronesincludepiezo-aswellasflexoelectricmaterials.Finally,twootherimportantapplicationsofXFEMaregiven:Inverse analysisand(topology)optimization.Forthoseproblems,XFEMallowsan exact–implicit–representationofthetopologythroughlevelsetfunctions andhenceemployingalwaysthesamemeshduringtheiterationswhile maintainingoptimalconvergencerates.Theperformanceofthemethod isdemonstratedforseveralchallengingproblemsintheassociatedsection forselectedproblem.Thephantomnodemethod,another“variation”of XFEM,inChapter 4 isnotbasedonenrichmentfunctionsbutoverlappingelements.Ithastheadvantageofbeingeasilyimplementableintoan existingfiniteelementcodebutcanbeappliedonlytofractureproblems. Conceptsforhowtoincorporatemultiplecracksandspecificcracktipelementsaredevisedinthischapter.

ThetopicofChapter 5 isextendedmeshfreemethods.Firstly,thebasic conceptofmeshfreemethodsisexplained.Subsequently,wepresentseveral popularmeshfreeapproximationsincludingtheSmoothedParticleHydrodynamics(SPH)andimprovementsofitsuchastheReproducingKernel ParticleMethod(RKPM).Themostlyrationalshapefunctionsandlack oftheso-calledKronecker-deltapropertyimposeadditionalchallengeson

meshfreemethodscomparedtoFEM.Inthiscontext,weprovidedifferent approachesofspatialnumericalintegration,impositionofessentialboundaryconditionsaswellassolutionstoavoidinstabilitiesofdifferentsources whichoccurinmanymeshfreemethodssuchasSPH.Wesubsequently presentclassicalmethodsonhowtoincorporatestrongdiscontinuities andfinallymodeldiscretefracturewithinmeshfreemethodsincludingthe visibility,diffractionandtransparencymethod.Differentextendedmeshfreemethodsarepresentedwhicharebasedeitheronanintrinsicoran extrinsicenrichment.Inanintrinsicenrichment,theenrichmentfunctionsareincludedinthepolynomialbasisusedtoconstructthemeshfree shapefunctionswhileextrinsicenrichmentschemesaremainlyfocusedon partition-of-unityenrichments.Twoclassesofdiscretefracturemethods arepresented.Thefirstclassensuresacontinuouscrackpathandrequires specialapproachestorepresentthecrack’stopologyandcracktracking algorithmswhilethesecondclass,theso-calledcrackingparticlemethods,representthecrackassetofcracksegmentsandavoidcracktracking algorithmsandmethodstorepresentthecracksurfaceentirely.Theperformanceofvariousextendedmeshfreemethodsarecomparedforseveral classicalbenchmarkproblemsmostlyinlinearelasticfracturemechanics.

Chapter 6 presentsformulationsbasedonextendedIsogeometricAnalysis(XIGA).WefirstdescribepopularIGAbasisfunctionsincluding B-splines,NonUniformRationalB-splines(NURBS)andPHT-splines, whichareusefulforh-adaptiverefinementprocedureswithinIGA.ImplementationdetailsofIGAforproblemsinlinearelastostaticsareprovided beforedifferentXIGAapproachesforweakdiscontinuitiesarediscussed indetail.Wededicatedanentirechaptertoextendedfiniteelementand meshfreemethodsformodelingfractureinplatesandshells.Formulations basedonMindlin-ReissneraswellasKirchhoffLoveshellsaredevised.In thelattercase,thehigher-ordercontinuityoftheassociatedmeshfreeor IGAmethodhasbeenexploited,sothatnoadditionalrotationaldegrees offreedomareneeded,whichdrasticallyfacilitatestheenrichmentstrategy requiringfulfillmentofaconstraintcondition.Wealsopresentamethod thatcanefficientlydealwithfluid-drivenfractureduetofluid-structure interaction.

