MaterialsScienceandEngineeringA259(1999)5364EffectofweldingparametersonthesolidicationmicrostructureofautogenousTIGweldsinanAl CuMgMnalloyA.F. Norman *, V. Drazhner, P.B. PrangnellManchester Materials Science Centre, Uni6ersity of Manchester/UMIST, Gros6enor Street, Manchester, M1 7HS, UKReceived4November1997; receivedinrevisedform7August1998AbstractTheweldmetal microstructuresofautogenousTIGweldshavebeeninvestigatedforarangeofweldingconditionsusinganAl CuMgMnalloy. Itwasfoundthatacombinationofhighweldingspeedsandlowpowerdensitiesprovidethethermalconditions required for the nucleation and growth of equiaxed grains in the weld pool, providing heterogeneous nucleation sitesare available. The most likely origin of the nucleants is from a combination of dendrite fragments and TiB2 particles that surviveintheweldpool. Thenestmicrostructurewasobservedinthecentreoftheweldandisattributedtothehighercoolingrateswhich operate along the weld centreline. Composition proles across the dendrite side arms were measured in the TEM and werefoundtofollowaScheil typesegregationbehaviourwherethereisnegligiblebackdiffusioninthesolid. Themeasuredcoreconcentrationofthedendritesidearmswasfoundtorisewithincreasingweldingspeedandwasattributedtotheformationofsignicant undercoolings ahead of the primary dendrite tip, which enriched the liquid surrounding the dendrite side arms. 1999ElsevierScienceS.A. All rightsreserved.Keywords: TIGwelding; Al CuMgMnalloys; Grainstructures; Microsegregation1.IntroductionHeat treatableAA2000series aluminiumalloys areamongthe most widelyusedmaterials for structuralapplications wherehighstrengthtodensityratios areimportant.Inmanyapplications,thedominantjoiningprocess is through the use of mechanical fasteners(rivets). Typically a large numbers of fasteners are used,which makes assembling primary structures an ex-tremely tedious and time consuming process. Addi-tionalproblemsarisebecauseawideangeisrequiredandthesectionthicknessofeachcomponentisusuallyincreasedaroundthejoint producingafurtherweightpenalty.Rivetedjointsarealsorelativelyinefcient,interms of strength, andact as sites for rapidfatiguecrackinitiation[1].Ifweldingcouldbeutilisedwithoutsignicantlyin-creasingthesectionthicknessofthejoint, thenacon-siderable advantage in the manufacturing processwouldresult, bothinterms of weight savings andareductioninthefabricationtime. Atpresenttheweld-ability of the most widely used AA2000 series alu-miniumalloys is knowntobe poor [1,3]. This is incontrast to alloys used in the former Soviet Union,whereanumber of weldablealuminiumcompositionshavebeendeveloped, andareusedinbothaerospaceandautomotiveapplications[46]. Furthermorethereis also some evidence to suggest that welded jointsperformbetterinfatiguethanrivetedjoints[2].Thefusionweldingof highstrengthAA2000seriesaluminium alloys was given some attention in the early1960s[7,8]. Despitethefact that at thetime, weldingtechnologywasmoreprimitive, several problemsasso-ciatedwiththe weldingof AA2000series aluminiumalloys were identied. These included the poor strengthandductilityoftheweldmetal,solidicationcracking,andgrainboundarymeltingintheheat affectedzone(HAZ) [7]. Recently the macroscopic grain structures ofAl alloy TIG welds have been investigated by a number* Correspondingauthor. Tel.: +441612003588; fax: +441612003586; e-mail: andy.norman@umist.ac.uk0921-5093/99/$-seefrontmatter1999ElsevierScienceS.A. All rightsreserved.PIIS0921-5093(98)00873-9A.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 54Table1WeldingconditionsforthesamplesunderinvestigationWeldingcurrent(A) Powera(W) SampleNo. Weldingspeed(mms1) Powerdensityb(Jmm3)100 630 1 14.8 7819 12.3 130 13 2160 1008 3 10.4 19190 1197 4 8.3 25aCalculatedfromQ=pIV, whereQisthetotal heatinput, Vistheweldingvoltageand, pistakentobe0.70[10].bThevolumeofweldpersecondisestimatedfromwidththicknessweldingspeed.of researchers [912]. It is widely accepted thatequiaxeddendriticgrainstendtoforminthecentreofthe weld, at high welding speeds, whereas columnarstructures (otherwise knownas axial or straygrains)usuallyformatlowweldingspeeds.Onlyalimitednumber of researchers havefocusedon the microsegregation behaviour in Al alloy TIGwelds [13,14]. However, understandingthe redistribu-tion of solute during solidication is particularly impor-tantinheattreatablealloysastheweldmetalstrengthwill depend on the solute supersaturation, which deter-mines the subsequent ageing response and yieldstrength. In order to develop models capable of predict-ing the weld metal properties, it is equally important tobe able to predict the grain structure and dendrite armspacing and volume fraction of eutectic, which willdominate the toughness and ductility. Anumber ofworkers have demonstrated that a relatively