FiniteElementsinAnalysisandDesign106(2015)41–55 ContentslistsavailableatScienceDirect Finite Elements in Analysis and Design journal homepage: www.elsevier.com/locate/finel Efficient 3D analysis of laminate structures using ABD-equivalent material models Goldy Kumarn, Vadim Shapiro SpatialAutomationLab,MechanicalEngineering,UniversityofWisconsin,Madison,USA a r t i c l e i n f o a b s t r a c t Articlehistory: Laminatecompositesarewidelyusedinautomotive,aerospace,andincreasinglyinconsumerindustries, Received28August2014 duetotheirreducedweightandsuperiorstructuralproperties.However,structuralanalysisofcomplex Receivedinrevisedform laminatestructuresremainschallenging.2Dfiniteelementmethodsbasedonplate/shelltheoriesmay 17July2015 be accurate and efficient, but they generally do not apply to the whole structure and require Accepted27July2015 identificationandpreprocessingoftheregionswheretheunderlyingassumptionshold.Fullyautomated structural analysis using solid 3D elements with sufficiently high order basis functions is possible in Keywords: principle,butisrarelypracticedduetothesignificantincreaseinthecostofcomputationalintegration Composite overalargenumberoflaminateplies. Laminate We propose a procedure to replace the original laminate by much simpler new virtual material Finiteelementmethod models.Thesevirtualmaterialmodels,undertheusualassumptionsmadeinlaminationtheory,havethe Materialmodel sameconstitutiverelationshipasthecorresponding2Dplatemodeloftheoriginallaminate,butrequire onlyasmallfractionofcomputationalintegrationcostsin3DFEA.Wedescribeimplementationof3D FEA using these material models in a meshfree system using second order B-spline basis functions. Finally,wedemonstratetheirvaliditybyshowingagreementbetweencomputedandknownresultsfor standardproblems. &2015PublishedbyElsevierB.V. 1. Introduction commonpracticeistoassumethatthelayersarepermanentlyfused together and ignore any fluctuation in stress–strain fields at the 1.1. Motivation interfaces of layers [3–6]. These assumptions allow to approximate laminate'sglobalbehaviorasthatofaplateorashell.Interlaminar Laminatecompositesarenowusedwidelyinvarietyofindustries, stresses and strains may be significant in boundary regions and including aerospace, automobile, medical and sports [1–3]. Lami- regions of discontinuities [7] where full three-dimensional and/or natesarelightweightandstiffwithcustomizablematerialproperties, layered methods should be used, but the plate/shell assumptions resulting in structures superior to those made of homogeneous givesufficientlyaccuratestressandstrainestimatesforregionsaway materials[2–4].Highstiffness-to-weightratioisachievedusingfiber from those regions [8]. In this paper, we will show that the same reinforced plies. These plies, when fused together under high plate/shell assumptions, when applicable, may be used within a temperatureandpressure,formcomplexmonolithiclaminateparts. general 3D finite element analysis to dramatically speed up the Thefiberreinforcements,laidusingtechniquesrangingfrommanual analysisprocedure. to fully automatic, are generally parallel and unidirectional and, Structural analysis of laminates can be carried out using therefore, result in plies which are anisotropic in nature. Material differentfiniteelementmethods,andsomeofthemareillustrated properties are customized by varying fiber angle within each ply, foratypicallaminatepartinFig.1.Duringfiniteelementanalysis controlling the number of plies, and adding additional materials (FEA), stiffness matrix K for each element must be computed, e between plies such as cores and fillers. The presence of numerous whichingeneralformisgivenas[9] Z plies,however,leadstocomplexgeometryandmaterialdistribution inlaminatestructures,and,therefore,structuralanalysisoflaminates Ke¼ BT(cid:2)Q(cid:2)BdΩ; ð1Þ Ω bytreatingeachplylayerindividuallyisprohibitivelyexpensive.The e where B is the strain–displacement matrix, Q is the material Ω constitutiverelationmatrix,and istheelement'sdomainover e whichintegrationisdone.Sincetherearenumerousplies,mesh- nCorrespondingauthor.Tel.:þ16083349646. ing each ply independently requires a large number of elements E-mailaddresses:[email protected](G.Kumar), [email protected](V.Shapiro). andis,therefore,prohibitivelyexpensive.Amuchsmallernumber http://dx.doi.org/10.1016/j.finel.2015.07.009 0168-874X/&2015PublishedbyElsevierB.V. 42 G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 Fig.1. Figure shows a structure made of 3 laminates analyzed using different finite element methods. (a) Plate/shell element. (b) Element of 3D conforming mesh. (c)Elementofa3Dnon-conformingmesh.