• Aucun résultat trouvé

Effective behavior of long and short fiber-reinforced viscoelastic composites

N/A
N/A
Protected

Academic year: 2021

Partager "Effective behavior of long and short fiber-reinforced viscoelastic composites"

Copied!
13
0
0

Texte intégral

(1)

ContentslistsavailableatScienceDirect

Applications in Engineering Science

journalhomepage:www.elsevier.com/locate/apples

Effective behavior of long and short fiber-reinforced viscoelastic composites

O.L. Cruz-González

a

, A. Ramírez-Torres

b

, R. Rodríguez-Ramos

c

, J.A. Otero

d

, R. Penta

b,

, F. Lebon

a

aAix Marseille Univ, CNRS, Centrale Marseille, LMA UMR 7031, Marseille, France

bSchool of Mathematics and Statistics, Mathematics and Statistics Building, University of Glasgow, University Place, Glasgow G12 8QQ, UK

cFacultad de Matemática y Computación, Universidad de La Habana, San Lázaro y L, Vedado, La Habana CP 10400, Cuba

dEscuela de Ingeniería y Ciencias, Tecnológico de Monterrey, Campus Estado de México, Atizapán de Zaragoza EM CP 52926, Mexico

a r t i c le i n f o

Keywords:

Homogenization Viscoelasticity

Fiber reinforced composites Power-law model Transverse isotropy Finite elements

a b s t r a ct

Westudythehomogenizedpropertiesoflinearviscoelasticcompositematerialsinthreedimensions.Thecompos- itesareassumedtobeconstitutedbyanon-aging,isotropicviscoelasticmatrixreinforcedbysquareorhexagonal arrangementsofelastictransverselyisotropiclongandshortfibers,thelatterbeingcylindricalinclusions.The effectivepropertiesofthesekindofmaterialsareobtainedbymeansofasemi-analyticalapproachcombining theAsymptoticHomogenizationMethod(AHM)withnumericalcomputationsperformedbyFiniteElements(FE) simulations.Weconsidertheelastic-viscoelasticcorrespondenceprincipleandwederivetheassociatedlocaland homogenizedproblems,andtheeffectivecoefficientsintheLaplace–Carsondomain.Theeffectivecoefficients arecomputedfromthemicroscalelocalproblems,whichareequippedwithappropriateinterfaceloadsarising fromthediscontinuitiesofthematerialpropertiesbetweentheconstituents,fordifferentfibers’orientationsin thetimedomainbyinvertingtheLaplace–Carsontransform.WecompareourresultswiththosegivenbytheLo- callyExactHomogenizationTheory(LEHT),andwithexperimentalmeasurementsforlongfibers.Indoingthis, wetakeintoconsiderationBurger’sandpower-lawviscoelasticmodels.Additionally,wepresentourfindingsfor shortfiberreinforcedcompositeswhichdemonstratesthepotentialofourfullythreedimensionalapproach.

1. Introduction

Materials characterized by both a viscoelastic response and a composite-likegeometrical arrangement arefoundin several biolog- icalcontextsdrivenbynaturalevolution,see,e.g., (Atthapreyangkul etal., 2021; Ojanen etal., 2017; Sherman etal., 2017). Especially, longandshortfiber-reinforcedcomposites arebeingincreasinglyex- ploitedin avariety of engineering andmanufacturingprocesses be- causeoftheircapabilityofoptimisingpropertiessuchaslightweight, stiffness,andstrength.Ontheonehand,highperformancecomposites aretypicallymadeoflongcontinuous fibresembeddedinapolymer matrixandexhibitviscoelasticproperties(see,e.g.,OrnaghiJr.etal., 2020;Wangetal.,2020).Ontheotherhand,reinforcementviashort fibers(i.e.,fiber-shapedinclusions)canprovideavaluablealternative inthemodellingoffailuresappearingincompositesreinforcedbylong fibers,see,e.g.,Cepero-Mejíasetal.,2020;NonatoDaSilvaetal.,2020. Materialcompositesreinforcedbyshortfiberscanalsoprovidesignifi- canteconomicalandmanufacturingadvantagesovercontinuousfiber- reinforcedcompositeswithoutcompromisinghighperformance,aslong

Correspondingauthor.

E-mailaddress:Raimondo.Penta@glasgow.ac.uk(R.Penta).

astheaspectratioishighenoughtosupportloadtransfer(Huangand Huang,2020;WangandSmith,2019;Yuetal.,2014).

The multiscale modellingof viscoelastic composites hasbeen in- creasinglyaddressedinrecentcontributions.Inthisrespect,microme- chanicalmodelsareparticularlysuitablewhenevertheaimistodeter- minetheeffectiveresponseofmaterialsonthebasisofindividualcon- stituents’properties,suchasviscoelasticmoduliandfiberspropertiesin termsofgeometricalarrangement,volumefractionandorientation.For instance,inSevostianovetal.(2016),theeffectiveviscoelasticproper- tiesofshortfiberreinforcedcompositesareinvestigatedbymeansof thefraction-exponentialoperatorsof ScottBlair-Rabotnov.Moreover, inKernetal.(2019)afrequency-domainfiniteelementsimulationsare consideredtodeterminetheeffectivemoduliofviscoelasticcoatedfiber- reinforcedcomposites.Theinvestigationofthepolymeralignmentwith theaidofdirectnumericalsimulationsoftheturbulentchannelflow ofaviscoelasticFENE-PfluidisconductedinPereiraetal.(2020).In addition,theeffectiveviscoelasticcreepbehaviorofalignedshortfiber compositesisobtainedinWangandSmith(2019)viaaRVE-basedFi- niteElementalgorithm.InOrnaghiJr.etal.(2020),theAuthorseval-

https://doi.org/10.1016/j.apples.2021.100037 Received30December2020;Accepted5February2021 Availableonline17February2021

2666-4968/© 2021TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/)

(2)

uatethecreep,recovery,andviscoelasticpropertiesofunidirectional carbon/epoxyfilament woundcomposite laminatesunder controlled stress,time,andtemperature.Recently,aprobabilisticmicromechan- icsdamageframeworktopredictthemacroscopicstress-strainresponse andprogressivedamageinunidirectionalglass-reinforcedthermoplastic polymercompositeshasbeenproposedinChenetal.(2021).

Amongthemostusedtechniquesaddressingthecalculationofthe effectivepropertiesofviscoelasticheterogeneousstructureswefindthe AsymptoticHomogenization Method(AHM).Forexample,analytical closedformexpressionsfortheeffectivecoefficientsoffibrousviscoelas- ticcomposites areobtainedin Rodríguez-Ramosetal.(2020), Otero etal. (2020)bymeansof thetwo-scaleAHM.The theoreticalbases ofthemethodarefoundinthecontributionsofseveralauthors(see, e.g., Allaire andBriane,1996; Auriault etal., 2009; Bakhvalov and Panasenko,1989;Bensoussanetal.,1978;Sanchez-Palencia,1980).In general,thetechniquetakesadvantageofthescalesseparationassump- tionfordecouplingthespatialvariablesintoamicroscopicandamacro- scopicone.Thus,thesolutionoftheoriginalproblematthemacroscale isapproximatedbyconsideringthesolutionofthecorrespondinglocal problemsonaperiodiccellandofthehomogenizedproblem.Thisper- mitstodecreasethecomputationalcomplexityoftheproblemathand, andtoencodetheinformationatthemicroscaleintotheeffectivecoef- ficientsofthemacroscalemodel.

