You are on page 1of 7

Journal of Materials Processing Technology 77 (1998) 233 239

The prediction of the temperature distribution and weld pool


geometry in the gas metal arc welding process
M.A. Wahab a,*, M.J. Painter b, M.H. Davies a
a

Uni6ersity of Adelaide, Adelaide, SA 5005, Australia


b
CSIRO-DMT, Wood6ille North, 5012, Australia

Abstract
The main objective of this study is to generate new understanding and improve computer methods for calculating the thermal
cycles in the near weld region during gas metal arc (GMA) welding. Predicting the thermal cycle also provides an estimate of the
depth of weld penetration, the geometry of the weld pool and the cooling rates. Both 2D and 3D finite-element models have been
developed using the solution of heat-transfer equations. The accuracy of the predicted cooling times, weld penetrations and the
lengths of the weld pools are compared with experimentally obtained values for bead-on-plate welds. A mechanical weld pool
ejection rig developed in this study provided a quick and ready means of defining a full 3D weld-pool shape. A new numerical
approach, split heat source model has been developed to model the arc heat. This gives good agreement with experimental data
for a range of welding conditions. 1998 Elsevier Science S.A. All rights reserved.
Keywords: Temperature distribution; Weld pool geometry; Gas metal arc welding process

1. Introduction
The study of the fusion-welding process covers various important aspects such as the metallurgical properties of the weldments, the effect of arc temperature on
the development of post-weld residual stresses, distortion in the base material, the physics and behaviour of
the arc with the formation and maintenance of the weld
pool. All of these activities are aimed at improving the
integrity of weld and welded structures. The development of post-weld residual stresses and distortion in a
welded structure and microstructural changes are based
intrinsically on the thermal cycle from the welding
processes. The thermal cycle also provides information
on weld penetration and heat-affected zone geometry.
Therefore, controlling the thermal cycle is critical to
good quality welding. Computer-based simulative
mathematical welding models that can predict process
outcomes without the need for costly pre-production
trials or experimentation will increasingly become essential tools for welding engineers.
The development of these models which attempt to
predict process outcomes began in the 1940s with ana* Corresponding author.
0924-0136/98/$19.00 1998 Elsevier Science S.A. All rights reserved.
PII S0924-0136(97)00422-6

lytical solutions by Rosenthal [1] but it grew rapidly


during the 1980s with the onset of powerful computers.
In his work Goldak [2] indicated that welding engineers
want a simple way of predicting the weld pool information, but it is far from a simple problem, and currently
most methods rely on regression analysis of experimental data. However, developed numerical models can
provide an increased understanding of the process, and
can provide a tool to improve processes, process control and productivity.

2. Background

2.1. The gas metal arc welding (GMAW) process


A schematic representation of the driving forces in
the GMA welding process is shown in Fig. 1. The
electric potential established between the electrode and
the workpiece causes current flow, which generates
thermal energy in the partially ionised gas. The current
also creates the surrounding magnetic field, which interacts with the diverging current field to induce an electromagnetic force which accelerates the plasma,
generating a flow towards and across the workpiece

234

M.A. Wahab et al. / Journal of Materials Processing Technology 77 (1998) 233239

surface. Intense heat generation and high plasma velocities are created, with a rapid transfer of heat into the
molten weld pool. In the GMA welding process, heat is
generated both by resistive heating and by heat transfer
from the arc current. The tip of the wire is melted and
molten droplets are formed and driven into and
through the plasma jet into the weld pool. A detailed
description of the physical process is given by Lancaster
[3].

2.2. Mathematical models: conducti6e heat transfer


Early attempts [1] to calculate the thermal cycle for a
given weld involved an analytical solution for heat
conduction within a plate under a set of idealised
welding conditions, e.g. assuming that the weld could
be represented by a simple point of line heat source,
and that the plate was semi-infinite. Almost all of the
constraining assumptions that were considered to limit
the accuracy of the analytic solutions [1] can be avoided
by using direct numerical methods to solve the heatconduction equation. Non-linear, temperature-dependent conductivity and diffusivity can be used and the
latent heat at phase changes allowed for. Contemporary
numerical models generally use finite-element methods
of solution considering the welding process as transient
conductive heat transfer, with the welding arc represented as a complex distributed heat source [4,5]. This
approach reduces the problem of finding a solution for
a single energy-balance equation. The disadvantages are
that assumptions have to be made about how the heat
flow from the arc is represented numerically and, in
addition, both the mass flow in the weld pool and the
convective flow within the molten metal, which are
known to influence the final weld pool shape, can only
be approximated. In such methods the governing equation of heat transfer is written in 3D transient heat
form as,
k