(d)Elementofanon-conformingmeshwithcurvedlaminateinside. of elements is needed if layered element method [10–12] is used composite laminate structures that is as general as 3D FEA and (Fig. 1b and c), wherein multiple plies can cross through an asefficientas2DFEAwhendimensionalreductionmakessense. element. Such finite element methods (also called 3D FEA since Specifically,weproposeamethodtoreducetheexcessivecostof integration domain Ω is 3D) may be further classified into integrationforlayeredelementsbytakingadvantageofplate/shell e conforming and non-conforming (or meshfree), depending on natureoflaminates,wheneversuchassumptionsarereasonable.To whether the finite element mesh conforms to the geometry of this end, we have devised a procedure to obtain material models thelaminate.Threedimensionalelementsmayexhibitlockingand whicharesimplerbutareequivalenttotheoriginallaminate,under ill-conditioning of stiffness matrix when used for laminates that theassumptionmadeinlaminationtheories.Werefertothesenew are thin in the plylayup direction [9,13]. These problems can be materialmodelsasABD-equivalent material models, asthey result alleviated or eliminated by using higher order hierarchical1 basis in the same ABD matrices as the original laminate and, therefore, functions[14,15]. can replace the original laminate during integration if plate/shell All3DFEAmethodsarecomputationallyexpensive,astheabove assumptionsapply.We demonstrate the effectiveness oftwosuch integrationhastobecarriedoverlargenumberofplies(tensoreven material models—a 3-ply and a graded material model—in a non- hundreds). Integration is performed using quadrature rules that conforming FEA system using layered solid elements. We validate depend on the geometry of the element as well as the degree of thetwoABD-equivalentmaterialmodelsbyusingthemtoanalyze theintegrand,andamountstosamplingtheintegrandatanumberof several benchmark problems, and compare obtained results from quadraturepoints[9].Togetanideaofthehighcostofintegration knownresults.Thefullyimplementednon-conformingFEAsystem forlaminates,weconsiderthelayeredelementusedinreference[10] uses layered solid elements with second-degree B-spline basis to analyze a laminate made of 100 plies. The element used is an functionsthatarehierarchicalinnature. eight-nodebrickelementwithtri-linearbasisfunctions,which,fora Abriefoutlineofthepaperisasfollows.InSection2,wesurvey homogeneousmaterial,isfullyintegratedusing2integrationpoints therelatedwork.Section3developstheconceptofABD-equivalent ineachdirection,or8integrationpointsintotal[10].However,ina material models and proposes two specific examples of such laminate,8integrationpointsareneededforeachply,whichresults models.Implementationoftheproposedapproachincombination ina100foldincreaseforour100-plylaminate.Sinceintegrationcost with a non-conforming finite element method is described in represents a significant portion of the overall solution procedure, Section 4. Its effectiveness is demonstrated using a number of analysis of composite laminates using layered elements is an benchmark problems in Section 5, followed by conclusions and expensiveproposition. futureworkinSection6. Plate or shell assumptions reduce the computation cost and increase accuracy of FEA for laminates owing to their thinwalled nature.Theseassumptionsmayleadtodifferentlaminationtheories, 2. Relatedwork where the material matrices Q of allthe plies are replacedbythe so-called ABD matrices [4,16]. The structure and the integration The finite element methods for simulating global behavior of domainΩ effectivelyreducestoa surface (Fig.1a),whichiswhy laminatestructures[16–18]maybebroadlyclassifiedasa2Dora e thismethodisalsocalled2DFEA.However,2DFEAisnotvalidin 3D FEA. For the purposes of this paper, we only consider those regions near boundaries and discontinuities (Fig. 1), which have methodswhichignoreinter-plyphenomena,butwenotethatthe significant 3D stresses and, therefore, plate/shell assumptions are inter-layer stress–strain can be partially predicted from global invalid.Inthissense,2DFEAmethodsarenotgeneral,becausesuch deformation[4,19]. regionsarecommoninlaminatestructures. 2.1. Two-dimensionalFEA 1.2. Contributionsandoutline Laminates usually behave as plates or shells, and are analyzed using 2D FEA. Depending on the strain field assumed in the Basedontheabovediscussion,thechoicebetween2Dand3D laminate's thickness direction, different lamination theories exist FEAamountstoatrade-offbetweengeneralityandcomputational [5,17,16], and can be classified as one of the following: Classical efficiency. We seek to develop an approach to analysis of Lamination Plate Theory (CLPT), First Order Shear Deformation Theory(FSDT),orHigherOrderShearDeformationTheories(HSDTs). CLPT assumes that laminates undergo only stretching and pure 1A basis function is called hierarchical when a higher order basis function bending (Fig. 2C): in-plane strains vary linearly in the thickness containsallthelowerorderbasisfunctions;forexample,B-splinesarehierarchical basisfunctions. direction, and out-of-plane strains are absent. On the other hand, G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 43 Fig.2. (A)Plywithfiberangle1201.(B)xzcrosssectionofalaminatewithpliesatangles45/(cid:4)60/120/0.(C)Linearvariationofstrainϵxalongzwhilelaminateisin stretchingandpurebending. Althoughtheseelementsdonotrequiremid-surfacesexplicitly,they mustconformtothegeometryofthelaminate,withtheirzdirection aligned to the transversal direction of structure's offset thickness, because the in-plane and out-of-plane behaviors are assumed or enhanceddifferently.Werefertosuchelementsasconforminglayered elements(Fig.1b).Theseelements,justlikelayeredelements,arestill expensiveforFEAoflaminates. Fig.3. (A)Alapjointbendingunderin-planeloading,whichleadstohighstress Aligning element's z direction to the laminate's transverse concentrationnearthejoint[20,21].