Characterisation and improvement of the quality of mixing of recycled thermoplastic composites

(1)

ContentslistsavailableatScienceDirect

Composites

Part

C:

Open

Access

journalhomepage:www.elsevier.com/locate/jcomc

Characterisation

and

improvement

of

the

quality

of

mixing

of

recycled

thermoplastic

composites

Guillaume

A. Vincent

a,b

_,

_Thomas

_A.

_de

_Bruijn

c

_,

_Sebastiaan

_Wijskamp

b

_,

Mohammed

Iqbal

Abdul

Rasheed

a

_,

_Martin

_van

_Drongelen

a,∗

_,

_Remko

_Akkerman

a a Chair of Production Technology, University of Twente, Enschede 7522 NB, the Netherlands

b ThermoPlastic composites Research Center, Palatijn 15, Enschede 7521 PN, the Netherlands

c ThermoPlastic composites Application Center, Saxion University of Applied Sciences, M.H. Tromplaan 28, Enschede 7513 AB, the Netherlands

a

r

t

i

c

l

e

i

n

f

o

Keywords: Long ﬁbre Thermoplastic composites Mixing Spatial statistics Image analysis

a

b

s

t

r

a

c

t

Alow-shearmixerwasusedtoblendthermoplasticcompositescrapmaterialintoamoltenmixeddough,which wasthencompressionmoulded.Thisprocessisakeystepinanovelrecyclingsolutionforthermoplastic com-posites.Astudywascarriedouttocharacterisethequalityofmixing(QoM)oftheblendeddoughstounderstand howtoimprovetheQoMofmixeddoughstowardsfurtherimprovementandimplementationoftherecycling solution.Inordertoachievethis,theeffectofmixingparametersandfibrelengthontheQoMwerestudied.This studyusedshreddedC/PPSflakes,originatingfromconsolidatedlaminatescrap.Theseflakesareabout20mm insize,andcontainwovenfabricreinforcement,makingthemfardifferentfromregularpellets,andtherefore moredifficulttomix.TheQoMwascharacterisedbymeansofimageanalysisofalargesetofcross-sectional microscopyimages,basedonwhichthescaleandintensityofsegregationofthefibreclusterswereevaluated. BundlesizedistributionwasdeterminedbyapplyingDelaunaytriangulationstoclusterthefibrecentres.These methodswerefoundtobesuitableforcharacterisingtheQoMofsuchdoughs.Increasingthemixingtimeand mixingspeedwereidentifiedaskeywaystoimprovethemixingprocess.Withthecurrentmixingmachine,it isalsosuggestednottousefibreslongerthan15mmonaverageinordertolimitintra-doughvariability.For doughsmadeoffibreslongerthan15mm,improvementsonthemixingdevicecouldsufficientlyincreasethe QoM.

1. Introduction

Theapplicationoflongfibrethermoplastics(LFTs)attracted consid-erableinterestinthepastdecades[1,2].Ononehand,theyofferand allowforanincreasedgeometricalcomplexity,fromadesignpointof view,comparedtocontinuousfibrecomposites.Ontheotherhand,they exhibitmechanicalpropertiesintermediatetoshortandcontinuous fi-brecomposites.SeveralcategoriesofLFTshavebeenbroughttomarket includingthermoplasticbulkmouldingcompounds[3,4],glassmat ther-moplasticsandLFTpellets[2,5,6].Duringthesameperiod,the produc-tionanddemandforcontinuousfibrethermoplasticcomposites(TPCs) hasincreasedconsiderably,leadingtoariseinTPCproductionscrap. Theeconomicbenefitsofrecyclingthisscrap,inconjunctionwith envi-ronmentalandlegislativeincentives,haveencouragedthedevelopment ofrecyclingsolutionsforTPCs.Recently,bothDeBruijnetal.[7,8]and Janneyetal.[9]separatelydevelopedarecyclingrouteforTPCscrap inspiredbytheprocessingofLFTpellets.Aschematicviewofthe recy-clingsolutionimplementedbyDeBruijnisshowninFig.1.Production

∗_{Corresponding}_author.

E-mailaddress:m.vandrongelen@utwente.nl(M.vanDrongelen).

offcuts,suchastrimsandnestingscrap,arefirstcollectedandshredded intoflakes.Mostoften,theseflakesaremulti-layeredmaterialandmay consistofwovenfabric,asin[8,10]andasshowninFig.1.Flakesare thenprocessedinablendertoformamoltenmixeddoughthatis trans-ferreddirectlytoapressforcompressionmoulding.Themixingphase usedinthisstudyissimilartothemixingofLFTpellets.Thepelletsor flakesareprocessedinadevicethatmeltsthepolymeranddisperses thefibres.However,mixingofflakesdescribedhere canbe substan-tiallymoredifficultthanmixingLFTpelletsthathavebeendesignedto disperseeasily[5].

Thesubsequentprocessingstep,aftermixing,involvessqueezingthe mixeddoughintoamould.Itsflowbehaviourdependsontheamount offibres,fibrelength,theinteractionoffibrebundlesinthedough,and thusthedispersionofbundles.Onceacomponenthasbeenmoulded,its mechanicalpropertiesarealsolinkedtofibrelength,orientationandthe dispersionoffibrebundles.ApreviousstudybyDeBruijn[7]showed thatmixingthemulti-layeredTPCflakessignificantlyimprovesthe me-chanicalproperties ofthemouldedpanels, comparedtocompression

https://doi.org/10.1016/j.jcomc.2021.100108

Received16October2020;Receivedinrevisedform18December2020;Accepted15January2021

(2)

Fig.1. Schematicﬂowchartofthechosen recycling solution.Thetoprow illustrates thevariousprocessingstepswhilethe bot-tomrowshowsthematerialstatebetween eachstep.

mouldingofnon-mixedflakes.Thisimprovementismainlydrivenbythe micro-structuralchange:fromstacked,thickflakestohomogeneously entangledfibrebundles.Additionally,therecyclingsolutionpresented herepreventsfibreattritionduringmixingbyusingalow-shearmixer,in ordertolimitthereductioninmechanicalproperties,whichdependon thefibrelength.Themixingofflakes,andtheresultingmicro-structure, isthereforecrucialfortherecyclingprocess.Consequently,aproper un-derstandingofthemixingstepisrequired,aswellasagooddefinitionof thequalityofmixing(QoM)ofmixedflakes.Asuitablecharacterisation methodforQoMmustbedevelopedandimplemented.

Thepresentarticlewilldealwiththemixingstepintherecycling solutiondevelopedbyDeBruijnetal.andcharacterisetheQoMofthe mixedmaterial, withtheobjectiveof understandinghowtoimprove theQoMofmixeddoughs.Inordertoachievethis,severalaspectsof theQoMthatcanbeappliedtofibrereinforcedcompositeswillbe se-lectedfromliterature,improvedandimplemented.Finally,theeffects ofmixingsettingsandfibrelengthonQoMwillbestudied.

2. Literaturereview

Thestudyof mixingandsegregationis afieldof spatialstatistics that,overthepastdecades,hasbeenwidelyappliedtovariousfields ofscience,rangingfromecologytothematerialblending[11–13].A firstcomprehensivedefinitionofQoMwasproposedbyDanckwertsin 1952[14]byintroducingtwoaspects:theintensityofsegregation(IoS) andthescaleofsegregation(SoS).TheIoScorrespondstothe coeffi-cientofvariation(CoV)oftheconcentrationofspeciesinthesystem, alsoknownasevenness[15].Withregardtoblendingfibre compos-ites,itreferstotheCoVofthelocalfibrefraction.TheSoSrepresents acharacteristiclengthofsegregation,orclustering,inthesystem.SoS represents,forinstancethemaximumstriationthicknessformixed flu-ids,orthesizeoffibreclustersforfibrecomposites.Fig.2showsavisual representationofgoodandpoorIoSandSoS,witheachsub-figure rep-resentingacross-sectionalmicrographoffibrecomposites.Thestudies thatfollowedDanckwertsextendedhisinitialworkanddefinedother aspectsofsegregation[15].Exposurecorrespondstotherateofchange insegregation,whichhassomeimportancealongthelengthofscrew extruders,forinstance.Densityisameasureofthemassorvolumeof oneof theconstituentsperunit volumeorarea.Lastly, centralisation correspondstotheaggregationofaconstituentawayfromitscentre.

Thefirsttwoaspectsofsegregationwerecontinuouslymeasuredin variousfieldsrelatedtospatialstatistics[11–13]byadaptingthe mea-surementmethodsforIoSandSoStospecificsituations.IoSandSoSare particularlywell-suitedforthemixingofdiscontinuous-fibrecomposites andhavebeenstudiedbyvariousauthors[13,16–29].Thelatterthree aspectsofsegregation– namelyexposure,densityandcentralisation– havebeenscarcelyusedforfibrecompositesandpolymerblends.

Theproblempresentedinthispaperissimilar,inaway,toother mixingproblemsfordiscontinuous-ﬁbrecompositesorpolymerblends, andcanbesummarisedasfollows.Twoimmisciblespecies,ﬁbresand

polymer,areinitiallyhighlysegregatedintheformoflargeclustersof flakes(largemulti-layeredflakes)andpolymergranulates.During mix-ing,theflakestructuredelaminatesanddisentanglesintoloosebundles, eventuallyleadingtoamoreevendistributionoffibresinthedough. MeasuringIoSandSoSonlyrequiresasnapshotofthefinalstateofthe doughtobemeasured,whiletheexposurecanbedifficulttodetermine formixeddiscontinuous-fibrecompositesbecauseitisusuallyonly pos-sibletocapturethefinalstate.Regardingthecentralisation,Kukukova etal.[15]statethat‘itdoesnotaddinformationaboutthemixing prob-lembeyondthescaleandintensityofsegregation’.However,poor mix-ingoffibrebundlescanleadtoasituationinwhichtheyarelocallywell dispersed(goodIoSandSoS),butaccumulateinsomeareas.