ThemaindisadvantageoftheAHMisthattheanalyticalsolutionof thelocalproblemscanbederivedforafewofsimplecompositestruc- tureswhichcanbereducedtooneortwodimensionsanalysis(asthose forlaminatedcompositesandlongfibers).Forinstance,instudiesre- latedwithviscoelastic composites forlaminatedandfiberreinforced composites(Cruz-Gonzálezetal.,2020a;Oteroetal.,2020).Forthis reason,inordertohandlemorecomplexmicrostructures,numericalap- proachesbasedontheFiniteElements(FE)providearobustalternative tosolvethelocalproblemsformoregeneralmicrostructures.Basedon theseconsiderations,inthepresentwork,weaimtostudytheeffec- tivepropertiesofviscoelasticcompositesbymeansofthecombination oftheAHM andtheFE method,thelatter allowsustofind thenu- mericalsolutionofthemicroscaleperiodiclocalproblemfordifferent three-dimensionalarrangements.Thisapproachprovidesanewandef- ficientcomputationalplatformforcomputingtheeffectivepropertiesof viscoelasticcompositesinthreedimensions.

Themainaimofthisworkistocalculatetheeffectivepropertiesof non-aging,linearviscoelasticcompositesinthreedimensions.Forour purposes,weemploythemodelingapproachintroducedinourprevious work(Cruz-Gonzálezet al.,2020b),wherein acombinedframework basedontheoreticalandcomputationaltechniquesforcomputingthe effectivepropertiesofviscoelasticcompositesisemployed.Specifically, inCruz-Gonzálezetal.(2020b),weanalyzedseveraltypesofcompos- itestructuresreinforcedbyfibers andinclusions.Here,however,we gofurtherinourinvestigationsandconsiderhexagonalperiodiccells andunidirectionallyalignedshortfibers.Furthermore,inthepresent framework,weconsiderthefibers tobe transverselyisotropic, while inCruz-Gonzálezetal.(2020b),thestudywasfocusedoncomposites withisotropicconstituents.Thisextensionrequiresasuitablegenerali- sationofthecomputationalsetupoftheproblemfound,forexample, inPentaandGerisch(2016)forelasticcomposites.Inparticular,the interfaceloadsrelatedtotheauxiliarylocalcellproblemsareobtained formoregeneralorthotropicmaterials,andthenspecialisedtotrans- verseisotropicconstituentsinourcalculations.Anotherextensionofthe presentworkwithrespecttoCruz-Gonzálezetal.(2020b)isthat,inthe presentstudy,wetakeintoconsiderationdifferentfiber’sorientations.

Wefurthernoticethat, tothebestof ourunderstanding,thema- jornovelty of this work withrespect to othersexistent in theliter- ature and focused on thestudy of viscoelastic composites (see, e.g.

Amiri-Radetal., 2019; Ornaghi Jr.etal., 2020; Oteroet al., 2020;

Rodríguez-Ramosetal.,2020;Sevostianovetal.,2016;TangandFe- licelli,2015;Tranetal.,2011;WangandPindera,2016a;Wangand Smith,2019;YanceyandPindera,1990;Yietal.,1998) isthathere,

Fig.1.(a)Macroscale:viscoelasticheterogeneousmaterialwith(b)square(i,ii) orhexagonal(iii,iv)arraysofnon-overlappinglongandshortfibers,respec- tively.(c)𝜀-structurallevel.Microscale:periodiccellforlongandshortfibers inclusionsthatdonotintersecttheboundaries.

usingasemi-analyticalapproach,wearecapabletocomputetheeffec- tivepropertiesofthree-dimensional,non-aging,viscoelasticcomposite materials.Particularly,itshouldbepointedoutthatalthoughwecon- sidered someof theresultsgivenin YanceyandPindera(1990)and WangandPindera(2016a,b)forcomparisonwithours,ourmethodol- ogyprovidesfurtherdevelopmentsbecauseofthepossibilityofconsid- eringmorecomplexcellgeometries.Indeed,wecalculatetheeffective propertiesforaviscoelasticcompositereinforcedwithperfectlyaligned shortfibersandthisgeometricalconfigurationisnottreatableunderthe two-dimensionalformulationreportedinWangandPindera(2016a,b). The manuscriptis organizedas follows.In Section2, we present thegeometricaldescriptionofthemodelandweformulatethelinear viscoelastic heterogeneousproblem. InSection3,we applythe two- scaleAHM toobtainthemacroscalefunctionalform oftheeffective viscoelasticcoefficients.InSection4,wecalculatetheeffectiveprop- ertiesbysolvingappropriatelocal(cell)problemsinthreedimensions.

Thetheoreticalframeworkisillustratedingeneralbyconsideringthat bothphasesareviscoelastic,althoughtheresultsarethenpresentedby consideringpurelyelasticfibersembeddedinaviscoelasticmatrixfor thesakeofcomparisonagainstpreviousanalyticalresultsandrelevant experiments.InSection5,wecompareourresultsagainstalternative homogenisation techniquesinthecaseof longfibers,andemphasise thepotentialofournewapproachbyillustratingtheresultsinthecase ofcompositesreinforcedbyshortfibers(i.e.,cylindricalinclusions).Fi- nally,inSection6,wesummarizeourfindingsandhighlightthelimi- tationsofthecurrentmodelandpossiblefurtherdevelopmentsofthe work.

2. Modeldescription

Weidentifytheheterogeneous,linearviscoelasticmaterialwithan open,boundedsetΩ3(seeFig.1(a)).Inparticular,weconsiderΩ asatwo-constituentcompositemadeofamatrixreinforcedbysquare (Fig.1(b)(i,ii))orhexagonal(Fig.1(b)(iii,iv))arraysofunidirectional andperiodicallydistributedlongandshortfibersinΩ(seeFig.1(b)).

Furthermore,weconsidertheexistenceoftwodistinct,well-separated lengthscales𝓁and𝐿whicharerelatedwiththecharacteristicsizeofthe periodicmicro-structureandthatofthewholecomposite,respectively (seeFig.1).Inthisframework,weintroducethedimensionlessscaling parameter𝜀asfollows,

𝜀= 𝓁

𝐿≪1, (1)

andthemicroscopicspatialvariable 𝑦=𝑥

𝜀, (2)

where𝑥issaidtobethemacroscopicspatialvariable.

(3)

Inparticular,wesetΩ =Ω𝜀1∪ Ω𝜀2 withΩ𝜀1∩ Ω𝜀2𝜀1∩ Ω𝜀2=∅,and whereΩ𝜀2 denotesthe matrix andΩ𝜀1=∪𝑁𝑖=1𝑖Ω𝜀1 represent the inclu- sionswith𝑁∈ℕ.Additionally,theinterfacebetweenΩ𝜀

1andΩ𝜀

2 (see Fig.1(b))isdenotedbyΓ𝜀.Referring toFig.1(c),theunitaryperi- odiccell𝑌 isconsideredtobeconstitutedbyafiber(longorshort)𝑌1

inthematrix𝑌2sothattheperiodiccellisgivenby𝑌 =𝑌1𝑌2with 𝑌1𝑌2=𝑌1𝑌2=∅.Atthisscale,theinterfacebetween𝑌1and𝑌2 is denotedbyΓ𝑌,seealsoDiStefanoetal.(2020)foranillustrationof variousperiodiccellarrangementinthecontextofelectro-activecom- posites.