(B)Whenanalyzedasa2Dstructure,bending thicknessdirectionalsosimplifiesvolumeintegration.Ifnotaligned, inlapjointisnotcapturedatall. pliescanintersectelementsatarbitraryanglesandrequirecompu- tation of intersections between individual plies and elements, FSTDandHSDT,asthenamessuggest,assumenon-zeroout-of-plane which is both non-trivial and computationally expensive. For this shearstrains;FSDTassumesconstantvariationwhileHSDTsassume reason,3DFEAoflaminatesusingnon-conformingmeshbecomes linear or even higher order [19] variation for out-of-plane shear lessattractive,despiteitsadvantagesoverusingconformingmesh strains.In-planestrainsareassumedtobeidenticaltoCLPT,andout- [26]. For completeness, we mention that several other methods of-plane normal strain is absent. Since our goal in this paper is to havebeenproposed[5,16,17,27],buttheyeitherlackgenerality,or showthatABD-Equivalentmaterialmodelscanbeusedtosimplify sufferfromoneormoreofthechallengesmentionedabove. theactualmaterialand,therefore,reducethecostof3DFEA,weonly focus on CLPT and FSDT and not higher order theories. Fig. 2 illustrates a laminate and the variation of strain field in the x 3. Materialmodelsforcompositelaminates directionalongthethicknessobtainedusingCLPT. Asdiscussedabove,although2DFEAisefficient,itisnotgeneral. Inthissection,webrieflysummarizetheclassicaltheoriesused Inaddition,2DFEAsuffersfromadditionaldrawbacksthatlimitits toestablishtheconstitutiverelationshipsinalaminate,andderive applicability in complex laminate structures. First, it requires the ABD-equivalent material models which remain valid under iden- structuretobedimensionallyreducedtoasurface,whichcouldbea ticalassumptions. complex task in itself [22]. Even if the structure is successfully reduced,modelinganassemblyofmultipleplatesandshellscanbe 3.1. Constitutiverelationsfororthotropicplies problematic [6]. In addition, due to dimensional reduction, 2D FEA cansometimescompletelymissa3Dphenomenon.Forexample,in Plies made of uni-directional fibers are generally modelled as thelapjointproblemshowninFig.3,2DFEAmissesthemoments transversely isotropic materials, but we are considering a more due to eccentric forces when the lap joint is reduced to a surface. general orthotropic model for the plies [4]. To characterize a Finally,differenttheoriesforthinandthickplatesandshellsfurther material in linear elasticity, stiffness matrix C is used, which complicate2DFEA[23]. requires 9 independent elastic constants in case of orthotropic materials.Theconstitutiverelationbetweenstressandstraintakes 2.2. Three-dimensionalFEA ageneralformgivenby[4] σ ¼C (cid:2)ϵ; i;j¼1;2;‥;6; ð2Þ In theory, three-dimensional FEA using layered elements will i ij j σ ϵ accurately simulate deformation in laminate structures. In practice, where, as usual, is stress and is strain. The equation is in however,usingsolidelementsisexpensive,andtheyalsomightgive contractednotationwherei;j¼1;2,and3arex,y,andzcoordinate inaccurate results due to locking. As a compromise, several hybrid axes, while i;j¼4;5, and 6 are yz, zx, and xy planes respectively. methodsthatincorporatetwo-dimensionalplateandshellbehaviors Unlikeisotropicmaterials,Cisdirectiondependentfororthotropic in3DFEAhavebeenproposed.Forexample,solid-shellelementsare3D materials and needs to be transformed from its principal material elements that use Assumed Natural Strain method to deform like directionstoelement'scoordinatedirections.FullmatrixformofEq. platesandshells[6,24].Theirthreedimensionalnatureiswellsuited (2),includingthetransformationofCfromitsprincipaltoageneral for interfacing with other solid elements in assemblies. These ele- coordinatesystem,isgiveninAppendixA.1. ments, however, still require mid-surface extraction and alsocannot The plane–stress constitutive relationship for dimensionally simulatebehaviorsotherthanplateandshellbehaviors. reduced laminates is characterized by a 3(cid:3)3 stiffness matrix Q Continuum solid-shell elements, unlike solid-shell elements, are [4]andisgivenas standard displacement based elements, but use advanced finite σ ¼Q (cid:2)ϵ; i;j¼1;2;3; ð3Þ elementtechniqueslikeAssumedStrainMethodandEnhancedStrain i ij j Methodtoimprovetheirperformanceforthinstructures[10,11].The where i;j¼1 and 2 stand for x and y axes respectively, while higherthenumberofassumedandenhancedparameters,thebetteris 3 stands for the xy plane. Full matrix form of the equation is in the element's performance, but at the expense of generality [10,25]. AppendixA.2. 44 G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 ϵ ϵ In thick plates, out-of-plane shear strains and are ABDmatrixmodelforlaminates: yz xz significant,andtherefore,out-of-planeshearstiffness,inaddition (cid:2)N(cid:3) (cid:2)A B(cid:3) (cid:2)ϵo(cid:3) to in-plane stiffness, are needed to characterize a ply. In an ¼ (cid:2) κ : ð8Þ σ σ M B D arbitrarycoordinatesystem,shearstresses and arerelated yz xz toshearstrainsϵ andϵ as[5]: The individual coefficients of A, B and D matrices for indices yz xz "σ # " # "ϵ # i¼1,2,and3aregivenas Q Q σ4 ¼ 44 45 (cid:2) ϵ4 ; ð4Þ Z t=2 Z t=2 Z t=2 5 Q45 Q55 5 A ¼ Q dz; B ¼ Q zdz; D ¼ Q z2dz: ð9Þ ij ij ij ij ij ij (cid:4)t=2 (cid:4)t=2 (cid:4)t=2 whereindices4and5standforplanesyzandxzrespectively. ThedetailsofthederivationcanbefoundinAppendixB.Intuitively, Let us now assume that, in a laminate, z direction is the A and D are extensional and bending components of the stiffness thickness direction for both the laminate and its plies, which is respectively, while B couples stiffness between bending and alsothethirdprincipaldirectionof theorthotropicplymaterials. stretching that occurs in a laminate if its material properties are Recall that plate theory assumes that the thickness of a plate in asymmetrical about its mid-plane. If B is a non-zero matrix, a stretchingandpurebendingremainsconstant,orinotherwords, ν ν normalpullinxorydirectioncanleadtobendingandviceversa. Poisson's ratios and are zero[6]. This assumption reduces thegeneralstressx–zstrainryezlationinEq.