Numerical Heat Transfer

ISSN: 0149-5720 (Print) (Online) Journal homepage: http://www.tandfonline.com/loi/unht19

ENTHALPY-POROSITY TECHNIQUE FOR MODELING

CONVECTION-DIFFUSION PHASE CHANGE:
APPLICATION TO THE MELTING OF A PURE METAL

A. D. Brent , V. R. Voller & K. J. Reid

To cite this article: A. D. Brent , V. R. Voller & K. J. Reid (1988) ENTHALPY-POROSITY

TECHNIQUE FOR MODELING CONVECTION-DIFFUSION PHASE CHANGE: APPLICATION
TO THE MELTING OF A PURE METAL, Numerical Heat Transfer, 13:3, 297-318, DOI:
10.1080/10407788808913615

To link to this article: http://dx.doi.org/10.1080/10407788808913615

Published online: 16 May 2007.

Submit your article to this journal

Article views: 782

View related articles

Citing articles: 348 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=unht20

Download by: [University of Sydney Library] Date: 15 August 2017, At: 12:44
 Numerical Heat Transfer, vol. 13, pp. 297-318, 1988

ENTHALPY-POROSITY TECHNIQUE FOR

MODELING CONVECTION-DIFFUSION PHASE
CHANGE: APPLICATION TO THE MELTING
OF A PURE METAL

A. D. Brent, V . R . Voller, and K . J . Reid

Mineral Resources Research Center, Universiry of Minnesota,
Minneapolis. Minnesota 55455
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

The melting of pure gallium in a rectangular cavity has been numen'cally investigated
using the enthdpy-porosity approach for modeling combined convection-diffusion
phase change. The major advantage of this technique is that 1 allows a frred-grid
solution of the coupled momentum and energy equ&'ons to be undertaken without
resorting to variable transformations. In this work, a two-dimensional dynamic model
b used and the influence of laminar natural-convectionflow on the melling process
is considered. Excellent agreement exists between the numeriEol predictions and ex-
perimental results available in the literature. The enthalpy-porosity approach has
been found to Converge rapidly, and is capable of producing accumte results for both
the position and morphology of the melt front at dgferent times with relatively modest
computational requirements. These results may be mken to be a sound v a l M o n of
this technique for modeling i s o t h e m l phase changes in metallurgical systems.

INTRODUCTION

Phase-change processes play an important role in both extraction and fabri-

cation operations in the metallurgical industries. These processes involve complex
heat and mass transfer phenomena that determine the ultimate strength and integrity
of metallic components by governing the microstructures formed. The ability to pre-
dict phase-change phenomena in metallurgical systems is therefore of considerable
engineering interest. A significant effort has recently been directed to the theoretical
modeling of phase-change problems [ 1- 121.
Phase change in a metallurgical system represents a moving-boundary problem.
A number of techniques are available for the solution of these problems, and a com-
prehensive review can be found in Crank [I]. Much of the early work in this area
was concerned primarily with phase change in which conduction was assumed to be
the principal mechanism of energy transfer. In most practical situations involving a
liquid/solid phase change, however, convection effects (either forced or free) are
also of importance and in fact may totally dominate over conduction in some cases.

Computations were carried out using an AT&T 3B2/400 microcomputer and a Cray 2/1 super-
computer. The AT&T 3B2/400 was pan of a computer donation by AT&T lnformation Systems, Inc.,
to the Civil and Mineral Engineering Department of the University of Minnesota. Time on the Cray 2/
1 was provided by a resource grant from the Minnesota Supercomputer Institute. The support by both
AT&T lnformation Systems. Inc.. and the Minnesota Supercomputer Institute is gratefully acknowl-
edged.

