• Aucun résultat trouvé

A new strategy for the numerical modeling of a weld pool

N/A
N/A
Protected

Academic year: 2022

Partager "A new strategy for the numerical modeling of a weld pool"

Copied!
20
0
0

Texte intégral

(1)

HAL Id: hal-01902116

https://hal.sorbonne-universite.fr/hal-01902116

Submitted on 23 Oct 2018

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Distributed under a Creative Commons Attribution| 4.0 International License

A new strategy for the numerical modeling of a weld pool

Yassine Saadlaoui, Eric Feulvarch, Alexandre Delache, Jean-Baptiste Leblond, Jean-Michel Bergheau

To cite this version:

Yassine Saadlaoui, Eric Feulvarch, Alexandre Delache, Jean-Baptiste Leblond, Jean-Michel Bergheau.

A new strategy for the numerical modeling of a weld pool. Comptes Rendus Mécanique, Elsevier,

2018, 346 (11), pp.999-1017. �10.1016/j.crme.2018.08.007�. �hal-01902116�

(2)

Contents lists available atScienceDirect

Comptes Rendus Mecanique

www.sciencedirect.com

Computationalmethodsinweldingandadditivemanufacturing/Simulationnumériquedesprocédésdesoudageetdefabricationadditive

A new strategy for the numerical modeling of a weld pool

Yassine Saadlaoui

a

, Éric Feulvarch

b,

, Alexandre Delache

c

, Jean-Baptiste Leblond

d

, Jean-Michel Bergheau

b

aUniversitédeLyon,ENISE,UJM,LTDS,UMR5513CNRS,58,rueJean-Parot,42023Saint-Étiennecedex2,France bUniversitédeLyon,ENISE,LTDS,UMR5513CNRS,58,rueJean-Parot,42023Saint-Étiennecedex2,France cLMFAUMR5509CNRS,sitedeSaint-Étienne,UniversitédeLyon,UniversitéJean-MonnetdeSaint-Étienne,France

dUPMC,UniversitéParis-6,UMR7190,InstitutJean-Le-Rond-d’Alembert,Tour65-55,4,placeJussieu,75252Pariscedex05,France

a r t i c l e i n f o a b s t r a c t

Articlehistory:

Received6March2018 Accepted15May2018 Availableonline23August2018

Keywords:

Surfacetension Freesurface Finite-elementmethod Weldingprocess

Welding processes involve high temperatures and metallurgical and mechanical conse- quences that must be controlled. For this purpose, numerical simulations have been developed tostudythe effectsofthe processonthefinalstructure.Duringthe welding process, the material undergoes thermal cycles that can generate different physical phenomena,likephasechanges,microstructurechangesandresidualstressesand distor- tions.Buttheaccuratesimulationoftransienttemperaturedistributionsinthepartneeds tocarefullytakeaccountofthefluidflowintheweldpool.Theaimofthispaperisthus toproposeanewapproachforsuchasimulationtakingaccountofsurfacetensioneffects (includingboththe “curvature effect” andthe “Marangoni effect”),buoyancy forces and freesurfacemotion.

Theproposedapproachisvalidatedbytwonumericaltestsfromtheliterature:asloshing test andaplatesubjectedtoastaticheatsource.Then,the effects ofthe fluidflowon temperaturedistributionsarediscussedinahybridlaser/arcweldingexample.

©2018Académiedessciences.PublishedbyElsevierMassonSAS.Thisisanopenaccess articleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Weldingisaveryoldassemblyprocessdatingfromthemetalage.Differenttechniqueshavebeendevelopedovertime:

assembliesbyheatingandhammeringatthebeginning,arcwelding,resistancewelding andmorerecently,electronbeam welding,andlaserwelding.Thelargeuseofweldingprocessesandtheneedtoimprovethemtoachieveabettercontrolof theweldingqualityledtothedevelopmentofseveralnumericalsimulationmethods,whichmayprovidevaluableassistance tostudytheeffectsofweldingonthefinalstructure.

Dependingon the objectivestargeted by thesimulation,different approacheshavebeen developed.Theseapproaches canbe classifiedintotwo groups,comprisingrespectivelythermo-mechanicalsimulations topredictthemetallurgicaland mechanicalconsequencesinducedbywelding,andsimulationsoftheprocessitselfincludingthefluidflowintheweldpool.

Industrialcompaniestook alargepartinthedevelopmentofthermo-mechanicalsimulations sincetheobjectivesofthese simulations are toestimate the residualstresses anddistortions[1–5], thus providingvery usefulinformationforfatigue lifetimepredictionsorweldingsequenceoptimization[6].Thethermalpartofthiskindofsimulationisgenerallysimplified

*

Correspondingauthor.

E-mailaddress:eric.feulvarch@enise.fr(É. Feulvarch).

https://doi.org/10.1016/j.crme.2018.08.007

1631-0721/©2018Académiedessciences.PublishedbyElsevierMassonSAS.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/).

(3)

Fig. 1.(a)Simplifiedschematicoftheweldingprinciple,(b)simplifiedschematicofthematerialstatesandfluidflowinthemoltenpoolduringthewelding process[23].

inordertoreducethecalculationtimesinceitisusedtosimulatelargestructures[7–9].Theheatinputisthenrepresented by anequivalent heatsource,theparametersofwhichmustbepreviouslyidentified onexperimentalheataffectedzones.

The thermo-mechanical computation is thereforeachieved intwo steps:a coupledthermal andmetallurgicalsimulation, followed bya mechanicalcomputation takingaccountofthetemperatures andmetallurgicalphasespreviously computed [10,11].

However,thistypeofsimulationisnottotallypredictiveasfarastemperaturedistributionsareconcerned.Theirpredic- tion indeedrequirescarefullysimulatingthefluidflow, whichplays an importantrole,particularlyonthemorphology of theweldpool.