2.1. Statementoftheproblem

Forthesakeofsimplicity,weneglectinertiaandexternalvolume forcesinthemodel,andimposecontinuityconditionsfordisplacements andtractionsontheinterfaceΓ𝜀,i.e.,thematrixandthesub-phasesare inidealcontact.Therefore,thebalanceoflinearmomentuminΩ⧵Γ𝜀 togetherwiththeinterfaceconditionsread

∇⋅𝝈𝜀(𝑥,𝑡)=𝟎 in(Ω⧵Γ𝜀)×ℝ, (3a)

𝒖𝜀(𝑥,𝑡)=𝟎 onΓ𝜀×ℝ, (3b)

𝝈𝜀(𝑥,𝑡)𝒏(𝑦)=𝟎 onΓ𝜀×ℝ, (3c)

where𝝈𝜀 represents thesecond-order stresstensorand𝑢𝑢𝑢𝜀 is thedis- placementfield.Moreover,𝑛𝑛𝑛(𝑦)denotestheoutwardunitvectortothe interfaceΓ𝜀,andtheoperator𝜙𝜀describesthejumpof𝜙𝜀acrossthe interfaceΓ𝜀betweenΩ𝜀1 andΩ𝜀2.Noticethatthesuperscript𝜀isused toindicatethenotation 𝜙𝜀(𝑥,𝑡)=𝜙(𝑥,𝑦,𝑡)(refer toDiStefano etal., 2020foracomprehensivediscussionregardingthisnotation).Thesys- temofEqs.(3a)–(3c)hastobesupplementedwithboundaryconditions on𝜕Ω ×ℝandinitialconditionsinΩ × {0}.However,theseconditions donotplayaroleinthederivationoftheeffectivecoefficients,andthey aretypicallytobespecifiedexplicitlyonlywhentheaimistoobtaina specificsolutionofthemacroscalesystemofhomogenizedPDEs,which isnotthecasehere.

Inthepresentframework,thecompositebehavesasanon-agingvis- coelasticmaterialsothat(Christensen,1982)

𝜎𝜎𝜎𝜀(𝑥,𝑡)=

𝑡 0

𝜀(𝑥,𝑡𝜏)∶𝜕𝜉𝜉𝜉(𝑢𝑢𝑢𝜀(𝑥,𝜏))

𝜕𝜏 𝑑𝜏, (4)

where𝜀isthefourth-ordertensorofrelaxationmoduli,whichhereis scale-dependent.Thediscontinuitiesofthepropertiesbetweenthehost mediumandthesubphasesareencodedinthetensor𝜀and,thus,its componentsareassumedtobesmoothfunctionsof𝑥in(Ω⧵Γ𝜀)×ℝ, butdiscontinuousonΓ𝜀×ℝ.Furthermore,wenoticethat,inEq.(4),𝜉𝜉𝜉 denotesthesecond-orderstraintensorforsmalldisplacements,namely 𝜉𝜉𝜉(𝑢𝑢𝑢𝜀(𝑥,𝑡))=1

2

(∇𝑢𝑢𝑢𝜀(𝑥,𝑡)+(∇𝑢𝑢𝑢𝜀(𝑥,𝑡))𝑇)

, (5)

andwerequirebothminorandmajorsymmetrypropertiesfor,i.e.,

𝜀𝑖𝑗𝑘𝑙=𝜀𝑗𝑖𝑘𝑙=𝜀𝑖𝑗𝑙𝑘=𝜀𝑘𝑙𝑖𝑗. (6) TheintegralinEq.(4),standingforthestress-strainrelationshipfor non-aging,viscoelasticmaterials,canbemanipulatedbymeansofin- tegraltransforms.Inparticular,theLaplace–Carsontransform,whichis givenby

̂𝜙𝜀(𝑥,𝑝)=𝑝

0 𝑒𝑝𝑡𝜙𝜀(𝑥,𝑡)𝑑𝑡,𝑡≥0, (7)

where𝑝isthevariableintheLaplace–Carsonspace,reduces(4)toan algebraic equationrepresenting theconstitutiverelationsin classical elasticitytheory(see,forinstance,Lakes,2009).Thismethodologyorig- inallyproposedbyHashin(1965)andknownastheelastic-viscoelastic

correspondenceprinciple,continuestogaininterestinthescientificlit- erature (seeLiu etal., 2020; Vilchevskayaet al., 2019; Yangetal., 2019andreferencestherein).Hence,basedontheaboveconsiderations, theoriginalsystem(3a)–(3c)writtenintheLaplace–Carsondomainis givenby

∇⋅[̂𝜀(𝑥,𝑝)∶𝜉𝜉𝜉(

̂𝑢 𝑢𝑢𝜀(𝑥,𝑝))]

=000 in(Ω∖Γε)× [0,+∞) (8a)

𝑢𝑢̂𝑢𝜀(𝑥,𝑝)=000 onΓ𝜀× [0,+∞) (8b) [̂𝜀(𝑥,𝑝)∶𝜉𝜉𝜉(

̂𝑢 𝑢𝑢𝜀(𝑥,𝑝))

]𝑛𝑛𝑛(𝑦)=000 onΓ𝜀× [0,+∞). (8c) 3. Asymptotichomogenizationapproach

Inthissection,wesummarizethemethodologydescribedinCruz- Gonzálezetal.(2020b)andwritethelocalandhomogenizedproblems resultingfromtheapplicationoftheAHMtoEqs.(8a)–(8c).

Beforegoingfurther,weremarkthatusingthechainrulethefollow- ingrelationforthespatialderivativesholds

𝜕̂𝜙𝜀𝑖(𝑥,𝑝)

𝜕𝑥𝑗 = 𝜕̂𝜙𝑖(𝑥,𝑦,𝑝)

𝜕𝑥𝑗 +1 𝜀

𝜕̂𝜙𝑖(𝑥,𝑦,𝑝)

𝜕𝑦𝑗 . (9)

Moreover,Eq.(5)becomes, 𝜉𝑘𝑙(̂𝜙𝜀(𝑥,𝑝))=𝜉𝑘𝑙(̂𝜙(𝑥,𝑦,𝑝))+1

𝜀𝜉𝑘𝑙(𝑦)(̂𝜙(𝑥,𝑦,𝑝)), (10) wherewehaveintroducedthenotation

𝜉𝑘𝑙(𝑦)(̂𝜙(𝑥,𝑦,𝑝))=1 2

(𝜕̂𝜙𝑘(𝑥,𝑦,𝑝)

𝜕𝑦𝑙 +𝜕̂𝜙𝑙(𝑥,𝑦,𝑝)

𝜕𝑦𝑘 )

. (11)

TheAHM(BakhvalovandPanasenko,1989;CioranescuandDonato, 1999)proposesthesolutionoftheviscoelasticheterogeneousproblem (8a)-(8c)asaformalseriesexpansioninpowersof𝜀.IntheLaplace–

Carsondomainitreads

̂𝑢 𝑢 𝑢𝜀(𝑥,𝑝)=

+∞

𝑖=0

𝜀𝑖𝑢𝑢̂𝑢(𝑖)(𝑥,𝑦,𝑝), (12)

wherethecoefficients𝑢𝑢̂𝑢(𝑖)(𝑥,𝑦,𝑝)areassumedtobeperiodicinthemicro- scopicvariable𝑦.Thus,followingthestandardprocedureinasymptotic homogenization(seeBakhvalovandPanasenko,1989;Cioranescuand Donato,1999),aftersubstitutionoftheseriesexpansion(12)intheorig- inalproblem(8a)–(8c)andbyequatingtheresultinthesamepowers of𝜀,weobtainthat,inthelimit𝜀→0,

̂𝑢𝜀𝑚(𝑥,𝑝)=̂𝑢(0)𝑚(𝑥,𝑦,𝑝)+̂𝑢(1)𝑚(𝑥,𝑦,𝑝)𝜀+𝑜(𝜀)