Copyright 0 1988 by Hemisphere Publishing Corporation 297

 298 A. D. BRENT ET AL.

NOMENCLATURE
coefficients in the discretized energy length of rectangular cavity
equation height of rectangular cavity
porosity function for the momentum thermal diffusivity, k / c
equations volumetric coefficient of thermal
source term in the discretized energy expansion
equation nodal latent heat
specific heat porosity of computational cell, LW/L
morphology constant relaxation factor
latent heat function dynamic viscosity
acceleration due to gravity density
sensible enthalpy
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

thermal conductivity
latent heat of fusion Subscripts
Rnndtl number, c p / k
effective pressure abs quantity absorbed over a time step
Raleigh number, gp(T, - T,.)Y'p2/ap error error in conservation balance
buoyancy source term for v momentum hot heated wall conditions
equation int initial and cold wall conditions
source term for energy equation in m melting point
terms of enthalpy n iteration counter
source term for energy equation in nb neighboring node points
terms of temperature P pth node point or control volume
source term for u momentum equation ref reference conditions
source term for v momentum equation
temperaturn
velocity vector Superscript
velocity components in x and y
directions 0 time-dependent coefficients, also old
x, y coordinates values at orevious time steo

The modeling- methodologies developed for melting or solidification studies

may broadly be classified into three groups:

1. Empirical: A possible engineering approach is to apply a conduction treat-

ment and account for the enhanced energy transfer resulting from convec-
tion by boosting the liquid conductivity. This approach has been applied
successfully to some metallurgical problems 113-151.
2. Classical: A more rigorous approach is to fully account for convection in
the liquid region. This may involve deriving suitable governing heat and
mass transfer equations and the development of accompanying numerical
techniques for their solution. Difficulties arise if the governing equations
are based on the classical Stefan formulation, i.e., a temperature-based heat
transfer equation. The main reason for this is that in the vicinity of the phase
change, conditions on temperature, velocity, and latent heat evolution have
to be accounted for. This effectively rules out the application of a fixed-
grid numerical solution, as deforming grids or transformed coordinate sys-
tems are required to account for the position of the phase front. Examples
of combined convection/diffusion phase-change studies employing these
techniques may be found in Sparrow et al. 121, Ramachandran et al. 131,
 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE 299

Gadgil and Gobin [4], Albert and O'Neill 151, and Sparrow and Ohkubo
161.
3. Enrhalpy: An alternative to the empirical and classical approaches outlined
above is the so-called enthalpy formulation, which allows a fixed-grid so-
lution to be undertaken. This approach removes the need to explicitly satisfy
conditions at the phase front and is therefore able to utilize standard solution
procedures for the fluid flow and energy equations directly, without re-
sorting to mathematical manipulations and transformations. This approach
is particularly attractive in cases where phase change may constitute only
part of a complex problem. Examples of fixed-grid solutions of convective/
diffusion phase-change problems may be found in Morgan [7], Gartling [8],
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

and Voller et al. [9- 121.

The enthalpy methods account for the latent heat in the energy equation by
assigning a nodal latent heat value to each computational cell according to the tem-
perature of the cell. Upon changing phase, the nodal latent heat content of the cell
is adjusted to account for latent heat absorption or evolution, this adjustment being
reflected in the energy equation as either a heat source or sink. A distinctive ad-
vantage of this arrangement is that no explicit conditions for energy conservation at
the solid/liquid interface need to be accounted for. A major problem, however, with
fixed-grid solution procdures, is accommodating the zero velocity condition, which
is required as a liquid region turns to solid. Various methods have been used to
"switch off" velocities in computational cells that are freezing (or 'switch on" ve-
locities in the case of melting). Velocities may simply be set to zero in a compu-
tational cell when the mean latent heat content reaches some predetermined value
between 0 (cell all solid) and L (cell all liquid), where L is the latent heat of the
phase change 171. A somewhat more subtle approach is employed by Gartling [8],
in which the viscosity of a cell is driven to a very large value as the latent heat
content of the cell falls to zero. This increasing viscosity provides the necessary
coupling between the physical state of the material in the cell and the momentum
equations, thereby driving velocities to zero in cells that are solid. Voller et al. [9-
121 have examined various methods for dealing with the zero solid velocity condition
in fixed-grid solutions for freezing in a thermal cavity, and have proposed an alter-
native but similar approach to that adopted by Gartling. Computational cells that are
undergoing a phase change (i.e., 0 < AH < L) are modeled as pseudo-porous media,
with the porosity, e, being a function of AH and ranging between 1 (fully liquid
cell) and 0 (fully solid cell). The present work involves the application of this "en-
thalpy-porosity" technique to the two-dimensional melting of a pure metal in a rect-
angular cavity. A comparison is made between the results predicted by this method
and experimental values available in the literature in order to validate the applica-
bility of this technique for metal systems.