Several authorshavethereforedeveloped numericalapproaches[12–22], theaim ofwhichis tosimulatethe physical phenomena involvedby theprocessitself(Fig.1).Among thesephenomena,thefluidflowinthemoltenpool,particularly relatedtothesurface tension,hasadirecteffectonthefieldtemperatureandweldpoolmorphology[24–26].Thisiswhy severalstudieswerecarriedouttomodelthesurfacetensioninthemoltenpool[27,28].Theeffectofthesurface tension on thefluid flowarisesfromtwofactors: first,surfacetensiongeneratesanormalforceontothefreesurface ofthefluid, proportionaltothecurvatureofthissurface(“curvatureeffect”),andseconditgenerates,throughatemperaturegradient,an additionaltangential forceontothefreesurface(“Marangonieffect”).Theclassicalimplementationmethodsofthesurface tension in finite-element codes consider these two effects in a naturalway. They includethe loads applied on the free surface ofthefluidamongtheexternalforcesoftheproblem[29,30].Inthistypeofapproach,theimplementationofthe tangential effectdoesnotraiseanyproblem.However,thenormaleffectisverydifficulttoincorporate,since itrequiresa directcalculationofthisforcethatdependsonthefreesurfacemotion[31,32].

Severalmethodshavebeendevelopedinordertomodelthefreesurface duringthefinite-elementcomputation. These methods can be classified into two categories: the Lagrangian approaches(Arbitrary Lagrangian Eulerian: ALE) [33] and Eulerian approaches(volumeoffluid: VOF[34] and level-set [35]).The twoEulerian methodscan compute the“keyhole collapse” andhelp predictdefect formationsuch asporosity [36,37]. However, takingintoaccount thesolid deformation using these two methods is very difficult.For this, the ALE approach is often used during the numerical simulation of welding [38–40].It consistsincalculatingthe displacementoftheboundary nodesphysicallyandthatof thebulk nodes arbitrarily. Thismeans that the solutionsinside the domainare computed withan Eulerianformulation. Thus, the mesh movementismodifiedinordertoimprovethemeshquality.

The aimofthispaperisto proposeanewapproachforthesimulationoftheweld pooltakingaccountofthesurface tensioneffects(includingboththe“curvatureeffect”andthe“Marangonieffect”),buoyancyforces,andfreesurfacemotion.

(4)

Thissimulationwillbeusedinfurtherstudiestosimulatetheinteractionbetweenthefluidflowandthesoliddeformation.

Forthispurpose,theapproach developedbyLeblondetal.[41] tomodelthesurface tensionisused.Ithasbeenapplied uptonowto simulatesimpleacademicbenchmarks.Inthiswork,afirstapplicationtoanindustrialproblemisproposed (weldingprocess). Thismodeling ofsurface tension consistsinincorporating the normalandtangential effects ina very simple andefficientwaycompared toclassical methods. Inaddition,a pragmaticALEapproach is developedto simulate thefreesurfacemotion.Thechoiceofthisapproachisbasedontwocriteria:thesimplicityofitsimplementation,andthe possibilitytotakeintoaccountthesoliddeformationinasimpleway.

Thepaperisorganizedasfollows:

– insection2,themodelingofthethermal-fluidproblemispresented,itcontainsthreeparts:anillustrationofthefluid andthermalformulation,themodelingofthesurfacetension,andthemodelingofthefreesurface;

– insection3,avalidationofthemodelingofthesurfacetensionandfreesurfaceiscarriedoutusingnumericalvalidation tests;

– insection4,theresultsofthethermal-fluidsimulationoflaser/archybridweldingarepresented.

2. Modelingthethermal-fluidproblem 2.1. Strongformulation

Let usconsider a bounded fluid domain Ω ofboundary ∂Ω. The coupled heat transfer and fluid flow problems are governedbythefollowingthreeequations.

Massconservationequation

Thefluidissupposedthermallydilatableandmechanicallyincompressible:thevolumevariationisonlyduetotempera- turevariations,thestresseshavenoeffect.FromtheEulerianstrainrateD=12(grad v+(grad v)),letusdefinetheEulerian mechanicalstrainratebytheformulation:

Dm

=

D

+ ρ ˙

3

ρ

I (1)

Themassconservationequationthenwrites:

divv

+ ρ ˙

ρ =

tr

(

Dm

) =

0 (2)

In practice, the variations of

ρ

are only due to the thermal dilation effects. It follows that ρ˙

ρ ∼= −βT˙, where β is the coefficientofvolumetricexpansion.

Energybalanceequation

ρ

H

t

+

v

·

gradH

div

gradT

)

r

=

0 (3)

Here

ρ

isthedensity, H,thespecificenthalpy,v,thefluidvelocity,λ,theheatconductivity,T,thetemperature,andrthe volumeheatsource.Theenthalpy Hisafunctionoftemperatureincludingboththesensibleheatandthelatentheatdue tophasechange,and

ρ

andλcanbetemperature-dependent.Theassociatedboundaryconditionsare:

T

=

T0 on

∂Ω

T (4)

λ

gradT

·

n

=

q on

∂Ω

q (5)

T0 istheknowntemperatureon∂ΩT andqatemperature-dependentheatfluxon∂Ωqwhoseoutsideunitnormalisn.

Momentumequations(Navier–Stokesequations)

gradp

+

div s

+

fv

= ρ

v

t

+

grad v

·

v

(6) The fluid flow is supposed to be laminar. Here p is the fluid pressure, s, the stress deviator, and fv, the volume force.

Buoyancyforcescanbeconsideredusingthevolumeforceswhicharegivenbythefollowingequation:

fv

= ρ

0

1

β(

T

T0

)

g (7)

gbeingthegravitationalaccelerationand

ρ

0themassperunitvolumeatthereferencetemperatureT0.