(2)to The out-of-planΓe shearΓstresses can also be reduced to mid- 2 3 2 3 2 3 planeshearforces 4and 5usingaprocedureverysimilartothe σ ϵ 66σx77 66QQ11 QQ12 QQ13 00 00 00 77 66ϵx 77 Eoqnse.u(7se)dantodr(e8d),ulceeadthineginto-planestressestomid-planeforcesNiin 6666666τττxyyyz7777777¼6666666Q01123 Q02223 Q01333 Q044 Q045 00 7777777(cid:2)6666666γγγxyyyz7777777; ð5Þ "ΓΓ45#¼K(cid:2)"AA4445 AA4555#(cid:2)"ϵϵ45#: ð10Þ 4 xz5 4 0 0 0 Q45 Q55 0 5 4 xz5 ϵ ϵ σ ϵ Here,strains and areassumedconstantinthezdirection,and z 0 0 0 0 0 E3 z anydeviation4fromth5eactualfieldiscorrectedusingacorrection wheretheindicesCijarereorderedtomatchthoseofQij.Also,note factorK[5].TheextensionalshearstiffnesscoefficientsA44,A45,and that C33 reduces to E3, Young's Modulus in the third principal A55 aredefinedasAijinEq.(9). materialdirection. Setting transverse Poisson's ratios to zero is invalid when 3D 3.3. ABD-equivalentmaterialmodelsoflaminate stresses,includinginterlaminarstresses,inlaminatestructuresare being computed. However, the assumption is valid for ABD- IfweacceptthatABDmatricesareanaccurateapproximation equivalentmaterialmodelsbecausetheyarebasedonplate/shell ofalaminate'sbehavior,itstandstoreasonthatanytwomaterial assumptions for laminates which ignore interlaminar effects. As models that result in identical ABD matrices should be deemed mentioned in Section 1.1, plate/shell assumptions are reasonable equivalent. In fact, multiple material models with identical ABD assumptionsforregionsawayfromboundariesanddiscontinuities matrices exist, forming an equivalence class of material models. inlaminates,whereinterlaminarstressesarenotsignificant. Fromthisclass,wecanseeksomeofthesimplestmodels,which canbeusedtoreplacetheoriginallaminateinthe3Dintegration procedureinFEA(Fig.4). 3.2. Constitutivemodelforlaminates,orlaminationtheories Any new ABD-equivalent model must satisfy the following equivalencerelationship: The classical lamination plate theory (CLPT) assumes that Z Z slfaotrmraeii,nnsatϵrteoiasianncϵadinactuoarnvnlyaytuuprnoeidneκtriiganotttshhtereelatmcmhidiinn-pagtlaeanncedanapnbudereirseblgaeitnvededninliagns.eTa[4hrl]eyrteo- Aoij¼Z(cid:4)tt=t==222Qoijdz¼ Z(cid:4)t=tt=2=22Qnijdz; Bo¼ Qozdz¼ Qnzdz; ϵi¼ϵoiþz(cid:2)κi; i¼1;2;3; ð6Þ ij Z(cid:4)t=2 ij Z(cid:4)t=2 ij t=2 t=2 where z is the distance of the point from the mid-plane. Fig. 2C Do¼ Qoz2dz¼ Qnz2dz; ð11Þ showstheplotofϵ1atatypicalcrosssectionofalaminate. ij (cid:4)t=2 ij (cid:4)t=2 ij Sincetheamountsofstretchingandbendinginaplatedepend wheretheoriginallaminateQo definesthematricesAo,BoandDo, onlyonthenetforcesandmoments,thestressesinthelaminate's andthenewmaterialmodelQijn resultsinthesameABijDmij atriceisj. cross-section can be reduced to just mid-plane forces N and ij i Since the above integral equations are a system of three σ moments M. This is done by integrating stresses over the i i equations for each entry of ABD matrices, theycan be completely thicknessofthelaminatefrom (cid:4)t=2tot=2andisgivenas determined by a material model Qn that varies in the thickness Z t=2 Z t=2 direction,andthevariationisfullyspijecifiedby3ormoreindepen- Ni¼ σidz and Mi¼ σizdz: ð7Þ dent coefficients. There are infinitely many such models, and any (cid:4)t=2 (cid:4)t=2 two of them are interchangeable if the assumptions made in Combiningthisequationwithplane–stressconstitutiverelation- lamination theory hold. In other words, for the purpose of finite shipfromEq.(3)andstrainfieldfromEq.(6)leadstotheso-called elementanalysis,anyarbitrarilycomplexlaminatewithnumerous Fig.4. OriginallaminatewithmaterialpropertiesQoijcanbereplacedbyanewmaterialmodelQnijthatisABDequivalenttoQoij. G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 45 plies can be replaced by a much simpler material model yielding given laminate. Again, there is a unique quadratically varying identicalresults.Furthermore,sincethenewmodelisvirtual,itis gradedmaterialforthegivenABDmatrices. not subject to manufacturingconstraintsand does not necessarily havetobeply-based. 3.3.3. TransversematerialpropertiesoftheABD-equivalentmaterial Fordemonstration,wechoosetwosuchmaterialmodels:a3- models ply laminate and a quadratically graded material model. In the In addition to in-plane, we also require out-of-plane material nexttwosubsections,thein-planematerialpropertiesofthetwo propertiestocompletelycharacterizetheABD-equivalentmodels. ABD-equivalent models respectively are derived, followed in These out-of-plane, or transverse, material properties can be Section3.3.3bythederivationofout-of-planematerialproperties derived using approaches similar to above, and are common for thatarecommonforboththemodels. boththeABD-equivalentmodels. From Eqs. (9) and (10), the transverse shear properties of a 3.3.1. Three-plymaterialmodel laminatearegivenbytheextensionalstiffnessmatrixA fori;j¼4 ij In the equivalence class of material models with given ABD and 5. Since we are only interested in the equivalent material matrices,a3-plymaterialmodelisthesimplestply-basedmodel. behavior, transverse shear stiffness Qnk can be assumed constant ij Fig.5Ashowsa3-plymodelwithQnkasthematerialpropertiesof along the laminate's thickness. When substituted in Eq. (11), Qn ij ij