<<

Key Engineering Materials Vol. 383 (2008) pp 35-52 online at http://www.scientific.net © (2008) Trans Tech Publications, Switzerland

Influence of Porosity on the Interlaminar Shear Strength of Fibre-Metal Laminates

Cláudio S. Lopes 1, a , Joris J.C. Remmers 2,b and Zafer Gürdal 1,c 1Faculty of Aerospace Engineering, Delft University of Technology, PO Box 5058, 2600 GB, Delft, The Netherlands 2Department of Mechanical Engineering, Eindhoven University of Technology, PO Box 513, 5600 MB Eindhoven, The Netherlands [email protected], [email protected], [email protected]

Keywords: Delaminations, Cohesive Elements, Fibre-Metal Laminates, Glare, Porosity, Fracture Toughness.

Abstract. Structures manufactured in fibremetal laminates (e.g. Glare) have been designed considering ideal mechanical properties determined by the Classical Lamination Theory. This means that among other assumptions, perfect bonding conditions between layers are assumed. However, more than often, perfect interfaces are not achieved or their quality is not guaranteed. Wheninlaboratory,highqualityfibremetallaminatesareeasilyfabricated,butintheproduction linethecomplicatedmanufacturingprocessbecomesdifficulttocontrolandtheoutcomeproducts maynotmeetthequalityexpected.Oneoftheconsequencesmaybethepooradhesionofmetal prepregorprepregprepregastheresultofporosity.

Theinterlaminarshearstrengthoffibremetallaminatesdecreasesconsiderably,duetoporosity,as the result of insufficient adhesion between layers. Small voids or delaminations lead to concentrations at the interfaces which may trigger delaminationpropagation at the aluminium prepregandprepregprepreginterfacesatloadlevelssignificantlylowerthanwhatisachievablefor perfectly bonded interfaces. Mechanical experiments show a maximum drop of 30% on the interlaminarshearstrength.

Inthepresentwork,theeffectsofmanufacturinginducedporosityontheinterlaminarshearstrength offibremetallaminatesarestudiedusinganumericalapproach.Theindividuallayersaremodelled by continuum elements, whereas the interfaces are modelled by cohesive elements which are equippedwithadecohesionlawtosimulatedebonding.Porosityisincludedinthegeometryofthe interfacebysettingsomeoftheseelementstoapredelaminatedstate.

Introduction Porosityisoneofthemostcommonmanufacturinginduceddefectsincompositelaminates.Small porositiesareformedduetotheentrapmentofairinthematrixduetothemoistureabsorbedduring materialstoring,processingandapplication.Larger,elongatedvoidsareformedattheplyinterfaces duetoinadequatecuringcycles[1].Moreover,inserviceconditionsmaycontributetothegrowthof thesedefects.Porosityhasadetrimentaleffectontheloadingresponseoflaminatesespeciallyon the interlaminar shear strength (ILSS), compressive strength and strength that are associatedwithmatrixdominatedmechanicalproperties[2,3].Sincethelargervoidsarelocatedat theplyinterfaces,thelargerinfluenceisexpectedontheILSSoflaminates. TheproductionofFibreMetalLaminates(FML’s)followsmethodssimilartotheonesemployed intheproductionoftraditionalcompositelaminates.Hence,porosityisanissuewithFML’saswell

All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of the publisher: Trans Tech Publications Ltd, Switzerland, www.ttp.net. (ID: 132.229.212.106-23/05/08,16:47:04) 36 Progress in Fracture and Damage Mechanics

(Figure1).ItseffectontheILSSofFML’scouldbeascertainedbycarryingoutmechanicaltests andemployinganalyticalmethods.Inthepast,Gürdaletal.[4,5]studiedtheinterlaminartensile and bending strengths of AS4/35016 graphite/epoxy laminates by conducting an extensive experimentalprogrammewhichincludedthecharacterizationoftheporesintermsofporevolume fraction,geometry,sizeandorientation.Theporositydatawerethenusedinanempiricalmodelto predictthelaminatestrengthasafunctionofporosity.Jeong[2]andCostaetal.[3]investigatedthe influence of porosity on the ILSS of graphite/epoxy systems by means of experimental and analyticalapproaches.Inbothworks,ashearfracturecriterionwasusedsuccessfullytocorrelatethe fracturestrengthofporouslaminates. Althoughpossible,itisdifficulttoexactlycontroltheamountandconfigurationofporosityat theinterfacesofthetestspecimens[4,5].Anindepthexperimentalinvestigationoftheeffectsof porosityinmultipleinterfacesofFML’sisoutofthescopeofthisprogramme.Byrelyingonafew experimental data and observations, the influence of various degrees of interfacial porosity is studied by means of numerical methods developed in the framework of Damage and Fracture Mechanics and implemented in the Finite Element Method (FEM) package ABAQUS ® [6]. Nowadays, finite element (FE) analysis represents a relatively fast and inexpensive tool in the analysisofstructurallyloadedcomponents. Over the past recent years a great effort has been made to implement fracture mechanics phenomenainthesimulationofdelaminationgrowthinlaminatedmaterialsonamesoscopiclevel ofobservation[710].Thisefforthasledtothedevelopmentofreliablenumericaltoolscapableof dealing with delamination onset and propagation under a large variety of loading modes. In the presentworktheseadvancedtoolsareusedtostudythestructuraleffectsofinterfaceporosityby actually including voids in the geometry of the interface. As opposed to the use of adhesion degradation parameters at a material level, this approach leads to the correct prediction of the locationofdelaminationonsetandacorrectpredictionoftheILSS.

Figure1.SEMpicturesofGlare.Left:specimenwithnoporosity.Right:Specimenwithporositydueto inadequatecuringcycle. Delamination growth in the different prepregprepreg and aluminiumprepreg interfaces is simulatedbycohesiveelements[810].Theseareanumericalimplementationofthe cohesive zone approach[7]inwhichtheenergyinvolvedinthedelaminationprocessisbeingdissipatedinasmall "processregion"aheadofthecracktip.Theopeningofthecohesivezoneisgovernedbyamixed mode delamination model [810], which accounts for both mode I (normal) and mode II (shear) delaminationsandcombinationsthereof. ThesimulationoftheILSStestisexemplifiedbymodellingaspecimenofGlare34/30.4with and without porosities included. Porosity is modelled by predelaminating a number of interface elements.Informationaboutthesize,positionanddistributionofporositiesattheinterfacesofan Key Engineering Materials Vol. 383 37

ILSS specimen was roughly obtained from a number of SEM pictures taken from actual test specimens(Figure1).