Thefluidissupposed tobehaveasaNewtonianfluid.TherelationbetweentheEulerianmechanicalstrainrateDm and thestressdeviatorscanthereforebewrittenas:

(5)

Fig. 2.P1+/P1 element.

s

=

2

μ

Dm (8)

where

μ

isthefluidviscosity.

Theassociatedboundaryconditionsare:

v

=

vd on

∂Ω

v (9)

σ ·

n

= ϕ

d on

∂Ω

ϕ (10)

Here vd and

ϕ

dare theprescribedvelocityandsurface forceon∂Ωv and∂Ωϕ respectively,and

σ

=spI istheCauchy stresstensor.

2.2. Finiteelementformulation

Theweakformulationoftheenergybalanceequation aswellasthefiniteelementformulationofthethermalproblem arequiteclassical,anddetailscanbefoundin[42,43].Wethereforefocusonthefluidproblemformulationinthesequel.

Thefiniteelementformulationofthefluidproblemrestsonamixedvelocity–pressure(v,p)variationalformulation.

Theproblemisnowtofind(v,p)(Vad×Pad),t∈ [0,T],suchas(v,p)(Vad0 ×Pad):

Ω v

ρ

v

t

+

grad v

·

v

dv

Ω

D

:

2

μ

Dmdv

+

Ω div

v

pdv

+

Ω

v

·

fvdv

+

∂Ωϕ

v

· ϕ

dds

=

0

Ω p

divv

+ ρ ˙ ρ

dv

=

0

(11)

withtheinitialconditionv(x,t=0)=v0(x). InEqs. (11),D=12(grad v+(grad v))and:

Vad

= {

v

H1

(Ω) |

v

=

vdon

∂Ω

v

}

Vad0

= {

v

H1

(Ω) |

v

=

0on

∂Ω

v

}

Pad

= {

p

L2

(Ω) }

P1+/P1 tetrahedralelements(Fig.2)areusedforthediscretizationofEqs. (11).

Velocityapproximation

The velocityfieldisdecomposedintoapartvs calculatedfromthevelocitiesofthevertexnodesgatheredinthearray Vs,usingtheshapefunctionsNs oftheclassicalfour-node tetrahedron[42],plusanotherpartvb calculatedfromavector bofadditionaldegreesoffreedomofthebubble(internal)nodeandtheassociatedshapefunctionNb:

v

=

vs

+

vb (12)

with

vs

=

Ns

.

Vs

=

4

n=1

NsnVsn vb

=

Nb

.

b

(13)

withNb=minn=1to4Nsn.

Notethat b doesnot representthevelocity ofthecentroidoftheelementbecausetheshape functionsofthevertex nodesNsdonotvanishatthispoint.

(6)

The virtual velocity v isdecomposedin thesame way.Notethat the “bubblevelocity” vb vanishes oneach element boundary∂Ωe (bubblenodeprinciple)andthereforeatanypointof∂Ω.

Pressureapproximation

ThepressureiscalculatedfromthevaluesofthevertexnodesPs:

p

=

Ns

.

Ps

=

4

n=1

NsnPsn (14)

andsoisalsothetemperature.

The Eulerianmechanical strain rateDm,aswell asits virtual counterpart Dm, are then decomposedin thefollowing form:

Dm

=

Dms

+

Dmb

Dms

=

Ds

+ ρ ˙

3

ρ

I

=

1 2

grad vs

+ (

grad vs

)

+ ρ ˙

3

ρ

I

Dmb

=

Db

=

1 2

grad vb

+ (

grad vb

)

(15)

Equations(11) with(8),(12) and(15) canthereforeberewritten,foranyvs,vb,andp:

Ω vs

ρ

v

t

+

grad v

.

v

dv

Ω

Ds

:

2

μ

Dmsdv

Ω

Ds

:

2

μ

Dmbdv

+

Ω div

vs

pdv

+

Ω

vs

·

fvdv

+

∂Ωϕ

vs

· ϕ

dds

=

0

Ω vb

ρ

v

t

+

grad v

.

v

dv

Ω

Db

:

2

μ

Dmsdv

Ω

Db

:

2

μ

Dmbdv

+

Ω div

vb

pdv

+

Ω

vb

·

fvdv

=

0

Ω p

div

(

vs

+

vb

) + ρ ˙ ρ

dv

=

0

(16)

Ds andDs areconstantover eachelementΩe astheyinvolvethederivativesoftheshape functionsoftheclassicalfour- node tetrahedron. Assuming that ρ˙

ρ and

μ

are constant over each element (in practice ρ˙

ρ and

μ

are calculated at the centroidofthetetrahedron),Dms isalsoconstantovereach element.Since vb vanisheson theelementboundary∂Ωe,it followsthat:

Ωe

Ds

:

2

μ

Dmbdv

=

2

μ

Ds

:

Ωe

Dmbdv

= μ

Ds

:

Ωe

grad vb

+ (

grad vb

)

dv

= μ

Ds

:

∂Ωe

n

vb

+

vb

n ds

=

0

(17)

Inthesameway,onecanshowthat:

Ωe

Db

:

2

μ

Dmsdv

=

0 (18)

Animplicitbackwardschemeforvt andalinearizationoftheconvectiontermareusedforthetimeintegrationofEqs. (16).

Itisalsosupposedthatthebubblenodedoesnotcontributetotheacceleration(justthevertexnodesaretakenintoaccount fortheinterpolationofacceleration).