simpleapproach, basedonestimatingthecoolingratecanbeused to reliably predict dendrite arm spacings in modelaluminiumalloys [15], but this has not yet beenat-temptedwithmorecomplexalloysofnearcommercialcomposition. Furthermore, there is little experimentaldata available in the literature comparing measure-mentsof soluteprolesacrossdendritestomodelsofsolute redistribution. Conictingresults exist betweenmeasured solute proles; some workers suggesting aScheil behaviour [16], where there is effectively nodiffusion in the solid phase, and other proles based onthemodel of BrodyandFleming[17], wherethereislimitedbackdiffusioninthesolid.In this paper, results are presented on the commercialaluminium alloy AA2024 which has been autogenouslyTIGwelded using a range of processing conditions.Although in practice it is unlikely that structures wouldbeweldedwithouttheuseofllermaterials, theaddi-tionof a ller material makes the predictionof themicrostructure inthe weldmore difcult. The weldswere characterisedusingoptical metallographytore-veal the different grain structures, produced with differ-ent weldingconditions. Thesub-grainstructures werestudiedusing a combinationof techniques. Scanningelectronmicroscopy(SEM) was usedtomeasure thedendrite secondary arm spacing at different positions ineachweld, whereas transmissionelectronmicroscopy(TEM), EDX analysis, and X-ray diffraction were usedto measure the effect of the welding speed on thesegregationbehaviour of themajor alloyingelements,and the phases which formed on solidication. Theresultsarecomparedtoexistingmodelstoseeif it ispossibletopredict themainfeatures of theweldmi-crostructure in multi-component alloys, as a function ofthe welding parameters, without resorting to thermody-namiccalculations whichrequireextensivecomputingpower.2.ExperimentalThe commercial aluminiumalloy AA2024 (Al 4.4wt%Cu1.5wt%Mg0.8wt%Mn, andFe andSiimpurities) was TIGwelded autogenously, using therangeofconditionsgiveninTable1. Thesheetthick-ness was 1.6 mm and a simple butt geometry was usedfor each experiment. The experiments were designedsuch that, as the welding speed was increased, thewelding current was also increased to just maintain fullpenetrationof the weld. The data inTable 1 showsthat, withincreasingweldingspeed, theoverall powerdensity required to maintain full penetration decreases.After welding, the samples were sectioned and mountedeitheralongthedirectionof theweldinplanvieworsectioned through the weld (transverse view). Eachsample was examinedusing conventional optical mi-croscopy. Samples for SEM examination were preparedbypolishingto1mmnishandelectropolishinginasolution of 10 vol% nitric acid in methanol at 20 V, and30C.ThinfoilsforTEMobservationwerepreparedfromtheweldsbysparkeroding3mmdiscsfromdifferentpositions in the weld sample. The discs were ground toathicknessof ~100mmandjet-polishedusingasolu-tionof30vol%nitricacidinmethanolat 30Cand12V. The foils were examinedinaPhilips CM200analytical TEM operated at 200 kV. Quantitative EDXanalysis was performed by traversing across severaldendrite side arms for eachweldingspeed. The coreconcentrationof the side arms was determinedas afunction of welding speed (approximately 15 arms weremeasuredforeachspeed).Becauseofthestronglikeli-A.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 55hoodof dendrites not being sectionedthroughtheircentreinarandomsection, onlythewidest dendriteswereanalysedandthelowestcoreconcentrationmea-suredwas usedas the most representative value. AllEDXdatawasZAFcorrectedandcompositionsweremeasuredacrossdendritesidearmsinpositionswherethe eutectic phase hadbeenpolishedout of the foil,thus eliminating the possibility of unrepresentative mea-surementsnearthedendriteboundary.3.Macroscopicgrainstructures3.1. FormationoffusionzonemicrostructuresInautogenous TIGwelding, themacroscopicgrainstructure is controlled by a combination of the thermalconditions that prevail at the solidliquid interface andthecrystal growthratewhichisdirectlyrelatedtotheweldingspeed[14]. Thethermal conditions aredeter-mined by the heat input and the weld speed for a givensheetthickness. Furthermore, theconditionsvarycon-siderably depending on the position at the solidliquidinterfacealongthetrailingedgeoftheweldpool. Forthin sheet Al alloys, the high thermal diffusivity ofaluminiumfavourstheformationof anelliptical weldpool shape, even at relatively high heat inputs andweldingspeeds[18].The different types of grain structures observed, as afunction of the welding conditions, are shown in Fig. 1andaresummarisedinFig. 2. Forthelowestweldingspeed considered (7 mms1) a fewextremely longcolumnar-dendriticgrainswereobserved(oftentermedaxial [11]), which grew along the centre of the weld inthe directionof the motionof the heat source (Fig.2(a)). The overall width of this weld was ~4 mm and thecolumnar-dendritic region was around 1 mm wide. Theaxial grainstructure showninFig. 