You are on page 1of 7

International Journal of Modern Physics B

Vol. 23, Nos. 6 & 7 (2009) 1584–1590


 World Scientific Publishing Company

MULTISCALE MODELING OF THE VACUUM ARC REMELTING PROCESS


FOR THE PREDICTION ON MICROSTRUCTURE FORMATION

LANG YUAN*, GEORGI DJAMBAZOV†, PETER D. LEE*,‡ and KOULIS PERICLEOUS†


*
Department of Materials, Imperial College London,
Int. J. Mod. Phys. B 2009.23:1584-1590. Downloaded from www.worldscientific.com
by NORTH CAROLINA STATE UNIVERSITY on 03/23/13. For personal use only.

London, SW7 2AZ, UK



p.d.lee@imperial.ac.uk

School of Computing and Mathematics, University of Greenwich,
London, SE10 9LS, UK

Received 28 November 2008

The final solidification structures of Vacuum Arc Remelting (VAR) ingots depend on the
temperature distribution and fluid motion within the molten pool. In this paper, a three-dimensional
multi-physics macroscale model for VAR is developed, based on the modular CFD software
PHYSICA. This model is used to provide estimates of process parameters and to study complex
physical phenomena, such as liquid metal flow with turbulence, heat transfer, solidification, and
magnetohydrodynamics in the VAR process. The macromodel is coupled to a microscale
solidification model. The micromodel combines stochastic nucleation and a modified decentred
square/octahedron method to describe dendritic growth with a finite difference computation of solute
diffusion. The resulting multiscale model allows prediction of the formation of microstructures in the
solidifying mushy zone. This gives a better understanding of the whole VAR process from
operational conditions to final ingot microstructures, as well as an essential first step in defect
prediction
Keywords: Vacuum arc remelting; multiscale modeling; magnetohydrodynamics; dendritic growth.

1. Introduction
Vacuum arc remelting (VAR) is commercially employed for producing fully dense and
homogeneous ingots of reactive and segregation sensitive alloys,1 such as INCONEL
718. The VAR process consists of melting a consumable electrode under a high vacuum
to produce a secondary ingot (Fig. 1). The pool dynamics and solidification conditions
are controlled by the process parameters, such as the arc current, voltage and electrode
gap. Fluctuations in these parameters will alter the motion of melt metal in the pool and
lead to thermal and compositional perturbations in the mushy zone. Thus, various defects
can occur in the remelted ingot such as tree ring patterns2, freckles3 and white spots4.
Multiscale and multi-physics numerical techniques have been used to enhance the
understanding of the wide range of physical and chemical phenomena including fluid
flow, heat transfer, electromagnetic effects and the formation of microstructures in the
VAR process. Macroscale VAR models5, 6 not only allow the visualization of the process
but also provide the field variable inputs required by microscale solidification models to

1584
Multiscale Modeling of the Vacuum Arc Remelting Process 1585

simulate corresponding microstructures. Xu et al.7 combined a 2D axisymmetrical model


with a CA mesoscale model of the grain nucleation and growth. By coupling the
predicted thermal profile, the multiscale model successfully studied the formation of tree-
rings. Nastac et al.8 developed a similar model to analyze the effects of production rate
and ingot size on the resultant secondary phases and freckles. Noticeably, current
macroscale VAR models are all 2D axisymmetric. However, observed arc motion and
pool surface motion lead to transient non-axisymmetric behavior of the liquid pool.9, 10
Thus, a 3D transient model is required to study instabilities in the process. On the other
hand, in order to study the mechanisms of defect formation in depth, a finer scale model
is required to simulate dendritic growth during the VAR solidification process.
Int. J. Mod. Phys. B 2009.23:1584-1590. Downloaded from www.worldscientific.com
by NORTH CAROLINA STATE UNIVERSITY on 03/23/13. For personal use only.

Fig. 1. Schematic principle of the VAR process Fig. 2. Boundary conditions for 3D transient VAR model.
(after Ref. 11).

In this paper, a 3D transient macroscale VAR model (based on the modular finite
volume software package, PHYSICA12, 13) was developed, allowing heat transfer, fluid
flow, turbulence, and magnetohydrodynamics (MHD) to be considered. A microscale
solidification model was coupled with the macroscopic multi-physics model to simulate
dendritic growth during the solidification process.