Thecharacterisationof theQoMofﬁbre-reinforcedplastics is di-verse.Mostresearchisbasedonimageanalysisofcross-sectional mi-croscopy. Severalmethodshavebeen implementedandcan be cate-gorisedasfollows.

Analysesbasedonquadrats

Several authors worked with cross-sectional microscopy of fibre composites dividedintosquaregridsof varioussizes.Thefibre frac-tion orthepresence of fibrein eachcellis measuredandrecorded. Fig.2(a)to2(d)representquadratsfilledwithfibresatvariousdegrees ofevennessandclustering.Inoueetal.[16]andSpowartetal.[17] mea-suredQoMbydeterminingabox-countingdimensionofthepresenceof fibresinthecells.ThisisindirectlyrelatedtotheSoSofthefibre clus-ters.Lietal.[18]alsoworkedoncross-sectionalmicrographsof fibre-reinforcedconcreteandmeasuredtheCoVofthelocalfibrefractionin thecells.UsingtheCoVoflocalfibrefractionisfoundtoleadtothe closestestimationoftheIoSforsuchasystem,sufficiently characteris-ingthedispersionoffibresinthesystem.NormalisingittotheCoVof thenon-mixedmaterialallowsforthemeasurementoftheimprovement ofevennessduetomixing.ThisisabettermeasurethanjusttheCoV forcomparingmixersorforidentifyinghowvariousmaterialsbehave inamixer.Besides,thevariationinCoVpercellsizehasscarcelybeen studied,andmayleadtonovelresults.

Analysesbasedonmicrostructuralfeatures

Le Baillif et al. [19] compounded cellulose fibre reinforced polypropyleneusingatwinscrewextruderandcharacterisedthe dis-persionoffibresusingcross-sectionalmicroscopy.Theymeasuredthe distributionofthesizeofthefibreclusters,whichisanestimationof theSoS.Yazdanbakhshetal.[13]formulatedanewdefinitionof par-ticledispersion,dispersivework,astheminimumamountof workper domaintomoveallparticlestoauniformstateofequallydistant par-ticles.Itisthereforebasedonthedistributionoftheparticlelocations inthedomain.Wangetal.[20]mixedcarbonfibrereinforcedcement andmeasuredthedispersionoffibreswiththeaidof3Dtomographic images.AdispersioncoefficientwasdefinedastheCoVofthesizeof fibreclusters.

Variousresearchersdeterminedﬁbredispersionbyeither consider-ingthesizeofagglomeratesortheirCoV.Theformerseemstobeagood measureoftheSoS,whereasthelattercharacterisesthepoly-dispersity

(3)

Fig.2.Schematiccross-sectionsofﬁbrereinforcedcompositeswithvariousdegreesofevenness(IoS)andclustering(SoS).Fig.2(a)to2(d)showscross-sections overlaidwithsquarequadratsthataresubdividedinto4× 4cells.Fig.2(e)to2(h)showsthesamecross-sectionsandtheirVoronoidiagrams,generatedfromthe centreoftheﬁbres.ThedualgraphoftheVoronoidiagram,theDelaunaytessellation,isonlyshowninFig.2(e)forclarity.

offibreclustersinablendbuthardlyindicatestheSoS,oranyaspect ofQoMdefined in[15].Dispersivework,asdefinedbyYazdanbakhsh etal.,maybeagoodmeasureofeitherclustering,evennessor centrali-sationinsomesituations.Nonetheless,itisnotcapableofdistinguishing systemsmadeupoflargeclustersfromthosewithpoorcentralisation.

AnalysesbasedonDelaunaytriangulation

Severalresearchers characterised the QoM of discontinuous-fibre compositesbasedonphotographs(madebyopticalorelectron micro-scopes)usingtheDelaunaytriangulationoritsdualgraph,theVoronoi diagram[21–26].Voronoidiagrams areconstructedon thecentroids oftheparticles,fromwhichseveralstatisticalquantitiesaremeasured. Fig.2(e)to2(h)showstheVoronoidiagramsoffourschematic cross-sectionsoffibrecomposites,withvariouslevelsof IoSandSoS.The figureshighlightthedifferencesinVoronoicellsizesandcell distribu-tionsforthesefoursituations.Inmostcases,thedistributionoftheareas oftheVoronoicellswasused.Additionally,thedistributionofthemean near-neighbourdistance,whichcanbeoutputfromtheDelaunay trian-gulation(seeFig.2(c)),wascomputed.ItwasfoundthattheCoVofthe meannear-neighbourdistanceortheCoVoftheVoronoicellsizesarea powerfultooltocharacterisethehomogeneityofthesamples.However, suchameasurehardlydifferentiatesbetweentheevennessofparticle dispersion(IoS)andtheirclustering(SoS).Consideringtwosystems,one withpoorevennessandasecondhighlyclustered,onecanfindmany fibreswithbothcloseanddistantneighbours,leadingtoasimilar mea-sure(seeFig.2(g)and2(f)).Additionally,inallaforementionedstudies themeannear-neighbourdistancewasmeasuredastheEuclidean dis-tancebetweenfibrecentroidsin2Dmicroscopicimages.Discontinuous cylindricalfibrescanorientthemselvesinanydirectioninacomposite specimen,whichmeansthatacross-sectionalmicrographofsucha spec-imenshowsfibresasellipsesofvariouseccentricitiesandorientations. Theactualdistance betweenfibres,however,isthedistancebetween theircentrelines,whichdependsontheellipseparameterswhencutin a2Dplane.Therefore,theEuclideandistancebetweenfibrecentroids leadstointerpretationerrorsiftheellipses’eccentricitiesand orienta-tionsarenottakenintoaccount.

Analysesbasedonphysicalmeasurements

Severalstudiesondiscontinuous-fibrereinforcedcomposites char-acterised QoM by performing global physical measurements. Yang etal.[27]andOzyurtetal.[28]workedwithcarbon-fibre-reinforced cementandmeasuredtheelectricalresistanceoftheirspecimensto de-termine fibredispersion. Similarly,Yenjaichon etal.[29]used elec-tricalresistancetomographytodeterminetheQoMofkraftpulp sus-pensions.Althoughthesemeasurementmethodscanbenon-destructive, theystruggletodistinguishthevariousaspectsofsegregation,in partic-ularIoSandSoS.

Outcome

The previous studieshave sought totacklethe problemof char-acterising the QoM of fibre reinforced plastics with various meth-ods[13,16–29],mainlybasedonimageanalysisofmicroscopyimages. Mostofthesemethodswereinspiredbyalargecorpusonspatial statis-tics[12,14,15,30,31],butthemethodsimplementedforfibre compos-itesoftenlackthemathematicalrigourofthelatter.Especially,mostfail todistinguishtheevennessofparticledispersionfromfibreclustering. ItiscrucialtostudybothIoSandSoS.

Cross-sectionalmicroscopyappearstobearelevantmethodof char-acterisingtheQoMofdiscontinuousfibrecomposites.Thestochastic na-tureofthesematerialsrequiresaconsiderablenumberofmicroscopic photographstoprovidestatisticallysatisfactoryresults.TheCoVofthe localfibrefractionisfoundtobeapertinentmeasureofIoS,which rep-resentstheevennessofasample.However,theCoVcharacterisesthe material, notthemixingtechnology,whichis whyitshouldbe nor-malisedtotheCoVofthelocalFVFsintheinitiallynon-mixedmaterial tomeasuretheimprovementinevennessthankstothemixingstep.This measurementmethodwasintroducedbyDanckwerts[14]andlaterused forvariousmixingproblems[15].ThenormalisedCoVappearstobea bettermeasuretocomparedifferentmaterialsaftermixing,ordifferent mixingsystems,anditwillthereforebeexaminedinthisstudy.Close attentionwillalsobepaidtotheeffectofcellsizeonIoS,since,tothe knowledgeoftheauthors,ithasnotbeenreportedinrelevantliterature. NopreviouslyinvestigatedmeasureofSoSseemedadequatefor fibre-reinforcedcomposites,whichiswhyoneclosetothegeneraldefinition mentionedinKukukovaetal.[15]willbeimplemented.

(4)

Fig.3.Insideviewofafour-shaftshredder(imagecourtesyofUNTHAGmbH).

Fig.4.FLDsofthevarioustypesofﬂakesusedinthisarticle.

Themeasurementof fibreclusteringusingDelaunaytriangulation appearstobevaluablefordiscontinuous-fibrecomposites.Theexact de-terminationofthefibreclustersmightnotbetheprimeobjectivewhen characterisingQoM,butthisdoesdrivethemechanicaland rheologi-calpropertiesoftheLFTs.Asofyet,littleattentionhasbeenpaidto correctingthedistancesandareaswiththeellipses’orientationand ec-centricity.Thiscorrectionisimplementedinthisarticletohelpmeasure thesizeoffibreclustersinamorepreciseway.