Eachoftheabovementionedmethodsarecapableofdealingwith discretefractureefficiently.However,theyallrequireafracturecriterion whichdeterminetheorientationand“length”ofthecrack.Therefore, Chapter 8 isrelatedtostate-of-the-artfracturecriteriaandcracktrackingalgorithmsformethodsrequiringacontinuouscracksurface.Fracture

criteriaforproblemsinlinearelasticfracturemechanics(LEFM)aswell asnonlinearcontinuaaredescribed.Whilethefracturecriterionprovides theorientationofthecracksurface,criteriaforpropagatingthecrackare neededaswell.Differentapproachesforhowtorepresentthecracksurface basedontriangularfacetsandlevelsetsareexplainedandrelatedtodifferentcomputationalmethods.Inthiscontext,wedescribeindetailefficient cracktrackingalgorithmsinthreedimensionsanddiscusschallengesand limitations.

Chapter 9 presentsdifferentmultiscalemethodsforfracturewhichare usefulforapplicationssuchascomputationalmaterialsdesign.Thefocus willbeonso-calledconcurrentmultiscalemethodsforfracturewherethe geometryofafine-scalemodelisdirectlyintegratedintothegeometryof thecoarse-scalemodel.Allthosemethodsarebasedonextendedfiniteelementmethodstorepresentfractureeitherononeortwolengthscales.Two approachesaredescribed:Inthefirstapproach,thefine-scaledomainand thecoarse-scaledomainiscoupledatadiscreteinterface.Thisapproach seemspromisingforstaticapplicationswhilethesecondapproachisbetter fordynamicfractureasartificialwavereflectionsareminimizedthrough ahandshakecouplingwhichcontainsboththefine-scaleandcoarse-scale domain.Efficientstrategiestocoarsegraincracksarepresentedwhichare requiredinadaptivemultiscalemethodstoguaranteecomputationalefficiency.Themultiscalemethodsforfracturearedescribedforcouplingtwo continuummodelsaswellascouplingatomisticandcontinuummodels.

Chapter 10 brieflygivesashortintroductiontocompetitiveandpopularalternativemethodsforfracture,i.e.thenumericalmanifoldmethod (NMM),peridynamics(PD)aswellasphasefieldmodelsforfracture. TheNMMsharessomefeaturesofthephantomnodemethodthough ithasbeenproposedmuchearlier,evenbeforetheextendedfiniteelement method.PDisaveryefficientmethodfordynamicfractureas–similarly tothecrackingparticlesmethod–itdoesnotrequirecracktrackingprocedures.ThecrackpathinPDisanaturaloutcomeofthesimulation.Phase fieldmodelsaresomehowrelatedtogradientdamagemodelsandsmear thecrackoveracertainwidth.Thebeautyofthephasefieldmodelliesin thethermodynamicconsistentframework,whichallowsastraightforward implementationintoafiniteelementframework.Thoughtheybelongto theclassofcontinuousapproachesforfracture,weincludedtheminour bookduetotheirgrowingpopularity.

Thelastchapterofthebookisdedicatedtotheimplementationdetails ofthepresentedapproachesinthisbook,i.e.XFEM,extendedmeshfree

methodsandXIGA.ThefocusisonthedevelopmentofCAD-compatible formulationsintheframeworkof(X)IGAisbrieflyexplained.Somenumericalexamplesareincludedandalinktoanopen-sourcerepository containingcorrespondingMatlabcodewithadditionaldocumentationand explanationsisprovided.

Wewouldliketothankourcollaboratorsandcurrentandformerstudents,whoseresearchcontributedtothisbook,amongwhomare:Fatemeh Amiri,StéphaneBordas,P.R.Budarapu,C.L.Chan,ThanhChau-Dinh, LeiChen,G.R.Liu,MohammedMsekh,S.S.Nanthakumar,VinhPhu Nguyen,NhonNguyen-Thanh,HungNguyen-Xuan,HaroldS.Park, HuilongRen,MohammadSilani,HosseinTalebi,NavidValizadeh,Nam Vu-BacandGoangseupZi.Wewouldalsoliketoexpressourgratitudeto theElseviereditorsandtechnicalstaff,inparticularBrianGuerin,Sabrina Webber,ThomasvanderPloeg,andIsabellaC.Silvafortheirsupportin therealizationofthisproject.