=̂𝑣𝑚(𝑥,𝑝)+ ̂𝜒𝑘𝑙𝑚(𝑦,𝑝)𝜉𝑘𝑙(𝑣𝑣̂𝑣(𝑥,𝑝))𝜀+𝑜(𝜀), (13) where𝑣𝑣̂𝑣isasmoothvectorfunctionof𝑥and𝑝,andthethirdordertensor

̂𝜒𝑘𝑙𝑚isthesolutionofthe𝜀-localproblemgivenby

𝜕𝑦𝜕𝑗

[̂𝑖𝑗𝑠𝑞(𝑦,𝑝)𝜉𝑠𝑞(𝑦)(̂𝜒𝜒𝜒𝑘𝑙(𝑦,𝑝))+̂𝑖𝑗𝑘𝑙(𝑦,𝑝) ]

=0 in(𝑌 ⧵Γ𝑌)× [0,+∞), (14a) ̂𝜒𝑘𝑙𝑚(𝑥,𝑦,𝑝)=0 onΓ𝑌× [0,+∞), (14b) ̂𝑖𝑗𝑠𝑞(𝑦,𝑝)𝜉(𝑠𝑞𝑦)(̂𝜒𝜒𝜒𝑘𝑙(𝑦,𝑝))+̂𝑖𝑗𝑘𝑙(𝑦,𝑝)]𝑛(𝑗𝑦)=0 onΓ𝑌× [0,+∞), (14c)

̂𝜒𝑘𝑙𝑚(𝑦,𝑝)=0 in𝑌× {0}, (14d)

where

𝜉𝑠𝑞(𝑦)(̂𝜒𝜒𝜒𝑘𝑙(𝑦,𝑝))= 1 2

(𝜕̂𝜒𝑘𝑙𝑠(𝑦,𝑝)

𝜕𝑦𝑞 +𝜕̂𝜒𝑘𝑙𝑞(𝑦,𝑝)

𝜕𝑦𝑠 )

. (15)

(4)

Theuniquenessofthesolutionofthelocalproblem(14a)–(14d)isguar- anteedbyenforcingeither,thecondition ⟨̂𝜒𝑘𝑙𝑚𝑦=0orbyfixingthe valueof ̂𝜒𝑘𝑙𝑚 atonepointofthereferenceperiodiccell𝑌 (seePenta andGerisch,2016;PentaandGerisch,2017).Inparticular,inthesub- sequentsectionswewillusethelatterbecauseofitsadvantagewhen dealingwithnumericalsimulations.Notethat,the𝜀-localproblemhas tobesupplementedwithaninitialconditionin𝑌× {0}.

Forcompletenessinouranalysis,wereportthehomogenizedequa- tionatthemacroscaleintheLaplace–Carsonspace,whichisobtained afterequatinginthesamepowersof𝜀0.Specifically,thiscanbewritten as

̂(∗)𝑖𝑗𝑘𝑙(𝑝) 𝜕

𝜕𝑥𝑗𝜉𝑘𝑙(𝑣𝑣̂𝑣(𝑥,𝑝))=0 inΩ× [0,+∞), (16a) where

̂(∗)𝑖𝑗𝑘𝑙(𝑝)∶=⟨̂𝑖𝑗𝑘𝑙(𝑦,𝑝)+̂𝑖𝑗𝑚𝑛(𝑦,𝑝)𝜉𝑚𝑛(𝑦)(

̂𝜒𝜒𝜒𝑘𝑙(𝑦,𝑝))⟩

𝑦. (17)

istheeffectiverelaxationmodulusintheLaplace–Carsonspace.InEq.

(17),thenotation⟨𝜙⟩𝑦denotesthecellaverageoperatorandisdefined bytheexpression

⟨𝜙⟩𝑦= 1

|𝑌|∫𝑌𝜙 𝑑𝑦, (18)

with|𝑌|beingthevolumeoftheperiodiccell𝑌. 4. Calculationoftheeffectiveproperties

Forsimplicityinourcalculations,weconsiderthattherelaxation modulus,̂𝑖𝑗𝑘𝑙,is𝑦-constantineachconstituentoftheperiodiccell𝑌, i.e.

̂𝑖𝑗𝑘𝑙(𝑦,𝑝)=

{̂(1)𝑖𝑗𝑘𝑙(𝑝), if𝑦𝑌1,

̂(2)𝑖𝑗𝑘𝑙(𝑝), if𝑦𝑌2, (19) wherethesuperscriptindicatethecorrespondingconstituent,“(1)” for thematrixand“(2)” fortheinclusion(seeFig.(1)(c)).Then,thelocal problem(14a)–(14d)isrewrittenasfollows,

𝜕𝑦𝜕𝑗

[̂(1)𝑖𝑗𝑠𝑞(𝑝)𝜉𝑠𝑞(𝑦)(

̂𝜒𝜒𝜒(1)𝑘𝑙(𝑦,𝑝))]

=0 in𝑌1× [0,+∞), (20a)

𝜕

𝜕𝑦𝑗

[̂(2)𝑖𝑗𝑠𝑞(𝑝)𝜉𝑠𝑞(𝑦)(

̂𝜒𝜒𝜒(2)𝑘𝑙(𝑦,𝑝))]

=0 in𝑌2× [0,+∞), (20b)

̂𝜒𝑘𝑙𝑚(1)(𝑦,𝑝)= ̂𝜒𝑘𝑙𝑚(2)(𝑦,𝑝) onΓ𝑌× [0,+∞), (20c) [̂(1)𝑖𝑗𝑠𝑞(𝑝)𝜉(𝑠𝑞𝑦)(

̂𝜒𝜒𝜒(1)𝑘𝑙(𝑦,𝑝))]

𝑛(y)𝑗

−[

̂(2)𝑖𝑗𝑠𝑞(𝑝)𝜉𝑠𝑞(𝑦)(

̂𝜒𝜒𝜒(2)𝑘𝑙(𝑦,𝑝))]

𝑛(y)𝑗 =𝑓𝑖𝑘𝑙(𝑦)(𝑝) onΓ𝑌× [0,+∞), (20d)

̂𝜒𝑘𝑙𝑚(𝑦,𝑝)=0 in𝑌× {0}. (20e)

Wenoticethat,inEq.(20d),thestressjumpconditionsleadtotheoc- currenceofinterfaceloads,i.e.,

𝑓𝑖𝑘𝑙(𝑦)(𝑝)=[

̂(2)𝑖𝑗𝑘𝑙(𝑝)−̂(1)𝑖𝑗𝑘𝑙(𝑝)]

𝑛(y)𝑗 , (21)

whichariseasaconsequenceofthediscontinuitiesinthecoefficients of̂ betweenthehostmediumandthesub-phases,andrepresentthe drivingforcetoobtainnontrivialsolutionsofthesixelastic-typelocal problems((𝑘,𝑙),𝑘𝑙)(seePentaandGerisch,2016;PentaandGerisch, 2017).Inparticular,whenthematrixandsubphasesareorthotropic materialsandconsideringVoigt’snotation,theinterfaceloads𝑓𝑖𝑘𝑙(𝑦)read