2. Model Theory
The Three-Dimensional VAR Macromodel, which takes into account fluid flow, heat
transfer, electromagnetic effects and solidification, is modeled by the governing
equations for mass, momentum, energy:

( )

∇ ⋅ ρu = 0 (1)

∂   
( ρ u ) + ∇ ⋅ ( ρ uu ) − ∇ ⋅ µ∇u = −∇P + Sf (2)
∂t
1586 L. Yuan et al.

∂ρT λ
( )

+ ∇ ⋅ ρ uT = ∇ 2T + Sh (3)
∂t cp

where u is the velocity vector in liquid, P is the pressure, µ is the viscosity augmented
by the effective turbulent viscosity, cp is specific heat, λ is thermal conductivity.
Additional forces were simulated via source terms: A Darcy type source for flow in the
semi-solid zone; electromagnetic field generated Lorentz force and buoyancy force:
 (1 − f l )2      T − T
Sf = −Π   u + J × B − g ρβ ref
(4)
 fl3  c
  p

where Π is the permeability of partially solidified regions. β is the volume expansion


Int. J. Mod. Phys. B 2009.23:1584-1590. Downloaded from www.worldscientific.com
by NORTH CAROLINA STATE UNIVERSITY on 03/23/13. For personal use only.

coefficient, f l is the fraction liquid. In this model,the fraction


 liquid is a function of the
local temperature, calculated as a linear function. J and B are the electric current density
and the magnetic flux density, respectively. The heat source term, Sh , accounts for the
latent heat and Joule heat produced inside a unit volume per unit time is defined by:
 2
∂ ( ρ fl L ) J
( )

Sh = − − ∇ ⋅ ρ u fl L + (5)
∂t σe
where L is the latent heat, σ e is electrical conductivity. The electromagnetic behaviour
for DC power  is determined by solving
 a Laplace equation for the electric potential φ .
The current J and magnetic field B that generate the Lorentz force are obtained by
Ohm’s Law and the Biot-Savart integral respectively (e.g. see Hughes14)
The system of equations was solved by the finite volume method on an unstructured
mesh covering the upper part of the ingot. For the electromagnetic calculation, the
domain was extended to contain also the lower part of the electrode and part of the
copper crucible, shown in Fig. 3a. A steady-state simulation assuming axisymmetric
distributions of current input was used as the initial condition for 3D transient runs. In the
transient model, the arc was assumed to have a time-dependent localized Gaussian
distribution. Its effective centre was located at a distance of 0.1 m from the ingot
centreline and it rotated counterclockwise around it with a period of 36 s. The boundary
conditions were given in Fig. 2. These assumptions reflect arc behaviour observed during
experiments9, 10. The same distribution was used for the heat inflow at the top surface.
The heat of the molten drops was part of the inflow but the momentum of the drops was
neglected. Simulation conditions and the thermophysical properties of INCONEL 718
can be found in Yuan15.
The Microscale Solidification Model combines a stochastic nucleation model and a
modified decentred square/octahedron method16 to describe dendritic growth with a finite
difference computation of solute diffusion. Assumptions and equations for the grain
nucleation and dendrite growth are given in detail in prior publications.17-19
Because of the scale difference between the micromodel and macromodel, a post-
processing link was developed to couple the two models.15 The temperature for
microstructure formation was calculated by spatial and temporal linear interpolation from
nodal values in the macromodel and was passed into the microscale solidification model.
Multiscale Modeling of the Vacuum Arc Remelting Process 1587

3. Results and Discussion


3.1. Prediction of the liquid pool behavior
According to the experimental measurements by Ward et al.10 , approximately 55% of the
total 6 kA electric current input flows through the crucible to the ingot and then to the
electrode. Therefore, 3.3 kA were distributed on the electrode face and the same amount
with opposite sign on the liquid pool surface immediately below the electrode. Because
part of the side-arcing heat is absorbed by the electrode and thus helps with its heating
and melting, it was assumed that 90 kW total heat power was applied to the pool top
surface (i.e. 65% instead of 55% as for the current). The predicted steady results are
Int. J. Mod. Phys. B 2009.23:1584-1590. Downloaded from www.worldscientific.com
by NORTH CAROLINA STATE UNIVERSITY on 03/23/13. For personal use only.

shown in Fig. 3.
By applying the movement of the assumed arc distribution and additionally solving
the full 3D electromagnetic field, the 3D transient phenomena in the VAR process were
simulated. As the transient run was initiated from the steady state simulation, it required
enough time to develop rotational Lorentz flow in the pool. Two time steps were selected
to show the typical behaviour, 54s and 72s, both after the arc has rotated one full circle.

