International Journal of Heat and Mass Transfer 125 (2018) 471–484

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

journal homepage: www.elsevier.com/locate/ijhmt

Study of mass transport in cold wire deposition for Wire Arc Additive
Manufacturing
Fatemeh Hejripour ⇑, Daniel T. Valentine, Daryush K. Aidun
Mechanical and Aeronautical Engineering Department, Clarkson University, 8 Clarkson Ave., Potsdam, NY 13699-5725, United States

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

Article history: Wire Arc Additive Manufacturing (WAAM) is a combination of an electric arc and wire feeding system
Received 22 February 2018 used extensively in building components and repair operations. The heat transfer, fluid flow and mass
Received in revised form 16 April 2018 transport were investigated in a numerical simulation of WAAM process with dissimilar substrate.
Accepted 18 April 2018
Experiments were performed to verify the numerical results. The predicted clad layer (1st layer) profile
Available online 26 April 2018
(width and height) is in good agreement with experiment. The cold wire transfer (CWT) impact on the
velocity field and mass transport were predicted around the cold wire immersion inlet in the weld pool
(WP). The effect of arc travel speed and wire feed rate on the homogenization process were studied. Both
the numerical and experimental results show that the increase of wire feed rate leads to homogenous
composition in fusion zone (FZ). The predicted composition distribution in the clad layer and measured
concentrations in experiments show a well-mixed region in middle of the clad layer.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction
Wire Arc Additive Manufacturing (WAAM) is a technology used
for prototyping by additive layers. In this method, the arc is used as
power source and the wire is the feeding system. WAAM has high
deposition rate, which is used for making large components with-
out material waste. Defect occurrence is unlikely because the wire
is completely melted at the time of deposition [1]. Since the depo-
sition procedure is similar to cladding process, cladding terminol-
ogy can be used for WAAM. There are functioning parameters in
the deposition process which affects mechanical properties. An
efficient process can be achieved by developing a numerical model.
Simulations could be used to optimize the process parameters and
predict the dimensions of deposited layer profile and chemical
composition distribution in the clad layer.
Gas Tungsten Arc (GTA) and Gas Metal Arc (GMA) processes are
employed in WAAM. The advantage of GTA welding (GTAW) is the
production of a smooth and spatter free clad layer. There are two
feeding systems containing hot wire and cold wire. In the hot wire
feeding system, the wire gets heat by electrical resistance heating
before entering the welding arc region. In the Cold Wire Transfer
(CWT) method, the welding arc heats the wire to reach the wire
melting point from the ambient temperature. There are some
experimental and numerical investigations on droplet transfer to
the Weld Zone (WZ). Chen et al. investigated two types of droplet
⇑ Corresponding author.
E-mail address: hejripf@clarkson.edu (F. Hejripour).
transfer modes include touching and free transfer modes in arcing-
wire GTAW [2]. The experiments showed the transfer period and
the droplets size are dependent to the distance from the initial
position of the wire to the Weld Pool (WP), the current on the wire
and the distance from the tungsten electrode to the WP in the
touching transfer. The bridging and globular transfers in GTAW
based additive manufacturing are studied by Geng et al. [3]. They
developed a mathematical model to optimize the wire feed flying
distance in the arc zone. They predicted the droplet landing posi-
tion in bridging transfer under a certain vertical distance of melting
wire tip to tungsten electrode.
Numerous researchers modeled WP behavior by adding wire to
the molten pool using laser as the heat source [4,5,6]. However,
limited studies investigated numerically the fluid flow in the
CWT weld zone in presence of the arc as the heat source. In arc
welding, electromagnetic field produced by the arc, affects the fluid
flow in the WP. Traidia et al. developed a 2D model to compute
Lorentz forces and Joule heating effects [7]. Then, the fluid dynamic
and heat transfer equations were solved under steady state condi-
tions for a 3D model that used the results of the 2D model. They
studied the effect of adding cold filler wire on the velocity field
and molten pool geometry. Pan et al. developed a 3D model of
WP and bead surface in high speed variable polarity GTAW [8]. A
mass source as a Gaussian distribution was added to the substrate.
The molten pool characterizations and transport phenomena
investigation showed the velocity of the molten flow becomes
stable after the initial stage of the WP formation. Bishal Silwal
and Santangelo used volume of fluid method to model bead on

https://doi.org/10.1016/j.ijheatmasstransfer.2018.04.092
0017-9310/Ó 2018 Elsevier Ltd. All rights reserved.
472 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484

Nomenclature

a material property T temperature field ½K ]

al material property in liquid state Tl liquidous temperature ½K ]
as material property in solid state Ts solidus temperature ½K ]
Am large number T0 ambient temperature ½K ]
Aw wire cross sectional area [m2] 
u relative velocity field [m s1]

B magnetic flux density [V s m2] Ut arc travel speed [m s1]
Ci concentration of the i th alloying element [wt%] Uw wire feed rate [m s1]
C Feclad concentration of iron in the clad layer [wt%] 
v inlet velocity field [m s1]
C Fesubstrate concentration of iron in the substrate [wt%] 
V electrical potential [V]
cp equivalent specific heat capacity [J kg1 K1] Va arc voltage [V]
cpl liquid specific heat capacity [J kg1 K1] a averaged mass fraction
cps solid specific heat capacity [J kg1 K1] bT thermal expansion coefficient [K1]
cpw wire specific heat capacity [J kg1 K1] c surface tension [N m1]
Di diffusion coefficient [m2 s1] Cw wire inlet boundary