Weimar,GermanyT.R.andC.A. Boulder,ColoradoJ.H.S. Hannover,GermanyX.Z. September,2019

Nomenclature

Thefollowingsymbolsarestandardinengineeringliteratureandarealsousedinthisbook. However,thenotationmightvarysomewhatamongthedifferentchaptersandsections.

Greeksymbols increment

δ variationorKroneckerdeltaorDiracdeltafunction

, strain boundary

λ LagrangemultiplieroreigenvalueorLamé’sfirstparameter domain

φ levelset potential

ψ enrichmentfunction

ρ density

σ, σ Cauchystress

Latinsymbols

B B-operator

b bodyforce

C (tangent)materialmatrix

d displacementvector

F deformationgradient

f force

I identitytensor

K stiffnessmatrix

M massmatrix

n normal

A area

a, a accelerations

E Young’smodulus

G shearmodulus

H Heavisidefunction

h meshsizeparameter

J Jacobian

KI , KII stressintensityfactor

N shapefunction

p polynomialdegreeofbasis

P , P firstPiolaKirchhoffstress

R supportsize

t , t traction

u, U displacement

V volume

v, v velocities

w quadratureweight

X , X materialcoordinates

x, x spatialcoordinates

1.1.Partitionofunitymethods

TheFiniteElementMethod(FEM)wasdevelopedinthe1950sand 1960sasaconvenientwaytosolvepartialdifferentialequationsarising fromvariousscientificandengineeringapplications.Rigorousmathematicalanalysisofthemethodstartedin1970sandsincethenthousandsof papershavebeenpublishedonthistopic.Thefiniteelementmethodisa Galerkinmethodthatapproximatesthesolutionofthepartialdifferential equation(PDE),posedinavariationalform.Themethodinvolvespartitioningthedomainintoafinitenumberof“elements”,andanapproximate solutionissoughtinafinitedimensionalspaceofpiecewisepolynomials, definedrelativetotheelements.Naturally,thequalityoftheapproximationdependsonthefineness(orcoarseness)ofthediscretization,i.e.onthe sizeoftheelements,thedegreeoftheunderlyingpolynomialsinthefinite elementspaceandontheregularityoftheexactsolution.

Duringthe1970sand1980s,ascomputationalresourcesbecameless expensive,thepopularityofthefiniteelementmethodgrewrapidlyanda largenumberofbothmathematicalresultsandcomputerprogramsweredeveloped[25].Nevertheless,evenwithincreasingcomputationalcapabilities, certainproblems(e.g.crackpropagationproblems,multi-scaleproblems, problemswithcomplexboundaries,etc.)were,andstillremaintooexpensive(duetosizeorcomplexity)tobesolvedsatisfactorilybytheclassical finiteelements.Thereforeinterestgrewintakingamoregeneralapproach, withessentiallytwofeatures.Thesefeaturesinvolve:(a)eithernotusing ameshatallorusingaverysimplemeshtodiscretizethedomain,and (b)suitablychoosingapproximationspacesthatarenotbasedonpolynomials.Theassociatedmethodsarebroadlydenotedas meshless or meshfree methods.

Thesemethodscanalsobegroupedintotwoclasses–theclassicalparticlemethods[52,53,55,56]andthemethodsbasedontheideaofdatafitting techniques[11,18].Theclassicalparticlemethodsweredevelopedfortime dependentproblemsorconservationlaws.Theyinvolveadiscretesetof pointscalledparticlestodiscretizetheunderlyingdomain,andthesolutions ofasystemoftime-dependentPDEsforalltheparticlesaresought.Onthe otherhand,themeshlessmethodsbasedondata-fittingtechniques,which

wereinitiallydevelopedforstationaryproblems,alsodiscretizethedomain byparticles.Here,eachparticleisassociatedwitha“patch”(anopenset), suchthattheunionofthepatchescoverstheunderlyingdomain.Suitablefinitedimensionalspaces(whichmaynotbebasedonpolynomialsand couldbeobtainedbydata-fittingtechniques)aredefinedoneachofthese patches.Finally,theassociated“shapefunctions”areusedinaGalerkinor collocationmethodtoobtainamatrixequation.Thesolutionofthislinear systemisthenusedtoobtainthesolutionofthemeshlessmethod.