1(a) has beenre-ported for other TIG welded Al alloys, such as AA3003(Al FeMn) and AA5454 (Al Mg) [9] which werealsoproducedat lowweldingspeedsandheat inputs.On either side of the central axial grains, columnar-den-driticgrains werefound, whichsolidiedonthebasemetal(epitaxial nucleation)andgrewtowardsthecen-tral axial region (Fig. 1(b)). The epitaxially growndendrites curve towards the heat source, sothat themaximum thermal gradients present at the solidliquidinterface, aremaintainedasgrowthproceeds. Forthesample with the lowest speed (7 mm s1), near the endof the weld, the central regionbecame slightlywiderandthedensityofgrainsincreased, producingamorestray-like grainstructure (Fig. 1(c)). This effect wasprobablyduetothesheetmetalincreasingintempera-tureastheweldproceededwhichreducedthethermalgradients at thesolidliquidinterface. Theweldwiththenexthighestspeed(13mms1)hadastraygrainstructure along the central region of the weld, through-out the entire weld length. It has been reported byGanaha et al. [9] that stray grain structures usuallyformat intermediateweldingspeeds andheat inputs,andarecharacterisedbythecontinual appearanceofnew anisotropic grains aligned along the central regionof the weld, in the welding direction. As the nucleationrate at the interface is verylow, manyof the grainsgrow a considerable distance in the direction of the heatsource, and therefore appear elongated in morphology.At all higher welding speeds, the central region of theweldconsistedof anequiaxed-dendriticstructurewithan average grain size of ~250 mm(Fig. 1(d)). Theremainingweldmicrostructurestillconsistedofepitax-ial columnar grains growing in from the weld edge. Theequiaxed region covered almost a third of the width oftheweldproducedat 19mms1, whichincreasedtoalmost half the weld width at 25 mm s1. These weldswere also examined in the transverse direction andshowthatforthetwohighestweldingspeeds,aregionofequiaxedgrainsformedthroughoutthethicknessoftheweld(Fig. 1(e)and(f)).3.2. Columnartoequiaxedtransition(CET)Inthiswork,atransitionseriesfromaxial,tostray,toequiaxedgrains, wasobservedastheweldingspeedwasincreased(showninFig. 2(d)). Thesametypeoftransitionhaspreviouslybeenreportedforarangeofother TIGwelded Al alloys; namely AA6061 [10],AA1100, AA5052, and AA7004 [9], and as can be seenin Fig. 2(d), the columnar to equiaxed transition occursat roughly the same welding speed/heat input combina-tions, despite the difference in alloy compositions. Hunt[19] has analysed the conditions necessary for thegrowthofequiaxedgrainsaheadofacolumnarinter-face duringdirectional solidication. Hunts analysis,whichassumesthata50%equiaxedvolumefractionissufcient to block columnar growth, results in thefollowing expression for predicting the columnar-to-equiaxed(CET)transition:GLB0.061N1/30

1(DTN)3(DTC)3nDTC(1)whereGListhemaximumthermal gradient, N0isthedensityofheterogeneousnucleants, DTNisthecriticalundercoolingforheterogeneousnucleation,and DTCisthe growth undercooling at the columnar front. Al-though these quantities cannot be easily established,Eq. (1)suggeststhattheformationofequiaxedgrainsinthe centre of the weldcanbe expectedprovidingthat; (i) there is a supply of nucleation sites from whichnew grains may develop, and (ii) thermal conditions arepresent which favour the nucleation and growth of newgrains. It should be noted that the transition fromcolumnar to equiaxed grains also requires a sufcientlyA.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 56Fig. 1. Examplesofthedifferentgrainstructuresfoundin2024autogenousTIGweldsforthedifferentweldingspeeds. (a)Axial, 7mms1(plan);(b)epitaxial,7mms1(plan);(c)stray,13mms1(plan);(d)equiaxed,25mms1(plan);(e)equiaxed,25mms1(transverse);and(f)equiaxed, 25mms1(transverse).highnucleationrate relative tothe growthrate. Thetwo conditions noted above are explained in moredetail below.3.2.1. SupplyofnucleantsInthe weldingof Al alloys, the formationof newgrains canoriginatefromtwosources; thegrowthofpre-existingnuclei createdbydendrite fragmentation,orbygraindetachment, andheterogeneousnucleationat second phase particles present in the melt, or impuri-ties in the weld pool [14]. It is well known that the grainstructureofcastingotsiscontrolledthroughthedelib-erateadditionof inoculants[20]. Duringweldingit ispossible for some of the inoculants tosurvive intheweldpool andact as nucleants for newgrains. Thealloy examined in this work also contains a smalladdition of TiB2 particles, which are used to control thegrainstructureduringDCcasting, andthereforeit isA.