E electrical field [V m1] e small number
F eq fluid volume force [N m3]
emissivity coefficient [W m3 sr1]
g gravitational acceleration [m s2] g arc efficiency
h convective heat transfer coefficient [W m2 K1] h liquid fraction
i unit vector in x-direction H angle between the x-axis and the wire feeding direction
Ia arc current [A] [rad]
J electrical current density [A m2] k Lagrange multiplayer

k unit vector in z-direction l dynamic viscosity [kg m1 s1]
k thermal conductivity [W m1 K1] lr magnetic permeability of the material [N A2]
kl liquid thermal conductivity [W m1 K1] l0 magnetic permeability of the free space [N A2]
ks solid thermal conductivity [W m1 K1] q material density [kg m3]
L latent heat [J kg1] ql liquid density [kg m3]
Lw wire latent heat [J kg1] qs solid density [kg m3]
MF mass fraction of alloying element qw wire density [kg m3]
n normal vector r Stefan-Boltzmann constant [W m2 K4]
p pressure field [Pa] re electrical conductivity [S m1]
Pa arc pressure [Pa] u weld free surface profile [m]
Pmax maximum arc pressure [Pa] ux first partial derivative of u in x-direction
q heat flux in considered domain [W m2] uy first partial derivative of u in y-direction
q8 rate of volumetric energy loss [W m3] Xl liquid domain
ri arc radius [m] Xs solid domain
ra effective radius of heat source [m] r gradient operator
rp effective radius of arc pressure [m] rs tangential gradient operator
rw wire radius [m]

plate and predict fluid flow in the WP and subsequent WP profile
[9]. They found that vibration of wire feeder reduced the droplet
release time in the cold wire GTAW. Their model showed that
the hot wire droplet resulted to a smaller width but with larger
penetration compared to the cold wire GTAW.
Mass transport study in dissimilar metal weld is important to
predict element distributions in the WZ achieve improved weld
properties. Bahrami et al. investigated mass transport phenomena
in linear dissimilar metals welding [10]. The alloying concentra-
tions in the numerical WZ had a reasonable agreement with the
experimental measurements. Regarding WAAM process, CWT pro-
duces a low dilution cladding and it is appropriate for dissimilar
metal joining. There is no study done on the investigation of fluid
flow and mass transport in CWT-GTAW with CWT on a dissimilar
substrate.
In the current study, a numerical simulation of the CWT using
GTAW process is developed for WAAM purposes. A 3D multi-
physics model is solved that includes electromagnetic, heat trans-
fer, fluid flow and mass transfer physics. The proposed model is
used to predict the clad geometry (width, thickness and penetra-
tion (or dilution)) as a function of process parameters such as tra-
vel speed, wire feed rate and material thermal properties. The
current model also predicts compositional mixing phenomena in
the clad layer in the case of building a multi-material part using
WAAM process.
2. Experimental details
The substrate used in this experiment was made of stainless
steel 410 and received in coupons with dimension of 180
30

7 mm. The wire with 1 mm diameter was made of Inconel 718; it
was used as deposited layer. The chemical composition provided
by vendor, is given in Table 1. The feeder is set at angle of 30° with
respect to the welding torch. The shielding gas used is pure Argon
with flow rate of 27 cfm. The welding mode is direct current elec-
trode negative (DCEN). The arc length is 5 mm and the electrode
tip angle is 30-degree. The temperature is measured during weld-
ing using two thermocouples attached in the center of the top and
bottom surfaces of the coupons. The weld coupons are sectioned
using a diamond saw and then, mounted and polished to prepare
the samples surfaces for performing metallography. The
microstructure was revealed using Glyceregia with instruction of
‘‘9 ml glycerol + 6 ml HCl + 3 ml HNO3”.
 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484 473

Table 1
Chemical composition of the used alloys (wt%).

Material Ni Cr Nb Mo Ti Al Co C Mn Si P S B Cu Fe
Substrate 410 0.21 12.33 – 0.02 0.001 0.002 0.02 0.12 0.42 0.36 0.018 0.001 – 0.03 Balance
Wire 718 55.0 21.0 5.50 3.30 1.15 0.8 1.0 0.08 0.35 0.35 0.015 0.001 0.006 0.3 Balance

Fig. 1. Schematic sketch of the model geometry.

1. The coordinate system selected allowed us to assume steady-

state conditions in the WP. The fluid flow is a laminar flow in
the molten pool. The liquid weld metal is assumed to be an
incompressible fluid.
2. The material loss associated with evaporation is not considered
in order to simplify the model.
3. The surface tension, arc pressure and gravity are in a static equi-
librium at free surface of the molten pool.

3.1. Model equations

The mixing phenomena in arc fusion welding is assumed to be

Fig. 2. Computational domain and subdomains.
3. Numerical model
A FEM model was developed using the COMSOL Multi-physicsÓ
5.3 software package. The model geometry and the coordinate sys-
tem are shown in Fig. 1. The dimension of the model geometry is
the same as welded coupons. The coordinate system selected is
attached to the arc welding torch. The welding torch and attached
wire feeder are moving at steady speed along substrate length. In
this model, the following assumptions are considered:
governed by four coupled physical phenomena that include elec-
tromagnetic, laminar flow, heat transfer, and mass transport,
which are described in the following sections.
3.1.1. Electromagnetic field
In this study, the arc is assumed to be stable and the wire feed-
ing has no distorting effect on the arc. Since the Lorentz force gen-
erated by the arc effects the fluid velocity field in arc fusion
welding, the calculation of magnetic field is required. The electrical
 can be expressed with respect to the electrical potential V,
field E
as follows:
 ¼ rV
E ð1Þ

Table 2
Material properties of alloys used in the model.