3. Materialsandmethods

3.1. Material

Thescrapmaterialusedfor thisstudywascollected atthe form-ingstageofvariousmanufacturingsites.ItconsistsoflaminatedTPC offcuts,madeofTenCateCetex® TC11005HScarbon/PPS(C/PPS)at 50%fibrevolumefraction (FVF),as shownin Fig.1(1.trims).The carbonfabriciswovenwithtowsthatcontain3,000fibreseach.The offcutswereshreddedin-houseusingatwo-shaftshreddermadeby UN-THAshreddingtechnologyGmbh,Austria(seeFig.3).Theoutputflakes oftypically20×20×3mmcan beseen inFig.1(2.shredded flakes). Somebatchesofshreddedflakesweresievedwithamulti-stage vibrat-ingsieve.Thefibrelength distribution(FLD)oftheoutputflakes,as wellastheshreddingandsievingprocess,werepreviouslycharacterised in[10,32].TheFLDsofthecategoriesofflakesusedinthisstudyare showninFig.4.Sieved-16mmandSieved-8mmcorrespondtothe flakesthatfirstpassedthroughthesieveswithanapertureof22.4mm and11.2mm,respectively,beforefailingtopassthroughthesieveswith anapertureof16mmand8mm,respectively.InadditionSieved-2.8mm &4mmconsistsofablendoftheflakesthatfailedtopassthroughthe sieveswithanapertureof2.8mmand4mm.C/PPSLFTpelletswith 3mmfibres wereusedtocomparetheQoMoftherecycledmaterial withcommerciallyavailablepellets.Thesepelletsareknownas

Luvo-com®1301/XCF/30andareproducedat30%ﬁbreweightfraction,or approximately24%FVF.Thisstudyalsousedpolymergranulatesto di-lutetheﬂakesorpelletsduringthemixingstage,namelyFortron® 0214 PPSgranulatesbyCelanese.Twoothermaterialswereusedfor compar-isonpurposes:

• Shortfibrepelletsthatwereinjectionmouldedat40%fibreweight fraction(oraround32%FVF)withfibresof200𝜇𝑚inlength. • Single-layerwovenfabricprepregsmeasuring10× 10mm2_,_which

werecompressionmouldedat50%FVF.Notethatthismaterialisnot mixedinanyway.Moreinformationonthematerialandprocesses canbefoundin[33].

3.2. Low-shearmixing

Aftertheoffcutshadbeenshredded,theresultingflakesorpellets werefedintoalow-shearmixertogetherwithPPSgranulatestodilute thefibresuspensionto20%FVF.Thedevicewasdevelopedand man-ufacturedattheCentreofLightweightStructuresintheNetherlandsin 1998[34].Contrarytotypicalscrewextruders,thelow-shearmixer con-sistsofaheatedhollowcylinderwitha70mmdiameter,andfeatures threeeccentrically-locatedheatedrodsrotatinginside.Anillustration ofthisdeviceisshowninFig.5.Thethreerodsareplacedatdifferent distancesfromtheaxialcentreofthecylinder(Fig.5ontheright).The mixerisfedbymeansofahopperlocatedonthecylinder.Afterfeeding, apistonpushesthematerialinsidethecylindricalcavity,whichcloses thefeedopening.Theheatedrodsthenstarttorotatearoundtheaxial centreataspeedvaryingfrom5rotationsperminute(rpm)to15rpm, for5to20min.Duringthis phase,thepolymerstartstomelt, caus-ingtheflakestodelaminate,thewovenstructureofthepliestoloosen, andeventuallythebundlestospreadintoclustersofvarioussizes.Atthe endofthemixingphase,thepistonpushesthematerialoutofthemixer, whichisnowamolten,mixeddough.Duringtheentiremixingprocess, thematerialisnotcompressedandconsequentlyonlyfillspartofthe cavity,rangingfrom20%to50%.Therestisfilledwithairand, possi-bly,volatileby-products.Asaresult,theextrudeddoughisveryporous, andevenswellsasmallamountwhenpushedout,reachingafinal di-ameterof100mmafterejectionfromthecylinderandgate,whichboth haveadiameterof70mm.Inthismixingdevice,therotationspeeds ofthelow-shearmixeraretypically10rpmandthedistancebetween therotatingrodsandthecylindricalcavityisabout10mm.Thisresults inverylowshearrates comparedtothosefoundintypicalscrew ex-truders.Asaconsequence,fibresarelesspronetobreakingduringthe mixingphase[35].Besides,thelow-shearmixerallowsforabatch-type processratherthanthecontinuousprocesstypicalforscrewextruders, asaresultofwhichtheresidencetimeofthematerialisequalforall particles.Thelow-sheardeviceisstillinitsprototypephaseandhasnot beenstudiedextensively.Thereforethemechanismsofmixinginsuch adevicearestillpoorlyknown.

Afractionalfactorialdesignofexperiments(DoE)ofresolutionIV wasperformedbyvaryingthemixingspeed,mixingtime,cavity ﬁll-ingratio andmixingtemperature.Twolevels werechosen forthese fourfactorsinadditiontoacentralpoint.Thedoughsandtheir mix-ingsettingsarelistedinTable1.Theresolution(IV)oftheDoEmakes it possibletoidentify thesigniﬁcance andstrength of themain fac-tors.However,thetwo-factorinteractionsforsucharesolutionare con-founded withother two-factorinteractions sothey willnot be anal-ysedin ordertoprevent interpretationerrors. Thereproducibilityof themixingprocesswasinvestigatedbyproducinganextradoughfor threeoftheDoEsettings(seetests#14,#15,#16inTable1).Some doughswerealsoproducedwithvariousFLDs– rows#10 to#13in Table1.

3.3. Microscopy

IntherecyclingprocessdevelopedbyDeBruijnetal.[7],thedough isdirectlytransferredfromthemixertoapressforcompression

(5)

mould-Fig.5. Aschematicofthelow-shearmixingdevice isshownontheleft.Ithighlightsseveralkey ele-mentsthatareexplainedinSection3.2.The cross-sectionA-Aisdrawnontheright-handsideofthe ﬁgure.

Table1

Listofthevariousdoughsmanufacturedandtested.

inginamatterofseconds.Forthisstudy,thestateofthematerialright aftermixingwasstudiedusingcross-sectionalmicroscopy.However, di-rectlyaftermixing,thedoughsaretooporoustobepreparedfor mi-croscopy.Therefore,thedoughswerehalf-pressedtoathickness vary-ingbetween15mmand20mm.Itwasfoundthatthecompressionof thesedoughsoccursintwostages.Intheinitial pressingstage,most voidpocketsclose,whilethedoughdoesnotflowoutwards.Duringthe secondstage,thedoughstartstoflowandtheremainingvoidpockets close.Adoughthicknessofabout15–20mmwasfoundtobetheend ofthefirstpressingstage.Note,inaseparatestudyitwasfoundthat, aftercompressionmouldingofadough,theporosityfractionisnullor negligible,exceptinspecificlocationssuchaslargethicknessincreases intheflowdirection.[35]

Itisassumedthatthestructureandarrangementof thefibresare identicalintheextrudeddoughsandthehalf-presseddoughs,except fortheout-of-planeorientationoffibres.Especially,itisassumedhere thatthelocalrelativevariationsinfibrevolumefractionandtheclusters offibresremainthesameasinthenon-presseddoughs.

Thestochasticnatureofthedoughmakesquantitativeanalysesof itsmicrostructurebymicrographydiﬃcult.However,anextensive mi-croscopystudycan providesuﬃcient datatoobtainreliable statisti-calmeasurements.Inthepresentstudy,thefollowingprocedure was adoptedtogatherdataformeasurementsoftheQoM.First,each

half-presseddoughwascuttoproducesectionsofroughly10mmwidth(see Fig.6).Sixsectionswerecutperdough– twofromthefront,themiddle andtherearareas– andwerethenembeddedandpolished.Twolarge cross-sectionalmicrographsweretakenforeachsectionata×700 mag-niﬁcationandwereautomaticallystitchedto20,000×20,000pixels each,representinganareaof6mm×6mm.Hence,twelvelarge micro-scopicimageswereobtainedperdough,representingananalysedarea ofroughly430mm2_._Parts_of_such_micrographs_are_shown_in_Fig.₇_(a)

and7(e).Microscopicimagesofthenon-mixedmaterial,i.e.theflakes andLFTpellets,werealsocapturedforthecalculationoftheIoS(see fol-lowingsection).Microscopicphotographsoftheothertwomaterials,the injectionmouldedshortfibrecompositesandthecompressionmoulded flakes(seeSection3.1),werealsomadetocomparetheirQoMtothose ofthedoughs.Sampleswerewatercutusingadiamondcoatedsawand driedat90◦Cfor12h.beforetheywereembeddedusinganepoxyresin (EpoFix,Struers)andpolishedusingasequenceof600-1200-2000-4000 sandingpapersandfinalyasilicaoxidesuspension.AKeyenceVHX dig-italmicroscopewasusedinalloccasions,stitchingwasperformedusing thebuilt-inautomaticstitchingfunction.

Imagesegmentationwasexecutedautomaticallyoneachpicture us-ingMatlabsoftware,usingthepeaklocationsofeachimagehistogram todeterminetheﬁbre,matrixandporosityfraction.Detailsonthis op-erationcanbefoundinAppendixAppendixB.

(6)

Fig.6. Photographsof adoughaftermixingand af-terbeinghalf-pressed.Thelocationsofthemicroscopy specimensinthedougharehighlightedinwhiteonthe half-presseddough.

Fig.7.VisualisationoftheimageprocessinganalysisthatcomputesthelocalFVFs.Theexampleislimitedtotwosectionsof2.4mm×2.4mmtakenfromthelarge microscopicimages.Fig.7(b)to7(d)are‘blurred’representationsofthemicrographin7(a),i.e.theyrepresentthedistributionofFVFs,forthreediﬀerentcellsizes, whereeachcelldisplaysitslocalFVFingrey.Thegrayscalerepresentstheﬁbrevolumefraction,rangingfrom0%inwhiteto100%inblack.Similarly,Fig.7(f)to

7(h)representtheFVFinthemicroscopicimagein7(e)forvariouscellsizes.

Atﬁrst,alarge-scalemeasurementoftheevennessofthedoughswas determined.Foreachdough,theFVFsofthetwelvemicroscopicimages wereﬁrstcalculatedandthengroupedbytheirlocationinthedough: front,middleorrear(seeFig.6).Thisresultedin anestimateofthe globalvariationsinFVF,whichwillbeanalysedinSection4.