Severalmeshlessmethods,basedontheideaofdatafittingtechniques describedabove,havebeendevelopedprimarilybyengineersandtheydifferessentiallyinthechoiceofthefinitedimensionalspacesonthepatches. ThefirstofthesemethodsisknownasShepard’smethod[61].Thisideawas furthergeneralizedintoamethodcalledSmoothedParticleHydrodynamics(SPH),[27,28,43,49,50,71]andtheMethodofClouds[16,17].Method ofCloudsusedmovingleast-squares(MLS)ideasbasedonpolynomialsto constructthefinitedimensionalspace.Manymethods,withslightvariationsoftheseideas,werealsodeveloped;forexampletheDiffuseElement Method[54]andMethodofSpheres[15].Later,aclassofmethodswere introducedwherethefinitedimensionalspaceswereconstructedusingthe ideasofreproducingkernels[38,40–42].Also,RadialBasisFunctionswere usedtoconstructmeshlessmethods,andtheapproachwasdevelopedby boththemathematiciansandengineers[23,24,35,36,73,74].Wemention thatthoughthemethodsdescribedabove(otherthanthosebasedonRBFs) werenotmathematicallyinvestigatedindetail,thegenericideasandtheassociatedmathematicalanalysisweregivenin70’sin[3,4,64].Foramodern mathematicaltreatment,wereferto[8].

The PartitionofUnityFiniteElementMethod (PUFEM)couldalsobe viewedasameshlessmethodinabroadsense.Ontheotherhand,itisnot aspecificmethod,butisageneralflexibleframework.Infact,theclassical FEMandmanymeshlessmethods,discussedbefore,couldbecastintothe PUFEMframeworkincertainsituations.

TheideaofPUFEMwaspartiallyintroducedin[6],wherepiecewise L-splines[72]insteadofpiecewisepolynomialswereusedtoapproximate locallythesolutionofaboundaryvalueproblemwith“rough”coefficients. However,thismethodcouldnotbedirectlygeneralizedtohigherdimensions.In1994,Babuška,CalozandOsborn[7]introducedtheso-called “SpecialFiniteElementMethod”wheretheyusedcertainnon-polynomial functionsforlocalapproximationon“finiteelementstars”(whichserved aspatches).Theselocalapproximationswere“pastedtogether”usingthe

finiteelementhatfunctions(whichformapartitionofunity)toobtain aglobalapproximation.Thisideawasusedtoapproximatethesolution ofaPDE,modelingaparticularunidimensionalcomposite.Itwasfurther refinedin[5,44]andageneralabstractframeworkwasdeveloped.Itwas shownthatthelocalapproximations(inthefinitedimensionalspacesdefinedlocallyonapatch)couldbepastedtogetherusinganypartitionof unity.Abasicapproximationresultwasobtainedin[44],whichindicated that“accurate”localapproximations,pastedtogetherusingapartitionof unity,yieldsanaccurateglobalapproximationoftheunknownsolutionof thePDE.TheseideaswereusedtoapproximatesolutionsofLaplaceand Helmholtzequations.ThebasicapproximationresultforPUFEMwasfurtherelaboratedin[9].Theframeworkpresentedin[5,44]wassuccessfully implemented[65–67]byStrouboulisetal.onengineeringapplicationsinvolvingcompositesmadeoffibersanddomainsinvolvingmultiplevoids andcracks.Thenameof GeneralizedFiniteElementMethod (GFEM)was firstusedinthesepapers.Severalissuesrelatedtotheimplementationof GFEM,e.g.hierarchicalconstructionofthepatches(opencover),numericalintegration,andsolvingthelinearsystem,wereaddressedinaseriesof papers[30–33]byGriebelandSchweitzer.Wenotehowever,thatthelocal approximationinthesepaperswerebasedonpolynomials.Ideassimilarto GFEMandPUFEMhavealsobeendeveloped,tosomeextentindependentlyandinparallel,underthenameof ExtendedFiniteElementMethod (XFEM),[62],[68].Thismethodisalsoanextensionofthestandard FEM,wherecertain“enrichment”functions(basedontheavailableinformationonthesolution)areusedforapproximation,locallyinspecially chosenfiniteelementstars.Theseenrichmentfunctionsarethenpasted togetherusingstandardfiniteelementhatfunctions.XFEMhasbeensuccessfullyimplementedforvariouscrackpropagationproblems[1,22,46–48, 62,69,70].WementionthatXFEMcouldalsobecastintheframeworkof PUFEM,wherestandardfiniteelementsareusedasthepartitionofunity. ThePUFEMframeworkcanbedescribedbythefollowingdiscretizationsteps:

• Weconsideranoverlappingfiniteopencover {ωi } oftheunderlyingdomain .Thesets ωi ,called patches,couldbetheinteriorsofthefinite elementstars,withrespecttoasimplefiniteelementmeshthat“triangulates”aregioncontaining .Wenotethatpreciselythesepatchesare usedinXFEMandwerealsousedbyStrouboulisetal.in[65,66]and [67].Theycouldalsobespherical,asusedintheMethodofSpheres andMethodofClouds.

• Relativetotheopencover {ωi },thePUFEMusesa partitionofunity {φi }. Standardfiniteelementhatfunctions,withrespecttoafiniteelement mesh,couldserveasthepartitionofunity,asareusedinXFEMand [65,66]and[67].Ontheotherhand,the“reproducingkernelparticle (RKP)shapefunctions”[38,40–42],canalsobeusedasapartitionof unity.

• Special,problemdependent,finitedimensionalspaces Vi ,definedon ωi ’s,areusedfor localapproximation inPUFEM.Thesefinitedimensionalspacesmaycontainconstantfunctionstogetherwithsingular functions,harmonicpolynomialsorotherspecialfunctionsbasedon theavailableinformation[9,44].Infact,aone-dimensional Vi containingonlyconstants,togetherwithreproducingkernelparticlefunctions asapartitionofunity,definemanymeshlessmethods.Thefunctions in Vi couldalsobeconstructedusingdata-fittingideas,e.g.MLSfunctions,asitwasdoneintheMethodofClouds.

WefinallymentionthatPUFEMisaGalerkinmethod,wherethetrial spaceisconstructedas

S PUFEM = j φj Vj ={v ∈ H 1 ( )

If {νji }n(j ) i=1 isabasisfor Vj ,then

S PUFEM = span{φj νji };

thefunctions φj νji willbereferredtoasshapefunctionsofPUFEM.

Withappropriatechoiceof φj and Vj ,thespace S PUFEM yieldsprecisely thetrialspacesusedinvariousmeshlessmethodsandXFEM.See[8,9].

ItisapparentthatPUFEMisessentiallyanextensionofthestandardfiniteelementmethod,butsomeaddedflexibilityisobtainedbytheselection ofthepartitionofunityfunctions φj andlocalapproximationspaces Vj .In factithasbeenshownthatthePUFEM/GFEMusingthehat-functionpartitionofunityand Vj = Pk (ωi ) isequivalenttothefiniteelementmethod usingLagrangeshapefunctionsincertainsituations[9].

OneofthemainadvantagesofPUFEMistheabilitytouseadiscretizationschemethatiseasytoconstructcomparedtothe“meshing”stepofthe standardFEM.Whiletremendousadvanceshavebeenmadeintheareaof meshgeneration,generatingasatisfactorymeshforproblemswithcomplicatedboundariesordiscontinuities,especiallyinthreedimensions,remains atime-consumingprocesswhichoftenrequireshumaninteraction.Thisis

particularlyinconvenientforproblemswherethemeshingneedstochange ateachtime-step,suchascrackpropagationorcertainfluidflowproblems.Thedifficultiesofmesh-generationhaveledtothedevelopmentof meshlessor“mesh-free”methods,whichwerebrieflydiscussedabove.