THE ENTHALPY-POROSITY APPROACH

In considering the phase change of an alloy, three distinct regions will be pres-
ent: a solid region, a totally liquid region, and a mushy region consisting of liquid
 300 A. D. BRENT ET AL.

dispersed among solid dendrites. To model such a phase change, the basic conser-
vation equations have to be solved throughout the calculation domain.

Governing Equations
The energy equation may be written in terms of the sensible enthalpy, h =
$;s,cuT as follows [12]:

Xph)
-+ div(puh) = div(a grad h) + S,,
at
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

where a is the thermal diffusivity ( k / c ) , and Sh is a source term that will be discussed
in more detail below. In the full liquid and mushy regions, fluid flow equations are
required. These equations consist of the conservation of momentum equations (one
for each velocity component) and a conservation of mass equation. Under the as-
sumptions of Newtonian laminar flow, these equations are

~(Pu)
-+ div(puu) = div(p grad u) -
a~ + Au
-
at ax

- ap
a(Pv) + div(puv) = div(p. grad v) - - + Av + S,
at ay

where u = (u,v) is the velocity, P is an effective pressure, and p is the viscosity.

Assuming the Boussinesq treatment to be valid-i.e., density is constant in all terms
except a gravity source term-natural-convection effects can be accounted for on
defining the buoyancy source term to be

where p is a thermal expansion coefficient and h,, and p,, are reference values of
enthalpy and density, respectively. The function of the parameter A in the momentum
equations will be discussed more fully below.

Source Terms
In using a fixed-grid approach for the analysis of solidification and melting
systems, a central difficulty is in accounting for mass and heat transfer conditions
in the vicinity of the phase change. The basic approach for overcoming this problem
is to define appropriate volume source terms for the governing equations. In the
enthalpy-porosity approach, the latent heat evolution is accounted for on defining
the energy-equation source term [Sh in Eq. (I)] to have the form [12]:
 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE 301

a(p w' + div(pu AH)

S* = -
ar

where AH = f(T), the latent heat content, is defined as a function of temperature.

In the case of an isothermal phase change, the t e r n div(pu AH) vanishes and the
function f(T) will be defined as follows:
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

where T, is the phase-change temperature and L is the latent heat of fusion. For a
mushy-region phase change, the temperature and latent heat relationship can be found
via experiment or one of the available metallurgical relationships; e.g., the Scheil
equation [I61 can be used.
The condition that all velocities in solid regions are zero is accounted for in
the enthalpy-porosity approach by appropriately defining the parameter A in Eqs. (2)
and (3). The basic principle is gradually to reduce the velocities from a finite value
in the liquid, to zero in the full solid, over the computational cells that are changing
phase. This can be achieved by assuming that such cells behave as porous media
with porosity E = the "element liquid fraction" [12]. The porosity can be computed
as AH/L (i.e., the fraction of liquid in the cell), where AH is the latent heat content
of the cell in question. In cells that are undergoing a phase change, A is defined so
that the momentum equations are forced to mimic Carman-Kozeny equations for flow
in a porous media [12]:

(1 -
grad P = [-C $]u.

where C is a constant accounting for the mushy-region morphology and u, is an

apparent velocity. In order to achieve this behavior, an appropriate definition of A
is

where b is merely a computational constant introduced to avoid division by zero.

In practice the effect of A is as follows: In full liquid elements, A is zero and
has no influence; in elements that are changing phase, the value of A will dominate
over the transient, convective, and diffusive components of the momentum equa-
tions, thereby forcing them to imitate the Carman-Kozeny law; in totally solid ele-
ments, the final large value of A will swamp out all terms in the governing equations
and force any velocity predictions effectively to zero. The validity of the Carman-
Kozeny equation in describing the flow in a dendritic mushy region has been in-
vestigated experimentally [17], and thus application of Eq. (9) has physical signif-
icance. Further details on the implementation of the enthalpy-porosity approach for
 302 A. D. BRENT ET AL.