Symbol Properties Unit Alloy 410 Alloy 718

Tl Liquidous temperature K 1803 1666
Ts Solidus temperature K 1733 1594
kl Liquid thermal conductivity W m1 K1 28 23
ks Solid thermal conductivity W m1 K1 24.9 11
cpl Liquid specific heat capacity J kg1 K1 847 778
cps Solid specific heat capacity J kg1 K1 460 652
ql Liquid density kg m3 7426 7580
qs Solid density kg m3 7730 7733
l Dynamic viscosity Kg m1 s1 6.3
103 7.5
103
L Latent heat kJ kg1 250 292
re Electrical conductivity S m1 1.75
106 0.7
106
474 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484

Table 3
Boundary conditions.

Domains Electromagnetic Heat transfer Fluid flow Mass transfer

Xs Front side Insulated T ¼ T0 – –
Rear side Insulated No flux – –
Top side Eq. (17) Eqs. (18) and (19) – –
Bottom side Grounded Eq. (19) – –
Lateral sides Insulated Eq. (19) – –
Xl Front side – – ¼0
u C ¼ C substrate
Rear side – – ¼0
u No flux
Top side – Eqs. (18) and (19) Eq. (22) C ¼ C substrate
Eq. (20) on Cw Eq. (21) on Cw on Cw
Bottom side – – ¼0
u No flux
Lateral sides – – ¼0
u No flux

The electrical current density is given as follow 3.1.3. Heat transfer

J ¼ re E
 The temperature field can be obtained by solving the conserva-
ð2Þ tion of energy equation given as follows:
where re is electrical conductivity of the material. The current con-  cp TÞ ¼ r
  ðkr
 TÞ þ q
 þ U t Þ  ðqr
ðu ð9Þ
tinuity equation in terms of electrical potential V is computed as 8

follows: where k is thermal conductivity. In this study, the thermal conduc-

  ðre VÞ
 ¼0 tivity is considered to be temperature dependent in transition zone
r ð3Þ
from solid to liquid state. In addition, equivalent specific heat
 which is required in Lor-
Finally, the magnetic flux density of B capacity cp is the summation of an averaged heat capacity and the
entz force calculation can be obtained from Maxwell’s equation as distribution of latent heat given as [12]:
follows:
1 da
Cp ¼ ½hql cpl þ ð1  hÞqs cps  þ L ð10Þ
r
B ¼ lr l0J ð4Þ 2 dT

where lr and l0 are magnetic permeability of the material and free where L is the latent heat due to the phase transformation. The
space, respectively. index notations of s and l represent the material properties in solid
and liquid phases, respectively. Also, a is the averaged mass fraction
3.1.2. Fluid flow which is expressed as follow [12]:
The Navier-Stokes equation describes the behavior of Newto- 1 ð1  hÞqs  hql
nian, incompressible, laminar steady state flow. The equations of a¼ ð11Þ
2 q
motion in the reference frame illustrated in Fig. 1 can be written
RT
as follows: And ddTa satisfies the L T sl ddTa dT ¼ L.
  ðqu
Þ ¼ 0 The feeding wire is not included as a domain in this model;
r ð5Þ
However, the impact of CWT is applied to the heat equation and
boundary conditions. The wire absorbs the volumetric heat q8 from
 Þu
qðu  r  ¼ r  ½p  u
 þ lðr  u
 þ ðr
T
 Þ Þ þ F eq  qU t :ru
 ð6Þ the molten pool to get melted and then, immerse into the pool
where u  is the relative velocity field of the fluid to the moving frame with corresponding liquidous temperature. The rate of energy loss,
(xyz) attached to the torch as shown in Fig. 1. Also, p  is the pressure q8 ; is assumed to be distributed uniformly through the liquid
field. q and l represent the density and viscosity of the liquid metal, domain with a negative value. The reduction of the WP energy
reduces the weld penetration. The rate of volumetric energy loss
respectively. U t is travel speed vector of the arc along the x-axis. F eq
is computed as follow:
denotes the sum of the body force due to gravity, the Lorentz force
due to electromagnetic field generated by the arc and the phases qw ½cpw ðT lw  T 0 Þ þ Lw U w Aw
q8 ¼  ð12Þ
related to the source term in the mushy zone (the region between 8
the liquidus and solidus temperatures) [11]. This sum is:
where qw , cpw and Lw are the density, specific heat capacity and
ð1  hÞ2 latent heat of the wire, respectively. T lw and T 0 are liquidous and
F eq ¼ qbT ðT  T l Þg þ J
B
þ 
Am u ð7Þ
h3 þ e ambient temperatures of the wire. U w and Aw are wire feed rate
and wire cross sectional area. 8 is the volume of the fluid domain
where bT and g represent thermal expansion coefficient and gravi- Xl (see Fig. 2).
tational acceleration, respectively. T and T l are the temperature
and reference temperature of the WP. The second term on the
3.1.4. Mass transport
right-hand side denotes the Lorentz force. In the source term, Am
The mass transfer is important to investigate the final distribu-
is dependent on the morphology of the mushy zone and is assumed
tion of chemical compositions in the weld zone after two dissimilar
to be 2
105 . Also, e is set to 0.001 to keep the source term nonzero. alloys are mixed. In this study, the substrate material is stainless
The liquid fraction (hÞ is defined as: steel and wire is made of Inconel 718. The mass transport equation
8 is:
>
< 0 T < Ts
h¼ TT s
T l T s
Ts < T < Tl ð8Þ  Ci Þ ¼ r
 þ U t Þ  ðr
ðu  CiÞ
  ðDi r ð13Þ
>
:
1 T > Tl th
where C i is the concentration of the i alloying element; and, Di is
where T s is the solidus temperature of the material. the diffusion coefficient of that element in the molten pool.
 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484 475

3.2. Computational domains and boundary conditions

The computational domain is divided into two subdomains.