ParameterIdentification Besides the wellknown elastic and plastic properties of aluminium 2024T3 and glassfibre prepregs, additional mechanical properties are needed to characterise the delamination process. These are the fracture toughness ( Gc) and ultimate tractions ( τu) of each interface at all possible loadingmodes.Specificstandardtestsweredevisedandareavailableintheliteraturetomeasure theseproperties for fullcomposite laminates, namely de Double Cantilever(DCB)testfor modeI,theEndNotchFlexure(ENF)testformodeIIandtheMixedModeBending(MMB)test formixedmodedelaminationpropagation[12].Afterafewadaptations,thesefractureenergytests wereconductedforFML’saswell[11].Theparametersoftheconstitutivemodelobtainedfromthe experimentsarecheckedagainstanalyticalcalculations[13]andnumericalsimulations.Inthisway, trustable numerical tools and mechanical material properties are used to model the ILSS experiment. Themostimportantvalueinthecharacterisationofdelaminationisthefracturetoughness,dueto itsglobalinfluenceinthefracturebehaviour.Theultimatetractionvalueshaveaminorinfluencein thedelaminationprocess(especiallyinmodeI),sincetheyinfluenceonlythedelaminationinitiation loaddisplacement point. Due to the difficulty in obtaining accurate measurements, approximate valuesof τuaretaken( τIu =75MPaand τIIu =90MPa),tunedinsuchawaytoproducethebestmatch betweenexperiments,analyticalandnumericalcalculations. In order to measure the mode I fracture energy of the interface, a DCB test procedure was followed[11].AssketchedinFigure2,itconsistsofpullingthetwotipsofa250x25x9.5mmpre crackedstripoflaminate.Thiscausesthecracktopropagate.Thespecimenispreviouslycrackedat agiveninterfaceinordertoavoidtheveryhighloadpeak,whichotherwisewouldbenecessaryto initiate delamination. The high energy dissipation at delamination initiation would hinder the accurateevaluationofthefractureenergy.Theaveragedfracturetoughnessistheenergydissipated per unit area of the new crack. This energy corresponds to the area underneath the traction displacementjumpcurve.Infullcompositespecimens,virtuallyalltheenergytransferredisstored inelasticbendingordelaminationpropagation.ThisisnotthecasewithFML's,whichcansustain plastic deformation in the metal layers. In such a case it would be very difficult to quantify the amountofenergyactuallyspentonthefractureprocess.Inordertomaintainthewholespecimenin theelasticregime,two4mmthickplatesofaluminium7075T6werebondedtotheGlare32/10.4 strip.

Al7075T6

Pre Glare3 2/1 0.4 crack Al7075T6

250mm Figure2.IllustrationofaDCBtest.Forthisconfigurationa50mmlongprecrackisincludedattheprepreg prepreginterface. ThemodeIIfractureenergywasmeasuredwiththeENFtest[11].Asimilar,initiallycracked specimen is used but now subjected to a 3point bending loadcase where the loading point is 38 Progress in Fracture and Damage Mechanics located at 80mm of each support. In this case the delamination is not progressive, but rather catastrophic, i.e. once the delamination is started, the crack rapidly grows to the centre of the specimen. Such event produces a sudden decrease in bending stiffness, which results in a remarkableloaddrop.Thismakesthemeasurementof GIIcmoredelicatethan GIc .

Figure3.MMBtestapparatus(after[10]). ModeIandmodeIIdelaminationsareparticularexamplesofthewiderangeofloadingsituations astructuralcomponentmayundergofromwhichmixedmodeloadingisthemostcommon.Most reallife delaminations initiate and propagate under the influence of combined normal and shear stresses. The mixedmode fracture propagation was investigated by means of the MMB test illustrated in Figure 3 [11]. Here the specimen is loaded by a combination of normal and shear forces, producing a specified ratio between the energies dissipated in mode I and mode II delamination,respectively.Forthesakeofbrevity,onlyanevencombination(50/50ratio)ofmodes was considered in these experiments. This means that 50% of energy is consumed in mode I delamination and 50% in mode II delamination. The fracture toughness values measured for the threetypesofinterfacesofGlare32/10.4arereportedinTable1. G G G Interface Ic IIc I/IIc [J/m 2] [J/m 2] [J/m 2]

AlLdirection/fibres0 ° 2960.8 1705.8 757.2 Fibre90 °/fibres0 ° 3545.5 1349.4 672.7 AlLdirection/fibres90 ° 3411.9 1623.1 622.7 Table1.ModeI,modeIIandmixedmodefractureenergyvaluesmeasuredforthethreetypesofinterfacesin Glare32/10.4(after[11]). ThefracturetoughnessvaluesreportedinTable1fortheseveralinterfacesofGlare32/10.4are somewhatdifferentfromwhatwasexpected.Ingeneral,forcarbonfibrereinforcedlaminates,G Ic is lower than G IIc and the mixedmode fracture values fall in between these. Glare 3 shows lower fracture toughness values for mode II loading than for mode I and the values corresponding to mixedmodeloadingareevenlower.Thisiscounterintuitivesince,whenloadedundermixedmode, theenergydissipatedbyacrackcorrespondstoaspecifiedratiobetweentheenergiesdissipatedin modeIandmodeIIloading.Korjakins[14]foundthatforglassfibrereinforcedplies,astheones usedinGlare3,theresultsoftheDCBtestarehighlydependentonthefibresurfacetreatment.Ifno surfacetreatmentisappliedtothefibres,thecrackpropagationisaccompaniedbyextensivefibre bridging.Thisphenomenonopposesthecrackpropagation,hence“artificially”increasingthevalues ofG Ic . Ontheotherhand,thecatastrophicfailureoftheENFandMMBtestsdoesnotallowforstable crackpropagation.TheFourPointBendingexperiment,describedin[15],isanalternativetestthat Key Engineering Materials Vol. 383 39 allowsthestabledelaminationpropagationofmodeIIandmixedmodecracksandconsequentlythe measurementofGIc andGI/IIc withhigheraccuracy.Nevertheless,thevaluesreportedinTable1are theonesusedforthepurposeofthepresentinvestigation.