mushy-region phase changes can be found in [9-121. It should be noted that the
constant C may significantly influence the morphology of the phase front, and care
must be taken in assigning a value to it. This constant effectively controls the degree
of penetration of the convection field into the mushy region. Further work needs to
be carried out in the area of mushy-region phase change to establish guiding prin-
ciples in assigning appropriate values for both the constants C and b in Eq. (9).
In the case of an isothermal phase change, however, the flow inside compu-
tational cells that are melting or freezing does not follow the governing laws for flow
in a porous media. In an isothermal phase-change system, the phase front is a well-
defined line and has no band or width associated with it as in the case of a mushy
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

phase change. In a discretized calculation domain, however, the phase front will
have a finite width associated with it. This width will generally be only that of a
single computational cell, and hence any convenient method for gradually extin-
guishing the velocities in a solidifying cell will be acceptable. A possible alternative
to applying the Carman-Kozeny equations is to use a linear function for A , that is,

One advantage of using a simple linear function is that it is slightly less ex-
pensive to compute than the Carman-Kozeny formula. The influence of A will again
be gradually to override the other terms in the discretized momentum equations in
solidifying control volumes, thereby forcing velocity predictions in solid regions to
zero. Although the value of the constant C in Eqs. (9) and (10) will not influence
the morphology of the phase front as significantly as in the case of a mushy-region
phase change, it should be sufficiently large to force velocity predictions in solid
regions essentially to zero. It should be mentioned that although any convenient
method may be acceptable for extinguishing velocities in solidifying cells in iso-
thermal phase-change systems, it is desirable that the method chosen allows a smooth,
gradual transition rather than a step change in velocity. Step changes in the mo-
mentum-equation source terms tend to retard convergence of their numerical solu-
tion, and sometimes lead to oscillations that may result in divergence. Inappropriate
switching methods for the velocities may also lead to reduced accuracy, particularly
when coarse grids are employed. A quantitative evaluation of the influence of dif-
ferent switching techniques has been conducted by Voller et al. [lo].

Numerical Implementation of the Enthalpy-Porosity Technique

Once a suitable expression has been selected for the velocity-switching function
A in the momentum equations, standard solution procedures for the coupled mo-
mentum and continuity equations may be used without any modification. In this
study, the SIMPLE procedure described by Patankar was applied [IS]. To ensure a
converged solution of the energy equation, however, special treatment of the en-
thalpy source term [Eq. (6)] is required. In the numerical solution of the energy
equation, the latent heat content of each control volume needs to be updated ac-
cording to the temperature values predicted from the energy equation after each it-
eration in the computational cycle. A reliable method for updating the latent heat
content of each computational cell has been developed by Voller et al. [lo, 121. This
 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE 303

method avoids oscillations in the iterative procedure that could result in the inability
to achieve a converged solution, and may be applied in both finite-element- and
control volume-based finitediierence numerical schemes. As a control volume-based
numerical scheme was used in this work, the following discussion will be limited
to the control volume approach. Following the control volume-based formulation and
notation in Patankar [18], the discretized energy equation [Eq. (I)] may be written
as follows:

where the subscripts indicate the appropriate node point values, the a's are coeffi-
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

cients that depend on the combined convection/diffusion fluxes into the pth control
volume, d contains energy sources and sinks for the pth control volume, and a; is
the unsteady coefficient linking the present h, value to its value at the previous time
step, h;. For this discretized energy equation, the nodal latent heat value for the pth
control volume is updated after the nth iteration according to Voller et al. [lo, 121
as

where f is the inverse of the latent heat function [Eq. (7) for an isothermal phase
change] used to describe the latent heat content of each computational cell. The
enthalpy update expression given by Eq. (12) is entirely general, i.e., it is applicable
to both mushy-region phase changes that occur over a temperature range as well as
to isothermal phase changes. It is only the form of the latent heat function f(T) that
would distinguish an isothermal phase change from a phase change occumng over
a temperature range. As this work is concerned with the application of the enthalpy-
porosity technique to a pure metal system, the following discussion will be restricted
to isothermal phase changes. It should be noted that during iterations it is possible
that use of Eq. (12) to update the nodal latent heat values may lead to nonphysical
values, i.e., nodal latent heat values greater than L or less than zero. The first of
these conditions will occur if more than an entire control volume changes phase in
a single time step, and the second may arise if a control volume is close to starting
the phase change at the beginning of a time step. To prevent the nodal latent heat
contents from assuming physically unrealistic values, the iterative update of [AH,],+,
should be restricted between zero and L. For an isothermal phase change, the com-
putational implementation of the enthalpy update given by Eq. (12) may thus be
simply accomplished as follows:

where A is a relaxation factor.