d 2T
d 2T
d 2T dQ
dT
+k 2 +k 2 +
=rC
2
dx
dy
dz
dt
dt

d 2T
d 2T
d 2T
dQ
dT
+ k 2 + k 2 u = urC
2
dx
dx
dx
dy
dz

(3)

where u (mm s 1) is the velocity in the x-direction

2.2.1. The double ellipsoidal heat source


representation
The welding arc and its heat energy (usually taken as
Q= hVI, where h is an arc efficiency, V is voltage and
I is the welding current) are represented as a distributed
heat flux. The most popular heat source definition has

Fig. 1. A schematic representation of the driving forces during the gas


metal arc welding process (GMAW).

(1)

with the boundary conditions represented as,


dT
k + h(T T0)+ se(T 4 T 40) =0
dn

(2)

where Q is the internal heat energy released or consumed per unit volume (J mm 3), T is temperature, T0
is ambient temperature, t is time, k is thermal conductivity (W mm 1C 1), r is density, C is specific heat (J
g 1C 1), h is a convection coefficient, s is the Stefan
Boltzman constant and e is emissivity. On assuming a
quasi-steady state situation, Eq. (1) can be rewritten in
the following form:

Fig. 2. The double ellipsoidal representation of the heat distribution


from a welding arc (Goldak [7,8]).

M.A. Wahab et al. / Journal of Materials Processing Technology 77 (1998) 233239

235

Fig. 3. The split heat source representation of the heat source distribution from a welding arc used in the 2D section and 3D quasi-steady-state
models.

been the double ellipsoidal source by Goldak et al.


[6,7], in which the arc energy is distributed in semi-ellipsoidal volume, as shown in Fig. 2.
This source is defined through six arbitrary parameters. The source dimensions are not known a-priori and
they have not been related to process parameters, being
simply chosen to give the best fit to experimental results. Despite the difficulties associated with this approach, considerable international effort has been made
to further its development, for such models are computationally fast and can give a reasonable prediction of
the general thermal cycle. However, for regions close to
the weld pool, the model predictions are less satisfactory because near to the source the resulting temperatures are more sensitive to the form of the heat source
distribution.
A further major difficulty is that this approach does
not directly include the transfer of heat by convective
movement in the liquid weld pool. This effect can only
be included by modifying the energy distribution of the
source or artificially manipulating the thermal conductivity of the liquid material. Typical results obtained on
GMA bead-on-plate welds using this heat source are
discussed in earlier work [8].

3. Numerical models
In this study two conduction models have been
developed:
(1) 2D section model
A 2D finite element section model, which follows
those used by other researchers [9,10] and was de-

scribed earlier in [4]. This 2D model calculates the


variation of temperature throughout a plane normal to
the direction of arc movement. This 2D approximation
is invoked by assuming that the welding speed is so
high that the transfer of heat along the workpiece can
be neglected. The calculation then determines the temperature at each point in the plane, as a time-dependent
heat source acts on it.
In this study, the governing equation of heat transfer
has been written in element discretisation form as described earlier [4] and has been solved using a modified
implicit, element-by-element product method originally
proposed by Hughes et al. [11], and an adaptation of
programs published by Smith and Griffiths [12]. A
parabolic weld bead is included by dynamically adding
mesh.
(2) 3D quasi-steady-state model (3DQSS)
A 3D quasi-steady state solution, which considers
that the heat source is fixed in space with the workpiece
effectively flowing past it. The calculation includes a
convective heat-transfer equivalent to this pseudo flow.
In effect, infinitely long workpieces are considered and
a steady-state solution is determined. The variation of
temperature with time is obtained by considering the
thermal history of a point as it moves through this
temperature field. The fixed heat source and unchanging mesh in this model means that the deposited weld
bead can be incorporated easily and the mesh appropriately refined near to the source to minimise the total
number of elements.
These models are evaluated using a newly developed
split heat source and their accuracy is judged by the
prediction of: (i) T8/5 (800500C) cooling times at the

236

M.A. Wahab et al. / Journal of Materials Processing Technology 77 (1998) 233239

weld centre line; (ii) weld pool depth; and (iii) the
length of the weld pool. These are compared with
experimentally measured data for bead-on-plate welds
for a range of welding conditions.