z (mm) Magnetic field, y=0 0.2 MA/m 2 z (mm) Liquid Fraction, y=0
50 mm/s
0
200

B (T ×10 )
3
fl
-100 Zone 3
100

Zone 2
0 -200 Zone 1

-100 -300

-200
-400
x (mm) -200 -100 0 100 200 x (mm)
-200 -100 0 100 200

(a) (b)
Fig. 3. Predicted results of the steady state simulation at the vertical plane when y = 0: (a) magnetic field and
current path on the vertical cut plane; (b) the liquid pool (Zone 1 to Zone 3 indicates the locations for
micromodel investigating the dendrite growth).

The temperature distribution, fraction of liquid and the motion in the pool are shown
in Fig. 4. From the top view, the position of the highest temperature is close to, but a little
behind, the arc center due to the fast movement of the arc. The large circulation in the
liquid pool was controlled by Lorentz force in Figs. 4(c) and (d), and the circulation was
moving with the arc center where the leading Lorentz force was generated. The flow
driven by buoyancy force can also be found at the upper corner opposite to the position of
the arc centre, sweeping the liquid metal downward at the crucible wall and back along
the solidification front. The buoyancy force is much smaller than the Lorentz force and
has less effect on melt transport in the liquid pool. The Lorentz force in the pool takes hot
liquid from the top surface down to the depth of the pool, causing local remelting. This
can slightly change the shape of the liquid pool by comparing Figs. 4(e) and (f ).

(b)
1588 L. Yuan et al.

Temperature, z = 0 Temperature, z = 0

y (mm)
50 mm/s (a) y (mm) 50 mm/s (b)
200
T( ) ℃ 200
T( ℃)
100 100

0 0

-100 -100

-200 -200

-200 -100 0 100 200 x (mm) -200 -100 0 100 200 x (mm)

Temperature, y = 0 Temperature, y = 0
z (mm)
(c) z (mm)
(d)
Int. J. Mod. Phys. B 2009.23:1584-1590. Downloaded from www.worldscientific.com
by NORTH CAROLINA STATE UNIVERSITY on 03/23/13. For personal use only.

50 mm/s 50 mm/s
0 0

T( )℃ ℃)
T(
-100 -100

-200 -200

-300 -300

-400
-400
-200 -100 0 100 200 x (mm) -200 -100 0 100 200 x (mm)

Liquid Fraction, y=0 Liquid Fraction, y=0


z (mm)
0
(e) z (mm)
0
(f )
fl fl
-100 -100

-200 -200

-300 -300

-400 -400
-200 -100 0 100 200 x (mm) -200 -100 0 100 200 x (mm)

Fig. 4. 3D transient VAR simulation results: (a) and (b) are temperature profiles of the ingot top; (c) and (d) are
temperature profiles of the vertical plane at y=0; (e) and (f ) are liquid fraction of the pool in the vertical plane at
y=0. (a), (c) and (e) are at 54s; (b), (d) and (f) are at 72s.

3.2. The formation of dendrites in the VAR process


Ni-based superalloy INCONEL718 is approximated as a Ni-Nb binary alloy
demonstrating the effect of dendritic growth without the high computational cost
associated with the full multi-component system. The properties used can be found in
Ref. 19. Three small domains (10 mm×20 mm) were selected from the middle, the inside
and the side of the ingot to investigate the microstructures in VAR. By interpolating the
temperature from the macroscale VAR model spatially and temporally, the temperature
profiles were obtained for each node on the microscale. The simulation results are shown
in Fig. 5.
Multiscale Modeling of the Vacuum Arc Remelting Process 1589

(a) (b) (c)

(d) (e) (f)


Int. J. Mod. Phys. B 2009.23:1584-1590. Downloaded from www.worldscientific.com
by NORTH CAROLINA STATE UNIVERSITY on 03/23/13. For personal use only.

Zone 1 Zone 2 Zone 3


Fig. 5. Predicted grain nucleation and dendritic growth in the mushy zone (zone 1 to zone 3 shown in
Fig. 3a). (a), (b) and (c) is the initial stage of the dendrite growth; (d), (e) and (f) is the near-final
solidified stage.

Directional solidification conditions were obtained for dendrites growing in the VAR
process. The average thermal gradient measured from the macromodel was more than
3500 K/m. Dendrites are prone to be formed in columnar morphology under this
environment. Columnar dendrites were simulated at the region near the crucible wall as
well as the middle radius part of the ingot (Figs. 5(e) and (f )). However, equiaxed
dendrites were extensively formed in the zone close to the central line of the ingot
(Fig. 5(c)). This can be understood by examining the local thermal gradient. The
thickness of the liquid pool at the center of the ingot is almost as twice that near the
crucible wall. Therefore, the thermal gradient close to the crucible wall is nearly double
that at the center. Then, it leads to the formation of columnar dendrites near the crucible
wall, which is under large thermal gradient, as exhibited by the simulations. In the center
region, equiaxed dendrites commonly mix with columnar dendrites and partially block
some individual columnar dendrites from growing. Columnar dendrites dominate near the
crucible wall. The morphology of dendrites predicted by the multiscale model is
consistent with the experimentally observed grain structures.7 Moreover, the solute
distribution in the region was also calculated, allowing the analysis of microsegregation
during the solidification process. This could provide key information on the formation of
1590 L. Yuan et al.

defects, such as freckles, which are recognized as macrosegregation, but also for
optimizing the upcoming processes, such as during heat treatment.

4. Conclusions
A 3D multi-physics macromodel for simulating the transient phenomena in the VAR
process was developed by solving heat transfer, fluid flow, turbulence, and MHD effects.
Experimentally measured arc behavior was applied for the boundary conditions. From the
predicted results, the motion of melt metal in the liquid pool was controlled by
electromagnetic force leading to thermal fluctuation in the solidifying mushy zone. A
microscale solidification model was coupled with the macromodel. Columnar dendritic
Int. J. Mod. Phys. B 2009.23:1584-1590. Downloaded from www.worldscientific.com
by NORTH CAROLINA STATE UNIVERSITY on 03/23/13. For personal use only.

growth was simulated in the middle radius and lateral parts of the ingot. The predicted
results show the same characteristics as experimental measurements. The information of
microsegregation can be helpful in understanding the formation of solidification defects
and may be used to optimize the upcoming processes.

Acknowledgments
The authors would like to acknowledge the EPSRC (EP/D50502 and EP/D505011) and
ORS for financial support, Corus, Special Metals Wiggin and Rolls-Royce plc for project
support and their colleagues at Imperial College London and University of Greenwich.

References
1. F. J. Zanner and L. A. Bertram, in Int. Conf. on Vac. Metal., ed. (Linz, Austria, 1985), 512.
2. X. Xu, et al., Metall. Mater. Trans. A 33, 1805 (2002).
3. A. Choudhury, Vacuum Metallurgy (ASM International, 1990).
4. L. A. Jackman, et al., Adv. Mater. Process 143, 18 (1993).
5. A. Jardy and S. Hans, in MCWASP-XI, ed. C. A. Gandin, et al. (Opio, France, 2006), p. 953.
6. G. Reiter, et al., in Liq. Metal Proc. Cast., ed. P. D. Lee, et al. (Nancy, France, 2003), p. 77.
7. X. Xu, et al., Metall. Mater. Trans. A 33, 1795 (2002).
8. L. Nastac and A. Patel, in MCWASP-XI, ed. C. A. Gandin, et al. (Opio, France, 2006), p. 961.
9. R. M. Ward, et al., in Liq. Metal Proc. Cast., ed. P. D. Lee. (Santa Fe, New Mexico, 2005),
p. 49.
10. R. M. Ward and M. H. Jacobs, in Liq. Metal Proc. Cast., ed. P. D. Lee, et al. (Nancy, France,
2003), p. 49.
11. A. D. Patel, et al., in Int. Symp. Superalloys, ed. K. A. Green, et al. (Champion, PA, 2004),
p. 917.
12. www.physica.co.uk.
13. M. Cross, C. Bailey, K. Pericleous, et al., JOM-e, 2002,
14. M. Hughes, K. Pericleous, N. Strusevich, Math. Modelling Weld Phenomena 6, 63-81 (2002).
15. L. Yuan, et al., in Liq. Metal Proc. Cast., ed. P. D. Lee, et al. (Nancy, France, 2007), p. 43.
16. C. A. Gandin, et al., Metall. Mater. Trans. A 30, 3153 (1999).
17. P. D. Lee, et al., Mater. Sci. Eng. A 328, 213 (2002).
18. P. D. Lee, et al., Mater. Sci. Eng. A 365, 57 (2004).
19. W. Wang, et al., Acta Mater. 51, 2971 (2003).

You might also like