𝑓𝑓𝑓(11𝑦)(𝑝)=[̂(2)11(𝑝)−̂(1)11(𝑝)]𝑛(1𝑦)𝑒𝑒𝑒1+[̂(2)21(𝑝)−̂(1)21(𝑝)]𝑛(2𝑦)𝑒𝑒𝑒2

+[̂(2)31(𝑝)−̂(1)31(𝑝)]𝑛(3𝑦)𝑒𝑒𝑒3, (22a) 𝑓 𝑓

𝑓(22𝑦)(𝑝)=[̂(2)12(𝑝)−̂(1)12(𝑝)]𝑛(1𝑦)𝑒𝑒𝑒1+[̂(2)22(𝑝)−̂(1)22(𝑝)]𝑛(2𝑦)𝑒𝑒𝑒2

+[̂(2)32(𝑝)−̂(1)32(𝑝)]𝑛(3𝑦)𝑒𝑒𝑒3, (22b)

𝑓𝑓

𝑓(33𝑦)(𝑝)=[̂(2)13(𝑝)−̂(1)13(𝑝)]𝑛(1𝑦)𝑒𝑒𝑒1+[̂(2)23(𝑝)−̂(1)23(𝑝)]𝑛(2𝑦)𝑒𝑒𝑒2

+[̂(2)33(𝑝)−̂(1)33(𝑝)]𝑛(3𝑦)𝑒𝑒𝑒3, (22c)

𝑓 𝑓

𝑓(23𝑦)(𝑝)=𝑓𝑓𝑓(32𝑦)(𝑝)=[̂(2)44(𝑝)−̂(1)44(𝑝)]𝑛(3𝑦)𝑒𝑒𝑒2+[̂(2)44(𝑝)−̂(1)44(𝑝)]𝑛(2𝑦)𝑒𝑒𝑒3, (22d) 𝑓𝑓

𝑓(13𝑦)(𝑝)=𝑓𝑓𝑓(31𝑦)(𝑝)=[̂(2)55(𝑝)−̂(1)55(𝑝)]𝑛(3𝑦)𝑒𝑒𝑒1+[̂(2)55(𝑝)−̂(1)55(𝑝)]𝑛(1𝑦)𝑒𝑒𝑒3, (22e) 𝑓

𝑓

𝑓(12𝑦)(𝑝)=𝑓𝑓𝑓(21𝑦)(𝑝)=[̂(2)66(𝑝)−̂(1)66(𝑝)]𝑛(2𝑦)𝑒𝑒𝑒1+[̂(2)66(𝑝)−̂(1)66(𝑝)]𝑛(1𝑦)𝑒𝑒𝑒2, (22f) where{𝑒𝑒𝑒𝑖}3𝑖=1representsthestandardvectorbasis.

4.1. Numericalapproach

Atthispoint,itispossibletosolvenumericallythesetofelastic- typelocalproblems(20a)–(20e)intheLaplace–Carsonspaceandthen, tocomputetheeffectiveviscoelasticpropertiesbyusing(17).Forthis purpose,weusethefiniteelementsoftwareCOMSOLMultiphysics®in conjunctionwithLiveLinkTMforMatlab®scripting(seeCruz-González etal.,2020b;PentaandGerisch,2016;PentaandGerisch,2017).Par- ticularly,once̂(∗)isknown,theeffectivecreepcompliance̂(∗)inthe Laplace–Carsonspacecanbecomputedthroughtherelationship

̂(∗)𝑖𝑗𝑚𝑛(𝑝)̂𝑚𝑛𝑘𝑙(∗) (𝑝)=𝐼𝑖𝑗𝑘𝑙, (23) where𝐼𝑖𝑗𝑘𝑙denotesthecomponentsofthefourth-orderidentitytensor (seeHashin,1972).

Theinversionoftheeffectivecoefficientstotheoriginaltimedomain isperformedbyemployingtheMATLAB’sfunctionINVLAP(seeJuraj, 2020;ValsaandBranĉik,1990andreferredheretoasValsa’smethod.

Thestepsaresummarizedasfollows,

(a) Discretizethetimeinterval𝑡=[𝑡1,𝑡2,...,𝑡𝑁]

(b) Foreach 𝑡𝑖,obtainthecomponents𝑝𝑗∶=𝛼𝑗𝑡𝑖and𝐵𝑗∶=𝛽𝑗𝑡𝑖for 𝑗=1,...,(𝑛𝑠+𝑛𝑑+1),where𝑛𝑠and𝑛𝑑areimplicitparameters,and 𝛼and𝛽aredefinedinValsa’smethod.

(c) Calculate̂(∗)(𝑝𝑗)and̂(∗)(𝑝𝑗)for𝑗=1,...,(𝑛𝑠+𝑛𝑑+1).

(d) UsethelaststepofValsa’smethodtodeterminetheeffectivecoeffi- cientsinthetimedomain

(∗)(𝑡𝑖)=

𝑛𝑠+𝑛𝑑+1 𝑗=1

Re[𝐵𝑗̂(∗)(𝑝𝑗)∕𝑝𝑗] for𝑖=1,...,𝑁, (24a)

(∗)(𝑡𝑖)=

𝑛𝑠+𝑛𝑑+1 𝑗=1

Re[𝐵𝑗̂(∗)(𝑝𝑗)∕𝑝𝑗] for𝑖=1,...,𝑁, (24b)

whereReindicatestherealpartofacomplexvariable.

Moreover, totake intoaccountthedifferent orientationsthatthe unidirectionalviscoelasticcompositesmayhave,werotatetheeffective viscoelastictensors(∗)and(∗)byanangle𝜃andobtain(∗)𝜃 and𝜃(∗). Inthisrespect,thefollowingtransformationsareuseful,

(∗)𝜃 =(∗)𝑇, (25a)

𝜃(∗)=(−1)𝑇(∗)−1, (25b)

(5)

Fig.2. MethodologysketchofAHMFE.

where

[]=

⎡⎢

⎢⎢

⎢⎢

⎢⎢

211212213 21213 21311 21112

221222223 22223 22321 22122

231232233 23233 23331 23132

2131223223332233+23322331+21332132+2231

3111321233133213+33123311+31133112+3211

1121122213231223+13221321+11231122+1221

⎤⎥

⎥⎥

⎥⎥

⎥⎥

and𝑖𝑗(𝑖,𝑗=1,2,3)arethecoefficientsoftheorthogonalrotationten- sor(seeRamírez-Torresetal.,2018;Ting,1996).

Toconcludewiththissection,itisworthtoremarkthatsteps(a)- (d)intheinversionprocessareequivalentstothestage(IV)inCruz- Gonzálezetal.(2020b). Here,weillustratethroughaflowchart(see Fig.2) themethodology describedin (I)-(IV)ofCruz-González etal.

(2020b)withmore details. Inthefollowing sections, werefer toas AHMFEthe semi-analytical approachproposed in thepresentwork, whichcombinestheAsymptoticHomogenisationMethod(AHM)and FiniteElements(FE)simulations.

5. Results

5.1. Instantelasticresponse

Tobeginwithouranalysis,inthissection,wecomputetheeffective instantelasticresponseofacompositewithahexagonalarrangement oftransversely,purelyelasticisotropiclongfibers(seeFig.1(b)-(iii)).

Althoughthetheoreticalframeworkintroducedintheprevioussection holds forviscoelasticfibersaswell,wefocuson purelyelasticfibers forthesakeofcomparisonofourresultswithalternativeanalytictech-

(6)

Fig.3.MeshdiscretizationsA,BandCforthehexagonalperiodiccellwith𝑉𝑓=0.6.

Table1

Elasticpropertiesoftheconstituents.

Materials 𝐸 𝐴(GPa) 𝐸 𝑇(GPa) 𝜇𝐴(GPa) 𝜇𝑇(GPa) 𝜈𝐴

AS4 graphite fiber 225 15 15 7 0.20

3501-6 epoxy 4.2 4.2 1.567 1.567 0.34

E-glass fiber 69.0 69.0 28.28 28.28 0.22

Boron fiber 420 420 175 175 0.20

Aluminum 69.0 69.0 25.94 25.94 0.33

niques.Inthiscontext,theelasticlimitcaseisreachedbyconsidering 𝑡=0inEq.(4),whichimpliesthat(∗)𝑖𝑗𝑚𝑛becomestheeffectivestiffness tensor.