thekthply,andforsimplicity,eachplyisassumedtobeofequal are obtained as the average values of Ao over the laminate's ij thickness. With these assumptions, Eq. (11) can be solved for thicknesst: materialpropertiesQnk,andintermsofABDmatricesofthegiven ij Ao laminate,aregivenas Qn¼ ij fori;j¼4;5: ð15Þ ij t 36Do(cid:4)18tBo(cid:4)t2Ao 13t2Ao(cid:4)36Do Qn1¼ ij ij ij; Qn2¼ ij ij; Transverse normal stiffness of laminates, which is not at all ij 8t3 ij 4t3 requiredfor2DFEA,isneededfortheABD-equivalentmodels.From Qn3¼36Doijþ18tBoij(cid:4)t2Aoij; ð12Þ Eq. (5), transverse normal stiffness for plate structures reduces to ij 8t3 Young's Modulus E . It is well known that for thin layered 3 wheretisthelaminate'stotalthicknessandindicesi;j¼1;2,and structures, the resultant out-of-plane Young's Modulus E3o can be approximatedastheharmonicaverageoftheYoung'sModulusEokof 3.Clearly,therealwaysexistsaunique3-plylaminatethatisABD- 3 theindividualpliesoftheoriginallaminate[28,4].Again,sincewe equivalent to the original laminate. Note that only the top and are only interested in the equivalent material behavior of the bottom plies depend on the B matrix and capture any material proposedABD-equivalentmodels,theirYoung'sModulusEn canbe asymmetry about the mid-plane. If the original laminate is 3 symmetrical, theBmatrixiszero,andthetwopliesQn1 andQn3 assumedconstantthroughoutthelaminatethickness.Thisassump- ij ij tionmakesEn identicaltoEo: areidentical. 3 3 1 1 kX¼nhk ¼ ¼ ; ð16Þ 3.3.2. Quadraticallygradedanisotropicmaterial En3 Eo3 k¼1Eo3k Insteadofaply-basedmodel,wecanalsoreplacetheoriginal laminate by a continuously varying, or graded, material. Since wherehkisthethicknessofthekthplyandnisthetotalnumberof there are 3 equations to be satisfied, a quadratic variation with pliesintheoriginallaminate. 3 independent coefficients Λk with k¼1;2;3 is sufficient. A To summarize, we can efficiently construct various virtual ij quadratically varying material model is shown in Fig. 5B and is material models that are ABD-equivalent to the original laminate. givenas If the usual lamination theory assumptions hold for the original laminate, these virtual models will result in identical stiffness Qnij¼z2Λ2ijþzΛ1ijþΛ0ij: ð13Þ matrices duringany finiteelementanalysisprocedure.Anyoneof Λk canbefoundfromtheequivalencerelationsinEq.(11),andare thesemodelscanbeusedduringanalysis;however,somemodels ij couldbeeasiertoimplementthanothersinaparticularsystem.For givenas example,the3-plylaminatemodelisstraightforwardtoimplement 15ð12Do(cid:4)tAoÞ 12Bo 3ð3tAo(cid:4)20DoÞ insystems which alreadysupport representationoflaminates. On Λ0¼ ij ij ; Λ1¼ ij; Λ2¼ ij ij : ð14Þ ij t5 ij t3 ij 4t3 theotherhand,thegradedmaterialmodelcanbeusedtoanalyze Interestingly,coefficientsΛk takedifferentrolesinthematerial laminates in systems that are meant for graded materials, but do model: together, the quadratiijc coefficient Λ2 and the constant notsupportlaminates.Inthenextsection,wediscussimplementa- coefficient Λ0 capture the bending and in-pliajne stiffness, while tionofthesetwoABD-equivalentmaterialmodelsanddemonstrate the linear coijefficient Λ1 captures the coupling stiffness of the theirvalidity. ij 4. Implementation Laminates,asproposedbythecurrentandemergingstandards, are commonly represented as a base surface and an associated layup table with an entry for each ply [29]. Base surfaces are generallythetoolingsurfacesonwhichpliesarelaid,andthetable specifiestheorder,materials,andfiberdirectionsoftheindividual plies.Insystemssupportinglaminatesbasedontheabovestandard, implementingthe3-plymodelisstraightforward:theoriginallayup table with any numberof entries is replaced bya new table with only 3 entries. However, for implementing the graded material Fig.5. (A)A3-plyABD-equivalentlaminatemodelwithstiffnessmatrixQinjkforthe model, instead of a table for plies, coefficients Λ defining the kth ply. (B) A graded ABD-equivalent material with its properties varying ij quadratically. quadraticallygradedmaterialinEq.(13)arerequired. 46 G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 We implemented the ABD-equivalent models in a meshfree systemcalledScanandSolve(SnS)[26].InSnS,displacementsand stresses are approximated using multi-variate B-spline functions thatareconstructedoverauniformCartesiangrid.Thischoiceof thebasisfunctionaddressestheconcernsofshearlockingaswell asnumericalill-conditioningofstiffnessmatrixforawiderangeof laminate thicknesses. More details about Scan and Solve can be foundintheReference[26]. Duringfiniteelementanalysis,theABD-equivalentmaterialmodels comeintopicturewhilecomputingelementstiffnessmatricesgivenin Eq.(1).Ifthemeshisconformingandtheelement'sz-axisisalignedto the laminate's thickness direction (Fig.1b), computing volume inte- gration for the 3-ply model is straightforward. This is because, in a Fig.7. (A)Anxycross-sectionoftheCartesiangrid,andanarbitraryfiberinthat conformingmesh,anindividualply'sexactlocationcanbecompletely cross-section.