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 57possibleforsomeoftheseparticlestoactasnucleantsintheweldpool. Recentlytherehasbeensomedirectevidence of Ti rich particles acting as nucleants forgrains in TIGwelds of Al alloys AA6061 [10] andAA7004 [9]. It is also interesting to note that for the Alalloy AA5083, which contains a signicant amount of Ti(0.024wt%),aregionofequiaxedgrainswasobservedat both low and high welding speeds [9]. Also, when thealloyAA2024isweldedwiththellerAA2319(whichcontains grain rening additions of Ti and Zr), thewelding speed at which the columnar to equiaxed tran-sitionoccursisreducedfrom15toaround6mms1[24].3.2.2. Thermal considerationsFor the welds studied in this work, an increase in theweldingspeedwasaccompaniedbyanincreaseinthewelding current so that full weld penetration was main-tained. However, thepower density(seeTable1) de-creased as the welding speed was increased. Thiscombinationof conditions cancauseadecreasing, ornearlyconstant thermal gradient inthe liquid, whilstpromoting an increase in the crystal growth velocity. Ifthe thermal gradient decreases, with increasing weldspeed, this could result in signicant undercoolingsahead of the solidliquid interface, due to a combinationof solutal anddendritetipcurvatureeffects (seealsoSection5)[21].In many of the previous studies on the grain structureof Al TIGwelds, thetemperaturegradient alongtheweldcentrelinehasnotbeenmeasuredexperimentally,but has beencalculatedusingthefollowingequation,which is derived from the Rosenthal thin plate solution[22,23]:GL=2yhzCtQ

26(TmT0)3(2)In Eq. (2), h is the thermal conductivity, z is the density,Qisthetotalheatinput,Cisthespecicheat,tistheplate thickness, Tm is the freezing temperature and T0 istheambienttemperature.The calculated values of GL for the different weldingspeedsaregiveninTable2.Thevaluesrangefrom81K mm1, at a speed of 13 mm s1to 73 K mm1, atthehighestweldingspeedof25mms1,andarethuspredictedtobenearlyconstant inthepresent experi-ment. According to observations made by Kou and Le[10], of experimentally determined thermal gradientsfromthermocouples, themeasuredresultsagreedwellwith the calculated values at low welding speeds and heatinputs, but were increasingly inaccurate at higher speedsandpowerdensities. Thisdifferencewasattributedtothe basic assumption that the heat source in the calcula-tion is treated as a line source and neglects the presenceof the weld pool, or heat ow by convection. Thereforethe calculatedvalues quotedinthis workshouldbetaken as an upper bound because it is likely that, due toconvectionintheweldpool, thereal valueswouldbesomewhat lower resultinginasignicantlydecreasingthermal gradient with increasing speed (as the observedtransitionsingrainstructuremightsuggest).In this work, both the welding current and the weld-ing speed were increased at the same time. If the effectof this was to keep the thermal gradient along the weldcentrelineconstant, as predictedbyEq. (2), theratioGL/Rat the weld centreline will still decrease withincreasing welding speed because R increases (see TableFig. 2. Schematicdiagrams illustratingthedifferent types of grainstructuresfoundinAlTIGwelds.(a)Axial,(b)Stray,(c)equiaxed,and(d) aplot of heat input versusweldingspeedforthedifferentweldsproducedinthiswork.A.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 58Table2SummaryofcalculationsinvolvingGLandRattheweldcentrelineGL/R GLR(Cs1) Calculatedbu2(mm) SampleNo. Measuredu2(mm) Weldingspeed(mms1) GLa(Cmm1)5.8 519 6.2 1 10.0 7 74.21059 4.9 2 13 81.5 6.0 4.93.9 4.4 1495 3 4.1 19 78.71835 4.1 4 25 73.4 2.9 3.5aCalculatedfromEq. (2).bCalculatedfromEq. (4).2) and the undercooling achieved in the liquid, ahead ofthe advancingsolidliquidinterface will be mainlyafunctionof the growthvelocity, whichfor columnargrain structures, is directly related to the welding speed.At low welding speeds this suggests that even if hetero-geneousnucleationsitesarepresent intheweldpool,thethermalconditionsdonotfavourtheformationofstablenuclei, andthat anygrainfragmentswouldre-melt, i.e. an axial grain structure would be expected. Astheweldingspeedisincreased, thegrowthvelocityofthe solidicationfront will alsoincrease producingahigherdegreeofundercooling, thusproducingtheob-served transition from axial to stray grain structures. Atthehighest weldingspeed, theundercoolingwill beatits greatest and can give rise to a signicant nucleationrateresultinginanequiaxedstructure.4.Grainsub-structures4.1. IntroductionTheobservationsmadeinSection3focusedonthetype of grain structure that formed with different weld-ing conditions. Of equal importance is the effect ofwelding parameters on the grain sub-structure thatformsduringwelding. TheresultsinSection3clearlydemonstrate that for the welding conditions consideredinthiswork, thealloyalwaysproducesaweldwhichcontains a dendritic sub-structure (irrespective ofwhetherthegrainstructureiscolumnarorequiaxed).Fig. 3(a) and(b) showexamples of adendritic sub-structuretakenattheweldcentrelinefortwodifferentspeeds. In both optical micrographs, a dendritic growthpattern can clearly be seen (light coloured) with a darkstructure present intheinterdendritic regions. At thealuminiumrichcorneroftheternaryAl CuMgsys-tem, the phase diagramis dominatedbythe ternaryeutectic reaction between Lh-Al +q(Al2Cu) +S(Al2CuMg) which is reported to occur at 503C with acompositionof 26.8 wt%Cuand6.2 wt%Mg [25].X-raydiffractionanalysisconrmedthatthedendriticmatrixwas h-Al andthe interdendritic regionwas amixtureof threephases, namely h-Al, q(Al2Cu), andS(Al2CuMg), i.e. the ternaryeutectic. InFig. 