3.3.1. Intensityandscaleofsegregation

The segmented cross-sectional micrographs are divided into a grid of various cell sizes (from 18.75 𝜇𝑚 to 600 𝜇𝑚). The FVF of each celliscomputed, excludingthefraction of porosity,i.e.FVF= Area_{𝑓𝑖𝑏𝑟𝑒𝑠}∕(Area_{𝑡𝑜𝑡𝑎𝑙}_𝑐𝑒𝑙_𝑙−Area_{𝑝𝑜𝑟𝑜𝑠𝑖𝑡𝑦}).Note thatiftheporosity fraction inany cellis largerthan30%,this cellisdiscardedfromthe analy-sis.TheCoVofthelocalFVFsisrecordedperimageforeachgridsize. Thesamecalculationwasperformedfortheinitiallynon-mixed materi-als,i.e.blendsofflakesandpolymerorpelletsandpolymer.TheIoSis definedastheratiooftheCoVbetweentwomicroscopicpictures:the half-presseddoughsovertheinitiallynon-mixedflakes-polymer com-bination(seeFig.8foranillustrationofthemethod).Fig.7(a)and7(e) displaytwomicrographswithacleardifferenceinQoM.Thetoppicture presentslarge clustersoffibres withmoderateevenness.Thebottom pictureshowstheopposite:smallclustersandconsiderableevenness.

Fig.7(b)to7(f)and7(h)showsthegridsofFVFsforvariouscellsizes (75𝜇𝑚,150𝜇𝑚and300𝜇𝑚)wherethegreyscalerepresentstheFVF, rangingfrom0%inwhiteto100%inblack.Itisclearfromthegrids thattheCoVofthelocalFVFswillgivediﬀerentresultsforboth micro-graphs,thusresultingindiﬀerentdegreesofIoS.

ThecalculationofSoSstartswiththegridoflocalfibrefractionwith cellsof18.75𝜇𝑚,computedfromthemicroscopyimages.Image segmen-tationisperformedonthegridintotwophaseswiththeFVFrespectively belowandabovethemedianFVF(seeFig.9).Consideringthenewly seg-mentedimageasamatrix,themaximumstriationthicknessofthehigh FVFclusters,i.e.themaximumheightofthewhiteregionsinFig.9(e), is measuredforeachcolumn.Theaverageofthemaximumstriation thicknessesofallcolumnsiscalculated,whichgivesthemeasureofthe SoS.Fig.9showsthegridofFVFsfromthetwomicrographsinFig.7(a) and7(e),aswellasthesegmentationbyFVF.Fig.9(b)hasmuchlarger clustersthanFig.9(d),whicharedetectedbythemeasureoftheSoS, 400𝜇𝑚and175𝜇𝑚respectively.Thismethodissensitivetothe orien-tationofthemicrographsinthecross-sectionalspecimen.However,all micrographsweretakenwiththesameorientationtopreventthisissue. Inaddition,anartificialrandomdistributionoffibreswascreated torepresenttheQoMofanalmostperfectlyevendough.Forthat,a2D

(7)

Fig.8. RepresentationofthecalculationoftheIoSofmixeddoughs.Inthecross-sectionalmicrographs,theﬁbresareinwhite,thepolymeringreyandporosities areinblack.ThesameprocedureisappliedforthedoughmadeofLFTpellets,forwhichtheCoVbeforemixingiscalculatedfrommicroscopicimagesoftheinitial LFTpellets.

Fig.9. Fig.9(a)and9(c)showthelocalFVFsofthemicroscopicimagesofFig.7.Fig.9(b)and9(d)arethesegmentedimagesof9(a)and9(c)respectively,where theareasofhigherﬁbrefractionareshowninwhite.TheSoSistheaverageofestriationthicknessesofthewhiteregions.Thebottomﬁgure,(e),illustratesthe calculationofSoSfromthesegmentedimages.

spacewasrandomlyfilledwithnon-overlappingellipsesat20% cover-age,fromwhichtheIoSandSoSweremeasured.Thisrandomfill rep-resentsamixingsituationthatisdifficulttoachieveinreality,butitis stillfarmoresegregatedthanaperfectlyuniformdistributionoffibres. Thisartificialrandomdistributionhelpstovisualisethetheoreticallimit ofevennessandclusteringachievableformixedC/PPSdough.

3.3.2. Bundlesizedistribution

Anadditionalmethodwasimplementedtodeterminethesizeofthe fibreclustersineachimage,whichcanrangefromasingleloosefibre to3000clusteredfibres.

Imagesofthesegmentedﬁbreswereusedforthispurpose.Inthe segmentedmicrographspresentedhere,adjacentﬁbresmayappearto

(8)

Fig.10. Visualisationoftheintermediatesteps ofthealgorithmimplementedtodeterminethe BSDofcross-sectionalmicrographs.Fig.10(a) shows the input cross-sectional micrograph while10(b)displaysthesegmentedﬁbresafter theﬁbresweredisjointedusingawatershed.

Fig.10(c)and10(d)showtheDelaunay trian-gulationbeforeandafterthereﬁnement tech-nique(Section3.3.2).Thereﬁnedtriangulation wasusedtocomputetheBSDineach micro-graph.

bejointbecauseofthesegmentationstep.However,anaccurate descrip-tionoftheclustersoffibresrequiresfullydisjointfibres.Awatershed algorithmwasappliedtothesegmentedimagestoovercomethisissue. Fig.10(b)showsthesegmentedimageafterapplyingawatershed anal-ysis.Mostfibresaredisjointedfromoneanother.However,flatellipses, whichcorrespondtofibresparalleltothemicroscopicplane,maybecut bythewatershedalgorithm,therebyoutputtingmultipledomainsofa singlefibre.Thesenumericalartefactsseemedlimitedcomparedtothe benefitsofapplyingawatershedalgorithm.

ADelaunaytriangulationwasthenperformedonallfibrecentreson eachimage(Fig.10(c)).Arefinementtechniquewasneededtocluster thefibresthatareclosetoeachother.Fig.10(d)illustratestheresult oftherefinementtechnique.Theclustersoffibresareclearlydisjointed fromoneanother,althoughsomefibresarewronglyattributedto an-othercluster.Thisissueartificiallyincreasesorreducesthenumberof fibrespercluster,buttheoveralleffectwasfoundtobelimited.The numberoffibresperclusteriscalculatedfromthelistofconnected com-ponentsintherefinedDelaunaygraph,whichisthenconvertedtothe bundlesizedistribution(BSD).

Inthe aforementioned refinementtechnique, all segmentslonger thanthreefibrediameters,correspondingtoaregularhexagonal con-figurationat10%FVF,arediscarded.Acorrectedfibre-to-fibredistance wascalculatedtotakeintoaccountellipseeccentricitiesand orienta-tions.A detaileddescriptionof thecalculationscan be foundin Ap-pendixAppendixC.

Toprovideadditionaloverviewonthevariousstepsperformedto obtaintheBSD, IoSandSoS (previoussection)from themicroscopy images,aschematicﬂowchartisincludedinAppendixAppendixD.

4. Results

The half-presseddoughs listedin Table 1 wereprepared for mi-croscopyandanalysedaccordingtothemethodspresentedinSection3. Initially,theDoEwasanalysedtodeterminetheeﬀectofmixing set-tingson QoM.Theresults ofthereproducibilitystudyarepresented next.Then,theeﬀectsofthevariousFLDsonQoMareanalysedand discussed.

4.1. Designofexperiments

Fig.11(a)showstheIoSforallcellsizes,ofthedistincttwelve mi-crographsmadefromdough#5.TheIoSisalwayssmallerthanone, whichindicatesamixedstatethatis,asexpected,lesssegregatedthan theinitialnon-mixedmaterial.Besides,theIoSoftherandomlyfilled 2Dspaceisplottedunderrandomfillwithadottedline(Fig.11).Itis mostprobablethattheIoSofalldoughsareboundedbyoneandtheIoS ofrandomfill.Ahatchedregionwasaddedtothefiguretorepresenta domainoftheIoSthatcannotbereachedforthedoughsinthisstudy. Onemaynotethelargescatterbetweenthesetwelvelines,whichiswhy agreyboundingregiondepictingthemaximumandminimumvaluesof

(9)

Fig.11. TheIoSofvariousdoughsanalysedfortheDoE.Fig.11(a)and11(b)showstheIoSofdough#5.In11(a),thetwelvesolidlinesaretheIoSofallmicrographs forthisdough,whilein11(b)theyarerepresentedasameanvalueandagreybandboundingtheextrema.Fig.11(c)highlightsthemeanIoSofseveraldoughs processedwithvariousmixingsettings.Fig.11(d)showstheIoSofanotherdoughwithitsmeanvalueandextremahighlightingthevariationsofIoSinadough.

theIoSareshownaroundthemeanvaluesplottedwithalineasshown inFig.11(b).Thisgraphillustratesthatthreemeasuresareimportant:

• themeanIoS:itrepresentstheevennessofthelocalFVF,asdeﬁned inSection3.3.1.Inordertomaketheanalysissimpler,onlytheIoS foracellsizeof150𝜇𝑚isconsideredindetailfortheDoE.Thisvalue waschosenduetoitssimilaritytotheinitialbundlesize, approxi-mately150𝜇𝑚inthicknessatnormalFVF.Thissizeisassumedto bearelevantscaletodeterminetheevennessofthismaterial.Lower IoSvaluesequalabetterevennessoftheﬁbresinthesample. • ThebandwidthoftheIoS(heightofthegreyband),which

corre-spondstotheintra-samplevariabilityofevenness.Theanalysisof theDoEwasperformedonlyforthebandwidthat150𝜇𝑚,similarly tothemeanofIoS.Asmallbandwidthindicateslimitedvariability ofevennessbetweenthetwelvemicrographs.