 304 A. D. BRENT ET AL.

Equation (13) illustrates the physical significance of the enthalpy update, i.e.,
the nodal latent heat values are adjusted according to the difference between the
nodal temperature predicted from the energy equation and the phase-change tem-
perature. In the case of freezing, the nodal latent heat value assigned to computa-
tional cells undergoing a phase change provides a heat source in the energy equation
to account for latent heat evolution. This source expires when the latent heat content
of the cell is exhausted. At this point AHp = 0 and the cell is totally solid. Con-
versely, in a melting situation the nodal latent heat value of a computational cell
undergoing a p h a ~ echange provides the energy equation with a heat sink to account
for the absorption of latent heat. This sink disappears when the total latent heat of
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

the phase change is reached. At this point Mp= L and the cell is totally fluid.
While iterating, corrections in the nodal latent heat values of computational cells in
which a phase change is occurring are driven by the difference between the tem-
perature predicted from the energy equation and the phase-change temperature. The
enthalpy update scheme attempts to annihilate this difference s o that at convergence,
the temperature of all cells in which both liquid and solid coexist will be that of the
phase-change temperature, T,,,.This scheme thus accurately reflects the physical state
of an isothermal phase change in a discretized domain, i.e., the phase front is 'smeared"
across the width of the computational cells in which the phase change is occurring.
The ratio ap/ai in the nodal latent heat update arises from the derivation of
Eq. (12) [I21 and is a "normalizing" factor that accounts for the number of grid
points and size of time step used. This ratio increases upon both grid refinement and
on increasing size of time step, thereby compensating for these changes in the en-
thalpy update equation and ensuring rapid convergence of the method. The physical
significance of the ap/aj term is as follows:

I. For a given size of time step, a larger number of conml volumes will change
phase during any particular time step if a fine grid is used. As the ap/aj
term increases as a function of the number of grid points, it forces more
rapid changes in the nodal enthalpy values near the phase front when a fine
grid is employed. This allows the true position of the phase front to be
quickly established in the discretized domain.
2. For a fixed grid size, increasing the size of the time step results in a larger
number of control volumes changing phase during each time step. As the
value of the ratio a p / a i increases according to the time step size, this term
ensures rapid changes in the nodal enthalpy values in the region of the phase
change when large time steps are used. This again quickly establishes the
true position of the phase front in the discretized domain.

MELTING OF PURE GALLIUM I N A RECTANGULAR CAVITY

T o verify the applicability of the enthalpy-porosity approach to metallurgical

systems, the modeling of an isothermal phase change was undertaken for which re-
liable experimental data exists in the literature. The study examined the two-dimen-
sional melting of pure gallium in a rectangular cavity with one heated wall. This
subject was chosen for the following reasons.
 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE 305

1. If the enthalpy-porosity technique is to be applied to metallurgical systems,

it is essential to verify the accuracy of its predictions for a substance that
is representative of metal properties. Comparisons to other materials such
as water/ice or organic phase-change materials (PCMs) may not be valid
if the technique is to be used for metallurgical applications. Large dissim-
ilarities between the thennophysical properties of organic compounds and
metals result in considerably different flow structures and phase-front mor-
phologies [19-2 I].
2. Gallium has a melting point close to the ambient (29.7g°C), and experi-
ments with it may therefore be well controlled and are able to yield complete
information about a melting or freezing front (e.g., the shape and position
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

of the front may be traced at different times). The high melting temperatures
of most practical structural metals (e.g., steel and titanium) are not con-
ducive to experimentation, and great difficulties exist with obtaining reliable
phase-change data for these materials. As a result of the convenience of
working with gallium, experimental data available in the literature may be
assumed to be both comprehensive and accurate.
3. A comprehensive experimental study on the melting of gallium in a rect-
angular cavity has been conducted by Gau and Viskanta [21]. The authors
present detailed traces of the morphology of the melt front at various times.
The narrow temperature range (lO°C) used by these investigators allows the
assumption of constant physical properties, which simplifies a computa-
tional simulation of this work. The results obtained by Gau and Viskanta
[21] also show a great deal of morphology of the melt front. It is believed
that if the numerical analysis is capable of accurately predicting the acute
morphology obtained in the experiments, this would represent a sound val-
idation of the enthalpy-porosity approach for isothermal phase changes in
metallurgical systems.