Here, we determine the instant elastic effective response of a graphite/epoxysystemwithhexagonalarchitectureanddifferentfiber volumefractions.Itisworthnoticingthatthefollowingresultsdiffer fromtheonesobtainedinCruz-Gonzálezetal.(2020b)sincethereinit wasconsideredasquarearrayofinclusionsinthematrixphasecom- prisingisotropicconstituents.Thevaluesoftheparametersreportedin Table1areobtainedfromWangandPindera(2016b).Thenotation𝐸𝐴 (𝐸𝑇)and𝜇𝐴 (𝜇𝑇)isusedfortheaxial(transverse)Young’sandshear moduli,respectively,and𝜈𝐴representstheaxialPoisson’sratio.

Beforeweproceedwiththeresultsoftheeffectivecoefficients,we gatherinformationontwomainfeaturesrelatedwiththecomputational approach.Thesolution’sconvergencebehavior,throughthreetypesof meshdiscretization,andtheexecutiontimesrequiredforthesecalcula- tions.Withthispurpose,weonlyprovidetheresultsofthecomputation oftheeffectivetransverseYoung’smodulus𝐸𝑇(∗)sincethoserelatedwith thetransversePoisson’sratio𝜈𝑇(∗)andtheaxialshearmodulus𝜇𝐴(∗)show similarcharacteristics.Furthermore,withregardstotheanalysisofthe solution’sconvergence,weusethemeshesA,BandCreportedinFig.3. InFig.4,weshow theeffectivetransverseYoung’smodulus𝐸(∗)𝑇 , normalizedbythecorrespondingmatrixYoung’smodulus𝐸𝑚.Specifi- cally,inFig.4(a)theeffectivetransverseYoung’smodulusiscomputed, foreachofthemeshesdiscretizationsA,BandC,asafunctionofthe volumefraction𝑉𝑓.Moreover,inFig.4(b)wecomparetheresultsob- tainedwiththepresentmodel(AHMFE)withthoseproducedbyWang andPindera(2016b)(seeFig.7inWang andPindera,2016b) using thefinite-volumedirectaveragingmicromechanics(FVDAM)theory.A closerlookatthedatainFig.4(b)revealsthattherelativeerrorbe- tweenthesolutionsthepresentapproach(AHMFE)andthatinWang andPindera(2016b)(FVDAM),namely

𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒𝐸𝑟𝑟𝑜𝑟(𝜙)∶=𝜙(𝐹 𝑉𝐷𝐴𝑀)𝜙(𝐴𝐻𝑀𝐹 𝐸)

𝜙(𝐹 𝑉𝐷𝐴𝑀) × 100%, (26)

Table2

Maximumrelativeerror(%).

𝐴𝐻 𝑀 𝐹 𝐸 𝐿𝐸𝐻 𝑇 𝐸 (∗)𝑇𝐸 𝑚 𝜈(∗)𝑇 𝜇(∗)𝐴𝜇𝑚 glass/epoxy 0.5844 0.3969 0.0767 graphite/epoxy 0.1971 0.1311 0.0604 boron/aluminum 0.3476 0.2088 0.0604 aluminum/porosity 1.4575 1.1796 0.0677

reacheshismaximumvaluewhenthecoarsermesh(MeshA)isconsid- eredandislessthan1.3%.Ontheotherhand,therelativeerrorcom- putedwiththemeshesdiscretizationsBandCislessthan0.2%.Then,we canconcludethatoursimulationsprovideagoodagreementwithFV- DAM,andthat,inthiscase,ourresultsdonotundergolargevariations aftersubsequentmeshdiscretizations.

Tocontinue withouranalysis, in Fig.4(c)weprovide thetotal executiontimesneededtoobtaintheentiresetofeffectivemoduliin relationtothethree meshdiscretizations.Thesimulationsaresetup totakeintoaccountafinitenumberoffibervolumefractionsranging from𝑉𝑓=0.05to𝑉𝑓=0.7withanincrementof0.05,andtheywere performedusingamachinerunningWindows10Professional64-bitop- eratingsystem,with32.0GBRAMandIntel(R)Core(TM)i5-8350UCPU 1.70GHz.AsillustratedinFig.4(c),theperformanceusingthemeshC couldbeconsideredinefficientintermsoftime,inpartduetothefact thatthemeshCissignificantlyfinercomparedtothemeshesAandB. Particularly,charts(d),(e)and(f)inFig.4offermoreprecisedetails onthecomputingtimeforeachvolumefractionandeachmesh.The differencesin theresultsareduetothefactthatdifferentvolumetric fractionsmodifythegeometryoftheperiodiccellforthecorrespond- ingfixedmesh.Wealsohighlightthecontrastinrelationtothemean computingtime.Takingintoaccountboththerelativeerrorsandthe executiontimes,weconcludethatthemeshBisthebestpossiblechoice foroursimulations.

Finally,tocomplete ouranalysis, in Fig.5, weshow thenumer- ical values of the effective moduli 𝐸𝑇(∗)𝐸𝑚, 𝜈𝑇(∗) and 𝜇(∗)𝐴𝜇𝑚 for a hexagonal array of unidirectional longfibers anddifferent contrasts in the constituents. In particular, we study the pairs glass/epoxy, boron/aluminum,aluminum/porosityandgraphite/epoxy.Wereferto Table1forthematerialpropertiesoftheseconstituents.Byreferringto Table 2, inwhichsummarizemaximumrelativeerrors, wenotethat our numerical results arein agreementwith theresults provided in WangandPindera(2016b)usingthelocallyexacthomogenizationthe- ory(LEHT).AsobservedinTable2themaximumrelativeerrorisless than1.46%.

(7)

Fig.4. Chart(a)showsthecomputationofthenormalizedeffectiveYoung’smodulusofagraphite/epoxysystemwithhexagonalarchitecture.Chart(b)displays therelativeerrorofAHMFEinrelationtoFVDAM.Chart(c)showsthetotalcomputingtimeandthethreecharts(d),(e)and(f)providethespecificcomputingtime foreachmesh,respectively.

Fig.5.Calculationoftheeffectivemoduliforunidirectionalcompositeswithdifferentcontrastintheconstituentsandhexagonalperiodiccell.Thecomparisonsare performedwithFig.9ofWangandPindera(2016b).

5.2. Viscoelasticresponse

Inthissection,we computetheeffectiveproperties oflinearvis- coelasticcompositematerials.Particularly,inSection5.2.1wecompute theeffectivepropertiesforacompositematerialmadeofisotropiccon- stituentswheretheviscoelastic behaviour ofthematrix isdescribed by means of a Burger’s model. We considera long fiber reinforce- mentwithsquareandhexagonalgeometricalarrangements,andcon- siderthecaseinwhichPoisson’sratioorthebulkmodulusofthevis- coelasticmatrixareconstants.Furthermore,inSection5.2.2,wedeal

withtransverseisotropiclongfiberswithdifferentorientationsandcon- sider thepower-law modelgiven in Yancey andPindera (1990)for the characterizationof thecreep compliance of theviscoelastic ma- trix.Indoingthis,weobtainedgoodagreementswithboththeLETH andexperimentalresults.Finally,inSection5.2.3,weaddressthecal- culationoftheeffectivepropertiesforacompositemadeofperfectly aligned shortfibers embeddedintoa viscoelastic matrix withtrans- verselyisotropicbehaviour.So,weshowthepotentialofourapproach inthesolutionoffullythree-dimensionalproblemsinvolvinginclusions, which cannot be addressedby meansof analyticalmethods suchas

(8)

Table3

Maximumrelativeerror(%)inthetimeintervalunderstudy.