3.1. Split heat source representation (Fig. 3)


In this heat source (as illustrated in Fig. 3), the total
heat content (hVI) is split into parts and distributed
roughly following the physical behaviour of the welding
process. One part of the heat input arises from the
inclusion of a deposited weld bead. In the transient
2D-section model this is simulated by dynamically
adding mesh as the welding arc passes over the meshed
section. Material is added at a temperature of 2400C
(based on experimental droplet temperatures), effectively giving a heat input in the region behind the arc
centre. The weld bead is static in the 3DQSS model so
that an equivalent uniform heat flux is applied to the
front of the weld bean to give a similar effect.
The remaining heat is then divided again into arc
and drop sources, with their heat contents in the ratio

Fig. 6. Typical data determined from weld-pool cavities.

of total heat to droplet heat. The part related to the


arc is distributed in an ellipsoidal volume at the plate
surface. The second part, from the drop, is distributed
below the plate surface at a depth proportional to the
arc current. This reflects the experimentally-observed
increase in droplet momentum and the increase in weld
penetration with increased arc current.
4. Experimental program

Fig. 4. Schematic diagram of the mechanical ejection device.

Fig. 5. Laser profiling device.

To provide experimental data for the validation of


the numerical models, the thermal cycles during GMA
bead-on-plate welding of steel were measured using
embedded PlatinumRhodium thermocouples. A range
of welds with heat inputs of from 0.5 to 1.5 kJ mm 1
at welding speeds of from 200 to 500 mm min 1 were
carried out on 10 mm thick mild-steel plate using 78.5%
Ar, 20% CO2 shielding gas. The fusion zone and HAZ
geometries were obtained by conventional metallography. In addition, equipment to rapidly remove the
workpiece from the arc, invert it and eject the weld
pool was devised and built (Fig. 4). Welding is carried
out on a spring-loaded plate, which is released at a
point during welding, rapidly accelerating and inverting
the empty weld-pool cavity. The mechanical ejection rig
works on the principle of rapidly accelerating and
decelerating the test plate being welded. A non-contacting laser transducer was used to scan and map the
weld-pool cavity. The device used to digitise the weld
pools is shown in Fig. 5 and the map of a typical
weld-pool cavity is shown in Fig. 6. Although simple,
this technique provides more information than a weldbead section, since the full pool-shape can be measured
and compared with 3D melting point isotherms predicted by the process models.

M.A. Wahab et al. / Journal of Materials Processing Technology 77 (1998) 233239

237

Fig. 7. (a) and (b) Comparison of predicted and experimental values of T8/5 and penetration from a 2D-section and 3D quasi steady-state models
for a bead-on-plate weld, using a split heat source representation. (c) Comparison of predicted and experimental weld pool length from a
2D-section and 3D quasi stead-state models for a bead-on-plate weld, using a split heat source representation.

5. Results and discussion


Fig. 7(a), (b) and (c) illustrates the contrast between
experimental and predicted values for both the 2D-section and 3DQSS models. The T8/5 cooling times are
slightly overestimated, Fig. 7(a), for both models, with
some suggestion of greater error in the 2D-section
model for larger T8/5 values that correspond with high
heat input welds (1.5 kJ mm 1). The weld penetrations
are also slightly overestimated, Fig. 7(b), in both
models with the 3DQSS values being closest overall to
the experimental values.

It is apparent that using an appropriate heat source


description, fairly simple and hence computationally
efficient models can give reasonable predictions for a
range of welding conditions. The important factor here
is the inclusion of a dependence on welding current,
both in terms of its influence on the volume of deposited material (size of weld bead) and on the depth of
the droplet source (parameter b, in Fig. 3). This approach results in calculated weld pools of the right
shape and reflects the formation of wine-glass shaped
weld pools at high arc current. Although more geometric factors are needed to define the source than with the
double ellipsoid, it is acceptable to use constant values

238

M.A. Wahab et al. / Journal of Materials Processing Technology 77 (1998) 233239

Fig. 7. (Continued)

for the arc efficiency, a, cb, and cf and the radius of the
droplet source, for this range of bead-on-plate welds.
Other parameters are then varied as a function of the
welding current.
The major discrepancy here is in the weld pool length
(Fig. 7(c)), which is a significant underestimate of the
measured values, particularly at high welding speeds,
corresponding to high arc current welds. It is probable
that this effect is caused by the failure of conductionbased models to adequately compensate for convective
flow within the pool.
In these conductive models, increased pool length
could be achieved by off-setting more heat energy behind the arc, although this must be done without
greatly reducing penetration. Here, it may be more
appropriate to artificially enhance the thermal conductivity in the molten material. This has been applied
isotropically in these models to a level of 5 10 times
the normal level, but the results suggest greater amplification may be appropriate, perhaps the use of an
anisotropic increase along the pool length, as suggested
by Mangonon and Mahimkar [13].