3(c), aTEMimage is shownof the typical eutectic growthmorphologyfortheweldingspeedof7mms1.4.2. DendritesecondaryarmspacingmeasurementsInthesolidicationofcaststructures, thereexistsawell establishedlinkbetweenthescaleofthedendriticsub-structure and the conditions under which solidica-tiontakesplace. Variousstudieshaveshownthat thedendrite armspacing, measuredafter processing, de-pends directly on a combination of the thermal gradient(GL)andgrowthrate(R)[14]. Thedependenceoftheprimary(u1) andsecondary(u2) armspacings, onGLandRtakethefollowingform:u1=a1(G2LR)n(3)u2=a2(GLR)n(4)wherea1anda2arecoefcients, whosevaluesdependonthealloysystem,andnisanexponentwhosevalueliesbetween1/4and1/2. Theprimaryspacingcannotbedirectlyrelatedtothecoolingrate(=GLR) sinceitsdependenceonGLandRhavedifferentexponents.However, the secondaryarmspacingdepends ontheproduct of GLR and can therefore be related directly tothecoolingrate.Thevaluesofa2=50 mm[Ks1]nandn=1/3havebeenestimatedforbinaryAl 4wt%Cualloyswithinthe range of cooling rates considered in this work(1105Ks1[15]). Using these constants, the sec-ondary arm spacing along the weld centreline has beenestimated from the data given in Table 2 for GL and R(assumingRisequal totheweldingspeed, thismightnot be the case for equiaxed microstructures as thegrowth direction does not necessarily correspond to theweldtraversedirection). ThisdataisplottedinFig. 4togetherwithsecondaryarmspacingsmeasuredalongthe weld centreline for the different welding speeds. Fig.4 shows that as the welding speed is increased, both themeasured and calculated secondary armspacing de-creases, andthatthereisgoodagreementbetweenthemeasured and calculated values, over the completerange of welding conditions, althoughthe calculatedvaluesgiveconsistentlylargespacingswhencomparedto the measured values. There are two reasons why theA.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 59Fig. 3. (a) and (b) Optical micrographs showing the dendritic (light) and interdendritic (dark) regions along the weld centreline for the speeds of7 and 25 mm s1respectively. (c) TEM micrograph showing the typical growth morphology of the ternary eutectic in the interdendritic regions.calculatedandmeasuredspacingsmightdiffer.Firstly,inSection3we suggestedthat the thermal gradientscalculated in Eq. (2) were to be considered as an upperestimate. If theactual values of thethermal gradientweresomewhat lower (duetoconvectionintheweldpool), thentheestimatedcoolingratesalongtheweldcentreline would be lower and the calculated arm spac-ingswouldbeevencoarser.Secondly,itisclearthatabetter t between the measured and calculated spacingscouldbeachievedbyadjustingthevaluesof thecon-stantsinEq.(2).However,theseconstantshaveyettobeestablishedforthecommercial Al alloy2024.According to Eq. (4), GLand Rhave the sameexponentandarethereforeequallyimportantindeter-miningthescaleof thedendriticsub-structure. How-ever, the calculatedthermal gradients alongthe weldcentreline are almost constant for the different weldingspeeds, so it is largely the increase in the welding speedwhich gives rise to the increase in cooling rate andreduction in the secondary arm spacing. Fig. 5 shows acomparison between the scale of the sub-structurealongtheweldcentrelineforthehighestweldingspeedand a region near the fusion boundary. The microstruc-turesclearlyshowthat nearthefusionboundary, thesub-structureisatitscoarsest, whichsuggeststhatthecooling rate at the fusion boundary is much lower thanthat observedinthe centre of the weld. This occursbecause althoughthe thermal gradients are higher atthe fusion boundary, the growth rate will be at itsminimumandtheoverall effect will betoproducealowcoolingratecomparedtothatobservedalongtheweldcentreline.5.Microsegregationbehaviour5.1. CompositionprolesTEM was used to study the segregation behaviour ofthealloyingelementsCuandMnwithinthedendritesidearms. Microanalyses wereperformedat differentpositions across dendrite side arms for the differentweldingspeeds, andthemeasuredvariationinCuandMnsolutecontents, fromthecentreofadendritesideA.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 60Fig. 4. Plotofmeasuredandcalculateddendritesecondaryarmspacing(alongtheweldcentreline)asafunctionofweldingspeed.armtowardstheinterdendriticregion. Foreachweld-ing speed, ve scans were made and the average valuesare shown in Fig. 6(a) and (b) respectively. The data inFig. 6 was obtained from regions where the intermetal-lic eutectic phase hadpolishedout of the sample onpreparing the specimen, sothat the results near theinterdendritic region would not be affected by theirpresence. Mg could not be reliably analysed due to theoverlapwiththeAlKhpeak.Thespatialresolutionofthemeasurements was ~0.2mmwhichis substantiallylessthanthedendritearmspacingsof between4and5.5 mmfor the welding speeds considered. Fig. 6(a)shows that, as expected, thelowest Cuconcentrationoccurs in the core of the dendrite side arm with the Cucontent increasing towards the interdendritic boundary.The prole in Fig. 6(b) also demonstrates that for Mn,the lowest concentration also occurs in the centre of theside arm, although it should be noted that the MncontentinthisalloyisalotlowerthanCu.The results clearlysuggest that the segregationbe-haviour is more similar tothe scenarioenvisagedbyScheil [16], wherethereisnegligiblebackdiffusioninthesolid, thanthat expectedbytheBrodyandFlem-ings model [17]. TheScheil equationfor soluteparti-tioning[16] isgivenby:Cs*=C0k(1fs)k1(5)where Cs* is the solid composition, C0 is the initial alloycomposition, fsis the fractionof solidandk is thepartitioncoefcient. WhiletheScheil equationcannotbe applied directly to multi-component alloys (for non-binary systems, an iterative approach should beadopted where the tie-lines are re-calculated at temper-ature increments using a thermodynamic database [26]),in a systemwhere one alloy component dominates,such as Cu does in this case, it can be used as a roughapproximation. The Scheil equation is based on thebasic assumptions that the liquid is of uniform compo-sition,localequilibriumismaintainedatthesolidliq-uid interface (k is constant), and that there is negligiblesolid-statediffusion.Whetherornotthereisasignicantcontributionofsolid state diffusion to the solute redistribution processcanbeestimatedfromtheparameter, p[17]:p=Dstfu22(6)whereDsisthesoliddiffusivityofthesoluteandtfisthelocal timeof freezing. For thewelds producedinthiswork, thevaluesoftfareintherange0.050.2swhich produces an h parameter in the range 2.41036104. As it canbe shownthat solidstatediffusion during freezing only begins to become animportant factor in the solute redistribution process forvalues of p\0.1 [17], the freezing conditions intheweldshouldbesimilartothosepredictedbytheScheilequation.5.2. VariationofcoreconcentrationwithweldingspeedAsthesoluteproleswerefoundtobeaffectedbytheprocessingconditions, theeffect of weldingspeedon the core composition of the dendrite arm was inves-tigatedinmore detail byanalysingthe centre of ap-A.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 61Fig. 5. Optical micrographs showing the difference in microstructural scale between (a) the weld centreline, and (b) near the fusion boundary foraweldingspeedof25mms1.proximately15sidearmsforeachweldingspeed. Be-cause of the difculties associated with the randomsectioningofdendritesidearmsduringtheproductionofTEMfoils,itispossiblethatmanyofthemeasuredconcentrations correspond to the composition somedistanceawayfromthecentre.Toreducethisrisk,thewidest side arms were analysed and the minimum mea-suredconcentrationfrom15arms was usedfor eachwelding speed. This data is plotted in Fig. 6(c) and usedin the subsequent discussion. Despite the difcultiesmentionedabove, theresultsinFig. 6(c) clearlyshowthat as the welding speed is increased, the dendrite coreconcentrationof bothCuandMnalsoincreases, butthiseffectismorenoticeableforCu.Evenhighercoreconcentrations have been observed for the laser weldingof AA2024where muchgreater weldingspeeds wereused[27].To try and explain this variation in the core composi-tion with welding speed, the arguments will be connedtoanalysingthebehaviour of Cu, themajor alloyingelement in2024, as this element producedthe mostreliabledata.Althoughthisisamulticomponentalloy,the behaviour is compared to that of the binary Al Cuphase diagram for simplicity, and the effect of the otheralloying elements on the microsegregation behaviour ofCu in Al are neglected. As under the welding conditionsused, the solid state diffusion of Cu in Al appears to benegligible(seeabove), itisassumedthatthecorecon-centrationinthedendritesidearmcorrespondstothedendritesidearmtipconcentrationduringgrowth.Undertrueequilibriumsolidicationconditions, thecompositionoftherstsolidtoformwill begivenbyC0k, where C0 is the initial alloy concentration (equal to4.46 wt% Cu for a binary Al Cu alloy of the same Cucomposition as AA2024) and k is the equilibrium parti-tion coefcient (for binary Al Cu alloys, k=0.17).Usingthesevalues, theequilibriumcoreconcentration(C0k) shouldbe 0.76wt%Cu, whichis muchlowerthanthat measuredinanyof theweldsamples. Fur-thermore, the results showninFig. 6(c) demonstratethat thecoreconcentrationvariesfrom~2C0kforthelowest weldingspeedto~3C0kfor the highest speedconsidered. Similar observations have been made byBrookes and Baskes [13] who studied GTAbinaryAl Cu welds containing between 1 and 2 wt%Cu.They alsoobservedcore concentrations greater thanequilibrium(around 2C0k3C0k) which is in goodagreement with the core concentration enrichment mea-suredinthiswork.To try and explain why the core concentration in thesidearmsincreasewithincreasingweldingspeed, it isnecessarytoconsiderhowthegrowthof thedendriteside arms are related to the growth of the dendriteprimary tip. Aschematic solute prole through theliquidinthemushyzonefordendritegrowthisshownin Fig. 7. In this gure, the composition in the bulk ofthe liquid is given by C0. It is possible that for the rangeof weldingspeeds consideredinthis work, signicantundercoolingscanbeachievedattheprimarydendritetip which will enrich the liquid at the tip, to a composi-tionCt. Microstructural observationshaveshownthata signicant volume fraction of eutectic phases areformedinthemushyzoneattheendofsolidication,i.e. at the root of the dendrites inthe interdendriticregions(seeFig. 3(c)). Therefore, thesoluteproleintheliquidwill changeasit movesthroughthemushyzone, from Ct to CE, the eutectic composition, as shownin Fig. 7. In the region where the secondary arms form,thelocalliquidcompositionwillthusbeenrichedwithsolute to a level somewhere between Ct and CE, i.e. thecore concentration of the dendrite side armwill begiven by k(Ct+DC), where DC depends on the relativedistancebackfromthetipwherethesecondaryarmsform(thought to be of the order of two to threesecondaryarmspacings[17]).A.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 62Fig. 6. EDX analyses across dendrite side arms, for (a) Cu and (b) Mn, and (c) a plot of the measured minimum core concentration for dendritesidearmsasafunctionofweldingspeed.A.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 63Fig.7.Schematicdiagramillustratingapossiblesoluteproleintheliquidduringdendriticgrowth.Fig. 8. Plot of primary tipvelocity versus calculatedprimary tiptemperatureforabinaryAl 4.46wt%CualloyusingthemodelsofBurdenandHunt[30], andKurzandFisher[31].model predicts that this will occur at a greater rate (Fig.8). UsingtheKurzandFishermodel, thevariationintip undercooling has been translated to a core composi-tionfor thegrowthof primarydendrites inabinaryAl 4.46 wt% Cu alloy, and the data is shown in Fig. 9together withthe measuredvalues for the secondaryarms. Byassumingthattheprimarytipvelocityisthesame as the welding speed, the model of Kurz andFisher demonstrates that as the welding speedis in-creased, so the primary tip concentration also increases,i.e. in agreement with the trends of this work. As mightbe expected fromFig. 7, the predicted compositionvalues for the primary tip (kCt) are somewhat less thanthemeasuredvaluesof k(DC+Ct), forthesecondaryarmcore compositions. However the secondary armcore composition does increase with increasing weldingspeedinlinewiththepredictedprimarytipbehaviour.Modelshavebeendevelopedtopredicttheprimarydendrite tip undercooling which can, in turn, be relatedtothe primarytipcomposition. AccordingtoBrody[28], the tipundercoolingcanbe estimatedfromthefollowingrelationship.DTtip=DTD+DTr+DTk(7)whereDTDisthediffusional undercooling, DTristhecontributionof theundercoolingduetocurvatureef-fects, andDTkisthekineticundercooling(thekineticundercooling is the smallest contribution to undercool-ing in an alloy, and according to Chalmers [29], can beneglected in this analysis). This approach has been usedby Burden and Hunt [30] to derive an expression for tipundercoolingbyassumingthe tipgrows at the mini-mumundercooling:DT=GDLV+32mV(1k)C0qDL(8)where V is the tip velocity, DL is the solute diffusivity inthe liquid, m is the liquidus slope, and q is the curvatureundercoolingcoefcient.More recently, Kurz and Fisher [31] have pointed outthattheminimumundercoolingcriteriaunderestimatesthechangeintiptemperature, andhaveinstead, usedthe criteria that the tip radius is equal to the wavelengthofinstabilityoftheinterface. Inbothapproaches, thetiptemperature is dominatedbythe growthvelocity,rather than the thermal gradient, when the growthvelocities are higher than 103cm s1and therefore inthe range of interest to welding. Both models show thatthetipundercoolingstartstorapidlyincreaseat highvelocities (~1cms1) althoughthe Kurz andFisherFig. 