(B)Figurezoomsinoneofthegridelementsandshowsatriangle determined by its position in the z direction. However volume thatisbeingintegrated.Fromthefiberorientation,thematerialprincipaldirections integration is more involved for a non-conforming mesh. Plies can 1and2arefound,whicharenotalignedtoelementdirectionsxandyingeneral. intersectagridelementatarbitraryangles(Fig.1candd),andalsoin structures made of multiple laminates, more than one laminate can AppendixA.1andinmatrixformisgivenas intersectanelement.Asaresult,computingintersectionofeachply Qn0¼GT(cid:2)Qn(cid:2)G; ð17Þ with an element can be both complicated as well as expensive. Therefore, for an ease of implementation, we approximate volume whereGisthetransformationmatrix. integrationbyintegrationoversurfaces:eachlaminateisreplacedbya Finally, we assume plate behavior everywhere in our simple setof surfaces paralleltothe toolingsurface (Fig.6).These surfaces, benchmark laminate problems, with no distinction between plate whichwecallintegrationsurfaces,canbeeasilygeneratedastooling and non-plate regions. This assumption is justified because we surface's offsets, a standard geometric operation in a CAD software. compareourresultswiththoseobtainedusingdimensionallyreduced Thelocationoftheseintegrationsurfacesinthelaminate'sthickness models of laminates. We skip rest of the implementation details as directioncanbeobtainedusingoneofthequadraturerules,andinthe theyarenotdirectlyrelevanttothecurrentwork. currentimplementation,weusetheLobattoquadraturerules[30].In addition to simplifying volume integration, this integration scheme also makes implementation of the ABD-equivalent graded model 5. Numericaltestresults much easier: since an integration surface is an offset at a constant distancefromthelaminate'smid-plane,coefficientsΛ ofthegraded In this section, we compare results of linear static analysis ij material (Eq. (13)) are also constant within that integration surface computed using ABD-equivalent materials to results in the stan- and, therefore, need to be computed once. Integration over each dard reference text [5], as well as to results computed using surface is done by first triangulating it, and then integrating the commercial software SolidWorks [31] and ANSYS [32]. Since we obtained triangles using Gauss quadrature rules. The triangles are areusing3DmeshfreeFEAimplementation,wedonotexpectitto constrainedtoconformtotheCartesiangrid,orinotherwords,each be more efficient than highly optimized commercial 2D FEA trianglecompletelylieswithinanelementoftheCartesiangrid. solutions. Our only goal is to demonstrate the accuracy of ABD- Whileintegratingoverthetriangles,wealsoneedtotransform equivalent material models in comparison with other analytical thematerialmatrixQn fromitsprincipalcoordinatesystemtothe andcomputationalmethodsthatrelyontheABDformulation.This elementcoordinatesystem.AsexplainedinFig.7,foreverytriangle, alsoexplainswhywedonotcomparetheaccuracyofourresults we transform Qn once for the triangle's centroid, and use the to more accurate models based on higher-order lamination the- transformed properties Qn0 for all the quadrature points of that oriesorfull3Delasticityformulations. triangle. The transformation relation is explained in detail in We are considering four benchmark problems: a rectangular plate,acylindricalshell,acylindricalroof,andalapjoint.Thefirst threestructuresaresinglemulti-plylaminates,whilethelapjoint is a bonded assembly of two laminates. For each of the test structures,weconsiderthefollowingthreeconfigurations: 1. Cross-plylaminates½0=90(cid:5) , n 2. Angle-plylaminates½(cid:4)45=45(cid:5) , n 3. 50-plylaminates(AppendixD). Cross-ply and angle-ply laminate configurations are often standard configurations.The50-plylaminatewaschosentotestiftheproposed method scales for a large number of plies. The plies in the 50-ply laminatewereselectedrandomly,andtheirorientationsaregivenin Appendix Appendix D. We restricted the number of plies to 50 because that was the maximum number of plies supported by SolidWorks.Thethreelaminateconfigurationsareasymmetricalabout theirmid-plane,andthereforeshowstretching–bendingcoupling,as predictedbytheclassicallaminationtheory.Allpliesineachlaminate areassumedtobemadeofoneofthetwomaterials: Material 1: E ¼25:0e6psi, E ¼E ¼1:0e6psi, ν ¼0:25, ν ¼ 1 2 3 12 23 Fig.6. LaminatestructureinFig.1replacedbyintegrationsurfaces. ν13¼0:0,G12¼G13¼5:0e5psiandG23¼2:0e5psi, G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 47 Fig.8. (A)Platewithgeometryparametersa¼10inandh¼1;0:1;0:01in,clampedfromallfoursideswithasurfacepressure,q¼100psi.(B)Platewitha¼10inand h¼0:1in,fixedononeendwithaforce,F¼1:0e4lbf ontheoppositeend. Table1 MaximumdisplacementvalueininchesforplateprobleminFig.8Afordifferentcases. Laminate Thickness a=h ANSYS SW SnS–3-Ply SnS–Graded (Numberofelements) 10k 1k 1k 3k 1k 3k 1 10 4.748e(cid:4)3 4.058e(cid:4)3 3.855e(cid:4)3 3.762e(cid:4)3 4.567e(cid:4)3 3.755e(cid:4)3 ½0=90(cid:5)5 0.1 100 1:543eþ0 1:550eþ0 1:532eþ0 1:543eþ0 1:532eþ0 1:543eþ0 0.01 1000 1:510eþ3 1:552eþ3 1:145eþ3 1:661eþ3 1:145eþ3 1:662eþ3 1 10 5.094e(cid:4)3 4.360e(cid:4)3 4.057e(cid:4)3 4.152e(cid:4)3 4.880e(cid:4)3 4.050e(cid:4)3 ½(cid:4)45=45(cid:5)5 0.1 100 1:629eþ0 1:620eþ0 1:597eþ0 1:611eþ0 1:598eþ0 1:611eþ0 0.01 1000 1:581eþ3 1:578eþ3 1:163eþ3 1:684eþ3 1:163eþ3 1:685eþ3 Material 2: E1¼7:5e6psi, E2¼E3¼2:0e6psi, ν12¼0:25, ν23¼ Table2 ν ¼0:0,G ¼1:25e6psi,andG ¼G ¼0:625e6psi, VonMisesstressatthemid-point(x¼0,y¼0)ofthetopfaceoftheplateinFig.8A. 13 12 13 23 ν whereE, andGareYoung'sModulus,Poisson'sRatioandShear Laminate Thickness a=h ANSYS SW SnS–3-Ply SnS–Graded Modulus respectively, and indices 1, 2 and 3 represent the three principalmaterialdirections.Thesematerialsareidenticaltothose (Numberofelements) 10k 1k 1k 3k 1k 3k ν ν usedin[5]thatweuseforcomparison,exceptweset and 23 13 tobezero. 