AnalyticalandNumericalAnalyses TheexperimentsaresimulatedusingtwodimensionalFEAmodelswheretheindividuallayersare modelled with standard geometrically nonlinear continuum elements. These elements canbehave accordingtoaplasticplanestrainconstitutiverelation.Duetotheirgeometricalsimplicity,itisalso possible to simulate the fracture energy tests analytically by using Beam Theory and Fracture MechanicsTheory[13].IntheFEmodels,thebondbetweeneachtwolayersissimulatedbymeans of a cohesive zone [7]. In this approach, the fracture behaviour (delamination) is lumped into a singleplane,whichisrepresentedbyinterfaceelementsplacedbetweentwolayers.Theseinterface elementsconsistoftwosurfaces,whichareattachedtotheadjacentcontinuumelementsmodelling thelayers(Figure4).Therelativedisplacementofthetwosurfacesisameasurefortheopeningof thedelaminationcrack.Theopeningiscontrolledbymeansofacohesiveconstitutiverelationthat completelycharacterisesthedelaminationprocess[810].

Figure4.IllustrationoftheFEsimulationofdelaminationinamultiplymaterial.Theindividuallayersofthe specimeninthetopleftpicturearemodelledbycontinuumelementsandtheadhesivethatbondsthelayersis modelledbyinterfaceelements(therighthandsidepictures).Therelativedisplacementoftheinterfaceelements (bottomleft)isameasurefortheopeningoftheinterface. AssuggestedbyCamanhoetal.[810],abilinearcohesiverelationisusedhere(Figure5).This model is based on the input of four materialparameters: mode I and mode II fracture toughness values( GIc and GIIc )andthecorrespondingultimatetractionvalues( τIu and τIIu )atwhichdebonding is initiated. A fifth parameter, η, is necessary to completely define the mixedmode propagation criterionasfunctionof GIc and GIIc only.Thisvaluemustbeextractedfromthecorrelationofthe testdatafordelaminationinmodeI,modeIIandmixedmodeloading.Whenloadinganinterface, the cohesive elements initially behave in a linearelastic way (point 1 in Figure 5). When the equivalenttractionexceedsalimitvalue,basedonthepuremodeIandIIultimatetractions τIu /τIIu (Point2inFigure5),damageisinitiated.Ifthedisplacementsarefurtherincreased,thestiffnessis gradually reduced to zero and the delamination starts topropagate. A cohesive elementbecomes fully delaminated when it is unable to transfer any further load (points 4 and 5 in Figure 5). However,itisnecessarytoavoidinterpenetrationofthecrackfaces.Theproblemisaddressedby reapplyingthenormalpenaltystiffnesswheninterpenetrationisdetected. 40 Progress in Fracture and Damage Mechanics

Figure5.Puremodeconstitutiveequations:(a)ModeIIThesheartractionasafunctionoftheshearopening forazeronormalopening.(b)ModeINormaltractionacrosstheinterfaceasafunctionofthenormal displacementwhenthesheardisplacementisassumedtobezero(after[8]).Notethatthereisnosofteninginthe caseofanegativenormalopening.Thisresemblesselfcontactoftwolayers.Thesurfaceunderthecurvesisthe fracturetoughnessofthematerial.

DCB,ENFandMMBTestSimulations TheDCB,ENFandMMBtestsaresimulatedinthecommercialFEcodeABAQUS ®[6]bymeans oftwodimensional,planestrainmodels.Thealuminiumandprepregsolidpartsaremodelledwith 4nodesolidelements.Theinterfaceismodelledbyuserdevelopedcohesiveelements[9,10].The solidelementsbehaveaccordingtoalinearelasticconstitutivelaw.Asexplainedpreviously,plastic behaviourwaspreventedbybondingtwothickplatesofstrongaluminiumtoeachsideoftheGlare 32/10.4 laminate specimen (Figure 2). For both aluminium alloys (Al 2024T3 from the Glare laminate and Al 7075T6 from the bonded plates) equal isotropic properties are defined, with properties:E=73GPaand ν=0.33.Theorthotropicpropertiesoftheglassfibreprepregweretaken from[16]andarereportedinTable2.

E11 [MPa] 55000 ννν12 0.195 G12 [MPa] 5500 E22 [MPa] 9500 ννν13 0.195 G13 [MPa] 5500 E33 [MPa] 9500 ννν23 0.06 G23 [MPa] 3000 Table2.Glassfibreprepregproperties(after[16]) In the DCB FE model, a minimum of two elements are required in thickness direction of the aluminiumlayersinordertoavoidthephenomenonofshearlocking.Alternatively,incompatible modeelementsmaybeusedbuttheserequiremorecomputationalresourcesthanstandardelements. In the longitudinal direction, the required number of elements is function of the length of the cohesive zone [17]definedby:

Gc lcz = E 2 . (1) τ u

InEquation1,EistheYoung'smodulus, Gcisthefracturetoughnessandτuistheultimatetraction. When the cohesive zone is discretised by a coarse mesh, the fracture energy is not accurately represented and the model does not capture the continuum field of a cohesive crack. Experience shows that three elements along the cohesive zone are sufficient to predict the propagation of delaminationinModeI.Thismeansthatinthepresentcasethelengthofthelengthoftheinterface elementsshouldnotexceed0.5mm.InModeII,acoarsermeshmaybeused. Key Engineering Materials Vol. 383 41