It should be noted that there is no fundamental difference between applying

the enthalpy-porosity technique to either melting or solidifcation problems. The
equations and numerical scheme developed for melting are equally applicable to freezing
situations; the particular type of phase change that occurs is determined only by the
initial and boundary conditions of the problem.

Calculation Domain
A diagram of the calculation domain and boundary conditions is given in Fig.
1. The values of Ti,,, Tho,,X, and Y were taken from one of Gau and Viskanta's
experiments [21] for which the melt front had been traced in detail at a number of
different times. With reference to Fig. 1, these values were Tin,= 28.3OC, Tho,=
3 8 T , X = 8.89 cm, and Y = 6.35 cm. The physical properties for pure gallium are
well documented [22-251, and the values used in this investigation are given in Table
1. The physical property values were chosen at 32°C. a temperature that is repre-
sentative of the temperature range of the experiment (i.e., 28.3-38°C). The Prandtl
number for liquid gallium at this temperature is 0.0216, and the situation shown in
Fig. 1 corresponds to a Stefan number of 0.039 and a Raleigh number of 6 X 10'.
 A. D. BRENT ET AL.

Top Wall (insulated)

------------------

Solid Gallium
Heated Wall (That)- -Cold Wall (Tint)
I n i t i a l Temperature (Tint)
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

I
I Bottom wall' (insulated)
I
I
!

Fig. 1 Calculation domain and boundary conditions.

It should be mentioned at this point that while gallium may be an ideal metal
for phase-change studies from an experimental point of view, the fact that the thermal
conductivity of solid, crystalline gallium is highly anisotropic complicates a theo-
retical analysis. The conductivity of the solid crystal ranges from 88.5 W . m-' . K - I
along the b axis to 16 W . m-' . K-I along the c axis [22]. The thermal conductivity
of polycrystalline gallium at the melting point (29.78"C) is 33.5 W .m-I . K-I [22].
However, the problem of anisotropic solid thermal conductivity is alleviated to a
large degree by examining a melting rather than a feezing problem. In melting, the
only disturbance the system is subjected to when the experiment is started at time t
= 0, is that one of the walls is heated to a temperature above the melting pint (see
Fig. I). Once melting is initiated against this wall, all the heat transfer into the
system occurs through a layer of molten liquid. It is therefore the physical properties
of the liquid rather than the solid that will govern the phase change in a melting
process. If freezing is studied, however, the major resistance to energy transport
resides in a solid layer growing on a cold wall, and it is then the thermal conductivity

Table 1 Physical properties of Pure Gallium

Density (liquid), p 6093 k g . m-'

Reference density, p, 6095 kg. m-'
Reference temperature, T,, 29.78"C
Volumetric thermal expansion coefficienl of liquid, P 1.2 X lo-'
Thermal conductivity, k 32.0 W .m-' . K"
Melting point, T, 29.78C
Latent heat of fusion, L 80160 J .kg-'
Specific heat capacity, C 381.5 J .kg-'
Dynamic viscosity. p 1.81 X lo-' kg. m-' . s-'
Prandtl number. PI 2.16 X lo-'
 ENTHALPY-POROSITY TECHNlQUE FOR PHASE CHANGE 307

of the solid that exerts a controlling influence on the phase change. The problems
associated with freezing studies on gallium have been noted by Gau and Viskanta
[21], who were unable to obtain reproducible results for the phase front position and
morphology when conducting solidifcation studies on the metal. It should be noted
that although the thermal conductivity of liquid gallium is also anisotropic (although
to a lesser degree than that of the solid) [22], the presence of convection currents
will tend to average out directional differences in thermal conductivity.