Fig. 2 illustrates the computational domain and subdomains. The
molten pool occurs in the smaller subdomain Xl . The fluid flow is
restricted to the subdomain Xl to save calculation time. The mate-
rial properties of the liquid domain are calculated using mass frac-
tion of alloying elements like iron concentration (C Fe ) as follows:
C Fe  C Feclad
MF ¼ ð14Þ
C Fesubstrate  C Feclad

al ¼ MFal clad þ ð1  MFÞal substrate ð15Þ

where MF denotes mass fraction, C Feclad and C Fesubstrate are concentra-
tion of iron in the clad layer and the substrate, respectively, and
al represents the material property of the liquid phase. The T l is
excluded from liquid domain properties and kept as liquidous tem-
perature of the substrate.
The bigger domain Xs is considered as solid material with the
substrate solidus material properties. The material properties tran-
sition at the phase change from solid to liquid is expressed as
follows:
a ¼ as þ hðal  as Þ ð16Þ
The material properties of the substrate and wire used in this
model are given in Table 2.
The boundary conditions of those domains are described for all
physics as follow:

3.2.1. Electromagnetic field

The current density with Gaussian distribution is assumed to
distribute on the top surface of the entire computational domain
as follows:
 
J  n 3I 3ðx2 þ y2 Þ
 ¼ a2 exp ð17Þ
pr i 2
ri
where ri is arc radius and Ia is arc current. In addition, the bottom
surface of the subdomain Xs is considered as ground (V  ¼ 0). n
 is
normal vector of the surface.

3.2.2. Temperature and flow velocity fields

The heat input is considered as Gaussian distribution at the top
surface of the entire domain as follows:
 
3gIa V a 3ðx2 þ y2 Þ
n
q ¼ exp ð18Þ
pr2a r 2a
where g and V a are the arc efficiency and arc voltage, respectively.
The radius r a denotes the effective radius of heat distribution.
The convective and radiative heat transfers on lateral and bot-
tom surfaces are given by

q  ¼ hðT  T 0 Þ þ r
ðT 4  T 40 Þ
n ð19Þ
where h, r and
, respectively, are convective heat transfer coeffi-
Fig. 3. Schematic diagram showing calculation steps for the CFD model and surface cient, Stefan-Boltzmann constant, and surface emissivity. The front
profile. boundary of subdomain Xs is subjected by a constant temperature
T 0 and a zero-heat flux is considered at rear boundary of that
subdomain.

Table 4
Welding parameters used in experiments and numerical model.

Case Arc current (A) Arc voltage (v) Travel speed (mm s1) Wire feed rate (mm s1)
1 150 14 3 12
2 150 14 3 18
3 150 14 4 18
476 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484

Fig. 4. Three dimensional of (a) temperature isoterms (b) velocity field and (c) mixing phenomena (iron mixture) in the weld pool of case 1.

As already mentioned, the cold wire is excluded from the model
domains. However, the feeding wire feed rate impacts the fluid
flow and temperature distribution of the molten pool. In the
absorbed heat from the WP Eq. (12), it is assumed that the melted
wire immerses the pool with wire liquidous temperature. How-
ever, the molten pool temperature is higher than liquid wire tem-
perature. Therefore, another thermal boundary condition is
assumed at the interface of wire immersion, Cw . In this model,
the wire is considered a virtual domain. The temperature at outer
boundary which is the molten pool temperature, T p , is known. The
inflow heat flux can be calculated through the boundary as follows:
Z Tp
: n ¼ q
q cp dT
T lw
At the same boundary, the wire is immersed into the WP with a
constant feed rate Uw . This speed affects the velocity field of the
molten pool. The wire is forced towards the WP in front of the
arc and the arc (reference framework) moves along the x-axis with
the constant travel speed Ut . The inlet velocity field is specified as
follows:
v ¼ ðUt þ Uw cos HÞi  Uw sin Hk
where H is the angle between the x-axis and the wire feeding
direction.
The top surface of the liquid subdomain Xl is subjected to Mar-
angoni shear stress except the interface of the wire and molten
pool Cw as follows:
 

@u @c @c
l ¼h rs T þ r s Cs
@z @T @Cs
where rs is gradient tangential to the surface and c is surface ten-
sion. The clad layer composition contains low amount of sulfur,
@c
then surface tension gradient (@T ) is assumed to be
1
104 Nm1 K1 [7]. In this study, the variation of sulfur con-
centration (rs Cs ) is negligible because the sulfur concentrations of
both substrate and wire are 0.001 wt%.
3.2.3. Mass transport
The lateral, top and bottom surfaces of the liquid subdomain Xl
has no convective flux. The concentration of the ith alloying ele-
ment of the substrate is applied at the front boundary of subdo-
main Xl . And, the concentration of the ith alloying element of the
ð20Þ wire is applied at the interface Cw . No outflow is considered at
the rear boundary of subdomain Xl .
Boundary conditions of all domains are summarized in Table 3.
3.2.4. Free surface
The WP surface displays a significant deformation under droplet
impact and arc pressure in spray transfer mode [13]. The top sur-
face shape of WP satisfies the equilibrium condition between sur-
face tension, hydrostatic pressure and arc pressure under steady
ð21Þ
state. The motion inside the WP is ignored. In this method, the pro-
file of WP, uðx; yÞ; will form a shape which minimizes the total
potential energy of the surface, Et , including surface energy, gravi-
tational potential energy and arc pressure work which are
described as follows:
Z  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  qg u2 
ð22Þ Et ¼ c u2x þ u2y þ 1  1 þ  Pu dxdy ð23Þ
S 2
 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484 477

Fig. 5. Velocity field in case 1 (a) top view, (b) side view, (c) cross sectional planes and (d) cross section views referred to the shown planes.

Z
where c denotes the surface tension which is assumed to be U w Aw
Et ¼ ðu  z0 Þdy ¼ ð25Þ
1.8 N/m for Inconel 718 [14], and a large number at solid state to A Ut
prevent deformation at that zone. S is top surface area of the entire
domain. P is the arc pressure. The droplet pressure is assumed to be where z0 is the z coordinate of the top surface which is zero in this
the same as the arc pressure term to avoid increase number of model. Variational calculus is applied to obtain the free surface pro-
unknowns. A Gaussian distribution is considered for arc pressure file. After using the Euler–Lagrange criterion and algebraic manipu-
([4]) as follows: lation, the following equation was derived:
! 0 1 0 1
Pa ¼ Pmax exp
ðx2 þ y2 Þ
ð24Þ @ B ux C @ B uy C
2r 2p  @c qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffiA  @c qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA þ qg u ¼ P þ k
@x u2 þ u2 þ 1 @y u2 þ u2 þ 1
x y x y
where P max is maximum pressure which is assumed to be 300 MPa ð26Þ
in this study. r p is the effective radius of the pressure distribution.
Since the change of the top surface shape depends on the depos- where k is Lagrange multiplayer used to satisfy the constraint Eq.
ited volume of the clad layer, the Eq. (24) must meet the following (25). The following boundary conditions should be applied to deter-
constraint mine the free surface profile in Eq. (26).
478 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484

Fig. 6. Velocity field in case 2 (a) top view, (b) side view, (c) cross sectional planes and (d) cross section views referred to the shown planes.

ux ¼ 0 at rear pool boundary electromagnetic, heat transfer and laminar flow are solved for a
ð27Þ
u ¼ 0 at front pool boundary non-deformed WP. The achieved temperature field is used for the
free surface analysis to ensure that the deformation of the free sur-
face occurs at the liquid pool. To this end, the surface tension is
3.3. Numerical method expressed in form of a step function as follow

The finite element method (FEM) is employed to solve the sys- 1:8 N=m h > 0
c¼ ð28Þ
tem of equations using COMSOL Multi-physics 5.3. The solid sub- 1000 N=m h ¼ 0
domain, Xs , is discretized using a free tetrahedral coarse mesh.
The maximum element size of free tetrahedral is 0.1 mm in the The next step is the calculation of the free surface profile and,
fluid subdomain, Xl . then, to extrude the profile to the top surface of the entire domain
The calculation steps are explained in a flowchart shown in using the deformed geometry (DG) utility in COMSOL. The Eq. (26)
Fig. 3. Since the arc efficiency and arc radius affect penetration can be rewritten in form of Helmholtz equation as follows:
0 1
and the width of the WP, a few experiments were done to achieve
the best match between numerical and experimental weld zone B ru C
r  @c qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA þ qg u ¼ P ð29Þ
profiles. Accordingly, arc efficiency and arc radius are selected to 1 þ u2x þ u2y
be 0.4 and 5 mm in this study, respectively. Then, the
 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484 479

Fig. 7. Velocity field in case 3 (a) top view, (b) side view, (c) cross sectional planes and (d) cross section views referred to the shown planes.

Table 5 4. Results and discussion

Maximum velocities and temperatures in the weld pools.

Maximum velocity (m/s) Maximum temperature (K) To verify the numerical results, three bead-on-plate depositions
Case 1 0.091 2347 are performed using the GTA-CWT welding parameters shown in
Case 2 0.099 2421 Table 4. The effects of change in travel speed and wire feed rate
Case 3 0.087 2271 on the mixing process are investigated.

The constraint, Eq. (25), is implemented as a global constraint in
the software and the Lagrange multiplayer, k, is calculated. The liq-
uid domain is meshed using free triangular element with maxi-
mum size of 0.2 mm. A coarser free mesh is distributed in the
remained domain. The achieved nodes displacements are linearly
extruded to top surface of the 3D computational domain using
deformed geometry interface. Finally, half of the computational
domains is taken account due to symmetry.
4.1. Weld pool temperature field, velocity field and diffusion field
A three-dimensional view of the numerical prediction of the
temperature field, velocity field and mass distribution for case 1
is shown in Fig. 4. In Fig. 4a, the arc travel speed causes compressed
isotherms at front of the WP and expanded isotherms behind the
WP. In cold wire feeding, the droplet temperature is lower than
the pool temperature. The isotherms are affected by the cold dro-
plet contact with the hot molten pool surface at the front part of
480 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484
Fig. 8. Comparison of macrograph of cross section of the weld with numerical prediction in (a) case 1, (b) case 2 and (c) case 3.
the WP. The free surface of the weld is deformed under arc pres-
sure while solidified zone is elevated due to wire material addition
as shown in Fig. 4a. The mushy zone is also visible along the fusion
boundary. The three-dimensional velocity field is illustrated in
Fig. 4b. It shows that the two toroidal vortices within the flow;
around the wire inlet the surface flow is toward the wire. This flow
pushes the molten wire material downward into the WP. The wire
material is then circulated toward the top surface of the WP in the
upwelling region that separates this secondary vortex from the pri-
mary vortex flow field that dominates the back of the WP. Subse-
quently, this strong vortex creates a well-mixed region at rear of
the WP which is solidified as the clad layer shown in Fig. 4c. The
velocity fields for the three cases examined are described in more
detail in the next section.
The velocity field of case 1 is presented in Fig. 5. The color bar
indicates the velocity magnitude in m/s. The normalized arrows
in this figure show the planar direction of the flow fluid. The length
of the WP (x-direction) is about 6.9 mm and the half width (y-
direction) is 2.5 mm. The molten metal circulates in the WP within
the two opposing toroidal vortices. Fig. 5a shows the velocity field
on the top surface of the WP. An inward liquid flow around the
interface of filler wire inlet (region A) is observed; this is the top
of the secondary toroidal (or ring) vortex. This inward flow occurs
in lower temperature region due to CWT immersion inlet. The flow
is driven outward to the fusion boundary as illustrated over the
remaining region of the top surface due to capillary effect (region
B); this flow is at the top of the primary toroidal vortex. The surface
flow velocity is a minimum near the stagnation point at the center
of the upwelling region between the primary and secondary vor-
tices. This occurs between regions A and B as shown in Fig. 5a.
The maximum velocity occurs on free surface at rear part of the
WP. To examine the flow in the WP, the velocity field of the molten
pool along xz-plane (y = 0) is illustrated in Fig. 5b. The velocity
field at three different planes parallel to the yz-plane (x = 0, 2
and 4 as shown in Fig. 5c) is shown in Fig. 5d. According to
Fig. 5b and 6d, a three-dimensional pair of vortices is developed
in the WP; this is due to the combination of Marangoni convection
and Lorentz forces. The projection of front part of the primary vor-
tex is small and clockwise (CW), as shown in the front side of the
WP in Fig. 5b. This vortex is much stronger at rear side of the WP;
its projection is counterclockwise (CCW) as shown in Fig. 5b. This
strong vortex pushes the mixed liquid toward the elevated part of
the WP and, subsequently, the flow is driven back towards the cen-
ter along the bottom of the WP. The projections of the primary and
secondary vortices are also visible in the yz-planes (x = 0, 2 and 4)
in Fig. 5d. The liquid flow yz- plane under arc (x = 0) contains the
projection of the two vortices. The cross-section of the vortex clo-
ser to the fusion boundary is the projection of the primary vortex.
There is another vortex, the secondary vortex, near the center of
the WP which is a part of the toroidal vortex around the molten
wire inlet. This vortex is developed around the wire inlet due to
the thermal gradient that occurs in CWT. A part of this vortex is
visible as a counterclockwise (CCW) vortex just after wire inlet in
center of the WP in Fig. 5b. It circulates the metal in the WP from
middle of the pool toward the molten wire inlet. This vortex is
smaller in cross section at front side of the WP. The depth of the
WP below the wire inlet is affected by heat absorption from filler
wire as shown in Fig. 5b. Over all, the molten metal circulates from
the front part of the pool to the rear part as shown in Fig. 4b. The
effect of flow circulation on mass distribution is illustrated in
Fig. 4c. The molten clad flow begins to mix with the substrate just
under the wire inlet. The mixture of the solidified clad layer is
made up of about 70% of the wire material and about 30% of the
substrate material (this mixture of the cladding is about the same
for cases 2 and 3 as well). Hence, the solidified clad layer is the
result of an additive layer materials process.
Fig. 6 illustrates the velocity field of case 2. The fluid flow is sim-
ilar to case 1 as illustrated by the free-surface flow. The length and
 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484
Fig. 9. Comparison of experimental thermal cycle with calculated temperature at top and bottom surfaces in (a and b) case 1, (c and d) case 3.
half width of the WP is essentially the same as predicted for case 1
because the amount of heat input is equal in both cases. However,
the penetration is reduced from 0.88 mm in case 1 to 0.8 mm in
case 2. The penetration reduction is a result of the higher heat of
absorption required due to the larger volume of feeding wire and
higher elevation of the top surface at rear part of the WP.
Fig. 7 illustrates the fluid flow field in case 3; in this case the tra-
vel speed is higher as compared with the two previous cases. The
increase in travel speed caused a cold region near the front edge
of the fusion boundary at the free surface of the WP (region C).
Consequently, an inward flow loop occurs at the surface of that
region. This flow demarcates the existence of a recirculation zone
of opposite sense as the secondary vortex. The length and half
width of the WP are, respectively, 6.5 mm and 2.3 mm which are
smaller than the ones in cases 1 and 2. The penetration is
decreased to 0.65 mm as well.
For all cases the maximum velocity and temperature occur on
free surface of the molten pools and are compared in Table 5.
The temperature and velocity in case 2 are the highest among
the three cases because of higher amount of heat input and more
vigorous convection due to the higher feed rate.
4.2. Validation of the model
4.2.1. Weld geometry
The experimental weld profile is compared to numerical predic-
tions in Fig. 8; the shapes predicted are in reasonably good agree-
ment with the experimental measurement. The height of the clad
layer is measured from the substrate surface (z = 0) to top of the
elevated layer. The width and height of the welds are reasonably
well predicted. The numerical penetration values have
481
approximately 35% deviation. It most likely is a result of the differ-
ence between thermal properties values assumed in the model and
actual ones used in the experiments.
4.2.2. Thermal measurement
Two thermocouples are attached to the middle of the coupons
on both top and bottom surfaces. The thermocouple was in dis-
tance of 2 mm from the weld bead boundary at top surface. Since
the amount of heat input is the same in both cases 1 and 2 and dif-
ferent with case 3, the measurements were only performed for
cases 1 and 3. The thermal cycles are calculated by converting time
domain into x distance using travel speed of the arc. Fig. 9 presents
a comparison of temperature between experimental measure-
ments and numerical predictions for two different emissivity coef-
ficients 0.2 and 0.8. The cooling rate is known to be overestimated
numerically due to the emissivity impact as reported by Traidia et
a. [7]. In this study, the emissivity coefficient is assumed to be 0.8
in the model for steel with rough surface at 25 °C [15]. The peak
temperatures and heating rates are in good agreement as com-
pared with the thermal measurements and computational results
at top surfaces. The maximum temperature of heat affected zone
(2 mm from the weld boundary) on top surface (Fig. 9 a and c) is
elevated from 600 °C to 850 °C with decrease of the travel speed
from 4 mm/s to 3 mm/s. The deviation of numerical temperatures
at bottom surfaces is in result of the coarse mesh used in the solid
domain (see Fig. 9 b and d).
4.3. Mixing
To study the mass transport in the cold wire feed cladding,
three major chemical elements, viz., iron (Fe), nickel (Ni) and
482 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484