1 10 2.42e3 2.43e3 2.50e3 2.40e3 3.20e3 2.35e3 ½0=90(cid:5) 0.1 100 2.54e5 2.55e5 2.04e5 2.06e5 2.05e5 2.06e5 Finally we note that the elements available for analyzing 5 0.01 1000 2.50e7 2.60e7 1.60e7 1.94e7 1.60e7 1.93e7 laminatesinthethreesystemsaredifferent: 1 10 2.70e3 2.66e3 2.67e3 2.60e3 3.20e3 2.60e3 ½(cid:4)45=45(cid:5) 0.1 100 2.25e5 2.26e5 1.80e5 1.82e5 1.80e5 1.82e5 5 Solidworks: We used two-dimensional parabolic triangular shell 0.01 1000 2.25e7 2.22e7 1.52e7 1.70e7 1.52e7 1.69e7 elements. ANSYS: Weusedtwo-dimensionalShell181elementsforanalyz- ingsinglelaminatestructures,andSolid186elementsfor and Solve can successfully capture bending in thin structures, analyzingthelapjoint.Shel181are4nodeelementswith since conventional 3D basis functions tend to underestimate 3 displacement and 3 rotational degrees of freedom at bending deformations due to locking. The plate was made of each node. A penalty method is used to relate the Material1,andconsistedof10plieslaidincross-plyandangle-ply independent rotational degrees of freedom about the configurations. normal (to the shell surface) with the in-plane compo- Table 1 compares the maximum displacements using ANSYS, nents of displacements. Solid186 elements are 20-node SolidWorks (SW), and proposed method (SnS) for different lami- layered solid elements that exhibit quadratic displace- nates. Tests were done for three different aspect ratios: thin (a/ mentbehavior[32]. h¼1000), moderately thick (a/h¼100), and thick (a/h¼10). For Scan&Solve: Each benchmark problem is solved using 1000 and both cross-ply and angle-ply laminates with moderate thickness, 3000 second-order tri-variate B-spline functions on a SnS accurately predicts the maximum displacement values, and uniform Cartesian non-conforming grid. The Lobatto theresultsfromallthesystemsareincloseagreement.Thereare quadrature rule implies that 3 integration surfaces per morenoticeabledifferencesinthedisplacementscomputedbythe ply for the 3-ply laminate model, and 4 integration threesystemsforthinandthicklaminates,e.g.,SnSandSWtend surfaces for the quadratically graded laminate model, todifferby5%.Importantly,SnSisnotunder-predictingdisplace- aresufficientforfullintegration. mentsforthinplates,suggestingthatlockingisnotanissue. Wealsocompare,inTable2,theVonMisesstressesatthemid- point ðx¼0;y¼0Þ of the top face for different plates. For both 5.1. Clampedrectangularplate cross-plyandangle-plylaminates,stressesfromdifferentmethods areinagreementforthethickplatewithaspectratio10,butthereis Our first benchmark problem is a plate clamped on all four somedeviationforotheraspectratios,whichincreases withplate sides,andanormal pressureonthetopsurface (Fig.8A).Aplate aspectratio.Acomparisonoftheentirestressfieldforthemoderate under these boundary conditions shows pure bending, with thicknessplate(aspectratio100)isshowninFig.9forcross-plyand maximumdisplacementatthecenteroftheplate.Thisparticular angle ply laminates. The stress patterns match for the two lami- problemwaschosentotestifsecondorderB-splinebasisinScan nates, and the highest stress due to stress concentration are also 48 G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 Fig.9. FigurecomparesVonMisesstressfortopfaceoftheplateinFig.8withaspectratio100for(A1)½0;90(cid:5) analyzedinANSYS(Maxstress-5:83e5),(A2)½0;90(cid:5) analyzed 5 5 usingtheproposedmethod(Maxstress-4:91e5),(B1)½(cid:4)45;45(cid:5) analyzedinANSYS(Maxstress-4:08e5),(B2)½(cid:4)45;45(cid:5) analyzedusingtheproposedmethod(Maxstress- 5 5 3:44e5). Fig.10. Figureshowsdeformedplateandcolormapofout-of-planedisplacementobtainedusing(A)ANSYS(B)Scan&SolvewithABD-equivalent3-plymaterialmodel,fora laminateplatemadeof50plies.Asexpected,thein-planeloadleadstoout-of-planebending.(Forinterpretationofthereferencestocolorinthisfigurecaption,thereaderis referredtothewebversionofthispaper.) G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 49 Fig.13. Figureshowingbarrelvaultproblem.Verticalpressureisq¼0:625psiand thecurvedendsarefixed.Otherdimension:β¼801,R¼300in,a¼600in,h¼3;6 and15in. Fig.11. ClampedcylinderwithPo¼2:04ksi,R¼20in,a¼20in,h¼1in. z displacement are in agreement with 0:3730in in ANSYS and 0:3734ininSnS. Wealsocarriedoutatimeanalysisinordertoestimatethenet Table3 efficiency achieved using proposed method when using 3D FEA. Maximumradialdeflection(e(cid:4)1in)forcylindricalshellinFig.11.Q4andQ9results Complete analysis of the 50-ply laminate plate using the ABD- from[5].SW-SolidWorks,SnS-ProposedMethod. equivalent3-plymodeltook14.9s,outofwhich12.8swerespent Laminate Q4 Q9 ANSYS SW SnS–3-Ply SnS–Graded integrating 9 integration surfaces (3 surfaces per ply). Therefore, an average of 1.42s were spent integrating each surface. This 16 4 15k 1.2k 1k 3k 1k 3k implies that integrating over 150 surfaces in the original 50-ply ½0=90(cid:5) 1.870 1.803 1.706 1.848 1.820 1.773 1.820 1.773 modelwouldrequireroughly215sforthesameanalysis.Thegain ½(cid:4)45=45(cid:5) – – 2.204 2.350 2.356 2.291 2.355 2.290 inefficiencyisevenhigherwhenusinggradedmaterialmodel,as ½0=90(cid:5)5 – – 1.719 1.830 1.814 1.776 1.815 1.776 itneedsonly4integrationsurfacesincomparisonto9forthe3- ½(cid:4)45=45(cid:5) – – 2.210 2.340 2.334 2.282 2.334 2.281 5 ply laminate. The total time taken for analysis was only 6.8s, 50-ply – – 2.350 2.542 2.465 2.424 2.464 2.424 decreasingthetotalcomputationcostofanalysisbymorethan30 times. Similar speedup should be expected in any 3D FEA of within 15% of each other. Solid elements generally better capture laminatedstructuresrelyingonlayeredelements. thestressesnearedgesastheyexplicitlymodelthefinitethickness ofplates. 