(a)Delaminationinitiation

(b)Endoftestsimulation Figure6.IllustrationoftheDCBtestsimulation.Aprecrackexistsbetweenthe0◦andthe90 ◦prepreglayers. Thecrackpropagatesnotonlythroughtheprecrackedinterfacebutit"jumps"tothealuminium0◦prepreg interface. TwodeformationplotsfortheDCBtestsimulationsareshowninFigure6.Thesecorrespondto two different loading stages: (a) delamination initiation and (b) the end of crack propagation. A displacementof35mmisappliedtothetopadherenttipwhilemaintainingtheloweradherenttip constrainedintheXandYdirections.Themodelincludesaprecrackof45mminlengthbetween the0◦andthe90 ◦prepreglayers.Thiscrackstartstopropagatewhenthetipdisplacementisabout 5mmlong(Figure6(a)).Atacertainloadlevel,theinterfacebetweenthetopaluminiumlayerand the0◦prepreglayerstartstodebondaswell(Figure6(b)).Althoughfibrebridgingwasobserved duringtheexperiments,thiscrack"jump"wasnotnoticed.Thecomputedlongitudinalstressvalues inthe0 ◦prepreglayerareinexcessof1200MPa.Itispossiblethatthisplycouldfailatlowerstress values,howeverthenumericalmodelsdidnottakeintoaccountintraplyfailurephenomenonsuch as fibre failure. The experimental and numerical results for tests with precracks at different interfaces show similar behaviour to the simulations described above. The crack initiation starts approximatelyataloadof300N.Oncethedelaminationstartstopropagate,theloadsustainedby the specimen starts to decrease gradually. Both experiments and simulations were stopped at a prescribedtipdisplacementof35mm,correspondingtoacrackpropagationofabout100mm. TheloaddisplacementbehaviourfortheDCBtestisshowninFigure7(a).Thecrack"jumping" phenomenon and the delamination at two interfaces justify the unstable crack propagation. The maximumcomputedloadisaround690N,25%higherthanthe550Nachievedintesting.Abetter agreementwiththeexperimentsisobtainedifthecrackisnotallowedto"jump",i.e.onlythepre cracked interface delaminates. This can be achieved by slightly reducing the mode I ultimate tractionvalueofsuchinterface.Theresultisnotonlyabettermatchwiththeexperimentalload displacement curves but also a more accurateprediction of the maximum load: 614N; only 11% higherthanthevalueobtainedexperimentally.Itshouldberealisedthatperfectinterfacesarebeing simulated while in laboratory these are virtually impossibletoproduce,i.e.thereisalwayssome degreeimperfectionthatdegradestheadhesiveproperties.Despiteporositybeinggloballyreflected intheexperimentalmeasurementofthefracturetoughness,itslocalinfluenceonthedelamination initiationisunknown.Duetotheoccurrenceoffibrebridging,ahigherGIc thantherealvaluemay havebeenreportedaswell,asexplainedbefore.Insuchcase,moreenergyisnecessarytopropagate thecrackbythesamelength. 42 Progress in Fracture and Damage Mechanics

800 4000 1200 Numerical, del. 1 interface Numerical Results Numerical Results Fracture Mechanics Curve Fracture Mechanics Curve Fracture Mechanics Curve Beam Theory Curve Beam Theory Curve 1000 Experimental Results Beam Theory Curve 600 Numerical, del. 2 interfaces 3000 Experimental Results Experimental Results 800

400 2000 600 Load [N] Load [N] Load [N] 400 200 1000 200

0 0 0 0 5 10 15 20 25 30 35 0 0.5 1 1.5 2 2.5 3 3.5 0 2 4 6 8 Displacement [mm] Displacement [mm] Load Point Displacement [mm] (a)DCB(b)ENF (c)MMB Figure7.LoaddisplacementcurvefortheDCB,ENFandMMBtestssimulations. Delaminationonsetispredictedbytheanalyticalmodelsatthepointofintersectionbetweenthe BeamTheorycurveandtheFractureMechanicscurve[13].BeamTheorypredictsstifferspecimen behaviour as compared to experimental results and numerical calculations. For the sake of simplification,itisassumedthattheprecrackedpartoftheDCBdeflectsasabeambuiltinatone ofitstips(correspondingtothecracktip).Thisconstitutesanoverconstraintbecause,inrealitythe precrackedpartoftheDCBiscontinuumwiththebondedpart,i.e.ithasafinitestiffnessatthe cracktip. The Fracture Mechanics Theory curve exactly matches the delamination propagation curvepredictedbytheFEsimulation.However,becauseoftheoverstifflinearpart,delamination onsetispredictedatahigherloadthanexperimented. Figure7(b)depicts,asanexample,theexperimental,analyticalandnumericalloaddisplacement plots for mode II delaminationpropagationbetween the aluminium and the 90 ◦prepreglayersin Glare32/10.4.Theunstableandcatastrophicnatureofthecrackpropagationiscorrectlycaptured by the analytical 1 and numerical simulations, at very similar load and displacement values. The numericalmodeldoesnotusethemixedmodedelaminationenergyinadirectformbutratherby meansofafunctionwhichinterpolatesthefracturetoughnessvaluesformodeI( GI),modeII( GII ) andmixedmodeloading(GII /G T): η GII  GGGIC+() IIC − IC  , GGG T = I + II (2) GT  Inthisfunction,proposedbyBenzeggaghandKenane[12],ηshouldbetunedinsuchawaythat thebestinterpolationoftheexperimentalvaluesisachieved.Inthepresentcase, η=0.1producesthe best results. The loaddisplacement behaviour for the MMB test, for the propagation of delamination between the aluminium and the 0◦ prepreg layers in Glare 32/10.4, is shown in Figure7(c).Themaximumloadpointispredictedwithremarkableaccuracy,especiallybytheFE analysis.

ILSSBenchmarkTestSimulation Theinterfaceproperties,previouslydeterminedbymeansoftheDCB,ENFandMMBtestswere correlatedwithanalyticalandnumericalmodelsandmaybeusedinindependentsimulationsofthe ILSStests.Theresultsmaybedirectlycomparedtovalidatethemodel. 1NoticethesnapbackbehaviouroftheFractureMechanicscurve Key Engineering Materials Vol. 383 43

Figure8.Left:ILSSMechanicaltestsetup.Right:loadboundaryconditionsintheFEmodel. ThegeometryoftheILSSspecimenisshowninFigure8.Itconsistsofa50x10mmstripofGlare 34/30.4. The two supports, separated by a distance of 10mm, are modelled as constraints preventingdisplacementsintheverticaldirection.Asinglediscretenodaldisplacementreplacesthe die that loads the specimen at its centre line. In order to remove rigid body movement, the horizontaldisplacementofthisnodeisprevented.Mostmaterialpropertiesarethesameastheones used in thepreviously described models, except for the aluminium which is now simulatedby a bilinearelastoplasticbehaviourwithstrengthof325MPaandahardeningcurveslopeof12%. ThecompletebehaviourismodelledinaquasistaticmannerintheFEcodeABAQUS®[6].