• TheeﬀectofcellsizeontheIoSisdescribedbytheslopeoftheIoS asafunctionof cellsize.Itexpressestherateof changeof even-nesswithinthestudiedscalerange.Steepslopesindicatethatﬁbre bundlesbecomeevenlyspacedatalargescale.Gradualslopesshow thatthelocalvariationsinFVFhardlydecreasewhenincreasingthe scale.

AnanalysisofvarianceofthethreemeasuresoftheDoEwas per-formed.Forallcases,alevelofsigniﬁcanceof0.05wasselected,asitis

commonpractice,todistinguishtheeffectsthatsignificantlyinfluence themeasuresoftheQoM.

TheeﬀectsofthemixingsettingsonthemeanIoS,thebandwidth andtheslopeareillustratedinFig.11.ThemeanIoSofseveraldoughs isdisplayedinFig.11(c)wheredoughs#1and#5wereprocessedata lowmixingspeedandforashortamountoftime,whereasdoughs#8 and#7wereprocessedatahighmixingspeedandforvaryingamounts oftime.

Aﬁrstconsiderationshowedthatincreasingthemixingspeed de-creasedthebandwidth.Boththemixingspeedandmixingtimewere foundtodecreasethemeanIoS.Themixingspeed,mixingtemperature andmixingtimewereallfoundtodecreasetheslopeofthemeanIoS.

Further,Fig.11(a)showstheintra-samplevariability,whichcanbe ofasimilarmagnitudeasthevariationsbetweentheIoSofdoughsmade withvarioussettings(Fig.11(c)).Thisurgedustoreconsiderthe relia-bilityofconsideringthemeanIoSandbandwidthseparately.Therefore, theanalysisoftheDoEwascarriedoutforsixrepetitions, correspond-ingtothesixsectionsineachdough.Theresultsshowagainthatmixing speedandmixingtimeinfluencetheIoS;withamoresignificanteffect (p-value5to10timeslower)ascomparedtotheanalysisofthemean IoS.Thestrengthofthesignificantfactorsissimilarforbothstudies, withmixingspeedhaving strongerinfluencethanmixingtime.Asa conclusion,althoughintra-samplevariabilityishigh,themixingspeed andmixingtimehaveastrongandsignificanteffectontheIoS. Fur-thermore,theanalysisofvarianceshowedthattheeffectsofthefilling

(10)

Fig.12. LargescalevariationsoftheFVFfortheninedoughsanalysedintheDoE(leftblock,orderedaccordingtotheDoEinTable1),andthethreerepetitions (rightblock)forthereproducibilitystudy.Theblacklinesandthegreybandsshowthemeanandtheextremarespectively.

fractionandmixingtemperaturearenotsignificant.Asaconsequence, theireffectisnotdetailed.Theeffectofinteractionsbetweenmultiple parameterscouldnotbedeterminedwiththeDoEcarriedout(resolution IV).

Next,theSoSs ofthe doughswereanalysed.The averageSoS of theninedoughswasfoundtovarybetween260𝜇𝑚and420𝜇𝑚(see TableA.3intheAppendix),whichisasmallrange,whencomparedto theSoSoftherandomﬁll(101𝜇𝑚)andindicatesthatﬁbre-dense re-gionsareof comparablesize.However,theintra-sample variationof SoSvariesconsiderably,from150𝜇𝑚to430𝜇𝑚.Again,mixingtime andmixingspeedwerefoundtoreducetheSoS.

Thestudyof theeffectofthemixingsettings ontheIoSandSoS showedthatboththetotalamountofshear(mixingtimeandmixing speed)andtheshearrate(mixingspeed)significantlyimproveQoMon multiplelevels:theyreducetheIoS,SoSandtheintra-samplevariability. However,thelow-shearmixerdoeshavesomelimitations.Acomparison oftheIoSandSoSobtainedforthebestmixeddoughswiththoseofthe randomfillconfirmsthattheevennessandclusteringofthedoughscan stillbeimproved,atleastfromatheoreticalpointofview.

Studyingthemicroscopic imagesalsorevealedthatfibres arefar fromevenlydistributedoraggregatedinsmallclusters,evenafterlong mixingtimesandhighmixingspeeds,althoughitisalsoclearthatlonger mixingtimeandhigherspeedimproveevennessandclustering.This phenomenonishighlightedinFig.12whichalsoshowsthevariations ofFVFatalargescale,wheretheFVFofeachmicrographiscalculated andgroupedaccordingtoitsdoughanditslocationinthisdough.The solidlineandthegreyband representthemeanandextremaofthe FVFsforeachlocationrespectively.Thedoughsprocessedat5rpmfor 10minand20min– #1&#5,and#2&#6respectively– allshow verylargevariationsinFVF,whereas FVFvariationsaremuchlower forthedoughsprocessedat15rpm(#3,#4,#7and#8).Thisseems toindicatethatmixingspeedhasamorepronouncedeffectthan mix-ingtime.Theresultsoftheanalysisofvariancealsoconfirmedthisfor theIoSandtheSoS.Additionally,themicrographsshowthata combi-nationoflargeandsmallfibreclustersarefoundindoughsmixedat 5rpm.Fewerlargeclustersarefoundindoughprocessedat15rpm. ThisresultisvisualisedinFig.13(b),whichplotstheBSDofseveral doughs.ThisfigureistheoutputoftherefinedDelaunaytriangulation andshowsthevolumefractionofbundlesofeverygivensize.The frac-tionsoflargebundles(⩾ 1,000fibres)andsmallbundles(⩽ 100fibres)

indicatewhethertheinitialtowsof3,000ﬁbreswerewelldispersed. Theﬁgureshowsthatdoughs#1and#6(5rpm)hadverysimilarBSDs withmorethan30%oftheirvolumeconsistingoflargebundles,while theBSDof#8(20min,15rpm)resultsinamuchlowerfractionoflarge bundles.

In conclusion, the analysis of variance shows that mixing time andspeedimproveallmeasuresofQoM,withmixingspeedhavinga strongereﬀectthanmixingtime.Thisobservationwasconﬁrmedby analysingtheBSD,thelarge-scalevariationsofFVFandthemicrographs ofdoughs.

4.2. Reproducibility

Thepreviousresultsshowedthatintra-samplevariabilityisofgreat importance,andthesamemightapplytointer-samplevariability.Three doughsmadewithvariousmixingsettings(#1,#6and#8)were repli-cated(#15,#16and#14,respectively)toinvestigatethe reproducibil-ityofthemixingprocess.ThemeanIoSofthesixdoughsisplottedin Fig.13(a).Theirlarge-scalevariationsinFVFaredisplayedinFig.12. Theﬁndings fromtheprevious sectionindicatethatdough#1hasa moderateQoMforallmeasures.Dough#6showedsomeimprovement duetoalongermixingtime.Dough#8demonstratesthatitisamong thedoughswiththehighestQoMinthestudyduetoalongmixingtime andahighmixingspeed.

Limitedinter-samplevariabilitywouldleadtoacleargroupingofIoS bymixingsettings:#1&#15,#6&#16,and#8&#14.However,the measurementsinFig.13(a)shownon-groupablelinesforalldoughs.The situationisdifferent,however,whenlookingatthelarge-scalevariation ofFVF(Fig.12).Clearly,doughs#1and#15haveveryhighvariations ofFVF.Doughs#6&#16showreducedvariability,whereas#8&#14 exhibitverylimitedvariationsofFVF.Additionally,theBSDwasalso analysedforthesesixdoughs(Fig.13(b)).Twoseparategroupscan eas-ilybeidentified.TheBSDsofdoughs#8&#14areclearlydistinctfrom theotherfour:theyhaveahigherfractionofsmallbundles(smallerthan sixteenfibres)andasmallfractionoflargebundles(largerthan1000 fibres).Theotherfourdoughscannot,however,begroupedbymixing settings,aswasthecaseforthemeanIoS.Thefractionoflargebundles inthesedoughsisnonethelessconsiderable,rangingfrom20%to40%. Dough#16isparticularlystriking,asithasthesecondworstIoSand

(11)

(a)

(b)

Fig.13. TheIoSandtheBSDofthedoughsanalysedforthereproducibilitystudy.

(a)

(b)

(c)

Fig.14. (a)TheIoS,(b)theBSDand(c)thelargescalevariationsofFVFofthedoughsprocessedwithvariousFLDs.

thelargestfractionoflargebundles,butasmalllarge-scalevariationof FVF(Fig.12).Nevertheless,itsSoSandtheslopeofthemeanIoSare average(seeTableA.3intheAppendixsection).

4.3. EﬀectofFLDs

Thelastsetofexperiments concernstheeﬀectof theFLDonthe QoM.SievedﬂakesandLFTpelletswereusedasdescribedinSection3. Fig.14aggregatestheIoS,theBSDandthevariationsofFVFforthefour

doughs#10,#11,#12and#13.Table2summarisesthemeasuresof theQoMforthefourdoughs.Italsoshowsthestandarddeviationofthe localFVFforthetwocomparisonmaterialsandtherandomlyfilled2D space.Asareminder,theIoSisdefinedastheimprovementinevenness ofthelocalFVFduetomixing.Theresultsshowthattheevennessof LFTpelletsimproveslessthanfortheotherthreedoughs.However,its standarddeviationofFVFislowerthanforthedoughs,showingamore evendistributionoffibres.ThestandarddeviationsinTable2also high-lights thedifferencesbetween thecompressionmouldedflakes

(12)

(non-Table2

SummaryofthemeasuresoftheQoMforthedoughsprocessedwithvariousFLDs.ThestandarddeviationsoflocalFVFarealsolistedfortwocomparisonmaterials andtherandomlyﬁlledspace.

dough material mean SoS at

18.75 𝜇𝑚 [ 𝜇𝑚 ] mean IoS at 150 𝜇𝑚 [ ×10 −1 ]

bandwidth at 150 𝜇𝑚 [ ×10 −1 ]

slope of the IoS [-] standard deviation of the FVFs at 150 𝜇𝑚 [ ×10 −1 ] #10 LFT pellets 177 2.0 0.64 − 0.67 0.30 #11 sieved 16 mm 310 1.6 0.54 − 0.53 0.79 #12 sieved 8 mm 224 0.94 0.35 − 0.72 0.52 #13 sieved 2.8 & 4 mm 207 0.67 0.13 − 0.80 0.42

compression moulded non-mixed ﬂakes 1.49

injection moulded short ﬁbre composites 0.35

artiﬁcial random ﬁll 0.17

mixed), the moulded LFT and short fibre pellets (see Section 3.1), andthedoughsmadefrom sievedflakes: theevennessof the small-est flakes is almost as good as the moulded short and long fibre composites.

TheeffectofflakesizeandFLDonQoMisevidentforall measure-ments.ShorterfibresreduceIoSandSoS,andleadtoasteeperslopeof theIoS.Theintra-samplevariationsforIoSandSoSsteadilydecrease withthereductioninFLD(seeTable2).Additionally,inthedough pro-cessedwiththesmallestflakes,only2%ofbundleswerelargerthan 1,000fibres,whichis farlowerthanthe20%-40%rangeobtained forlargeflakes(Figs.14(b)and13(b)).However,thereisstillalarge differencebetweentheBSDsofthedoughmadefromLFTpelletsand fromthesmallestflakes,butthiswasexpectedduetothecompletely differentnatureoftheinputmaterial.Inaddition,thelarge-scale vari-ations of FVF mainlyreduce from large tomedium-size flakes. The doughs#10,#12 and#13all show similarlarge-scale variationsof FVF,forwhichnoimprovementisnoticeablewithareductioninfibre length.Thismaybealimitfortheprototypemixerusedinthisstudy (Fig.14(c)).

5. Discussion

Whenconsidering the wholerecycling route studiedin this arti-cle, theQoMobtained during themixingphase affectsthe flow be-haviourofthematerialinthemouldanditsmechanicalproperties after-wards.Theindustrialapplicationofthisrecyclingtechnologyrequires thatmouldedcomponentshaveconsistentmechanicalproperties,and hencelowintra-partvariabilityandgoodreproducibility.Additionally, theflowbehaviourofthedoughispartiallydeterminedbythenumber ofbundle–bundleinteractions,whichtranslatestothebundleaspect ra-tioamongothers[36,37].Thehomogeneityandreproducibilityofthe flowbehaviourofthedoughsarealsoimportantfactors,whichcanbe improvedbyreducingintra-andinter-samplescatter.Theconditions necessarytoaddresstheseissuescanbederivedfromdifferentaspects oftheQoM.ThisleadstoseveralmeasuresofQoMthataretobeeither minimisedorbeoptimised.

• LimitedlargescalevariationsofFVFarerequired.Figs.12and14(c) demonstratethatvariationsofjustafewpercentcanbeachieved. • TheIoSandtheintra-samplevariabilityofIoSareotherproperties

tobe minimised.Alocally evendistributionofFVFhelpsreduce variabilitybothintermsofrheologyandpartperformance. • AhighreproducibilityofQoM,allaspectscombined,isalsoakeyto

thesuccessfulapplicationofsuchatechnology.

• BSDhasamajoreffectonbothflowbehaviourandmechanical per-formance.Onone hand,aconsiderablefraction ofsmall bundles drasticallyincreasesthenumberoffibre–fibreinteractions,which increasestheapparentviscosity[36],wheretheupperlimitissetby theflowdistanceinthemouldandthesizeoftheintricatefeatures tofill.Ontheotherhand,a substantialfraction oflarge bundles reducestheaveragebundleaspectratio,whichtends tolimitthe theoreticalmechanicalpropertiesofthemouldedcomponent[38].

Nonetheless,itisratherdifficulttodrawaconclusiononthe opti-mumBSDwithoutaproperstudyofitseffectontheflowbehaviour ofdoughsaswellasonthemechanicalpropertiesofmouldedparts. Regardingthelistofcriteriaabove,wellmixeddoughs(long mix-ingtime,highmixingspeed)doshowimprovementofIoS,intra-sample variabilityofIoS,andlarge-scalevariationsofFVF.Itwas,however, foundthatthereproducibilityofdoughsismoderateforallmeasured aspectsofQoM,althoughonlytwodoughswereanalysedpersetting. Thislackofreproducibilitycanalsobecausedbythelimitationsin sam-plesize:twelvemicroscopicimagesof6× 6mm2_per_dough._In_terms

oftheprocesswindow,theoxidationofthepolymerismoreseverefor longmixingtimesinthepresenceofoxygen,whichshouldbeprevented inordertolimitthedegradationofthepolymer.Theresultsalsopointed outthatincreasingthemixingspeedisamoreeffectivewaytoimprove QoMthan increasingmixingtime.Thus,theauthorsrecommend in-creasingthemixingspeedtolimitpolymerdegradation,althoughhigh shearratesmaybreakfibresduringmixing.Thishasnotbeenaproblem yetwiththecurrentmixingprototype,whichhastechnicallimitations thatpreventitsuseathigherrpm.Inaddition,itwasfoundthattheSoS ofallthedoughswereclose,andofthesameorderofmagnitude;in com-parison tootherfinely-clusteredmaterials, i.e.theinjection-moulded short-fibrecompositesandtheartificialrandomdistributionoffibres. Onthecontrary,theBSDvarieddrasticallywithrespecttomixing set-tingsandFLDs.Hence,itissupposedthatBSDandSoSdonotexactly characterisethesameaspectsofQoM,ordonotcharacterisethemwith thesamestrength.BSDisanimportantmaterialproperty,asexplained inthepreviousbulletlist.Thismayindicatethatthemethodchosento measureSoSisnotthemostsuitableforthistypeofmaterialormixing technology.

ConcerningtheeffectofFLDontheQoMofdoughs,large improve-mentsarenoticedbetweendoughs#11/sieved16mmand#12/sieved8 mmforallmeasuresofQoM.Theseimprovementsseemtosaturate be-tween#12/sieved8mmand#13/sieved2.8&4mm;andthelarge-scale variationsofFVF,inparticular,arecomparable(Fig.14(c)).Therefore, theauthorsrecommend using smallflakesinthis process,especially withameanfibrelengthsmallerthan15mm.

6. Conclusions

Anexperimental studywas carriedouttounderstandhowto im-provetheQoMofmixed doughs,madeoflong-fibre-reinforced poly-mers.ContrarytootherLFTprocessesandmaterials,theflake mate-rial andthemixer used inthis studyareunconventional.The mate-rialconsistedofmulti-layered,woven-fibrereinforcedcompositesplus polymergranulates.Themixerhadaverylowrotationspeedto pre-vent fibrebreakage,thereby limitingblendingefficiency. Several as-pectsofQoMwereselectedfromliterature,namelyevennessand clus-tering,andimplementedforthisstudy.Theexperimentalmethods in-volvedimageanalysisofalargesetofcross-sectionalmicrographs.The analysis wasbased on quadratstomeasureIoS andSoS,andon re-finedDelaunaytriangulationtomeasureBSD.Choicesweremadeto

(13)

measurethefollowingaspectsfromtheextensivesetofcross-sectional micrographs:

• IoS,orevennessofthelocalFVFs.

• SoS,representingameasureoftheclusteringofﬁbres. • Intra-samplevariationsofIoS,SoSandFVFs.

• BSD,describingthesizedistributionofﬁbreclusters.

Themethodsdevelopedandimplementedinthispaperwerefound adequateforthematerialinthisarticle,aswellastocharacterisethe QoMof otherLFTs.Therefinementtechniqueperformedon the De-launaytriangulation is definitely beneficial, as itquantifies a mate-rialpropertyofLFTs:bundlesizes.Thispropertycanalsobeusedfor themicro-mechanicalmodellingof LFTs.TheIoSis agood measure tocompareothermixingdevicestotheoneusedinthisstudy,orto comparethedoughswithothermaterials.Furtherdevelopmentsofthis recyclingtechnologyorofLFTmaterialshouldconsiderthe measure-mentoftheIoSasabenchmark,similartothecurrentpracticefor poly-merblends[39].Finally,itisnotsurewhetherthecurrent implemen-tationof SoSmeasurementprovidesinformationbeyondtheIoSand BSD.

Theeffectsofmixingtime,mixingspeed,temperatureandfilling ra-tioonQoMwereanalysedbyperformingafractionalfactorialDoE.It wasfoundthatthemixingtimeandmixingspeedarethemaindrivers toimproveQoMforallmeasures.ThereproducibilityoftheQoMin doughswasinspectedandwasfoundtobemarginal.Ontheotherhand, theeffectoftheFLDsonQoMwasclearforallmeasures:smallerflakes significantlyimprovetheQoM.Therecommendationsforbetter mix-ingforthisparticularmixingsetuparefirsttoshredscraptoasmaller size,yieldingameanfibrelengthunder15mm.Inaddition,further im-provementscanbemadetothemixingmachinebyblendingathigher shearrates.Thiscanbeachievedbymodifyingthemixerdesign,but specialcareshouldbetakentominimisefibrebreakage.Itisstill un-clearwhetherthebestQoMobtainedinthisstudyissufficient.Further workisrequiredtounderstandtheeffectofQoMontheflowbehaviour of doughs and, subsequently, on mechanical properties of moulded parts.

DeclarationofCompetingInterests

Theauthorsdeclarethattheyhavenoknowncompetingﬁnancial interestsorpersonalrelationshipsthatcouldhaveappearedtoinﬂuence theworkreportedinthispaper.

Acknowledgments

ThisprojectwasﬁnancedbytheDutchOrganisationofApplied Re-searchSIA,throughtheprojectgrantSIA-RAAK2014-01-72PRO.The authorsaregratefultotheprojectpartners:TorayAdvancedComposites, GKNFokker,CatoCompositeInnovations,DutchThermoplastic Compo-nentsandNidoRecyclingTechniek.TheauthorsarethankfultoSotiris KoussiosandWaqasAliforinternalreviewofthemanuscript.

AppendixA. ListoftheQoMmeasurements

AppendixB. Imagesegmentationandcorrection

Theimagesegmentationofallcross-sectionalmicrographswas au-tomaticallyperformedbythepeaklocationsoftheimagehistogram.A sectionofamicrographisshowninFig.B.15(a);itshistogramisshown inFig.B.15(b).Thehistogramcontainsagreylevelpeakfortheepoxy (mountingresinforpolishing),thePPSmatrixandtheﬁbres.The av-eragegreylevelbetweenepoxy&matrix,andbetweenmatrix&ﬁbres

ischosentosegmenttheimagesinthethreephases(seetheimage seg-mentationaxisinFig.B.15(a)).

Themagnifiedfibreatthebottom-leftcornerofFig.B.15(a) empha-sizesablackcontouraroundthefibre,beinganopticalartefactcaptured bythemicroscope.Suchadefectsystematicallylowersthemeasured FVFoftheimage.Therefore,aglobalcorrectionwasperformedforthe FVFofeachimage.Itwasassumedthatalargenumberofmicrographs wasenoughtorepresentthecorrectFVFofthesamples,20%on aver-age.TheglobalaverageFVFofthesixteendoughs×twelvemicrographs wassetto20%.TheresultsofthiscorrectionareshowninFigs.12and 14(c).

AppendixC. Centre-to-centredistancecorrection

Themicrographsstudiedinthisarticlearecross-sectionsof discon-tinuousfibrecomposites.Thecross-sectionsofcylindricalfibreswiththe micrographyplaneareellipseswithvariouseccentricitiesand orienta-tions.InSection3.3.2,theDelaunaytriangulationoftheellipsecentres isrefinedbydiscardingcentre-to-centresegmentslongerthana thresh-olddistance.Thisrefinementtechniqueisrequiredtoclusterthevarious fibrebundles.However,theEuclideancentre-to-centredistanceinthe planeofobservationdoesnotrigorouslymeasurethedistancebetween thefibres.

Considerthefollowingsituation;twoellipsesincontactalongtheir majoraxesinthemicroscopyplanehavealongerEuclidean centre-to-centredistancethantwotouchingfibreswhosecross-sectionsarecircles. (seeFig.C.16(a)).Yet,thetruedistance betweenthecentresisone fibrediameterinbothcases.Therelationwiththeobserveddistance dependsontheorientationsandeccentricitiesofthetwoellipses.Thus,a correctedcentre-to-centredistanceisrequiredtoaccuratelydistinguish thevariousclusters. Theobjectiveofthis sectionis todefinesucha correcteddistance.

Thestrategyimplementedhereis toconsiderasleeveof polymer growingradiallyaroundtwofibresuntilthesleevestouch.Whenthe fibresarecutinthemicrographyplane,thecross-sectionsofthesleeves arealsoellipticalwiththesameorientationandeccentricityastheir parentfibre(seeFig.C.16(b)to(d)).Inthemicrographyplane,a non-normalorthogonalbasiscanbeassociatedwitheachfibreanditssleeve (seeFig.C.17).Theellipticalsleevesgrowradiallyandisotropicallyin theirrespectivebases.

Atthejunctionpointofthetwosleeves(Fig.C.16(b)to(d)),the sleeveshavegrownbythesameamountintheirrespectivebases.The correctedcentre-to-centredistance,noted𝑑𝑐,correspondstothesizeof

thesesleevesofpolymerandisinallcases≤𝑑0.Thisdistanceisrelated

tothechange-of-basismatricesofeachellipse’sbasisandthedirection of𝑃⃖⃖⃖⃖⃖⃗𝑃′_,_as_elaborated_{mathematically}_in_the_following_paragraphs.

FigureC.17illustratesthecross-sectionoftwofibres𝐹𝑃 and𝐹𝑃′,in theplaneofobservation.Thefibresareorienteddifferentlyfromeach other,leadingtodifferentellipseorientationsandeccentricitiesinthe cross-sectionalplane,whosecentresarenoted𝑃and𝑃′_._Each_ellipse_is

determinedbyitsmajoraxis,minoraxisandorientationinthereference orthonormalbasisofthecross-sectionalplane,0:(a,b,𝜃)and(a’,b’,𝜃’)

respectivelyfor𝐹_𝑃 and𝐹_𝑃′.Forallﬁbresanalysedinthispaper,𝑏′= 𝑏=ﬁbrediameter.Let⃖⃖⃗Ωbetheunitvectorin0collinearto⃖⃖⃖⃖⃖⃖⃖⃗𝑃𝑃′.

Additionally,consideranorthogonalbasisforeachellipse,and′_,

whoseunitvectorscorrespondtotheaxesoftheellipses.The change-of-basismatrixfromto0is:

𝐓= (𝑎 𝑏cos(𝜃) 𝑎𝑏sin(𝜃) −sin(𝜃) cos(𝜃) ) (1) Therefore,𝐓−1⃖⃖⃗_Ω_is_the_unit_vector_Ω⃖⃖⃗_expressed_in__,_and_{||⃖⃖⃖⃖⃖⃖⃖⃗}_𝑃_𝑃′_{|| ×}

||𝐓−1_Ω⃖⃖⃗_{|| represents}_the_length_{||⃖⃖⃖⃖⃖⃖⃖⃗}_𝑃_𝑃′_{|| expressed}_in_basis__._Similarly,

||⃖⃖⃖⃖⃖⃖⃖⃗𝑃𝑃′_{|| × ||𝐓}′−1_Ω⃖⃖⃗_{|| is}_computed_for_the_basis_′_.

Asstatedabove,theellipticalsleevesgrowradiallyandisotropically intheirrespectivebases.Ifoneoftheellipseswereacircle,thesleeve

(14)

TableA1

SummaryofthemeasurementresultsoftheQoMforthesamplesstudiedinthisarticle.

(a)

(b)

Fig.B1. Sectionofacross-sectionalmicrographontheleft,showingthatfibres,matrixandfilledporositieshaveadifferentgreylevel.Aclose-upofafibreshows itsspuriousblackenvelopeThethreegreylevelsarehighlightedinthehistogramoftheimageontheright.Automaticsegmentationisperformedbasedonthepeak locationsofthehistogram.

wouldgrowproportionallyto||⃖⃖⃗Ω|| (with||⃖⃖⃗Ω|| =1).However,forthe ellipsecentredin𝑃,theradiusalongΩ⃖⃖⃗grows||𝐓−1_Ω⃖⃖⃗_||−1_faster_(this

expressionis≥1).Thustheaveragegrowthfactorsofthetwosleeves’ radiiareproportionalto:

𝑟=||𝐓

−1_Ω⃖⃖⃗_||−1₊_||𝐓′−1⃖⃖⃗_Ω_||−1

2 (2)

with𝑟≥1.Fromthis,itisstraightforwardthattheratio𝑑0∕𝑑𝑐istheratio

of𝑟over||⃖⃖⃗Ω||.Here𝑑0=||⃖⃖⃖⃖⃖⃖⃖⃗𝑃𝑃′||,i.e.themeasureddistancebetween𝑃

and𝑃′_in_the_orthonormal_basis_

0ofthemicrographyplane.𝑑𝑐isthe

correctedcentre-to-centredistancementionedearlier.Hencethisratio is:

𝑑0

𝑑𝑐 =

||𝐓−1_Ω⃖⃗_||−1₊_||𝐓′ −1⃖⃗_Ω_||−1

(15)

Fig.C1. Schematicsofcross-sectionsofcylindricalfibresinamicrographyplane.ThefewdrawncaseshighlightthedifferencesbetweentheEuclidean centre-to-centredistance𝑑0andthetruedistancebetweenthefibres𝑑𝑐incomparisontothefibrediameter𝑑𝑓.

Fig.C2. Illustrationoftwofibresofdifferentorientationsandeccentricities.Theircentrepointsare𝑃and𝑃′_,_and_their_related_fibre_parameters_are_used_for_the

calculationofthecorrectedcentre-to-centredistance.Notethatand′_are_orthogonal_non-normal_bases,_because_their_unit_vectors_correspond_to_the_length_of

theellipseaxes.

andtheexpressionof𝑑𝑐is:

𝑑𝑐 = 2||𝑃⃖⃖⃖⃖⃖⃗𝑃

′_||

||𝐓−1_Ω||⃖⃗−1₊_||𝐓′ −1⃖⃗_Ω||−1 (4) Notethatthisexpressionof𝑑_𝑐isalsotheharmonicmeanof||⃖⃖⃖⃖⃖⃖⃖⃗𝑃𝑃′_{|| ×}

||𝐓−1⃖⃖⃗_Ω_{|| and} _{||⃖⃖⃖⃖⃖⃖⃖⃗}_𝑃_𝑃′_{|| × ||𝐓}′−1_Ω⃖⃖⃗_||. _This _corrected _{centre-to-centre}

dis-tancewascalculatedforallsegmentsoftheDelaunaytriangulationand usedforthereﬁnement.