UsingEqs. (17) and(18),Eqs. (16) canthereforeberewrittenatthefinalinstantofanytime-step[t,t+t],foranyvs, vb,andp:

(7)

Ω vs

ρ

vs

vts

t

+

grad

(

vs

+

vb

) ·

vt

dv

Ω

Ds

:

2

μ

Dmsdv

+

Ω div

vs

pdv

+

Ω

vs

·

fvdv

+

∂Ωϕ

vs

· ϕ

dds

=

0

Ω vb

ρ

vs

vts

t

+

grad

(

vs

+

vb

) ·

vt

dv

Ω

Db

:

2

μ

Dmbdv

+

Ω div

vb

pdv

+

Ω

vb

·

fvdv

=

0

Ω p

div

(

vs

+

vb

) + ρ ˙ ρ

dv

=

0

(19)

whereallthequantitiesareconsideredattimet+t,exceptforvt,whichrepresentsthevelocityattimet.

UsingnowtheVoigtnotationfortheEulerianstrainrateanddefining,from(13) and(15),Bs andBbmatricessuchthat Ds=Bs·VsandDb=Bb·b,Eqs. (19) leadtothefollowingexpressionsofthenodalresidualvectors:

Ts

= −

Ω

Ns

·

vs

vts

t

ρ

dv

Ω

Ns

· ρ

grad

(

vs

+

vb

) ·

vtdv

Ω

Bs

·

2

μ

Dmsdv

+

Ω

(

grad Ns

)

pdv

+

∂Ωϕ

Ns

· ϕ

dds

+

Ω

Ns

·

fvdv

=

0

Tb

= −

Ω

Nb

·

vs

vts

t

ρ

dv

Ω

Nb

· ρ

grad

(

vs

+

vb

) ·

vtdv

Ω

Bb

·

2

μ

Dmbdv

+

Ω

(

grad Nb

)

pdv

+

Ω

Nb

·

fvdv

=

0

Tp

=

Ω Ns

div

(

vs

+

vb

) + ρ ˙ ρ

dv

=

0

(20)

Note that thesecond equation ofsystem(20) defines,at theelement level,thefollowing linearrelationbetween Vs,b andPs:

Kbs

·

Vs

+

Kbb

·

b

+

Kbp

·

Ps

Fb

=

0 (21)

with:

Kbs

=

Ω

Nb

· ρ

grad Ns

·

vtdv

+

Ω Nb

· ρ

tNsdv Kbb

=

Ω

Bb

·

2

μ

Bbdv

+

Ω

Nb

· ρ

grad Nb

·

vtdv

Kbp

= −

Ω

(

grad Nb

)

·

Nsdx

Fb

=

Ω

Nb

·

fvdv

+

Ω Nb

·

vts

t

ρ

dv

(22)

Equations(20) aresolvedusingaNewton–Raphsonmethod.AftereliminationofbfromEq. (21),ateachiterationcorrec- tions δVs andδPsarecalculatedthroughthefollowingequations:

Kss

Ksb

·

Kbb1

·

Kbs Ksp

Ksb

·

Kbb1

·

Kbp

Kps

Kpb

·

Kbb1

·

Kbs

Kpb

·

Kbb1

·

Kbp

δ

Vs

δ

Ps

=

Ts

Tp

(23) with:

(8)

Ksb

=

Ω

Ns

· ρ

grad Nb

.

vtdv

Kss

=

Ω

Bs

·

2

μ

Bsdv

+

Ω

Ns

· ρ

grad Ns

·

vtdv

+

Ω Ns

· ρ

tgrad Nsdv Ksp

= −

Ω

(

grad Ns

)

p

·

Nsdx

(24)

Kps

=

Ksp

Kpb

=

Kbp (25)

The processis repeatedaslongasTs>

ε

s andTp>

ε

p,where

ε

s and

ε

p are precisionthresholds prescribed bythe user.

2.3. Modelingthesurfacetensioneffects

Oneofthemaineffectstobeaccountedforinweldpoolsimulationsisrelatedtosurfacetension.Surfacetensioninduces apressurenormalto thefreesurface ofthe fluid,proportionaltothecurvatureofthissurface,anditsgradient generates a tangential surfaceforce (Marangoni effect).As farastheimplementation ofthesurface tensioninfinite-element codes [44,45] isconcerned,thesetwoeffectsaretakenintoaccountinanaturalwaythroughtheboundaryconditionsasdefined inEq. (7) byincludingtheloadsappliedonthefreesurfaceofthefluidamongtheexternalforcesoftheproblem.Insuch anapproach,iftheMarangonieffectcanbeintroducedinastraightforwardway,theimplementationofthecurvatureeffect is farmore complicated, asit requires the preliminarycalculation of themean curvature (traceof thecurvature tensor) ofthefreesurface [31,32].Butthepositionofthefreesurface isgivenbythefluid motion,andthecurvaturecalculation requiresthe evaluationofthe spatial secondderivatives ofthedisplacement field. Thiscalculation canbe achievedquite accuratelyusingC1elements,buttheseelementsareveryheavyandtime-consuming.WhenclassicalC0elementsareused, anotherwayconsistsincalculatingthe divergenceoftheunit normaltothe freesurface, andthespatial firstderivatives ofthe shapefunctions[46,47].Butthe calculationofthe unit normal,generallyobtainedby averaging thenormalofthe surfaceelementsconnectedtoeachnodeconsidered,caninducesignificanterrorsintheresults.

Inthisstudy,we usetheefficientapproachsuggestedin[27,44,45] forthecurvatureeffectandgeneralizedin[41] for both the curvature andMarangoni effects. The approach is basedon the equivalence betweenthe effects of the surface tensionandthoseofafictitiousstressedmembranebondedtothefreesurface.Thesurfacetensionisthereforetakeninto accountthroughasurfacehydrostaticstateofstress

τ

=

γ

0 0

γ

,where

γ

isthesurfacetension,appliedtomembrane-type (skin)elements (Fig.3)thatdonothaveanymaterialproperties[41].Thetransmissionofthesurfacetensioneffectsfrom theskinelementstothe3Dfluidelementsiscarriedoutnaturallythroughthecommonnodesofthe2Dand3Delements.

Theweak formulationofthefluidproblemisnowmodified by addingthe term−

∂ΩfreeDs:

τ

ds,where∂Ωfree isthe freesurfaceofthefluid,intheleft-handsideofthefirstofEq. (19),whichisnowwritten:

Ω vs

ρ

vs

vts

t

+

grad

(

vs

+

vb

) ·

vt

dv

Ω

Ds

:

2

μ

Dmsdv

+

Ω div

vs

pdv

∂Ωfree

Ds

: τ

ds

+

Ω

vs

·

fvdv

+

∂Ωϕ

vs

· ϕ

dds

=

0

(26)

The proposedapproach doesnotrequirecalculating themeancurvatureofthe freesurface, whichmakes it efficientand accurate.Inaddition,itispossibletoaccountforbothnormalandtangentialeffectsinanaturalandsimplewaybysimply consideringtemperature-dependentvaluesof

γ

.Itissufficienttoconsiderthecorrectvalueofthesurfacetensionateach Gausspointofthemembraneelements.Thisallowsustoavoidthenumericalproblemsofclassicalapproaches.

2.4. Modelingthefreesurfaceevolution

Totake the free surface evolution into account, differentmethods havebeen proposed. Combined withfullyEulerian formulations,thevolumeoffluid[34] orlevelset[35] techniquesaregenerallyused.Alternatively,theArbitraryLagrangian–

Eulerian(ALE)approachconsistsinimposingthefreesurfacemotiontothemesh[38–40].Butinordertokeepagoodmesh quality, thenodesinside theweldpool mustinturnbe repositioned, thusdefining ameshmotionwhich mustbe taken intoaccountinthefiniteelementformulation.

The algorithm used in the presentwork is summarized in Fig. 4.The first stepconsists in recovering the mesh and the results for a given time step of the thermal-fluid computation. Here, a strong coupling between fluid andthermal

(9)

Fig. 3.Mesh containing skin elements (red 1D elements with sky blue nodes).

Fig. 4.Proposed algorithm of the ALE approach.

computations isused.Then thestatus ofeachnode (freesurface, fluid, solid)isdefined. Thefree-surfacenodes arethen displaced inordertofollowthefreesurface motion.The fluidnodesinsidethemoltenpoolare finallyrepositioned using anelasticcalculationwherethesolidnodesremainfixed.

TheelementshavingameantemperatureabovethemeltingtemperatureTliqareconsideredasfluidelements,theothers as solid elements.For each node,depending on its positionandthe statusof theelements it isconnected to,four-node statusescanbedefined:

– solidnode(rednode):connectedtoatleastonesolidelement;

– solid/free-surfacenode(purplenode)–solidnodeconnectedtooneoftheskinelementsdefiningthefreesurface;

– fluidnode(greennode):connectedonlytofluidelements;

– fluid/free-surfacenode(skybluenode):fluidnodeconnectedtooneoftheskinelementsdefiningthefreesurface.

Thenormalatafree-surfacenodeisdefinedasthemeanvalueofthenormaloftheskinelementscontainingit(Fig.5).

The normalvelocitiesandthedisplacements ofthefree-surfacenodesare thencalculated usingthefollowingexpres- sions:

vnk

(

t

+

t

) =

nk

·

vk

(

t

+

t

)

·

nk (27)

(10)

Fig. 5.Definition of the normal at a node of the free surface and projection of material velocity.

Fig. 6.The mesh before updating the node positions.

Fig. 7.The mesh after updating the node positions.

unk

(

t

+

t

) =

unk

(

t

) +

vnk

(

t

+

t

) ·

t (28)

Here vnk and unk are respectivelythe normalvelocity andthe normaldisplacementof thefree-surfacenode k,vk isthe velocityofthisnode,andnkisitsnormal(Fig.6).

Once the displacements of the free-surface nodeshave been calculated, differentmethods can be used to reposition theinternalfluidnodes.Forexample,Heuzéetal.[48] usedabarycentricreplacement.Inthepresentapproach,anelastic computation is carried out inorder to findthe arbitrary displacements ofthe fluid nodes.In thiscomputation, the dis- placementsofthesolidnodesandthefree-surfacenodesareprescribed.Finally,themeshisupdatedusingthecomputed displacements(Fig.7).

Allthesestepsareperformedateachtimestepofthethermal-fluidcomputation.

Remark. For the sake of simplicity, the mesh velocity has not been included in the presentationof the finite element formulationabove.Butofcourse,themeshvelocityistakenintoaccountinalltheadvectiontermssuchasgrad(vs+vb)·vt inEq. (19),whichisnowwrittengrad(vs+vb)·(vtvtmesh),wherevtmeshisthemeshvelocity.

(11)

Fig. 8.Schematization and mesh of the sloshing problem.

3. Validationtests 3.1. Sloshingproblem

The sloshingproblemwasdevelopedbyRamaswamyetal.[49] tovalidate anALEformulation.Itconsistsinfollowing the oscillations of a liquid ina container. Sloshing was initially considered to be dueonly to gravity. This problemwas generalizedbyother authorsbyaccountingformorephysicalphenomena.Forexample,Popinetetal.[50] havesimulated in2Dtheoscillationsproducedbythesurfacetension.Theyhavecomparedtheirnumericalresultswiththeanalyticresults givenbyProsperettietal.[51].TheproblemwasconsideredlaterbyEl-Sayedetal.[25] usinga3Dmesh.

Inthiswork, thesloshingproblemwas usedtovalidatethe ALEapproachofthefree surfaceandthenormaleffectof thesurfacetension.Theaimofthenumericalmodelistofollowtheoscillationsofasingle liquidinacontainerthathasa squaregeometry.Initially,thefreesurfaceoftheliquidhasasinusoidalshapegivenbythefollowingequation:

H

(

x

) =

1

+

a0sin

π (

0

.

5

x

)

(29) wherea0=0.01 istheinitialamplitude.Inthisequationalldistancesareexpressedinm.

Withregardtotheboundaryconditions,thebottomsurfaceisclampedandthenormalvelocitiesaresuchthatv·n=0 on the side walls. The free surface is thetop one. The mesh (Fig. 8) contains 4223nodesand 19,736 linear tetrahedral elements. Thefreesurface ismeshed with1002Dtriangular“skin”elements (redelements inFig. 8)toapplythesurface tension.The liquidpropertiesare: relative density

ρ

=1 kg·m3,kinematicviscosity

η

=0.01 m2·s1,andgravity g= 1m·s2 (artificialvalue).

Afirstsimulationiscarriedouttofollowthesloshingofaliquidundertheeffectofitssoleweight(gravityeffect).The resultsofthissimulationandtheanalyticalsolution[51] areprovidedbyFig.9.

Regarding the other simulation,the sloshingof the liquid is generated by its weight as well asthe surface tension.

Values ofthesurface tensionbetween0.001and1 N·m1 areused.The resultsshowproportionalitybetweenthevalue of the surface tensionand the intensityof liquid sloshing(Fig. 10). The comparisonbetween the numericalresults and the analytical solution shows a good correlation:the maximal error is lessthan 0.006%. We note that the value of this erroris verysimilarto thosefoundinthe literature.Forexample,Rabieretal.[32] havefound amaximal errorequalto 0.005%.However,theymeshed withadiscretizationtwicefinerthanours.Inadditiontothat,theyusedatimestepequal to0.0025 s,thatis 18 ofthevalueusedinoursimulation(0.1 s).

3.2. Benchmarkproblem

TheBenchmarkconsistsofsomenumericalsimulationofthespotwelding“GTWA”process(Fig.11).Itwasdevelopedin a2DaxisymmetricoptionbyGirardetal.[52],ina3DoptionbyHamideetal.[53] andEl-Sayedetal.[25] tovalidatethe tangentialeffectofthesurfacetension.Inthiswork,apurelythermal-fluidcomputationofthisproblemwascarriedouton the quarterofa disk(Fig.11).A viscositydependingonthe temperatureallowsseparatingthefluidandsolid zones.The solidisthusschematizedasafluidwithahighviscosity(

μ

=106kg·m1·s1).Themeshincludes68,130lineartetrahedral elements,762skinelementsand13,617nodes.Thelengthoftheelementsedgesis30 μminthefluidzoneandmorethan 100 μminthesolidzone.

The material used issteel, supposed to be isotropic, homogeneous, Newtonian, thermally dilatable,and mechanically incompressible.ItspropertiesaregiveninTable1.

(12)

Fig. 9.The displacement of the free surface under the effect of the liquid weight.

Fig. 10.The displacement of the free surface under the effect of the liquid weight and the surface tension;γ=0.001 N·m1andγ=1 N·m1.

Fig. 11.Schematization [52] and mesh of the benchmark problem.

Withregardtotheboundaryconditions,thenodevelocitiesofthefreesurfaceandsymmetricboundariesaresuchthat v·n=0.Theheatfluximposedontheuppersurfaceofthediskisthesumofthreeterms:

total

= ∅

source

− ∅

convection

− ∅

radiation (30)

where:

source

(

x

) =

P 2

π

R20exp

R2 2R20

(31)

(13)

Table 1

Characteristicsofthematerial.

Solidus temperature TS=1696 K

Liquidus temperature TL=1740 K Relative density (solid) ρ=7500 kg·m3 Specific heat (solid) c=602 J·K1·kg1 Thermal conductivity (solid) λ=24 W·m1·K1 Relative density (liquid) ρ=6350 kg·m3 Specific heat (liquid) c=695 J·K1·kg1 Thermal conductivity (liquid) λ=20 W·m1·K1 Dynamic viscosity μ=2.5·103kg·m1·s1 Coefficient of expansion β=104K1

Fig. 12.Temperature evolution with positive derivative of surface tension,∂γT >0.

withR0=3·103m and P=790 W.

radiation

= σ

0

T4

T04

(32)

with

σ

0 theStefan–Boltzmannconstant, theemissivityequalto0.5,andT0=300 K.

convection

=

h

(

T

T0

)

(33)

withhthetransfercoefficientequalto15 W·m2·K1.

Thelateraledgesofthediskareadiabatic,andalineartemperaturedecreasefrom800to400 Kisenforcedonitslower side,fromthecentertotheedge(Fig.11).

Duringthissimulation,onlythetangentialeffectofthesurfacetensionistakenintoaccount,withoutthenormaleffect.

Twovalues ofthederivative ofthesurface tension, γ

T =104 N·m1·K1 and γT = −104 N·m1·K1, areimposed on the skinelements ofthe upperface ofthedisk. The firstthree secondsof theprocess are simulated,andthe time step is 0.1s during theformationoftheweldpool (thermalcomputation)and2·104 s after theactivationofthefluid flow (thermal-fluidcomputation).

A comparisonbetweenthe Benchmarkandournumericalresultsiscarriedout. Itconsistsinexamining thefollowing quantities:

– temperatureontheradiusoftheupperfaceT(z=0). – radialvelocity vr(z=0).

Figs. 12and13showrespectivelytheevolutionofthetemperaturefield andthevelocityvector withapositivederivative of the surface tension with respect to temperature. Figs. 14 and 15 illustrate the caseof a negative derivative. A good correlation betweenourresultsandthebenchmark resultsisfoundin bothcases.Withregard tothebenchmark results, weusetheaveragevalueoftheresultsgivenbyfivecomputation codes[25,53].Amaximalerrorlessthan2%isfoundfor thetemperature. Regardingthevelocity, amaximalerrorof10%isfound.Themagnitudesofthetemperatureandvelocity arealsocheckedonthesymmetryaxisandhereagain,agoodcorrelationwithprevioussimulationsisfound.

(14)

Fig. 13.Velocity evolution with positive derivative of surface tension,∂γT>0.

Fig. 14.Temperature evolution with negative derivative of surface tension, ∂γT<0.

Fig. 15.Velocity evolution with negative derivative of surface tension, ∂γT<0.

(15)

Fig. 16.Finite-element model: mesh.

4. Numericalsimulationofthehybridweldingprocess 4.1. FEmodel

In thissection,a 3D thermal-fluidsimulationofhybridlaser/arc welding ispresentedusingthe SYSWELD® code [54].

It takesintoaccount thefluid flowthrough thesurface tensionandthebuoyancy.Both equivalent doubleellipsoidaland Gaussian surfaceheatsources[55] areappliedintheY directiontoametalblock 30 mminlength,10 mminwidth,and 20 mm inheight. Themodelhasa symmetryabouttheY Z plane(Fig. 16).Itisnotedthatthemeshistakensufficiently largeinorderthatthemoltenpooldoesnotreachtheboundaryofthedomaincomputation.Thismakesitpossibletoavoid theproblemsrelatedtothedefinitionofthenormalatnodesofthemeshboundaries.Onlyhalfofthestructureismodeled, andasingle-passhybridweldisconsideredwithaweldingspeedequaltovwelding=6mm·s1.Themesh(Fig.16)includes 77,603 lineartetrahedralelements(P1+/P1),10,550lineartriangularelements (skinelements),and14,922 nodes.Inthe fluid zone, the elementsize is 500μm,and itcan reach 5000μmin the solidzone. The multiphasic ferritic steelgrade S355J2G3isused.Allthedatapertainingtothismaterialaregivenin[25].

Tostudytheeffectofthefluidflowonthetemperaturefieldandtheweldpoolmorphology,twosimulationsarecarried out. The first simulationis a thermal-fluid computation (with fluid flow). Here, the heating time is 1.4 s, including 1s for thecreationof themolten pool(thermal computation withstaticheatsources andwithoutfluid flow), and0.4s for the thermal-fluidsimulation (with fluid flow)with the motion of heat sources. The time step is equal to 0.1s for the creationofthemoltenpooland2.104 s aftertheactivationofthefluidflow.Atemperature-dependentsurfacetension

γ

=

γ

L+

α

(TTL) with

γ

L=0.15N·m1, TL=1500C,and

α

=104 N·m1·k1 is consideredto includeboth the normalandMarangonieffects.Thesecondsimulationisapurelythermalsimulation(withoutfluidflow)during1.4s.

In both simulations, heat exchangesdue toconvection andradiation on the boundarywith theexternal medium are takenintoaccount:

∅ = ∅

convection

+ ∅

radiation (34)

with

convection

=

h

(

T

T0

)

(35)

radiation

= σ

0

T4

T04

(36) whereh=40W·m2·K1isthetransfercoefficient,T0=293 Ktheroomtemperature,

σ

0 theStefan–Boltzmannconstant, and =0.7 designatestheemissivity.Theseconditionsareenforcedonallmeshfacesexceptforthesymmetryplane(Y Z), whereaninsulatingconditionisimposed.

4.2. Resultsanddiscussion

The thermal-fluidsimulation gives the distribution of the thermal field and fluid flow during the welding processes.

In thissimulation, the solid isschematized asa fluid witha high dynamic viscosity

μ

=106 kg·m1·s1.The fluid is supposed to be Newtonian, with a dynamic viscosity

μ

=2.103 kg·m1·s1. The fluid flow is taken intoaccount throughthesurfacetension(normalandtangentialeffects)andthebuoyancy.Thefluidflowisactivatedatt=1s,thetime necessarytoformasufficientquantityoffluidintheweldpool.

Inthissection,acomparisonoftheresultsofthepurely thermalandthethermal-fluidsimulationsare given(Fig. 17).

Then, some illustrations oftheresults ofthe thermal-fluidsimulation aregiven(Figs. 18–21). Fig. 17illustrates thefield

(16)

Fig. 17.Temperature field att=1.12 s (YZplane).

Fig. 18.Temperature field, and velocity vector of the fluid flow att=1.02 s (YZplane).

temperatureforpurelythermalandthermal-fluidcomputationsatt=1.12 s.Figs.18and19showrespectivelythetemper- aturefieldandthevelocityfieldofthefluidflowattwoinstants(t=1.02 s,t=1.13 s),thefirstwhentheheatsourcesare fixed,andthesecond duringthemotionoftheheat sources.Figs.20and21illustratetheevolutionofthefreesurfacein theweldpoolatthesametimes(t=1.02 s,t=1.13 s).Here,thedeformationofthefreesurfaceisinducedbythesurface tensionandthebuoyancyinthemoltenpool.

Theeffectofthefluidflowonthetemperaturefieldisverysignificant(Fig.17).Withregardtothethermal-fluidresults, we noteasignificant temperaturedrop.Thisisexplainedby thepositivederivative ofthesurfacetensionwithrespectto temperaturewhichgeneratesa penetratingweldpool,bringingthecoldmaterialfromtheedgetowardsthecenterofthe weldpool.

Duringthethermal-fluidsimulation,anaveragevelocityapproximatelyequalto0.3m·s1 isfound.Thisvalueisconsis- tentwithliteraturevalues[56,57],whereasamaximalvalueofvelocityequalto0.6m·s1 isnotedwhentheheatsources begintomove.Thisincreaseofvelocityislinkedtothehigh-temperaturegradient.Indeed,thegradientofthesurfaceten- sionisproportionaltothetemperaturegradient:therefore,whenthetemperaturegradientincreases,thevelocityincreases.

Also, thehigher γ

T, thehigher theMarangoni numberand thevelocity ofthe fluid flow. Thetwo effects ofthe surface tensionwith apositive γ

T are clearly visible inFigs. 18and19.These figures show that thefluid flow movesfromthe edgeofthemoltenzone toitscenter.Thus,a penetratingvortexisproduced.Thisvortexisresponsibleforthe changein theweldpoolmorphology.Tables 2and3showtheweldpoolmorphologyatdifferenttimesofthe“purely”thermaland thermal-fluidcomputations.Maximal gaps equalto15% forthelength, 9.6%forthe heightand1.6% forthewidthofthe weldpoolarefound.Apositive γ

T givesadeeperandnarrowerweldpool.Thiseffectofthefluidflowonthetemperature

(17)

Fig. 19.Temperature field, and velocity vector of the fluid flow att=1.13 s (YZplane).

Fig. 20.Vertical displacement (Uz) of the free surface att=1.02 s (YZplane).

fieldandtheweldpoolmorphologycaninfluencethemicrostructureandthemechanicalcomputations.Duringasimulation takingintoaccount onlythebuoyancyeffect(without thetensionsurface effects), amaximal valueequalto 0.03 m·s1 isfound.Therefore,itisabouttentimessmallerthanthatgeneratedbythesurfacetension.Thisshowsthatthefluidflow linkedtothesurfacetensionspeciallythetangentialeffect“Marangonieffect”isdominantinthiskindofprocess.Regarding thefreesurface,themaximalvalueoftheverticaldisplacementis0.5 mm,whentheheatsourcesarefixed(Fig.20),andit reaches0.8 mmduringthemotionofthesources(Fig.21).Thesevaluesaresimilartotheliteratureresults[18,58].

(18)

Fig. 21.Vertical displacement (Uz) of the free surface att=1.13 s (YZplane).

Table 2

Weldpoolmorphologyatt=1.05 s.

[mm] Thermal-fluid simulation Purely thermal simulation Gap

Length 10.85 10.88 0.3%

Width 4.78 4.86 1.6%

Height 2 1.93 3.5%

Table 3

Weldpoolmorphologyatt=1.4 s.

[mm] Thermal-fluid simulation Purely thermal simulation Gap

Length 13.7 16.14 15%

Width 4.78 4.86 1.6%

Height 3 2.71 9.6%

5. Conclusion

Inordertostudy theinteractionbetweenthefluidflow andthesolid deformationduring thenumericalsimulationof theweldingprocess,a newtransient formulationforthenumericalmodeling ofaweldpool wasdeveloped.Itconsistsin consideringthe fluidflowinthe weldpoolthroughthe surfacetensionandthebuoyancy.Inthiscontext,anewmethod was usedto simulatethesurface tension.Itincorporates boththe normalforce exertedonto thefree surface (“curvature effect”)andthetangential effect(“Marangonieffect”)inaverysimpleandefficientwaycomparedwithclassicalmethods.

However,takingintoaccountthenormaleffectofthesurfacetensionstill requiresthesimulationofthefreesurface.That isthereasonwhya newALEapproachwas developedformodeling theevolution ofthefree surface.Thismethodallows takingintoaccountthesoliddeformationinasimpleway,whichisnotthecaseforEulerianmethods.

This work was validated using two numerical examples. The first one is the sloshing test; it consists in simulating the oscillationsofa liquidin acontainer, andit allowed validatingthenormal effectofthe surface tensionandtheALE approach.Thesecondoneisthebenchmarkproblem,itrepresentsthenumericalsimulationofspotwelding“GTWA”,and itallowedvalidatingthetangentialeffectofthesurfacetension.

After validation, the 3D thermal-fluid simulation ofa hybrid laser/arc was carried out. It gives good resultsin terms of the velocity ofthe fluid flow andthe evolution of the free surface compared to the literature results. Tounderstand theeffectofthefluid flowduring thesimulation ofhybridwelding,acomparisonofresultsbetweenpurely thermaland thermal-fluidsimulationswas performed.Itshowsthesignificantinfluenceofthefluidflowonthetemperaturefieldand the morphology of the weld pool. Forexample, we note a significant temperaturedrop in thethermal-fluid simulation.

Thismaybeexplainedbythepositivederivative ofthesurfacetensionwithrespecttotemperature,whichbringsthecold

Références

Documents relatifs

In 2017, UMONS, along with UCLouvain and Multitel, initiated the so-called “Deep Learning Academy” with the aim of organising quarterly gatherings for dozens of researchers

We assume that the Lamé constants ofthe material constituting the plate vary as e~ 3 , the density varies as e~ , and the Lamé constants and density of the material constituting the

Abstract — In this paper, we study a simplified mathematical model that descnbes the stationary dx ^placements of a sohd body immersed in a transverse flow This model involves

We construct 2-surfaces of prescribed mean curvature in 3- manifolds carrying asymptotically flat initial data for an isolated gravitating system with rather general decay

Ajouter une question à cocher de votre choix et faites en sorte que la case &#34;Vous avez choisi la spécialité NSI soit cochée par défaut&#34;. Intro HTML / CSS : Travail sur

We expect that these two families (conic transforms and conic transforms of a line transform) should yield all Fano threefolds with same intermediate Jacobians as X, thereby

The immediate aim of this project is to examine across cultures and religions the facets comprising the spirituality, religiousness and personal beliefs domain of quality of life

It is very important that your meeting addresses the multisectoral aspect of viral hepatitis control and prevention, and that the national focal points seize the opportunity and