(a)Undeformedshape

(b)Deformedshape Figure9.FiniteelementmodelofILSStestspecimen. The undeformed FE mesh of the ILSS specimen is shown in Figure 9(a). It consists of 100 elementsinthelengthdirection.Inthethicknessdirection,oneelementperaluminiumandprepreg layerisused.Experienceshowsthatahighernumberofelementsintheverticaldirectiondonot provide a more accurate global solution. Fournodded elements equipped with an incompatible modes constitutivebehaviourareusedinordertopreventshearlocking. The deformed mesh of a perfectly bonded ILSS specimen, for a loadpoint displacement of 0.3mm, is depicted in Figure 9 (b). A closeup of one of the delaminated regions isprovided in Figure10.Cracksstartattheshearloadedinterfacialregionsbetweenthesupportpointsandthe loading point and they propagate progressively to the specimen tips. The onset of delamination occursattheinnerprepregprepreginterface,notonlybecausethesheartractionsarehigheratthe specimen midplane but alsobecause the mode II fracture toughness (GIIc ) of this interface is the lowestofthethreedifferenttypesofinterfaces. 44 Progress in Fracture and Damage Mechanics

Figure10.CloseupoftheILSStestspecimendeformedmesh.Delaminationsarevisibleattheinnerprepreg prepreginterface. Figure11showstheshearstressesatinterfaces1to8(countingfromthelowestsurfaceofthe model), for a prescribed displacement slightly higher than needed to the onset of delamination. Shearstressesreachmaximumvaluesatthetworegionsbetweentheloadingandsupportpoints. Outsidetheseregionstheyaremaintainedatnegligiblevalues.Shearstressvaluesincreasefromthe outerinterfacestotheinnerones,e.g.frominterfaces1to3.Theplotscorrespondingtotheinner interfaces,speciallyinterface5,showinflectionpoints.Intheseinterfacestheshearstressesstartto decrease after reaching the maximum shear traction values ( τIIu =90MPa). The regions where this phenomenonisobservedaredelaminating.Furtherloadincreaseleadstodelaminationpropagation awayfromtheloadpoint.

100 100 100 100 80 80 80 80 60 60 60 60 40 40 40 40 20 20 20 20 0 0 0 0 -20 0 10 20 30 40 50 -20 0 10 20 30 40 50 -20 0 10 20 30 40 50 -20 0 10 20 30 40 50 -40 -40 -40 -40 -60 -60 -60 -60 ShearStress [MPa] ShearStress [MPa] ShearStress [MPa] ShearStress [MPa] -80 -80 -80 -80 -100 -100 -100 -100 X [mm] X [mm] X [mm] X [mm] (a)Interface1(b)Interface2(c)Interface3(d)Interface4

100 100 100 100 80 80 80 80 60 60 60 60 40 40 40 40 20 20 20 20 0 0 0 0 -20 0 10 20 30 40 50 -20 0 10 20 30 40 50 -20 0 10 20 30 40 50 -20 0 10 20 30 40 50 -40 -40 -40 -40 -60 -60 -60 -60 ShearStress [MPa] ShearStress [MPa] ShearStress [MPa] ShearStress [MPa] -80 -80 -80 -80 -100 -100 -100 -100 X [mm] X [mm] X [mm] X [mm] (a)Interface5(b)Interface6(c)Interface7(d)Interface8 Figure11.Shearstressvaluesatspecimeninterfaces(numberedfromlowesttoupperlayer).Prescribed displacement=0.24mm,Load=2000N

NumericalAnalysesontheInfluenceofPorosity Inthissection,themodeloftheILSSspecimendescribedintheprevioussectionisusedtoanalyse the effect of porosity in Glare 34/30.4 interfaces. Porosity can be modelled by reducing the ultimatestrengthand/orthefracturetoughnessinafewelementsatspecificlocations.Nevertheless, itisnotobvioustowhatextenttheseparametersshouldbereduced.Inordertobeonthesafeside, theworstcasescenarioisassumed.Inthissense,theshearloadcarryingcapabilityoftheadhesive is locally reduced to zero as if entirely delaminated. In those interface elements, the original cohesive constitutive relation is replaced by a tractionfree relation. However, the normal compressiveresponseismaintainedinordertosimulatecrackclosureeffects. Key Engineering Materials Vol. 383 45

Threedistinctcaseshavebeenconsidered.Inthefirsttwocases,theporosityissmearedevenly inasingleandinallmodelinterfaces.Inthethirdcase,porosityisrandomlydistributedalongthe interfaces. The final case approximates the real situation where the size of the porous and the distancebetweenthemisvariable. PorosityinaSingleInterface Intheseanalyses,eachinterfacecontainsalevelofporosityof 25%.Ninesimulationshavebeencarriedout,eachtostudytheinterlaminarshearstrengthreduction caused by porosity in each of the nine interfaces in the Glare 34/30.4 specimen. A smeared distribution of porosity is considered, i.e. one of each four equallysized, consecutive interface elements is a priori set to a delaminated state. This means that 25 out of the total 100interface elementsarepredelaminated. Theloaddisplacementcurvescorrespondingtoeachoftheninesingleinterfaceporositycases areplottedinFigure12(a).Afteraninitiallinearpath,anonlinearbehaviourisobservedinallof thecases.Thisisduetothedelaminationpropagationintheporousinterface.Byincreasingtheload levelevenfurther,delaminationseventuallystartatotherinterfacesandthealuminiumlayersstart todeformplastically,reducingtheoverallstiffnessofthespecimen.Afteracertaindamagelevel, thespecimenisunabletosustainahigherload.Thismeansthatthespecimenmaximumstrength valueisreached.Fromthispointonthedelaminationsintheporousinterfacerapidlypropagateto neighbouring elements away from the centre of the specimen, drastically reducing its strength. However,thespecimensstillretainpartoftheirstrengthbecausethissuddendelaminationdoesnot propagatethroughtheentireinterfacespan.