9. Comparisonbetweenthepredictedcorecompositionsoftheprimarydendrite(Ct)takenfromthemodelofKurzandFisher[31]and the measured core concentrations from the dendrite side arms asa function of welding speed (it is assumed that the primary tip growsatthesamevelocityastheweldingspeed).A.F. Normanetal. / MaterialsScienceandEngineeringA259(1999)5364 646.ConclusionsA detailed understanding of the effect of the process-ing conditions is required to produce a good weldmicrostructurefortheautogenousTIGweldingofthecommercialaluminiumalloyAA2024.Athighweldingspeeds andlowpower densities, it is possibletopro-mote the formation of an equiaxed-dendritic mi-crostructureinthecentreofautogenouswelds. Thisisdue to the development of an undercooled liquid aheadof the moving solidliquidinterface, whichprovidesthecorrect thermal conditions for thenucleationandgrowthof newgrains. Themost likelyoriginfor thenucleants is expected to be froma combination ofdendritearmdetachment andTiB2particles that sur-viveinthemelt.Thenestmicrostructurewasobservedatthecentreof the weld and is attributed to the higher cooling ratesat the weld centre compared to those at the fusionboundary. As the welding speed was increased, thecoolingrate at the centre of the weldalsoincreasedproducing a reduction in the scale of the dendritesecondaryarmspacing.Both the dendrite side arm spacing and the microseg-regation behaviour have been investigated using quanti-tative EDXanalysis. The microsegregationbehaviourof thedendritesidearmshasbeenshowntofollowaScheil type behaviour in which there appears to benegligible back diffusion in the solid. However, the coreconcentrationwithinadendritesidearmsubstantiallychangeswiththeweldingspeedsusedinTIGwelding.This behaviour can be explained in terms of the forma-tionofsignicantundercoolingsaheadoftheprimarydendritetipwhichenrichestheliquidsurroundingthedendritesidearms.AcknowledgementsThisworkwassupportedbytheEPSRCundertheIMI Programme, Grant: GR/K66901, andis releasedwith the kind permission of the following partners:BritishAerospace, ShortBrothers, BritishAluminium,DERA, RollsRoyce, TWI, CraneldUniversity, Uni-versityof Liverpool andtheUniversityof Essex. Theauthors gratefully acknowledge the assistance of one ofthe referees in interpreting the data produced fromsoluteproles.References[1] J.R. Davis (Ed.), ASMSpecialityHandbook: AluminiumandAluminium Alloys, ASM International, Metals Park, OH, 1993,p. 376.[2] R. Onjukka, Weld. J. 75(1996)29.[3] S. Kou, Weld. Res. Council Bull. 320(1986)1.[4] R.V. Ilyushenko, Aluminium69(1993)364.[5] V.V. Grinin, A.V. Petrov, V.V. Ovchinnikov, Weld. Prod. 33(1986)19.[6] I.N. Fridlyander, in: T.H. Sanders, Jr., E.A. Starke, Jr. (Eds.),Proceeding of the 5th International Conference on Al Li Alloys,Williamsburg, Virginia, 1989, MCEP, Birmingham, 1989, p.1359.[7] D.E. Schillinger, I.G. Betz, F.W. Hussey, H. Markus, Welding J.42(1963)269S.[8] W.I. Pumphrey, J.V. Lyons, J. Inst. Metals74(1948)439.[9] T. Ganaha, B.P. Pearce, H.W. Kerr, Metall. Trans. A 11 (1980)1351.[10] S. Kou, Y. Le, Weld. J. 65(1986)305S.[11] Y. Arata, F. matsuda, A. Matsui, Trans. Jpn. Weld. Res. Inst. 3(1974)89.[12] M. Rappaz, C.A. Gandin, J.L. Desbiolles, P. Thevoz, Metall.Trans. A27(1996)695.[13] J.A. Brookes, M.J. Baskes, in: S.A. David(Ed.), Advances inWelding Science andTechnology, ASMInternational, MetalsPark, OH, 1987, p. 93.[14] S.A. David, J.M. Vitek, Int. Mater. Rev. 34(1989)213.[15] H. Jones, Monograph No. 8, Institute of Metallurgists, London,1982.[16] E. Scheil, Z. Metallkd. 34(1942)70.[17] H.D. Brody, M.C. Flemings, Trans. AIME236(1966)615.[18] . Grong, in: H.K.D.H. Bhadeshia(Ed.), Metallurgical Mod-ellingofWelding, InstituteofMaterials, London, 1994.[19] J.D. Hunt, Mater. Sci. Eng. 65(1984)75.[20] D.G. McCartney, Int. Mater. Rev. 34(1989)247.[21] R. Trivedi, W. Kurz, Int. Mater. Rev. 39(1994)49.[22] D. Rosenthal, Weld. J. 20(1941)220s.[23] D. Rosenthal, Trans. ASME68(1946)849.[24] A.F. Norman, K.E. Jakielski, G. Scott, P.B. Prangnell, in: J.Beech, H. Jones (Eds.), Proceedings of the 4th InternationalConferenceonSolidicationProcessing,UniversityofShefeld,December1997, p. 670.[25] S.L. Chen, Y. Zuo, H. Liang, Y.A. Chang, Metall. Trans. A 28(1997)435.[26] N. Saunders, Mater. Sci. Forum217-222(1996)667.[27] A.F. Norman, P.B. Prangnell, in: T. Sato, S. Kumai, T.Kobayashi, Y. Murakami (Eds.), Proceedings of the 6th Interna-tionalConferenceonAluminiumAlloys(ICAA-6),JapanInsti-tuteofLightMetals, Japan, 1998, p. 1435.[28] H.D.Brody,in:S.A.David(Ed),AdvancesinWeldingScienceand Technology, ASM International, Metals Park, OH, 1987, p.83.[29] B.A. Chalmers, Principles of Solidication, Wiley, NewYork,1964.[30] M.H. Burden, J.D. Hunt, J. Crystal Growth22(1974)109.[31] W. Kurz, D.J. Fisher, ActaMetall. 29(1981)11..