Constant Poissons ratio Constant bulk modulus AHMFE LEHT 22(∗) 44(∗) 66(∗) 22(∗) 44(∗) 66(∗) Hexagonal cell 0.2840 0.2605 0.2622 0.2707 0.2345 0.2901 Square cell 1.7953 0.0994 0.3588 2.2884 0.2034 0.3230

theLEHT.Wementionthatintheupcomingsimulations,weemploy MeshB.

5.2.1. Comparisonwithlocally-exacthomogenizationtheory

Inthissection,weanalyzetheinfluenceofthesquareandhexag- onalarrangementoflongfibers (seeFig.1) inthecalculationofthe effectivecreepcomplianceinpolymericmatrixcomposites.Inaddition, asinWangandPindera(2016a),weconsiderthateitherPoisson’sratio (𝜈𝑚)or thebulk modulus(𝐾𝑚)of theviscoelasticmatrixisconstant.

This assumption isjustified bythe necessityof modelingthepoten- tiallytime-independent responseofpolymeric matricesunderhydro- staticloadingWang andPindera(2016a).Ontheone hand,we take 𝜈𝑚=0.38asthevalueforthematrixandthereforethetime-dependent bulkmodulusisgivenby𝐾𝑚(𝑡)=𝐸𝑚(𝑡)∕(3(1−2𝜈𝑚)),where𝐸𝑚(𝑡)stands fortheone-dimensionalrelaxationmodulus.Ontheotherhand,ifweas- sumeaconstantbulkmodulusforthematrix,weusetheequation𝐾𝑚= 𝐸0∕(3(1−2𝜈𝑚))todetermineits value,andthen,thetime-dependent Poisson’sratio𝜈𝑚(𝑡)arisesfromtheequation𝜈𝑚(𝑡)=1∕2−𝐸𝑚(𝑡)∕(6𝐾𝑚). Here,𝐸0representstheinstantaneouselasticrelaxationmodulus.

In the following, we investigate the effective properties of unidirectionally-reinforcedglass/epoxycompositeswithlinearisotropic constituents,whereelasticglassfibersareembeddedintoaviscoelastic polymericmatrix(see,e.g.CavalcanteandMarques,2014;Chenetal., 2017;Cruz-Gonzálezetal.,2020b;WangandPindera,2016a).Here,in contrastwithCavalcanteandMarques,2014;Chenetal.,2017;Cruz- Gonzálezetal.,2020b,weconsiderhexagonalperiodiccellsinthecom- putations.ThemechanicalpropertiesofthefibersaregivenbyYoung’s modulus𝐸𝑓=68.77GPaandPoisson’sratio𝜈𝑓 =0.21.Furthermore,we describetheviscoelasticmatrixbyassumingtherelaxationrepresenta- tionofthefour-parametermodelorBurger’smodel,i.e.twoMaxwell elementssetinparallel(seeMainardiandSpada(2011)forfurtherde- tails).Specifically,theexpressionoftherelaxationmodulusisgivenas follows,

𝐸𝑚(𝑡)=𝐺1exp (

𝑡 𝜂𝜎,1

) +𝐺2exp

(

𝑡 𝜂𝜎,2

)

, (27)

where the material properties are taken from Wang and Pindera (2016a)bymeansofthescalarformof(23)(seeParkandKim,1999) andsome algebraic transformations.Thedata setis reported asfol- lows,𝐺1=1.12511GPa, 𝐺2=2.14489GPa, 𝜂𝜎,1=6999.34h and𝜂𝜎,2= 58.2551h, where 𝐺𝑛 (𝑛=1,2) represents the elasticmodulus of the springand𝜂𝜎,𝑛is arelaxationtimeMainardiandSpada(2011). Itis worthtoremarkthatfromeq.(27),weobtain𝐸0=𝐺1+𝐺2.

InFig.6,weshowthecurvescorrespondingtotheeffectivecreep compliances22(∗),44(∗)and66(∗)forconstantPoissonsratio(leftcharts) andconstantbulkmodulus(rightcharts).Specifically,inFig.6,wecom- pareourresultswiththoseobtainedinWangandPindera(2016a)via theLEHT.Inthesecomparisons,hexagonalandsquarearraysoffibers arestudiedforafibervolumefractionequalto0.6.Asitcanbenoticed, thereisagoodagreementbetweenthetwoapproaches,whichisfurther evidencedbythemaximumrelativeerrorsprovidedinTable3.

5.2.2. Power-lawmodel.Comparisonwithexperiments

Inthissection,wefollow theanalysisadoptedinWang andPin- dera (2016a), and compareour results with theexperimental creep measurements obtained in Yancey and Pindera (1990) for off-axis graphite/epoxy specimens, and with the numerical results given in

Table4

ElasticpropertiesofthetransverselyisotropicT300graphitefiberat room(22𝐶)andelevated(121𝐶)temperature.

Temperature 𝐸 𝐴(GPa) 𝐸 𝑇(GPa) 𝜇𝐴(GPa) 𝜈𝐴 𝜈𝑇 22 𝐶 202.82 25.30 44.12 0.443 0.05 121 𝐶 214.33 14.82 68.18 0.450 0.05

WangandPindera(2016a).InYanceyandPindera(1990),theauthors observed thatat22𝐶 and121𝐶 theT300 graphitefiberpresentan elasticbehavior,whereasthecreepresponseofthe934epoxymatrixis fittedbythepower-law

𝑆𝑚(𝑡)= 1 𝐸0

+𝐶𝑡𝑛, (28)

where𝐶and𝑛areexperimentallymeasuredparametersand𝐸0isthe instantaneouselasticrelaxationmodulus(YanceyandPindera,1990).

Beforeproceeding,itisworthmentioningthat,eventhoughinthe presentmodelwedonotdealwithfractionalviscoelasticity(werefer theReadertoAtanacković etal.,2016;Beltempoetal.,2019;Bouras etal.,2018;Mainardi,2010;MainardiandSpada,2011andtherefer- encestherein),wefindconvenienttousesomeoftheresultsgivenin theseworksforthecalculationoftherelaxationmodulus ̂𝐸𝑚(𝑝)which isneededinoursimulations.