2500 2500

2000 2000

1500 1500 Porosity in: interface 1 interface 2 Load[N] Load[N] 1000 1000 interface 3 interface 4 interface 5 No porosity 500 interface 6 500 50% porosity interface 7 25% porosity interface 8 25% porosity in interface 5 interface 9 0 0 0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25 Displacement [mm] Displacement [mm] (a)Smearedporosityindiscreteinterfaces(level:25%)(b)Severallevelsofsmearedporosityinallinterfaces Figure12.LoaddisplacementplotsforseveralILSSsimulationsincludingporosity.Interfaceswithseveral degreesofporosityarecharacterised. By comparing the different loaddisplacement curves and maximum strength points, it can be concluded that a 25% porosity level in the outer ILSS specimen interfaces does not lead to a significantinterlaminarshearstrengthreductionascomparedwiththenominalcase,whereasifthe porous interface is one of the inner ones, a remarkable decrease in strength is observed. This is because this level of porosity in the outer interfaces does not trigger delamination propagation. Instead,asintheperfectlybondedcasediscussedintheprevioussection,delaminationistriggered at the inner prepregprepreg interface where the combination of the highest shear strain and the lowest fracture toughness occurs. When one of the three inner interfaces isporous, delamination occurs at a lower load level that could be achieved if the specimen was perfectly bonded. A maximumstrengthreductionof15.2%isobservedwhenporosityislocatedattheinnerprepreg prepreginterface,againduetothecombinationofthehighestshearstrainandthelowestfracture toughness. Global Porosity In these simulations, global smearedporosity upto a level 50% of thetotal interfacialareaisconsidered.Thiscorrespondstospecimenmodelswheretheinterfacialstiffnessof 46 Progress in Fracture and Damage Mechanics upto50%ofallthecohesiveelementsisignored.These0.5mmlongpredelaminationsaresmeared alongalltheinterfaces.The50%porositylevelcaseisconsideredtheworstcasescenariooccurring inFML’s. Theloaddisplacementcurvescorrespondingtoperfectlybonded,25%porousand50%porous specimens are shown in Figure 12(b). The worstcase scenario leads to an interlaminar shear strengthreductionof30.8%.The50%porositycaseshowsaverygradualdecreaseinstrengthafter themaximumloadpointwhilethecasescorrespondingtolowerdegreesofporosityaremarkedby suddenfailure.Thisiscorrelatedwiththewaydelaminationspropagate.Asanexample,considering the25%porositylevelcase,theinnerprepregprepreginterfacestartstodelaminateataloadlevel around 1500N. The delamination propagates at this interface alone until the maximum shear strengthlevelisreached.Then,anotherinterfacesuddenlystartstodelaminateandequilibriumis achievedatalowerloadlevel.Forthe50%porositylevelcase,theonsetofdelaminationoccursat threeinterfacesbarelyatthesameloadlevel(1500N)followedbysmoothpropagation. It is interesting to noticethat,forthe25%porositylevel,thecasesofglobalporosityandthe porositysolelyattheinnerprepregprepreginterfaceproducesimilarresults;around15%reduction in shear strength. This means that, for this porosity level the delamination of this interface dominatestheshearfailureprocess. Figure13isillustrativeoftheeffectofporosityinallinterfaces.TheILSSofFML’sseemstobe linearlydependentontheoveralllevelofporosity,forporositylevelsrangingfrom0%to50%.The maximumloadcomputedfortheperfectlybondedILSSis2354N.Ontheotherside,theworstcase porosityscenarioresultsinafailureloadof1630N.

2600

2200

1800 Maximum Load [N] Maximum

1400 0 10 20 30 40 50 Porosity Level [%] Figure13.EffectoftheinterfaceporositylevelinthemaximumloadsupportedbyanILSSspecimen. RandomPorosityInreality,theoccurrenceofporosityinaperfectlysmearedfashioninoneor allinterfacesisratherunlikely.Mostprobably,therewillbesmallandlargevoidsandthedistance betweenthemwillvaryaswell.Theeffectsofthisrandomnessontheinterlaminarshearstrengthof FML’smaybequitesignificantandmaydeviatefromtheresultsobtainedforthesmearedporosity cases.Therefore,foracorrectassessmentoftheeffectsofreallifeporosityontheILSSofGlare, randomporositycasesareconsidered. Atotalof27casesofglobalandsingleinterfaceporosity(25%and50%porositylevels)were generated by randomly choosing 25 or 50 of the interface elements, corresponding to pre delaminations.Inthesingleinterfacescenario,porosityisincludedonlyintheinnerprepregprepreg interface(interface5)sincethisrepresentstheworstcasescenario.Voidslargerthantheinterface elementssize(0.5mm)areobtainedwhentwoormoreconsecutiveelementsoccurintherandomly generatedlistofelementnumbers.Similarly,ifthedifferencebetweentwoconsecutivemembersof thatlistishigherthanaverage,itmeansthattherewillbealargeperfectlybondedinterfacearea.On average,the25%porositylevelcasegenerated18.9voidsateachinterfacewithanaveragelength of0.668mm,whilethe50%porositylevelcasegenerated,onaverage25.7voidswithanaverage lengthof0.98mm.Theprobabilityoftheoccurrenceoftwoconsecutivepredelaminatedinterface elementsisobviouslymuchhigherinthe50%porositysetofcasesthaninthe25%set. Key Engineering Materials Vol. 383 47

The loaddisplacement curves corresponding to random porosity in the inner prepregprepreg interfaceareplottedinFigure14(a)and14(b),respectivelyforthe25%and50%porositylevels. Similarly,theresultsforglobalporosityaredepictedinFigure14(c)and14(d),respectivelyforthe 25%and50%porositylevels.Thefirstimportantconclusiontodrawfromtheseplotsisthatthereis aremarkablescatterintheILSSproducedbyrandomporosity.Thismaybeexplainedbythescatter inthenumberofpredelaminatedelementsintheregionofthespecimenbetweenthetwosupports. Asmentionedpreviously,thisisacriticalshearloadedarea.Theboundaryandloadingconditions heredominatethefailureprocessofthewholespecimen.Asanexample,indepthobservationof the single interface (50% level porosity scenario) reveals that the specimen corresponding to the poorer results (1491.3N) contains 13 predelaminated elements in this region, for an average probabilityofhavingonly10oftheseelements.Partoftheseelementsisgroupedintwovoidsof 2mmandoneof1.5mm.ThespecimenshowingthehighestILSS(1872.4N)containsonly6pre delaminated elements in this region and most of them are dispersed, except for a single 1mm porous.