AnotherattractivefeatureofPUFEMisthatitallowstheuseoflocal approximationspaceswhichcanbeadaptedtotheproblemathand.In general,thebetterthefunctionsin Vj approximatelocallytheexactsolutionofthePDE,thebetterwillbetheglobalapproximationqualityofthe PUFEMsolution.Thereforeitispossible(anddesirable)totailortheapproximationspaces Vj totheinformationavailableabouttheexactsolution. Forexample,ifitisknownthattheexactsolutionisharmonic,onecould choose Vj tobethespaceofharmonicpolynomialsratherthanallpolynomials.Also,iftheexactsolutionhassingularitiesduetotheboundary geometry,onecouldselecta Vj whosefunctionsaccuratelyapproximate thesingularbehavioroftheexactsolution(asinXFEM,see[9,62,68]).

Conversely,iftheexactsolutionisknowntobesmooth,onecan selectasmoothpartitionofunityandlocalapproximationspacesthatcontainsmoothfunctions(e.g.polynomials).Thentheresultingglobalshape functionsofPUFEMarealsosmoothandwillapproximatewelltheunknownexactsolution.Thisisanimprovementoverthestandardfinite elements,whereitismoredifficulttoobtainanapproximationwhichis C 1 -continuousorsmoother.

TheaddedflexibilityofPUFEMisnotwithoutsomecosts.Themain drawback,whichissharedwithmostoftheothermesh-freemethods,is theaddedcomplexityofthenumericalintegration.InPUFEM,insteadof integratingovershape-regularelements,onehastoconsiderthesupport ofeach φj whichcanhaveamorecomplicatedstructure.Forexample,if the φj haveradialsupports,thentheirintersectionisalens-shapedregion whichisdifficulttohandlebynumericalintegrationroutines.Moreover, PUFEMandXFEMallowtheuseofapproximatingfunctionsthatmay notbepiecewisepolynomialsoflowdegree,thereforeaccuratenumerical integrationwillrequiretheuseofalargenumberofquadraturepoints.In general,numericalintegrationrequiredintheassemblyofthelinearsystem becomesthemosttime-consumingpartofthesolutionprocess.Onthe otherhand,thisstepismoreamenabletoparallelizationandtherecent adventofparallelCPUarchitecturesislikelytomakecomputationalcosts manageableinamajorityofcases[33].

Anotherpotentialpitfallisthatcertainchoicesofpartitionofunity functionsandlocalapproximationspacescangiveriseto(almost)lin-

earlydependentPUFEMshapefunctions.Ifthatisthecase,theresulting linearsystemisconsistentbutthestiffnessmatrixwillbesingularandspecialalgorithmswillbeneededtodealwiththesesituations.Furthermore, multi-gridsolversandotherefficientmethodsavailableforthelinearsystemsarisingfromthestandardfiniteelementsmayneedtobemodifiedfor usewithPUFEM[32].

WealsomentionthatsolvingaDirichletboundaryvalueproblemby PUFEMwillrequirethatthefunctionsinthetrialspace S PUFEM satisfythe essentialboundaryconditions.Thisinturnrequiresthatthefunctionsin Vj ,for ωj ∩ ∂ =∅,satisfytheessentialboundaryconditions,whichmay beverydifficulttoaccomplish.Severaldifferentapproacheshavebeenproposedtoovercomethisproblem,suchascouplingtomesh-basedmethods closetotheboundary[37],penaltyorperturbationmethods[2,9,39],the Lagrangemultipliermethod[10,39,51],andamethodduetoNitsche[34, 59].Mostoftheseapproacheseitheraddanextralayerofcomplexitytothe problem,orlosesomeoftheadvantagesofPUFEM,orresultinalossof theoptimalrateofconvergence.Themostpromisingideaintheliterature seemstobeNitsche’smethodwhichworksbymodifyingthevariational formulationoftheproblemtoaccountfortheboundarydata.Thisapproachretainstheoptimalconvergencerateandthemesh-freecharacterof themethodandwasimplementedin[59].