Fig. 10. (a) The yz-plane location in solid state, the distribution of the element (b–d) Fe, (e–g) Ni and (h–j) Cr in the solidified weld zone at the yz-plane for case 1, case 2 and
case 3.

chromium (Cr), are the only constituents considered. Fig. 4c shows
a three-dimensional view of Fe distribution with the filler wire
immersed in the WP in case 1. The black line illustrates the fusion
boundary between solid and liquid. The concentration of Fe in the
substrate and wire are assumed to be 78 wt% and 20 wt%, respec-
tively. It is observed that the filler wire is a significant part of the
mixture with the substrate at rear part of the WP.
Fig. 10 represents the distribution of the element Fe, Ni and Cr
in the solidified WZs at yz-plane. This plane is located just behind
the weld pool where the material is in the solid phase. The
x-coordinate of the WP rear point of cases 1, 2 and 3 are shown
in Fig. 10a. It is obvious that the most non-uniform mixing occurs
in case 1 (see Fig. 10a). The computed concentrations for all ele-
ments show that the mixing process grows from the middle of
the weld (y = 0) toward the fusion boundary. The composition of
the WZ gets more uniform for cases 2 and 3 (Fig. 10 c and
Fig. 10d) because molten wire inlet velocity is higher in the latter
cases. The normal component of the inlet velocity (U w sin Hk) 
pushes the molten wire to the bottom of the WP; this results in
better mixing. The arc travel speed has less impact on mixing as
 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484 483

Fig. 11. The microstructure of the welds at right side, middle, left side of the fusion boundary of (a–c) case 1, (d–f) case 2 and (g–i) case 3.

Fig. 12. (a) SEM image and (b) EDS results of UMMZ.