Withthispurpose,weintroducethenotation𝜇=𝐸0∕2,𝜏=𝛽−1∕𝑛, 𝛽=𝐸0𝐶Γ(1+𝑛)and𝛼=𝑛,sothatEq.(28)canbeequivalentlyrewrit- tenas

𝑆𝑚(𝑡)=𝑆𝑀(𝑡)= 1 2𝜇

[

1+ (𝑡𝜏)𝛼 Γ(1+𝛼)

]

, (29)

whereΓdenotestheGammafunction,𝛼∈ ]0,1],and𝑆𝑀(𝑡)represents thefractionalcreepcompliance (Mainardi andSpada,2011).Hence, byreferringtotheresultsobtainedinMainardiandSpada(2011),the fractionalrelaxationmodulusisgivenbytheexpression

𝐸𝑀(𝑡)=2𝜇𝛼(−(𝑡𝜏)𝛼), (30)

where𝛼denotestheMittag–Lefflerfunctionoforder𝛼,whichisdefined bytheexpression(Gorenfloetal.,2014)

𝛼(−(𝑡𝜏)𝛼)=

𝑝=0

(−1)𝑝 (𝑡𝜏)𝛼𝑝

Γ(1+𝛼𝑝), 0<𝛼 <1, 𝜏 >0, (31)

andfor𝛼=1reducesto𝛼((−𝑡𝜏)𝛼)=exp(−t∕τ).Wenoticethat,from Eq.(30)andbyvirtueoftheparameteridentificationsestablishedfor obtainingEq.(29),obtain

𝐸𝑚(𝑡)=𝐸0𝑛(

𝐸0𝐶Γ(1+𝑛)𝑡𝑛)

. (32)

However,Eq.(32)doesnotprovide,inadirectway,anincremental solutionfortheprobleminthetimedomain.Therefore,werelyonthe expressionfortherelaxationmodulus ̂𝐸𝑚(𝑝),expressedwithrespectto theLaplace–Carsondomain,forthecomputationoftheeffectiveproper- ties.Thus,byemployingtheresultsgiveninMainardiandSpada(2011), theLaplace–CarsontransformofEq.(30)is

̂𝐸𝑀(𝑝)= 2𝜇(𝑝𝜏)𝛼

1+(𝑝𝜏)𝛼, (33)

whichimpliesthat

̂𝐸𝑚(𝑝)= 𝐸0𝑝𝑛

𝑝𝑛+𝐸0𝐶Γ(1+𝑛). (34)

Inthisway,theLaplace–Carsontransformfortherelaxationmodulus canbeanalyticallyobtainedinanexplicitform,whichhighlyreduces thenumericalcomplexityoftheproblem.

Inthefollowing,weconsideracompositewithhexagonalarrange- mentof longfibers wherethepropertiesof theconstituents,i.e. the elasticfibers andtheviscoelasticmatrixgivenin Wang andPindera (2016a)aresummarizedinTables4and5,respectively.

(9)

Fig.6. Charts(a)-(c)showtheresultsunder theconsiderationsofconstantPoisson’sratio whilecharts(d)-(f)assumeconstantbulkmod- ulus.We consider square andhexagonalar- raysoffibers.Ourresultsarecomparedwith Fig.7ofWangandPindera(2016a).

Table5

Materialpropertiesoftheepoxymatrixatroom(22𝐶)andel- evated(121𝐶)temperature.

Temperature 𝐸 0(GPa) 𝜈 𝐶(1/(GPa ×min)) 𝑛

22 𝐶 4.51 0.311 0.0135 0.17

121 𝐶 3.36 0.317 0.0250 0.20

Fig.7shows theeffectivecreepresponseof theviscoelastic com- positematerialgivenbythecoefficients11(∗)and22(∗).Similarlytothe previoussection,weconsiderthecasesofconstantPoisson’sratioand constantbulkmodulus.Inaddition,weanalyzetheinfluenceoftwodif-

ferenttemperatures,i.e.22𝐶(room)and121𝐶(elevated).Theresults correspondtocompositeswithfibersthatarerotated10and90coun- terclockwiseaboutthe𝑦2-axis.Inthiscase,thefibervolumefraction ofthefiberisfixedto𝑉𝑓=0.6.Acomparisonofourresultswiththose obtainedinWangandPindera(2016a)usingLEHTshowsagoodagree- mentbetweenbothapproaches.Additionally,thequalitativebehavior ofthecurvesisverysimilartotheexperimentaldata.InTable6,we providethemaximumrelativeerrorsbetweentheresultsobtainedvia twomethods.

5.2.3. Modelingofshortfiberreinforcement

Thelocally-exacthomogenizationtheory(LEHT)isbasedonatwo- dimensionalformulationwhichisonlycapableoftakingintoaccount

(10)

Fig.7. Calculationoftheeffectivecreepresponseinviscoelasticcompositeswithhexagonalarrayoftransverselyisotropicfiberwith10and90off-axisspecimens aboutthe𝑥2-axis.Weconsidertwodifferenttemperatures,i.e.22𝐶(room)and121𝐶(elevated).Inaddition,Charts(a)-(b)showsthecasesofconstantPoissons ratioandcharts(c)-(d)theconstantbulkmodulus.

Table6

Maximumrelativeerror(%)inthetimeintervalunderstudy.

Constant Poissons ratio Constant bulk modulus Hexagonal cell 𝜃= 10 𝜃= 90 𝜃= 10 𝜃= 90 AHMFE LEHT 11(∗) 22(∗) 11(∗) 22(∗)

22 𝐶 0.0798 0.3287 0.0870 0.3799

121 𝐶 0.1340 0.3847 0.1781 0.1666

AHMFE Experiments 11(∗) 22(∗) 11(∗) 22(∗)

22 𝐶 4.2505 3.3196 4.5503 5.2098

121 𝐶 5.6338 4.7281 6.7843 1.7558

longcylindricalfibers(see,e.g.Chenetal.,2017).Inthissection,we showthepotentialoftheAHMFEapproachinthemodelingofviscoelas- ticcompositesforthree-dimensionalgeometricalconfigurations.Partic- ularly,weconsideraviscoelasticcompositematerialwithsquareand hexagonalarrangementofperfectlyalignedshortfibers(seeFig.1(b) (ii,iv))represented bycylindrical inclusions.In addition,weassume

thattheconstituentsbehaveasthoseinSection5.2.2,thatis,wecon- sideraviscoelasticmatrix(934epoxy)withpower-lawcreepcompli- anceasgivenin(28),andreinforcedbytransverselyisotropicelastic fibers(T300graphite).

In this context we define the parameters𝛾1∶=1𝐻1 and𝛾2∶=

2𝐻2 whichrelatethelengthmeasurementofthefiberandthema- trixinthesquareandhexagonalperiodiccell,respectively(seeFig.8 (a)).Inaddition,weassumethefibers tobecenteredin theperiodic cells.Therefore,wenoticethat0≤𝛾1,𝛾2≤1,whereazerovaluerep- resentsahomogeneousmaterialmadeonlywiththematrixandaone valuereproducesthecaseoflongfibersasparticularcaseofthisap- proach.Moreover,asobservedinFig.8(a),westudycounterclockwise uniformrotationsoftheshortfibersabout𝑦2-axisof0≤𝜃 <𝜋.

InFig.8(b)and(c),weshowourfindingsinthecalculationofthe effectivemoduli𝐸1(∗)and𝜈(∗)32 forsquareandhexagonalperiodiccell.

Inparticular, we assumetheroom temperature(22𝐶) andthecon- stantbulkmodulusapproachdiscussedintheSection5.2.2.Thefiber volumefractionandtheratio𝛾 arefixedto𝑉𝑓=0.1and𝛾=𝛾1=𝛾2=

Références

Documents relatifs

A saber: Heriberto Cairo (Universidad Complutense de Madrid), Rosa de la Fuente (Universidad Complutense de Madrid), Javier Franzé (Universidad Complutense de Madrid), Gerónimo

conditions resulted in little decrease of elastic properties, it was observed that mechanical loading of the materials combined with water immersion resulted in a strong

In order to evaluate the influence of plasticizing on flax fibers, the dependence of their elastic modulus to water content was updated to fit the local Halpin-Kardos homogenized

As far as composites and their corresponding matrices have different viscoelastic responses (composites are yield stress shear-thinning flu- ids and MAPP/PP blends demonstrate

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

The influence of temperature, frequency, the type of rein- forcements and their orientations on the mechanical and viscoelastic behavior of composite laminates has been de- termined

PDG is obtained by setting all weight contributions to 1, except for predicted load misses which have a weight contribution equal to N, which is a fixed number that is larger than