2000

2000 1600

1600

1200

1200 Load[N] Load[N] 800 800

400 400

0 0 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 Displacement [mm] Displacement [mm] (a)Porosityininterface5(level:25%)(b)Porosityininterface5(level:50%)

2100 1500

1800 1200 1500

1200 900

900 Load[N] Load[N] 600

600

300 300

0 0 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 Displacement [mm] Displacement [mm] (c)Globalporosity(level:25%)(d)Globalporosity(level:50%) Figure14.Loaddisplacementcurvesfor27randomporositycases. The relations between the probability of failure of a specimen and the failure load values are plottedinFigure15,forthefourscenariosofrandomporositystudied.Itisassumedthattheresults follow statistical normal distributions and there is a significant number of occurrences in the universe of possible results. It can be observed that, for porosity at the inner prepregprepreg interface,95%ofthespecimensfailatloadsintheranges1631.62194.6Nand1473.21790.8Nfor the25%and50%porositylevelscenarios,respectively.Similarly,fortheglobalporosityscenario, 95%ofthespecimenshavemaximuminterlaminarshearstrengthintheranges1532.11992.8Nand 1052.41429.0N, respectively for the 25% and 50% porosity levels. Curiously, the scatter 48 Progress in Fracture and Damage Mechanics correspondingtothe50%porositylevelcasesishigherthanforthe25%porositylevelcases,either forthesingleinterfaceorglobalporosityscenarios.

0.003 0.006

0.004 0.002 P(failure) P(failure)

0.002 0.001

0 0 1393.8 1473.2 1552.6 1632.0 1711.4 1790.8 1870.2 1490.9 1631.6 1772.4 1913.1 2053.8 2194.6 2335.3 Failure Load [N] Failure Load [N] (a)Porosityininterface5(level:25%).Av.value= (b)Porosityininterface5(level:50%).Av.value= 1913.1N.St.deviation=140.7N.95%ofthe 1632.0N.St.deviation=79.4N.95%ofthespecimens specimensfailataloadintherange1631.62194.6N failataloadintherange1473.21790.8N

0.005 0.004

0.004 0.003

0.003

0.002 P(failure)

0.002 P(failure)

0.001 0.001

0 0 970.8 1062.4 1154.1 1245.7 1337.4 1429.0 1520.6 1416.9 1532.1 1647.3 1762.5 1877.6 1992.8 2108.0 Failure Load [N] Failure Load [N] (c)Globalporosity(level:25%).Av.value= (d)Globalporosity(level:50%).Av.value=1245.7N. 1762.5N.St.deviation=115.2N.95%ofthe St.deviation=91.6N.95%ofthespecimensfailata specimensfailataloadintherange1532.11992.8N loadintherange1052.41429.0N Figure15NormaldistributionsoftheILSSofGlare34/30.4correspondingtoseveralscenariosofporosity. TheILSSvaluesfortheconfigurationsanalysed,aswellastheamountofstrengthreductiondue toporosity,arereportedinTable3.Fortherandomporosityscenarios,averageresultsareshown. The worstcase porosity configuration, corresponding to a level of 50% of random porosity, produces46.5%reductioninthespecimenshearstrength.Withoutexception,therandomporosity scenariosleadtolowerILSSresultsthanthesmearedporosityscenarios.Actually,fortheglobal porosityscenariostheresultscorrespondingtothesmearedporositycasesdonotevenfallinthe intervalsof95%probabilityoffailureforthecorrespondingrandomporositycases.Thismeansthat specimenswithsmearedporosityareaveryparticularcaseofallporositycasesandleadtofailure resultsthathighlysurpasstheaverageexpectedvalue. Unlikethesmearedporosityscenario,randomporositydoesnotproducesimilarILSSvaluesfor thecasesofporosityintheinnerprepregprepreginterfaceandglobalporosity.Thismaybedueto thesmallnumberofspecimensanalysed. Key Engineering Materials Vol. 383 49

Specimenconfiguration ILSS[N] Strengthreduction[%] Perfectlybonded 2354 25%smearedporosityininterface5 1995 15.2 25%randomporosityininterface5 1913 18.7 25%globalsmearedporosity 2013 14.5 25%globalrandomporosity 1762(average) 25.1(average) 50%randomporosityininterface5 1632 30.7 50%globalsmearedporosity 1630 30.8 50%globalrandomporosity 1259(average) 46.5(average) Table3.AverageILSSofGlare34/30.4andshearstrengthreductionduetoseveralporosityscenarios.

ConclusionsandRecommendations TheexistenceofporositytriggersdelaminationsattheinterfacesbetweendifferentlayersinFML’s causingthemtofailatlowerappliedloadsthanotherwiseachievable.Thepresentreportdescribesa studyontheeffectsofsuchporosityontheinterlaminarshearresponseofGlare.Thisworkwas performedbymeansofanalyticalandnumericalsimulationsoflaboratorytests.Theobjectivewas to replicate the ILSS experiments previously performed and quantify the decrease in Glare shear strengthduetoseveraldegreesofporosity.TraditionalFEmethodswereusedincombinationwitha cohesive zone approach, developed by Camanho et al [910] to simulate delamination onset and propagationatmaterialinterfaces.However,itsimplementationinnumericalmodelsrequiresthe input of nontrivial mechanical material properties such as the interface fracture toughness. Experiments were carried out to find these properties [11]. Then, FE simulations and simple analyticalmodelsbasedonBeamTheoryandFractureMechanicsTheory[13]werecomparedto these experiments, for the sake of model validation and finetuning of the interfacial ultimate traction values. Remarkable agreement was found between the mode I loading delamination propagationtestresultsandrespectivesimulations.ThecatastrophicfailureoftheENFandMMB testsdoesnotallowforstablecrackpropagation.Thisunstablenatureiscorrectlycapturedbythe analyticalandnumericalsimulationsatremarkablysimilarloaddisplacementpoints. PerfectinterfacialbondingandseveralporosityscenariosweresimulatedwithmodelsofGlare3 4/30.4ILSStestspecimens.Theinnerprepregprepreginterfaceisthemostpronetodelaminate (hencethemostsensitivetoporosity),sinceitcombinesthehighestshearstrainswiththelowest fracture toughness. A remarkable agreement in the loaddisplacement behaviour was achieved betweenthesesimulationsandexperimentscarriedoutpreviously,butnotinthepredictionofthe maximumstrengthvalues.Thetestsrevealafailureloadof1650Nforthespecimenswiththebest interfacialquality,whilethenumericalmodelspredicta42%highervalue(2354N).Thiscouldbe explained by the fact that perfectly bonded interfaces, as simulated, are virtually impossible to manufacture,i.e.thereisalwayssomeporositythatdegradestheadhesiveproperties. Randomly generated porosity cases were simulated for a better agreement with reality. A significant drop in the ILSS is observed when comparing these cases with the smeared porosity casesbecauseoftheprobableexistenceoflargervoidslocatedinthemostcriticalspecimenregion forshearfailure,i.e.betweenthesupports.Theworstcaseporosityscenariosimulatedresultedina 46.5%reductionintheILSSofGlare. Severalrecommendationsshouldbegivenforfuturework.Firstly,intraplyfailurecriteriashould beincorporatedinthemodels.Thiswouldresultinabettermatchwithrealitysincedelaminations would be allowed to propagate at any of the Glare interfaces and the phenomenon of "crack jumping" could be analysed in more detail. Secondly, DCB tests should be repeated, this time avoiding fibre brigding. Also, the FourPoint Bending tests [15] should be carried out in replacement of the threepoint bending ENF and MMB test. These would result in stable crack 50 Progress in Fracture and Damage Mechanics propagationsinmodeIIandmixedmodeloading,thereforeallowingthemeasurementofaccurate fracturetoughnessvalues,consequentlybettercorrelationwithFEmodels.Thirdly,thedevelopment of three dimensional models for the investigation of porosity in test specimens and in critical structural components would better answer the needs of the industry. Three dimensional models wouldalsoallowthemodellingofvoidswithbettergeometricalresemblancewithreality. Finally, this investigation would benefit from more experimental data to which the numerical predictionscouldbecompared.Anextensivestudyontotheactualexactspatialdistributionofvoids intheporouslayers,similartotheworkconductedin[15]wouldimprovethedataavailableforthe simulations. Having more experimental data available, it can be interesting to compare this approach with a more traditional stress based approach to predict the initiation of damage [2,3]. ThismayprovetobeaneffectivewaytopredictfailureinitiationinFML’sandmaybemorerobust thannumericalmethodimplementinthepresentwork.