(16)

Fig.D1. SchematicoverviewofthevariousstepsinvolvedtowardsobtainingtheBSD,IoSandSoSfromthemicroscopyﬁgures.

References

[1] CompositesWorld, Evolution of tailored D-LFT, 2013, https://www.compositesworld.com/articles/evolution-of-tailored-d-lft , [Online; accessed 21-August-2018].

[2] M. Schemme, LFT - development status and perspectives, Reinforc. Plast. 52 (1) (2008), doi: 10.1016/S0034-3617(08)70036-5 .

[3] G. Tweed, Xycomp® DLF high-performance thermoplastic composite. www.gtweed.com , [Online; accessed 07-November-2018].

[4] T.A. Composites, Compression molded structural components. www.tencatecomposites.com/products/parts-and-services/compression-molded-parts , [Online; accessed 07-November-2018].

[5] W. Schijve, High performance at medium ﬁbre length in long glass ﬁbre polypropylene, Plast., Addit. Compd. 2 (12) (2000) 14–21, doi: 10.1016/S1464-391X(00)80121-X .

[6] F.W.J. Van Hattum, S. Van Breugel, LFT: The future of reinforced thermoplastics? Reinforc. Plast. 45 (6) (2001) 42–44, doi: 10.1016/S0034-3617(01)80208-3 . [7] T.A. De Bruijn , G.A. Vincent , F.W.J. Van Hattum , Recycling of long ﬁbre thermo-

plastic composites by low shear mixing, SAMPE Europe, 2016 .

[8] T.A. De Bruijn , G. Vincent , F.W.J. Van Hattum , Recycling C/PPS laminates into long ﬁbre thermoplastic composites by low shear mixing, in: 21st International Confer- ence in Composite Materials, 2017 .

[9] M. Janney , J. Ledger , U. Vaidya , Long ﬁber thermoplastic composites from recycled carbon ﬁber, 44th ISTC, 2012 .

[10] G. Vincent , T. de Bruijn , S. Wijskamp , M. Adbul Rasheed , R. Akkerman , Shredding and sieving thermoplastic composite scrap: method development analyses of the ﬁbre length distributions, Compos. Part B: Eng. (2019) .

[11] L. Mertes, D. Daniel, J. Melack, B. Nelson, L. Martinelli, B. Forsberg, Spatial patterns of hydrology, geomorphology, and vegetation on the ﬂoodplain of the amazon river in brazil from a remote sensing perspective, Geomorphology 13 (1995) 215–232, doi: 10.1016/0169-555X(95)00038-7 .

[12] A. Kukukova, B. Noel, S.M. Kresta, J. Aubin, Impact of sampling method and scale on the measurement of mixing and the coeﬃcient of variance, AIChE J. 54 (504) (2008), doi: 10.1002/aic .

[13] A. Yazdanbakhsh, Z. Grasley, B. Tyson, R.K. Abu Al-Rub, Dispersion quantiﬁcation of inclusions in composites, Compos. Part A: Appl. Sci. Manuf. 42 (1) (2011) 75–83, doi: 10.1016/j.compositesa.2010.10.005 .

[14] P.V. Danckwerts, The deﬁnition and measurement of some characteristics of mix- tures, Appl. Sci. Res. Sect. A 3 (4) (1952) 279–296, doi: 10.1007/BF03184936 . [15] A. Kukukova, J. Aubin, S.M. Kresta, A new deﬁnition of mixing and segregation:

Three dimensions of a key process variable, Chem. Eng. Res. Des. 87 (4) (2009) 633–647, doi: 10.1016/j.cherd.2009.01.001 .

[16] A. Inoue, K. Morita, T. Tanaka, Y. Arao, Y. Sawada, Effect of screw design on fiber breakage and dispersion in injection-molded long glass-fiber-reinforced polypropylene, J. Compos. Mater. 49 (1) (2015) 75–84, doi: 10.1177/0021998313514872 . [17] J.E. Spowart, B. Maruyama, D.B. Miracle, Multi-scale characterization of spa-

tially heterogeneous systems: implications for discontinuously reinforced metal- matrix composite microstructures, Mater. Sci. Eng. A 307 (1-2) (2001) 51–66, doi: 10.1016/S0921-5093(00)01962-6 .

[18] M. Li, V.C. Li, Rheology, fiber dispersion, and robust properties of engineered cemen- titious composites, Mater. Struct. (2012) 405–420, doi: 10.1617/s11527-012-9909-z . [19] M. Le Baillif, K. Oksman, The effect of processing on fiber dispersion, fiber length, and thermal degradation of bleached sulfite cellulose fiber polypropylene composites, J. Thermoplast. Compos. Mater. 22 (2009) 115–133, doi: 10.1177/0892705708091608 .

(17)

[20] Z. Wang, J. Gao, T. Ai, W. Jiang, P. Zhao, Quantitative evaluation of carbon ﬁber dispersion in cement based composites, Constr. Build. Mater. 68 (2014) 26–30, doi: 10.1016/j.conbuildmat.2014.06.035 .

[21] N. Yang, J. Boselli, I. Sinclair, Simulation and quantitative assessment of homoge- neous and inhomogeneous particle distributions in particulate metal matrix composites, J. Microsc. 201 (2) (2001) 189–200, doi: 10.1046/j.1365-2818.2001.00766.x . [22] J. Summerscales, F.J. Guild, N.R.L. Pearce, P.M. Russell, Voronoi cells, frac-

tal dimensions and ﬁbre composites, J. Microsc. 201 (2) (2001) 153–162, doi: 10.1046/j.1365-2818.2001.00841.x .

[23] M. Bertram , H. Wendrock , Characterization of planar local arrangement by means of the Delaunay neighbourhood, J. Microsc. 181 (1996) 45–53 .

[24] D.J. Bray, S.G. Gilmour, F.J. Guild, T.H. Hsieh, K. Masania, A.C. Taylor, Quan- tifying nanoparticle dispersion: application of the Delaunay network for objec- tive analysis of sample micrographs, J. Mater. Sci. 46 (19) (2011) 6437–6452, doi: 10.1007/s10853-011-5615-4 .

[25] D.J. Bray, S.G. Gilmour, F.J. Guild, A.C. Taylor, The eﬀects of particle morphology on the analysis of discrete particle dispersion using Delaunay tessellation, Compos. Part A: Appl. Sci. Manuf. 54 (2013) 37–45, doi: 10.1016/j.compositesa.2013.07.003 . [26] B. Bertoncelj, K. Vojisavljevi ć, J. Rihtar š i č, G. Trefalt, M. Huski ć, E. Ž agar, B. Mali č, A Voronoi-diagram analysis of the microstructures in bulk-molding compounds and its correlation with the mechanical properties, Express Polym. Lett. 10 (6) (2016) 493–505, doi: 10.3144/expresspolymlett.2016.47 .

[27] Y. Yang, Methods study on dispersion of ﬁbers in CFRC, Cement Concr. Res. 32 (5) (2002) 747–750, doi: 10.1016/S0008-8846(01)00759-1 .

[28] N. Ozyurt, T.O. Mason, S.P. Shah, Correlation of ﬁber dispersion, rheology and mechanical performance of FRCs, Cement Concr. Compos. 29 (2) (2007) 70–79, doi: 10.1016/j.cemconcomp.2006.08.006 .

[29] W. Yenjaichon, J.R. Grace, C. Jim Lim, C.P. Bennington, Characterisation of gas mixing in water and pulp-suspension ﬂow based on electrical resistance tomography, Chem. Eng. J. 214 (2013) 285–297, doi: 10.1016/j.cej.2012.10.057 .

[30] T. Mattfeldt , Explorative statistical analysis of planar point processes, J. Microsc. 220 (2005) 131–139 .

[31] L. Pernenkil, C.L. Cooney, A review on the continuous blending of powders, Chem. Eng. Sci. 61 (2) (2006) 720–742, doi: 10.1016/j.ces.2005.06.016 .

[32] G. Vincent , T. de Bruijn , M.I. Abdul Rasheed , S. Wijskamp , R. Akkerman , Fibre length distribution of shredded thermoplastic composite scrap, in: 21th International Con- ference on Composite Materials, 2017 .

[33] M.I. Abdul Rasheed , Compression molding of chopped woven thermoplastic composite ﬂakes: a study on processing and performance, University of Twente, 2016 Ph.D. thesis .

[34] A. Beukers, F. Wiltink, J. Van Breugel, Device and method for the preparation of a mixture comprisong ﬁbre-reinforced thermoplastics plastics pellets, 2000. Patent: WO 000002718.

[35] G. Vincent, T. de Bruijn, S. Wijskamp, M. Van Drongelen, R. Akkerman, Process- and material-induced heterogeneities in recycled thermoplastic composites, J. Thermo- plast. Comp. Mater. (2020), doi: 10.1177/0892705720979347 .

[36] C. Servais, A. Luciani, J.-A.E. Månson, Fiberfiber interaction in concentrated suspen- sions: dispersed fiber bundles, J. Rheol. 43 (4) (1999), doi: 10.1122/1.551014 . [37] C. Servais, J.-A.E. Månson, S. Toll, Fiberfiber interaction in concentrated suspen-

sions: dispersed ﬁber, J. Rheol. 43 (4) (1999), doi: 10.1122/1.551014 .

[38] J.L. Thomason, The influence of fibre length and concentration on the properties of glass fibre reinforced polypropylene: 5. Injection moulded long and short fibre PP, Comp. Part A: Appl. Sci. Manuf. 33 (12) (2002) 1641–1652, doi: 10.1016/S1359-835X(02)00179-3 .

[39] H. Meijer, M. Singh, P. Anderson, On the performance of static mixers: a quantitative comparison, Progr. Polym. Sci. 37 (10) (2012) 1333–1349, doi: 10.1016/j.progpolymsci.2011.12.004 .