1.2.Movingboundaryproblems

Movingboundaryproblemsareofmajorimportanceinengineeringandparticularlychallengingforcomputationalmethods.Theyincludemulti-phaseflow,fluid-structureinteraction,biofilm/tumorgrowth, shape/topologyoptimization,inverseanalysisorimageregistration,to namesomeoftheapplications.Fractureisanotherchallengingmoving boundaryproblemwhichwillbethefocusinthenextsection.Thereare twosubtledifferencestotheaforementionedproblems:(1)Acrackisnota closedboundarybutanopenboundaryand(2)boundarydoesnotpropagateorthogonaltoitsinterfacebutfromthecrackfront.

Therearetwowaystomodeltheboundary:(1)assmearedinterfaceand(2)assharpinterface.Thefirstcategorycommonlysimplifies theimplementationofcomputationalmethodsoncostofaccuracyand/or computationalefficiency.Agoodexampleisthe88linecodefortopology optimizationbasedonSolidIsotropicMaterialwithPenalization(SIMP). InSIMP,thematerialpropertiesareinterpolatedandthereforecontinuous.

Hence,theSIMPmethodcannotcapturejumpsinthestrain/stressfield astheyoccuratmaterialinterfaces.ThoughSIMPanditsimprovedversionshavebeensuccessfullyappliedtonumerouschallengingproblems,it hasitslimitations.Forinstance,lackofrobustnessandmeshdependency hasbeenreportedformulti-materialandnonlinearoptimizationproblems. Methodsbasedonlevelsetsforinstancecapturethesharpinterfacebutthe implementationeffortishigher;the129lineimplementationofthelevel setmethodfrom[14]isprobablyamongtheshortestavailable‘levelset’ codes.Notethatalsothesolutionprocedureoftheforwardproblemmight bemorecomplex.InFEMforinstance,theinterfacecanberepresented ‘explicitly’or‘implicitly’.Inotherwords:Whentheinterfaceisalignedto thediscretization,wecallitan‘explicitrepresentation’;otherwisean‘implicit’one.Thefirstscenariorequiresthegenerationofnewmesheswhen thetopologychangesduringtheiterationswhichinturndeterioratescomputationalefficiencywhentheentirestiffnessmatrixhastobereassembled. Whentheinterfaceistobecapturedwithinanelementbythelevelset function,enrichmenttechniquesasdescribedinthisbookareneeded.It isofcoursepossibletoavoidenrichmentschemesbyemployingasimplesub-triangulationtechniquesforintegrationandtheoutcomeofthe topologyoptimizationmightnotdiffermuchfromenrichmentschemesin manycases.However,thejumpinthestress/strainfieldcannotbecaptured andtheremightbeapplicationswherethisisimportant,forinstancewhen interfacephenomenaneedstobeaccountedforasintopologyoptimizationofnano-structures.Itisimperativetosaythatwithoutenrichment,the convergencerateswillbesub-optimalwhenmaterialinterfacesareinside anelement.

Anotherissueishowtodealwith moving interfaces.Thisapparently dependsontheapplicationandmethodemployed.Inlevelsetbasedtopologyoptimizationwheretheinterfaceisrepresented‘implicitly’bythelevel set,thenewinterface(i.e.thenewlevelset)isobtainedthroughthesolutionofthe(stabilized)Hamilton-Jacobiequationwherecommonlyshape derivativesareusedasvelocitynormaltotheinterface.Theprocedureis quitesimilarinfluidmechanicsproblemswhenlevelsetsareusedtoimplicitlyrepresenttheinterfacebetweentwofluids.However,inthiscase,the velocitynormaltotheinterfaceisthevelocityofthefluidwhichisinterpolatedfromthenodes.Inproblemsinvolvingfluid-structure-interaction, onecandistinguishbetween interphasecapturing methodsand interfacetracking methods.ThemostclassicalinterfacecapturingmethodisprobablyALE (ArbitraryLagrangianEulerian)whichadaptstheEulerian(fluid)meshto

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.