compared to the wire feed rate. The numerical outcomes of the
mixing phenomenon were confirmed by microstructural analysis
as described next. Fig. 11 shows the microstructure of the WZ
along the fusion boundaries for all the cases. It is observed that
an unmixed melted zone (UMMZ) is formed between the WZ and
the fusion boundary in case 1 (see Fig. 11b–d). This zone is partially
visible in case 2 (Fig. 11e–g) and not noticeable in case 3 (Fig. 11h–
j). UMMZ is a region along the fusion boundary, which is melted
during the welding but not mixed within the molten pool [16]. It
is solidified with a different microstructure compared to the heat
affected zone (HAZ) because of diffusion of other chemical ele-
ments from the liquid pool into that zone. Fig. 12a shows SEM
image of UMMZ in case 1 and the chemical compositions of that
zone using EDS analysis at point 1 shown in Fig. 12b. The measured
concentrations of the three elements indicate diffusion of Ni from
the WP into the UMMZ. This zone can be removed by the enhanced
convection; convection can be enhanced by increasing the Lorentz
forces inside the molten pool as shown by Hejripour and Aidun
[16] experimentally. In this study, UMMZ width is significantly
reduced by increasing the wire feed. This parameter intensifies
the flow velocity due to a higher convection in the WP producing
to a uniform mixed WZ; this was numerically predicted.
To verify the numerical predictions of the concentrations of the
three elements (Fe, Ni and Cr), EDS analysis was performed along a
484 F. Hejripour et al. / International Journal of Heat and Mass Transfer 125 (2018) 471–484
Fig. 13. Comparison of three elements concentrations (Fe, Ni and Cr) found numerically and experimentally in the weld zone of case 2 along vertical line.
vertical line for case 2. The vertical line is located at middle of the
WZ (y = 0) from the substrate to top surface of the clad layer. The
experimental and numerical results are compared in Fig. 13a–c
for Fe, Ni and Cr concentrations, respectively. As shown in the fig-
ures, the predicted concentrations of Fe, Ni and Cr are in a good
agreement with the EDS results; the results indicate uniform dis-
tributions of these alloying elements in the WZ.
5. Conclusion
In this study numerical simulations are compared with experi-
mental data; the mass transport in cladding of stainless steel sub-
strate with a clad layer of a nickel-based super alloy using GTAW
was investigated. The study has engineering implications in
WAAM processes. The main results are as follows:
 The flow field is dominated by two toroidal vortices. The
fluid flow at free surface of the WP is inward around the
wire inlet (the imprint of the ring vortex adjacent to the fil-
ler wire inlet flow) and outward at distances relatively far
from the filler inlet (the imprint of the primary toroidal vor-
tex). An upwelling flow separates the two vortices. The
inward flow is a result of local thermal gradients associated
with the fact that the molten CWT is relatively cool. Maran-
goni convection and Lorentz forces cause the outward flow
toward the fusion boundary of the WP. The ring-vortex flow
around molten wire inlet moves the molten wire material
from the location of its deposition toward the bottom of
the WP and, ultimately to the free surface and the fusion
boundary. The toroidal circulation due to Marangoni and Lor-
entz forces at rear part of the WP mixes the two materials.
The clad layer is mostly made up of the wire filler (about
70%) with only about 30% of the substrate mixed into a rel-
atively homogeneous clad layer.
 The predicted deposited bead shape is in a reasonable agree-
ment with experiment. The width of the clad layer is dependent
on the heat input. The penetration varied according to feed rate
and heat input.
 The simulation allowed us to study in detail the effects of cold
wire feeding on the dynamics of the WP; this included the mix-
ing phenomena and, hence, the composition of the clad layer in
the cladding procedure. The numerical predictions show that a
uniformly mixed weld metal occurs in middle of the weld zone
that leads to clad layers with relatively uniform composition.
We found that increasing the feeding velocity of the wire led
to higher mixing in the molten pool and led to removal of the
UMMZ. The arc travel speed has little impact on mixing in the
WP. The uniform mixing in the middle of the weld zone and
alloying element concentrations are in reasonable agreement
with experiment.
Declarations of interest
None.
References
[1] A. Busachi, J. Erkoyuncu, P. Colegrove, F. Martina, J. Ding, Designing a WAAM
based manufacturing system for defence applications, Procedia CIRP 37 (2015)
48–53, https://doi.org/10.1016/j.procir.2015.08.085.
[2] S. Chen, S. Zhang, N. Huang, P. Zhang, J. Han, Droplet transfer in arcing-wire
GTAW, J. Manuf. Process. 23 (2016) 149–156, https://doi.org/10.1016/j.
jmapro.2016.05.014.
[3] H. Geng, J. Li, J. Xiong, X. Lin, F. Zhang, Optimization of wire feed for GTAW
based additive manufacturing, J. Mater. Process. Technol. 243 (2017) 40–47,
https://doi.org/10.1016/j.jmatprotec.2016.11.027.
[4] E. Le Guen, M. Carin, R. Fabbro, F. Coste, P. Le Masson, 3D heat transfer model of
hybrid laser Nd:Yag-MAG welding of S355 steel and experimental validation,
Int. J. Heat Mass Transf. 54 (2011) 1313–1322, https://doi.org/10.1016/j.
ijheatmasstransfer.2010.12.010.
[5] Z. Nie, G. Wang, J.D. McGuffin-Cawley, B. Narayanan, S. Zhang, D. Schwam, M.
Kottman, Y. (Kevin) Rong, Experimental study and modeling of H13 steel
deposition using laser hot-wire additive manufacturing, J. Mater. Process.
Technol. 235 (2016) 171–186, https://doi.org/10.1016/j.jmatprotec.2016.
04.006.
[6] X.S. Gao, C.S. Wu, S.F. Goecke, H. Kugler, Numerical simulation of temperature
field, fluid flow and weld bead formation in oscillating single mode laser-GMA
hybrid welding, J. Mater. Process. Technol. 242 (2017) 147–159, https://doi.
org/10.1016/j.jmatprotec.2016.11.028.
[7] A. Traidia, F. Roger, E. Guyot, J. Schroeder, G. Lubineau, Hybrid 2D–3D
modelling of GTA welding with filler wire addition, Int. J. Heat Mass Transf. 55
(2012) 3946–3963, https://doi.org/10.1016/j.ijheatmasstransfer.2012.03.025.
[8] J. Pan, S. Hu, L. Yang, D. Wang, Investigation of molten pool behavior and weld
bead formation in VP-GTAW by numerical modelling, Mater. Des. 111 (2016)
600–607, https://doi.org/10.1016/J.MATDES.2016.09.022.
[9] Bishal Silwal, M. Santangelo, Effect of vibration and hot-wire gas tungsten arc
(GTA) on the geometric shape, J. Mater. Process. Technol. 251 (2018) 138–145,
https://doi.org/10.1016/J.JMATPROTEC.2017.08.010.
[10] A. Bahrami, B.T. Helenbrook, D.T. Valentine, D.K. Aidun, Fluid flow and mixing
in linear GTA welding of dissimilar ferrous alloys, Int. J. Heat Mass Transf. 93
(2016) 729–741, https://doi.org/10.1016/j.ijheatmasstransfer.2015.10.058.
[11] V.R. Voller, A fixed grid numerical modelling methodology for convection-
diffusion mushy region phase-change problems, Int. J. Heat Mass Transf. 30
(1987) 1709–1719, https://doi.org/10.1016/0017-9310(87)90317-6.
[12] Comsol Multiphysics, Heat Transfer Module User’s Guide, version 5.3, n.d.
[13] M. Ushio, M. Ushio, C.S. Wu, C.S. Wu, Mathematical modeling of three-
dimensional heat and fluid flow in a moving gas metal arc weld pool, Metall.
Mater. Trans. B. 28 (1997) 509–516, https://doi.org/10.1007/s11663-997-
0118-z.
[14] J.J. Valencia, P.N. Quested, Thermophysical properties, ASM Handb. Cast. 15
(2008) 468–481. doi:http://doi.org/10.1361/asmhba0005240.
[15] Mikron Instrument Company, Table of Emissivity of Various Surfaces, 2014,
pp. 1–13.
[16] F. Hejripour, D.K. Aidun, Consumable selection for arc welding between
stainless steel 410 and Inconel 718, J. Mater. Process. Technol. (2017), https://
doi.org/10.1016/j.jmatprotec.2017.02.013.