5.2. Clampedcylinderwithinternalpressure Next,wetestthesamerectangularplate,butmadeof50random plies.Theboundaryconditionsaredifferentfromtheprevioustests; Next, we consider a cylinder, made of Material 2, that is the plate is under in-plane load of 1:0e4lbf on one end and clamped at the two ends and has an internal pressure of P . o clamped at the opposite end (Fig. 8B). This particular problem DetailsofgeometryandboundaryconditionsareshowninFig.11. waschosentovalidatetwoimportantclaims:(a)proposedmethod ThemaximumradialdeflectionsfromReddy[5],ANSYS,Solid- can handle large number of plies, (b) proposed ABD-equivalent Works, and proposed method are comparedin Table 3.Elements material models successfully capture coupling behavior in lami- used by Reddy [5] are Q4 elements, a four-node (linear) quad- nates that are asymmetrical about the mid-plane. Due to this rilateralelement,andQ9elements,anine-node(quadratic)quad- coupling, the in-plane load F will lead to bending and produces rilateral element. The total number of elements used in different out-of-plane deformation. The proposed method did capture this methodsarespecifiedinthesecondrowofTable3.Thetestswere coupling phenomena accurately as shown in Fig.10, which com- performed forone aspect ratio,but 5 differentlaminates.Results paresthezdisplacementfieldsfromSnSandANSYS.Themaximum fromall the methods are clearly in agreement. Fig.12 shows the Fig.12. Figureshowsdeformedcylinderandthecolormapofradialdisplacementobtainedusing(A)ANSYS(B)Presentmethod,foralaminatecylindermadeof50plies. (Forinterpretationofthereferencestocolorinthisfigurecaption,thereaderisreferredtothewebversionofthispaper.) 50 G.Kumar,V.Shapiro/FiniteElementsinAnalysisandDesign106(2015)41–55 Table4 Barrelvaultproblem-maximumdeflection(in.)forcross-plyandangleplylaminates. Laminate R/h Reddy ANSYS SW SnS–3-Ply SnS–Graded (No.elements) 16 10k 1.2k 1 k 3k 1k 3k 100 2:339eþ0 2:407eþ0 2:460eþ0 2:307eþ0 2:415eþ0 2:307eþ0 2:416eþ0 ½0=90(cid:5) 50 5.082e(cid:4)1 5.291e(cid:4)1 5.659e(cid:4)1 5.480e(cid:4)1 5.810e(cid:4)1 5.496e(cid:4)1 5.503e(cid:4)1 20 7.292e(cid:4)2 7.449e(cid:4)2 7.560e(cid:4)2 7.877e(cid:4)2 8.067e(cid:4)2 7.956e(cid:4)2 7.016e(cid:4)2 100 3:597eþ0 3:871eþ0 3:866eþ0 3:411eþ0 3:743eþ0 3:413eþ0 3:699eþ0 ½(cid:4)45=45(cid:5) 50 6.760e(cid:4)1 7.652e(cid:4)1 7.170e(cid:4)1 6.675e(cid:4)1 7.157e(cid:4)1 6.671e(cid:4)1 7.780e(cid:4)1 20 1.205e(cid:4)1 1.397e(cid:4)2 1.130e(cid:4)1 1.061e(cid:4)1 1.127e(cid:4)1 1.131e(cid:4)1 1.386e(cid:4)1 100 1:415eþ0 1:434eþ0 1:564eþ0 1:542eþ0 1:593eþ0 1:540eþ0 1:593eþ0 ½0=90(cid:5) 50 2.940e(cid:4)1 2.979e(cid:4)1 3.270e(cid:4)1 3.335e(cid:4)1 3.412e(cid:4)1 3.337e(cid:4)1 3.412e(cid:4)1 5 20 5.234e(cid:4)2 5.246e(cid:4)2 5.370e(cid:4)2 5.361e(cid:4)2 5.398e(cid:4)2 5.361e(cid:4)2 5.399e(cid:4)2 100 1:818eþ0 1:836eþ0 1:955eþ0 1:821eþ0 1:912eþ0 1:821eþ0 1:912eþ0 ½(cid:4)45=45(cid:5) 50 4.096e(cid:4)1 4.082e(cid:4)1 4.089e(cid:4)1 3.796e(cid:4)1 3.940e(cid:4)1 3.799e(cid:4)1 3.940e(cid:4)1 5 20 1.004e(cid:4)1 9.727e(cid:4)2 8.959e(cid:4)2 7.856e(cid:4)2 8.009e(cid:4)2 7.857e(cid:4)2 8.088e(cid:4)2 50-ply 100 – 1.4106e0 1.478e0 1.391e0 1.445e0 1.391e0 1.445e0 Fig.14. Figureshowsdeformedshellandcolormapofout-of-planedisplacementobtainedusing(A)ANSYS(B)Presentmethod,foralaminateplatemadeof50plies.(For interpretationofthereferencestocolorinthisfigurecaption,thereaderisreferredtothewebversionofthispaper.) Table5 MaximumdisplacementvalueininchesforlapjointinFig.15.Secondrowshows thenumberofelementsused. Laminate ANSYS SnS–3-Ply SnS–Graded 760 1k 3k 1k 3k ½0;90(cid:5) 1.251 1.147 1.212 1.147 1.212 5 ½(cid:4)45;45(cid:5) 7.975 6.540 7.641 6.540 7.641 5 50Plies 3.304 2.937 3.194 2.938 3.195 deformed shape, aswell as, the color map of displacement value for the cylinder made of 50 plies. Maximum radial displacement usingANSYSis0:2350in,andusing3-plymaterialmodelinScan Fig.15. Figureshowsalapjointmadeoftwolaminatesidenticalingeometry,with andSolveis0:2419in,whichislessthan3%difference. a¼10in,b¼2in,c¼2:5inandh¼0:1in.Theleftendisfullyfixed,andtheright endisallowedtoslideinthexdirection.AforceF¼10e4lbisalsoappliedonthe 5.3. BarrelVaultproblem rightend. In this section, we consider another popular benchmark shell laminate (see Fig. 14). The maximum displacement values from problemknownastheBarrelvaultproblem[5]:acylindricalroof Scan&Solve using 3000 elements and ANSYS are within 2.5% of underitsownweight.ThestructureismadeofMaterial1,andthe eachother(Table5). detailedboundaryconditionsareshowninFig.13. As before, when analysing 50-ply laminates using layered 3D Inreference[5],maximumverticaldisplacementsaregivenfor elements,the3-plylaminateimprovesefficiencyroughly15times, cross-plyandangle-plylaminateswithdifferentaspectratios.We whereas the graded material was 30 times more efficient than usethesevaluesforcomparingvaluesobtainedusingSW,ANSYS, usingtheactuallaminatelayup. andproposedmethodinTable4.ReddyusedQ81elements,which areeighthorderelement(p¼8)with405degreesoffreedom.For 5.4. Multi-laminatestructure—lapjoint othermethods,elementsaresameasbefore.Again,thereisafairly closeagreementbetweentheresultsfromReddy[5],SolidWorks, Finally, to show that proposed method can be extended to ANSYS and proposed method in all cases, including the 50-ply structures made of multiple laminates, we analyze a lap joint
Description: