The numerical challenges in multiphysical modeling of laser welding with

arbitrary Lagrangian-Eulerian method

I. Tomashchuk 1 , I. Bendaoud2, , J-M. Jouvard 1, P. Sallamand 1
1. Laboratoire Interdisciplinaire Carnot de Bourgogne – UMR CNRS 6303, Université de Bourgogne –
Franche-Comté, Le Creusot, France
2. Laboratoire de Mécanique et Génie Civil, Université de Montpellier, Montpellier, France

Abstract: The interaction of high power laser beam apprroximation5) or by introducing temperature-
with metallic materials produces a number of dependent fictive viscosity (for example 100 Pa·s 3 or
interconnected phenomena that represent a serious 1000 Pa·s 2). The previous models differ in a way to
challenge for numerical modeling, especially for introduce the energy distribution and adsorption
creation of auto-consistent models. Additional coefficient, which is not perfectly known and depends
difficulty consists in lack of data on materials not only on the aggregate state of irradiated matter, but
properties at the temperatures superior to their melting also on local free surface curvature. Another difficulty
point. The present work summarizes the numerical consists in the lack of data about exact values recoil
challenges in creation and validation of free-surface pressure due to the partial condensation of metallic
models using ALE moving mesh coupled with heat vapor in the keyhole and about the variation of
transfer equation and Navier-Stokes fluid flow. materials properties in the liquid phase.
The motivation for the present work was determined
Keywords: laser welding, metallic alloys, moving by the interest to represent both keyhole formation and
mesh, multiphysics. collapse in a fully auto-consistent 3D model, for
further application to multimaterial welding. The
1. Introduction modelling in 3D accentuated the convergence
problems, mainly because of the limitation of minimal
mesh size. Our previous model of dissimilar welding2
ALE method has been successfully used for
suffered from noticeable lack of weld penetration,
multiphysical modelling of pulsed1,2 and continous2
compared with experimental results. Moreover, the
laser welding, drilling3, direct laser metal deposition4
influence of many numerical (inconsistent
and thermal matter ablation5. The displacement of
stabilization for Navier-Stokes equation, type of ALE
mesh in these models is based on the velocity field
smoothing, type of used Navier-Stokes equation) and
calculated with Navier-Stokes equation strongly
physical parameters (dynamic viscosity, adsorption
coupled with heat equation1-4 or on the property-
coefficient, and condensation coefficient) on the
depending condition5. 2D axisymmetric models with
convergence and on the resulting behavior of the
µ-metric mesh size allowed an accurate description of
impact zone was not fully comprehended. The present
the keyhole formation1,3. The main limitation of ALE
work attempts to bring some clarity about the
consists in the impossibility to represent the complex
influence of these factors on the simple example of
modifications of topology, like the formation of
standalone laser pulse applied to the plate of Ti6Al4V
bubbles or porosities; however, it provided quite
alloy.
accurate description of free surface movement and
good agreement with experimental dimensions of the
impact zone. The liquid phase is modelled as an 2. Governing Equations
incompressible2,3 or weakly compressible1 Newtonian
liquid that undergoes laminar flow under the influence 2.1 Heat transfer
of Marangoni convection, natural convection and
recoil pressure in the keyhole, produced by the Heat transfer equation was used in time-dependent
metallic vapor. The recoil pressure that is represented form:
by Clausius-Clapeyron equation or in form of adjusted  T
 c p eq 
   
 U  T     T   (1)
 t
spatial function is an important factor that determines
the depth of the keyhole. To suppress the parasite

displacement in the solid part of modelled domain, two Laser energy supply is represented by Gauss heat
approaches exist: to stop the displacement by applying source applied at the top of butt joint. Pulsed laser
an important frictional dissipation depending on the beam is represented as follows:
local fraction of liquid phase (Carman-Kozeny

Excerpt from the Proceedings of the 2018 COMSOL Conference in Lausanne

  x 2  y 2  2.3 Free surface
P A  
 t  t pulse  ,
r02 
qL  L 2 e  (2)
 r0 Present study uses ALE moving mesh approach for
where PL – laser power, r0 – beam radius and t pulse – representing the movements of free surface. The
pulse duration A - absorption coefficient of laser advantage of this method consists in possibility to
radiation by each individual material, represented as : neglect the vapor phase and thus lighten the
A = Asolid +(Aliquid -Asolid)  flc2hs(T-Tm, ∆T), (3) calculation. The weak point consists in limitations of
where Asolid = 0.4 and the absorption coefficient in free surface movement: the formation of droplets or
liquid phase increases when the surface temperature bubbles in the melted zone is forbidden by the
nears vaporization temperature: impossibility to break or unite individual meshes.
Aliquid = Asurf+(Akh-Asurf) flc2hs(T-Tv, ∆T), (4) Movement of free surface is determined by the
where Asurf = 0.25 and Akh = 0.8. velocity field in the melted zone.
Latent heat of fusion and evaporation are taken in
account by means of equivalent enthalpy approach6: 2.4. Numerical implementation

p  c p  Dm .Lm
c eq (5) Present model involves Heat Transfer in Fluids,
where Cp is heat capacity as function of temperature, Laminar flow and Moving Mesh modules that are
Lf – latent heat of fusion and Dm - Gauss function strongly coupled and solved with time-depended
normalized around melting temperature Tm: solver. Multiphysics coupling is ensured by the
 T Tm 2 interrelation between heat transfer and Navier-Stokes
T 2 equations: velocity field in heat equation is provided
e
Dm  (6)
by resolution of Navier-Stokes equation, when
 T 2 temperature field defines thermophysical properties of
where ∆T is smoothing interval of 100 K. the liquid involved in convection process. Consistent
Materials properties of Ti6Al4V are considered as (streamline and crosswind) stabilization and
interpolation functions of temperature7. inconsistent stabilization (isotropic diffusion) were
applied to Navier-Stokes equation. Consistent
2.2 Fluid flow stabilization was also applied to the heat transfer
equation. Moving Mesh solves surface deformation in
Liquid metal is assumed as incompressible function of locally calculated velocity field.
Newtonian liquid that undergoes laminar flow: Only one plate (3mm x 2mm x 2mm) is modelled
 U
     


 
 t


 U   U      pI   (T )U    U     F (7)
for symmetry reasons. Tetrahedral mesh of 250 µm
 t    maximal size is applied over calculation domain. The
 top surface of is meshed with 50 µm.
 U  0
Metal that has T<Tm is considered as highly The models were solved on work station with Intel
viscous fluid, that practically stops all convection 2.20 (2 processors, 44 cores) and 256 Gb RAM.
movements in solid materials. The transition between
solid and liquid material is provided with smoothed 3. Experimental study
Heaviside function:
(T) = solid +(liquid-solid) flc2hs(T-Tm, ∆T). (8) Welding experiments were carried out with pulsed
Convection movements are generated by Nd:YAG laser of 3 kW maximal power and focused
following phenomena: beam diameter () of 560 µm.
 natural convection described through Boussinesq Laser beam was focused on the surface of butt joint
approximation, between 2 mm thick Ti6Al4V plates. Standalone
 surface tension force, pulses with duration of 4 and 6 ms and power 1.5 kW
 Marangoni convection with  <0 introduced in were realized. Three standalone pulses per condition
were made.
form of weak formulation, The penetration of pulses at the joint line was
 recoil pressure of vapor filling the keyhole, measured by breaking the weld and observing
represented in form of function Ti6Al4V side of joint plane with optical microscope.
pr 
1     a  e  Tb c (9)
2 4. Discussion
where the condensation coefficient β = 0.5 and the
coefficients a,b, c for Ti6Al4V are given by Kaidalov8. 4.1 Comparison between the different methods of
mesh smoothing

Excerpt from the Proceedings of the 2018 COMSOL Conference in Lausanne

The main challenge in use of ALE for modelling of 4.2 The influence of numerical stabilization of
keyhole welding consists in rapid surface deformation, Navier-Stokes equation
which is associated with quite severe elongation of
meshes. In the first place, the performances of In absence of inconsistent Navier-Stokes stabilization,
different smoothing methods were compared (Figure the model is likely to diverge at the moments
1, Table 1). The Laplace smoothing (Figure 1.a) led to corresponding to sudden rise of velocity field, as, for
the rapid divergence after interaction time of only 1.5 example, during the collapse of the keyhole at the end
ms, associated with quite high distortion of the of the pulse. In general way, the increase of id value
elements. The Yeoh smoothing (Figure 1.b) allowed to promotes the smoothing of velocity field and better
extend the calculation up to only 3.2 ms, followed by convergence, but the excessive values can alter the
divergence, in spite of low estimated distortion of the observed phenomenology. The values of id of 1
elements. Parasite ripple-like surface deformation was (default) and 0.5 were compared (Table 2). The
observed in the solid around the melted zone. The calculations were performed with mean viscosity of
Winslow smoothing (Figure 1.c) allowed to reach the solid of 100 Pas. It can be concluded that melt depth
end of laser pulse (6 ms), however the aspect of free is not at all sensitive to the reduction of id value, and
surface is very rough and the width of the keyhole is makes small effect on melted zone width (W = -124
highly overestimated. Finally, the use of hyperelastic µm), maximal melt temperature (T = -16K) and
smoothing (Figure 1.d) allowed the most accurate
velocity (U = +0.2 m/s). The default value of id=1
keyhole representation, associated with quite low
was conserved for the rest of the calculations.
distortion of the elements. This formulation allows
attaining the maximal deformation of free surface, and Table 2: The effect of inconsistent stabilization
thus is the most appropriate for this kind of simulation, MZ (µm)
in accordance with conclusion of Bruyere et al.1 id Depth Width Tmax (K) Umax (m/s)
However, the important deformation of the meshes 1 611 1454 3380 1.6
situated at the center of laser pulse leads to the loss of 0.5 611 1330 3364 1.8
mesh quality at the bottom of the keyhole. These
preliminary calculations were performed with 4.3 The effect of viscosity formulations
inconsistent Navier-Stokes stabilization id = 1 and
mean viscosity of solid of 100 Pas. Another important numerical problem is the parasite
velocity field in the solid material. It was observed that
Table 1: The comparison of ALE smoothing methods the mean viscosity of solid of 100 Pas produces
Element zKHa unphysical relaxation of already solidified matter in
Smoothing Convergence
distortion (µm) the ring formed around the zone of the impact (Figure
Laplace No 12.84 120 2). Cutting off the velocity field in the solid area using
Yeoh No 1.10 301 temperature or viscosity-based condition in ALE
Winslow Yes 7.20 477 module is not enough, because, as it was found, the
Hyperelastic Yes 3.2 532 fictive viscosity of the solid makes important influence
a maximal keyhole depth
on interaction phenomenology. An insufficient value
T (surface) + velocity field (arrows) T (K) of solid viscosity makes an important influence on
Laplace Yeoh temperature evolution of the melt and fluid dynamics.
The evolution of the impact zone under standalone
laser pulse can be divided in four major stages: the
initiation of the keyhole (Figure 3.a), the progression
t = 1.5 ms t = 3.2 ms of the keyhole (Figure 3.b), the keyhole collapse
(a) (b) (Figure 3.c) and the solidification resulting in final
Winslow Hyperelastic surface profile with the ring around the impact zone
and slight depression in its center (Figure 3.d).
The main characteristics of the process and the
evolution of maximal temperature, velocity field and
keyhole depth for solid viscosity values of 100 and 200
t = 6 ms t = 6 ms Pas are illustrated in Table 3 and Figure 4
(c) (d) respectively. The increase of solid viscosity has
important impact on calculation time. Moreover,
Figure 1. The comparison of ALE smoothing methods higher values of solid viscosity (300 Pa·s) leads to
applied to the model of pulsed welding of Ti6Al4V (tpulse = the convergence problems.
6 ms, =560 µm, PL= 1.5 kW).

Excerpt from the Proceedings of the 2018 COMSOL Conference in Lausanne

 Relaxation of solidified ring Table 3: The effect of mean viscosity of the solid
Calculation
Relaxation of melted zone center solid MZ (µm) Tmax Umax time (h)
(Pas) (K) (m/s)
Depth Width
100 611 1454 3380 1.6 4h57
200 828 1324 4480 7.5 72h24
Hf a 698 1450 4870 10.0 144h52
(a) a Heaviside function , equation 10

5000 keyhole Viscosity of the solid

initiation
4000 keyhole
200 Pa*s
collapse 100 Pa*s

Tmax (K)
3000 end of solidification
(b) 2000
Tm
Figure 2. Unphysical relaxation of solidified weld between
t = 14 ms (end of solidification) and t = 20 ms (wireframe) 1000
end of the pulse (6 ms)
for solid viscosity of (a)100 Pas and (b) 200 Pas.
0
0 0.005 0.01 0.015 0.02
The main difference between the cases with solid of t (s)
100 Pas and of 200 Pas consists in the intensity of
keyhole initiation process. The application of high 8
keyhole Viscosity of the solid
viscosity of the solid impedes the “parasite” lateral 7 initiation
6 200 Pa*s
evacuation of the heat from impact zone, because the Umax (m/s)
residual convection in solid material is efficiently 5 100 Pa*s
reduced. This results in sudden progression of the 4 keyhole end of the pulse (6 ms)
keyhole depth starting from 1.5 ms with maximal 3 progression
temperature of 4480 K at the center of the impact and 2 keyhole
1 collapse
the velocity of the liquid reaching 7 m/s. The solid
viscosity of 100 Pas does not produce such peak of 0
0 0.002 0.004 0.006 0.008 0.01 0.012
temperature and velocity: during all keyhole life they
remain at the plateau of 3380 K and 1.6 m/s. The t (s)
sudden progression of the keyhole for the case of 200 700 keyhole progression Viscosity of the solid
Pas results in 80 µm more profound keyhole.
600 200 Pa*s
T (surface) + velocity field (arrows) keyhole
500
100 Pa*s
z cap (µm)

initiation
400
keyhole
300 collapse
200
100 end of the pulse
remaining depression
0
0
0.002 0.004 0.006 0.008 0.01 0.012
t (s)
Figure 4. The influence of solid viscosity on the evolution
of maximal temperature, melt velocity and keyhole depth for
standalone Nd:YAG laser pulse on Ti-6Al-4V plate (1,5 kW,
Ø = 560 µm).

Figure 3. Nd:YAG laser pulse on Ti-6Al-4V plate (6 ms, 1.5 Higher solid viscosity also produces more tardive
kW, Ø = 560 µm). solid = 200 Pas. completion of the solidification of the melt: 15.5 ms
compared to 14 ms for 100 Pas. It also modifies the
Two calculations produce the similar evolution of the proportion of the melt, making it more profound and
maximal temperature on the stage of keyhole less large (Table 3). In spite of quite similar keyhole
progression and collapse, however, higher solid depth (530 µm), the case with 200 Pas results in 217
viscosity results in little higher speed of collapsing µm deeper melt, when the thickness of the melted zone
liquid (2.45 m/s compared to 1.55 m/s). is reduced at 130 µm. The comparison with the

Excerpt from the Proceedings of the 2018 COMSOL Conference in Lausanne

average experimental dimensions of the melted zone Table 5: The effect of viscosity of the liquid
speaks in advantage of higher solid viscosity (Table 4). Melted zone Calculation
For both solid values, the depth of the melt is liquid dimensions Tmax Umax time (h)
(Pas) (µm) (K) (m/s)
underestimated, when the width is correct, however
Depth Width
higher solid viscosity results in deeper melt. The
experimental weld was produced on two distinct f(T) a 828 1324 4480 7.5 72h24
plates, when numerically only the condition of 410-3 725 1358 4400 7.5 82h23
a equation 11
symmetry is applied on the joint line, which can
partially explain such high difference between 10 Liquid viscosity
calculated and experimental values. The simplified f(T)
8
representation of absorption in the keyhole and the

Umax (m/s)
4 mPa*s
lack of information about recombination coefficient β 6
influencing the recoil pressure (equation 9), along with 4 Solid viscosity : 200 Pa*s
the difficulties of accurate numerical resolution of 2
keyhole bottom, are likely to be responsible for
underestimation of melt depth. The increase of solid 0
0
0.002 0.004 0.006 0.008 0.01 0.012
viscosity has an immediate effect on calculation time,
t (s)
multiplied roughly by 16 (Table 3). The attempt to
Figure 5. The influence of liquid viscosity on the evolution
apply the gradual decrease of fictive viscosity with of maximal melt velocity for standalone Nd:YAG laser pulse
temperature (Heaviside function, equation 10) can be on Ti-6Al-4V plate (PL= 1.5 kW, Ø = 560 µm).
used to stabilize the numerical solution and produces
the behavior of the melt similar to the use of high (200 4.4 The influence of keyhole absorption coefficient
Pas) solid viscosity, however, very important
calculation time makes this approach unattractive. The absorption coefficient in the keyhole makes a
strong influence on maximal surface temperature, and
solid = cold +(hot-cold) flc2hs(T-900, ∆T) (10) thus on the resulting recoil pressure (equation 9). The
parametric study for Akh values from 0.6 to 0.9 was
Table 4: The comparison with experimental weld carried out. In these calculations, dynamic viscosity
dimensions was considered as function of temperature (equation
Experimental dimensions of the melted zone (µm) 11) for the liquid and 200 Pas for the solid.
Test 1 1055 1286 1269 1553 It was found that the shift from 0.7 to 0.8 produces an
Test 2 1059 1349 1259 1426 important shift in keyhole depth, associated with rise
Test 3 1008 1332 1222 1379
of maximal temperature and velocity field (Table 6).
Average 1041 1322 1250 1453
For Akh  0.7, the first peak in velocity field associated
Relative error (%)
solid with rapid digging of the keyhole is absent, and the
after 4 ms after 6 ms velocities associated with the collapse of the keyhole
(Pas)
Depth Width Width Depth are also lower (Figure 6). To reach the realistic values
100 57.0 10.2 8.0 51.0 of penetration, it is needed to use high adsorption
200 47.0 9.6 9.0 34.0 coefficients 0.9-1, but it this cases the convergence
Hfa 48.6 3.2 0.3 44.2 was not reached for the moment.
a Heaviside function

The effect of the viscosity in the liquid phase also was

analyzed: the temperature-dependent formulation
(equation 11 considered by default in all calculations)
was compared with constant value if T>Tm (Table 5).
As the viscosity of liquid decreases with temperature,
in the first case the digging of the melted zone is
facilitated and consequently the higher depth of the
melt is obtained. However, no important effect was
noticed on the maximal melt velocity or temperature,
as well as on weld width. The behavior of the liquid
zone is quite similar (Figure 5). Figure 6. The influence of keyhole absorption coefficient on
- 4  10 -6  T + 0.0118 if Tm  T  2500 K (11) the evolution of melt velocity for standalone Nd:YAG laser
l   pulse on Ti-6Al-4V plate (1.5 kW, Ø = 560 µm).
 1.5  10 -3 if T  2500 K

Excerpt from the Proceedings of the 2018 COMSOL Conference in Lausanne

 Table 6: The effect of keyhole absorption coefficient 50
MZ (µm) β = 0.5
Akh T max (K) U max (m/s) 40 β = 0.6
Depth Width

pr (bars)
30 β = 0.7
0.6 3445 2.28 587 1308
β = 0.8
0.7 3460 2.28 587 1308 20 β = 0.9
0.8 4480 5.5 828 1324 β =1
0.9 No convergence 10
0
4.5 The influence of condensation coefficient 2000 2500 3000 3500 4000 4500 5000
(a) Tsurf (K)
The condensation coefficient represents the variation
of surface pressure in function of evaporation
conditions (equation 9): from  = 0.17 for strong
evaporation to  = 1 for full thermodynamic
equilibrium9. Strong evaporation diminishes the
surface pressure (Figure 7), which can influence the
keyhole formation. By default, the value of  = 0.5 is
= 0.5 = 1
admitted in our calculations. Lower values of 
resulted in convergence problems. The parametric (b)
Figure 7. The effect of condensation coefficient (a) on the
study was performed for  = 0.5 – 1.
rise of recoil pressure with temperature and (b) on the
It was found that for  > 0.5 the intense initiation of relative pressure field at the end of the pulse ( t= 6 ms).
keyhole formation is absent, which traduces by lower
weld penetration, temperature and melt velocity 4.6 Comparison of incompressible and weakly
(Table 7). Melted zone width is not very sensible to compressible Navier-Stokes equations
the variation of . For  = 0.6 – 1, quite similar depths
of the melted zone, maximal temperatures and Axisymmetric ALE-based model of keyhole digging
maximal velocities of the melt (associated with proposed by Bruyere et al.1 showed slight loss of mass
keyhole collapse) are observed. of the domain. The same problem was noticed in the
present 3D model. In the calculations performed with
Table 7: The effect of condensation coefficient incompressible Navier-Stokes formulation, the
MZ (µm) noticeable loss of mass produces during the creation of
Tmax (K) Umax (m/s) the keyhole (Figure 8). It is proportional to the
 Depth Width
maximal deformation of the domain and took place for
0 No convergence
both considered values of solid viscosity. After the
0.4 No convergence
0.5 828 1324 4480 7.5
collapse of the keyhole, the relaxation of strained
0.6 589 1308 3440 2.34 meshes allows to “recover” some missing mass, but
0.7 594 1312 3420 2.18 not completely, which results in the situation when the
0.8 597 1308 3400 2.04 depression in the center of solidified melt is not fully
0.9 600 1314 3390 2.04 compensated by the height of solidified ring around
1 600 1318 3380 2.04 the melt. Weakly compressible Navier-Stokes
equation offers good mass conservation (Figure 8),
It looks like the increase of  (and so the recoil comparable with result of Bruyere et al.1, however, the
pressure) slows down the digging of the keyhole, convergence of the models is more difficult due to
which seems unphysical. The observation of relative higher temperature and velocity gradients and more
pressures at weld cross-section at the end of the pulse rapid deterioration of element quality. For instance,
(Figure 7.b) shows the zones of maximal pressure at only the association of compressible Navier-Stokes
the bottom of the keyhole, which seems physical, but formulation with Heaviside function for solid
also around the ring of ejected matter, where the viscosity (equation 10) allowed to reach the
displacement of the meshes is blocked by the absence convergence along with good mass conservation.
of velocity field. This second maximum is a numerical The different values of solid viscosity produce
artifact with no physical significance, as well as the different effect in incompressible and weakly
zones of low pressure situated at the solid material compressible Navier-Stokes equations (Table 8). For
beneath. This local increase of pressure may impede solid viscosity of 100 Pas, the reduction of penetration
the ejection of liquid from the keyhole, which explains depth is negligible and the maximal residual
why low melt depth for  > 0.5. depression in the impact center increased by only 20

Excerpt from the Proceedings of the 2018 COMSOL Conference in Lausanne

µm. For higher solid viscosity, the melt becomes  problems. This problem can be treated with
130 µm less profound, and there is no impact on different remesing approaches that increase the
residual depression. time of calculation to the high degree.
5. The condensation coefficient of 0.5 seems the
0.6
most appropriate. Higher values impede the
Mass loss ()

0.5 digging of the keyhole because unphysical

0.4 accumulation of pressure around the ejection ring,
0.3 and lower values produce convergence problems.
6. The shift from incompressible or weakly
0.2
compressible Navier-Stokes formulation does not
0.1 alter the observed phenomenology and allows
0 better mass conservation. However, it is likely to
0 0.005 0.01 0.015 0.02 be more sensible to the convergence problems.
t (s)
200 Pa*s Incompressible References
100 Pa*s Incompressible
200 Pa*s, Compressible
100 Pa*s Compressible 1. V.Bruyere, C. Touvrey, P. Namy, Comparison
Hf, Compressible between Phase Field and ALE methods to model the
Figure 8. Mass conservation compared for incompressible keyhole digging during spot laser welding, COMSOL
and compressible Navier-Stokes formulations. Conference Rotterdam (2013).
2. I. Tomashchuk, I. Bendaoud, P. Sallamand, E.
Table 8: The effect of Navier-Stokes formulation Cicala, S. Lafaye, M. Almuneau. Multiphysical
solid Controlled dimensions (µm) modelling of keyhole formation during dissimilar laser
Navier-Stokes (Pas) Meted zone Impact center
welding, COMSOL Conference Munich (2016).
depth depression
3. S. Sharma, Y. Pachaury, S.N. Akhtar, J. Ramkumar,
Incompressible 100 611 125
200 828 150
A study of hydrodynamics of melt expulsion in pulsed
Compressible 100 596 144 Nd:YAG laser drilling of titanium. COMSOL
Hf a 698 150 Conference Pune (2015).
a Heaviside function 4. S. Morville, M. Carin, M Muller, M. Gharbi, P.
Peyre, D. Carron, P. Le, 2D axial-symmetric model for
Conclusions fluid flow and heat transfer in the melting and
resolidification of a vertical cylinder, COMSOL
1. Hyperelastic deformation seems the most Conference Paris (2010).
appropriate for the modelling keyhole evolution. 5. W. Frei, Modeling Thermal Ablation for Material
2. The velocity field not sensible to the inconsistent Removal, COMSOL blog, March 30, 2016.
stabilization coefficient for Navier-Stokes at the 6. C. Bonacina, G. Comini, A. Fassano, M Primicerio,
range 0.5-1. Numerical solutions of phase change problems, Int. J.
3. Low viscosity of the solid phase not only Heat Mass Transfer, 16, 1825-1832 (1973).
produces parasite relaxation of solidified domain, 7. S. Morville, Modélisation multiphysique du procédé
but can completely alter the phenomenology of de fabrication directe par projection laser, phD thesis,
laser/metal interaction. Thus at least 200 Pa·s Université de Bretagne-Sud, France (2012).
value is recommended. The solution is much less 8. A.A. Kaidalov, Electron beam welding and annexed
sensible to the formulation of viscosity in the technologies (in Russian), Kyiv, Technologia (2004).
liquid phase. However, temperature-dependent 9. R. Fabbro. Quelques processus physiques présents
viscosity promotes the digging of the keyhole. en régime de soudage par keyhole. Ecole thématique
4. Keyhole adsorption coefficient is the key CNRS Laser’ApE, 2-5 October 2017, Nouan-le-
parameter that controls surface temperature and Fuzelier, France.
melt depth. The shift from 70 to 80 adsorption
produces the qualitative change in the dynamics Acknowledgements
of keyhole progression, with much more intense
digging in the beginning of the pulse. Higher This work was carried out as a part of joint laboratory
value would create even more profound weld, and project LabCom FLAMme between Laboratoire
thus reach experimentally observed penetration, Interdisciplinaire Carnot de Bourgogne, University of
but it is associated with too severe mesh Bourgogne-Franche Comté and SME Laser Rhone-
deformation producing the convergence Alpes. This project is funded by French National
Agency of Research.

Excerpt from the Proceedings of the 2018 COMSOL Conference in Lausanne