Acknowledgements The funding of this work through the scholarship SFRH/BD/16238/2004 from the Portuguese FoundationforScienceandTechnologyisgratefullyacknowledged. Special acknowledgements are addressed at Dr. Mário Vesco for carrying out fracture energy tests [11] and making the results available to this study. Also the valuable ideas of Dr. Doobo Chung in the introduction of probability in the analysis of the effects of porosity in FML’s are kindlyacknowledged.

References [1]A.M.RubinandK.L.Jerina(1994), Evaluation of Porosity in Composite Aircraft Structures , MechanicsofCompositeMaterials,Vol.30,No.6,pp.587600. [2] H. Jeong (1997), Effects of Voids on the Mechanical Strength and Ultrasonic Attenuation of LaminatedComposites,JournalofCompositeMaterials,Vol.31,pp.276292. [3] M.L. Costa, S.F.M. de Almeida and M.C. Resende (2001), The influence of Porosity on the Interlaminar Shear Strength of Carbon/Epoxy and Carbon/Bismaleimide Fabric Laminates, CompositeScienceandTechnology,Vol.61,pp.21012108. [4]Z.Gürdal,A.P.TomasinoandS.B.Biggers(1991), Effects of Processing Induced Defects on the Laminate Response: Interlaminar Tensile Strength ,SAMPEJournal,Vol.27,No.4,pp.3949 [5] A. Mobuchon and Z. Gürdal (1991), Effects of Processing-Induced Defects on Failure of Graphite/Epoxy Angles , Mechanics and Mechanisms of Damage in Composites and Multi Materials,ESIS11,MechanicalEngineeringPublications,London,pp.95115. [6]ABAQUS,Inc(2007), ABAQUS Version 6.7 User's Manual ,Pawtucket,RI,USA. [7] J.C.J. Schellekens and R. de Borst (1994), Free Edge Delamination in Carbon-Epoxy Laminates: a novel numerical/experimental approach ,CompositesandStructures,Vol.28,pp.357 373. [8] P.P. Camanho, C.G. Dávila and M.F. Moura (2003), Numerical Simulation of Mixed-mode Progressive Delamination in Composite Materials ,JournalofCompositeMaterials,Vol.37,No. 16,pp.14151438. [9]A.Turon,P.P.Camanho,J.CostaandC.G.Dávila(2004),An Interface Damage Model for the Simulation of Delamination Under Variable-Mode Ratio in Composite Materials. NASA/TM2004 213277. Key Engineering Materials Vol. 383 51

[10]A.Turon,P.P.Camanho,J.CostaandC.G.Davila(2006), A Damage Model for the Simulation of Delamination in Advanced Composites under Variable-Mode Loading .MechanicsofMaterials, Vol.38,pp.1072–1089. [11] M. Vesco and J. Sinke (2005), Response and damage evolution in GLARE laminates under indentation loads – Experimental results ,ReportB2V0501,FacultyofAerospaceEngineering TUDelft. [12] M.L. Benzeggagh and M. Kenane (1996), Measurement of Mixed-Mode Delamination Fracture Toughness of Unidirectional Glass/Epoxy Composites with Mixed-Mode Bending Apparatus ,CompositeScienceandTechnology,Vol.56,pp.43949. [13]Y.MiandM.A.Crisfielfd(1996),AnalyticalDerivationofLoad/Displacementrelationshipfor the DCB and MMB and Proof of the FEA Formulation, Department of Aeronautics, Imperial College,London. [14]A.Korjakins(1997), Interlaminar Fracture Toughness of Glass Fibre Reinforced Polymeric Composites Influenced by Fibre Surface Treatment .Ph.DThesis.RigaTechnicalUniversity. [15] C. Schuecker and B.D. Davidson (2000), Evaluation of the accuracy of the four-point bend end-notched flexure test for mode II delamination toughness determination , Composites Science andTechnology ,Vol.60,pp.21372146. [16]Ad.Vlot(2001),FibreMetalLaminates(AnIntroduction),KluywerAcademicPublishers. [17]A.Turon,C.G.Davila,P.P.CamanhoandJ.Costa(2005), An Engineering Solution for using Coarse Meshes in the simulation of Delamination with Cohesive Zone Models ,NASATechnical Memorandum,NASA/TM2005213547.