6. Conclusions
1. It has been shown that relatively simple 2D-section
and 3D quasi steady-state conductive heat-transfer
models adequately simulate the gas metal arc welding process, provided that a suitable representation
of the welding heat source input is used.
2. A split heat source in which the total heat content
of the welding process is divided according to some
simple ratios that are dependent primarily on the

welding current, has been shown to produce reasonable predictions of T8/5 cooling times and weld
penetrations for a range of bead-on-plate welds
from 0.51.5 kJ mm 1 heat input.
3. Weld pool lengths are under-predicted due to, it is
suggested, the failure of these conduction models
to adequately compensate for convective flow.

Acknowledgements
The authors gratefully acknowledge the support received from the Cooperative Research Centre for Materials Welding and Joining, Australia, for this study.
Experimental work was carried out by Liew, D. Gunter, M. Sparrow and L. Jarvis their contributions being
acknowledged with thanks.

References
[1] D. Rosenthal, Mathematical theory of heat distribution during
welding and cutting, Welding J. Res. Suppl. 20 (5) (1941) 2205
2345.
[2] J. Goldak, M. Bibby, Computational and Thermal Analysis of
Welds: Current Status and Future Directions, Proceedings of
Modelling and control of Castings and Welding Processes IV, in:
A.F. Giamei, Q.J. Abbaschian (Eds.), Minerals, Metals and
Materials Society, May 1988, pp. 153 167.
[3] J.F. Lancaster, The Physics of Welding, IIW Publication, Pergamon, Oxford, 1984.
[4] M.J. Painter, L. Jarvis, Numerical Modelling the SAW process,
Institute Of Metals and Materials Conference On Modelling and
Control of Materials Processing, Wollongong, November 1992.
[5] C. Liew, M.A. Wahab, M.J. Painter, The Prediction of Temperature Distribution in Gas Metal Arc Welds, Proceedings Welding Technology Institute of Australia/Australian Institute of

M.A. Wahab et al. / Journal of Materials Processing Technology 77 (1998) 233239


Non-destructive Technology, Fabcon/Fabfair Conference, September 1993.
[6] J.A. Goldak, A. Chakravarti, M.J. Bibby, A new finite element
model for welding heat sources, Met. Trans. 13 (15B) (June
1984) 299 305.
[7] J.A. Goldak, A. Chakravarti, M.J. Bibby, A Double Ellipsoid
Finite Element Model for Welding and Heat Sources, IIW
Doc-212-603-85, January 1985.
[8] M.H. Davis, M.J. Painter, M. Wahab, A Comparison of Weld
Pool Shapes Predicted form Models incorporation Conduction
and Convection, 42nd Annual Welding Conference, Melbourne,
October 1994, paper 27.
[9] E. Friedman, Thermomechanical Analysis of the Welding Process Using the Finite Element Method, Trans. Am. Soc. Mec.

239

Eng. J. Pressure Vessel Tech. 97 (3) (Aug. 1975) 206 213.


[10] P. Tekriewal, J. Mazumdar, Finite Element Modelling of Arc
Welding Processes, in: S.A. Davies (Ed.) Proceedings Trends in
Welding Research 86 Advances in Welding Sci. and Tech., ASM,
May 1986, pp. 71 83.
[11] T.J.T. Hughes, I. Levit, J. Winget, Element by element implicit
algorithms for heat conduction, J. Mech. Eng. 109 (2) (1985)
576 585.
[12] I.M. Smith, D.V. Griffiths, Programming the Finite Element
Method, Wiley, Chichester, 1988.
[13] P.L. Mangonon, M.A. Mahimkar, A Three Dimensional heat
Transfer Finite Element Method for SAW of HSLA Steels,
Proceedings TWR 86 Advances in Welding Sci. and Tech., in:
S.A. David (Ed.), ASM, May 1986, pp. 35 47.

You might also like