Numerical Solution
The conservation equations assuming two-dimensional laminar flow in carte-
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

sian coordinates with constant physical properties are given in the appendix, along
with appropriate boundary conditions for this problem. These equations represent a
set of four nonlinear, coupled partial differential equations, which were solved it-
eratively using the control volume-based finite-difference procedure described by
Patankar [IS]. A fully implicit formulation was used for the time-dependent terms
and the combined convection/diffusion coefficients were evaluated using Patankar's
power law scheme. The SIMPLE algorithm was applied to solve the momentum and
continuity equations to obtain the velocity field. A line-by-line solver based on the
TDMA was used to iteratively solve the algebraic discretization equations. Details
of the numerical method and solution procedures may be found in Patankar [IS].
The nodal enthalpy update procedure described by Voller et al. [lo, 121 was used
to iteratively adjust nodal enthalpy values [i.e., Eq. (13)l. A flow chart of the com-
putational algorithm is given in Fig. 2.
After conducting a grid-refinement study, a uniform 42 X 32 grid was chosen
for this work. A time step of 5 s was applied for the first four time steps to resolve
the rapid initial changes in temperature near the heated wall, after which a constant
time step of 10 s was used. Further refinements in the grid or time step size did not
produce discernable improvements in accuracy. In order to be consistent with the
general methodology developed for mushy-region phase changes, the Carman-Koz-
eny formula [Eq. (9)] was applied for switching the velocities in cells undergoing a
phase change. A value of 1.6 X lo6 was used for the morphology constant C and
a value of 1 X for the constant b in Eq. (9). Underrelaxation factors of 0.5,
0.6, and 0.9 were used for the solution of the two momentum equations, the pressure
correction equation, and the energy equation, respectively. Underrelaxation of the
enthalpy update [Eq. (13)] was also required to avoid divergence in this highly non-
linear, coupled situation. It was found advantageous to ramp the underrelaxation
factor A in Eq. (13) from an initial value of 0.2 at the beginning of each time step
to a maximum of 0.4 as convergence was neared. The position of the phase front
at desired times was determined by plotting the L/2 nodal latent heat contour. This
was found to provide a smoother, more precise estimate of the phase-front position
than a plot of the melting-point isotherm. The reason for this is that all control vol-
umes in which both liquid and solid coexist will be at the melting point (T,,,), and
hence the T,,, isotherm represents a band rather than a well-defined line. The rationale
behind using the L/2 contour is the assumption that the phase front at any given
time passes through control volumes that posses half the latent heat of the phase
change.
 A. D. BRENT ET AL.

Initial temperature and

velocity fields at time t - 0

A t
set Toid -- "T
Uold
Void -
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

Step forward in time

t - t i d t

Calculate sources ST, S,

1
and S,
I
Update nodal Latent
heat velirea according
to temperature
Solve for u, v and T
Calculate overall momentum A
and energy balances

Write out results if

required for time t

NO
Y

Fig. 2 Flow chan of the computational algorithm.

Convergence Criteria and Computational Requirements

The convergence criteria for the numerical solution of the conservation equa-
tions given in the appendix were based on both the maximum local mass imbalance
within the cavity (reflecting convergence of the flow field) and on an unsteady over-
all energy balance (indicating convergence of the energy equation). As the energy
equation contains a source term (viz., the latent heat term p(aAH/ar)J that is not
coupled directly to the momentum equations, it is necessary to monitor both the mass
 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE 309

and energy balances to ensure a fully converged solution. The unsteady energy bal-
ance was calculated as the sum of the total energy flowing into the cavity at the two
end walls and the energy absorbed by the material in the cavity, over each time step
Ar, as follows:

where
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

and

The energy absorption term contains a contribution from the latent heat change
over the time step. The integrals in Eqs. (15)-(17) were evaluated numerically by
summing over the respective control volumes. The temperature derivatives in Eqs.
(15) and (16) were evaluated in a fully implicit manner to be consistent with the
numerical scheme used. Convergence was declared after the maximum local mass
imbalance became less than I X w 4 % of the total mass present in the cavity and
the absolute value of the energy balance error dropped below 1 X lo-*%. Increasing
the stringency of the convergence criteria beyond these values did not result in sig-
nificant improvements in accuracy. An average of 4 3 iterations per time step were
needed to achieve convergence. The computations were conducted on both an AT&T
3B2/400 microcomputer and a Cray 2/1 supercomputer. The longest simulation car-
ried out (i.e., a melt time of 19 min, spanning 116 time steps) required 26 CPU
hours on the microcomputer and 9 CPU min on the supercomputer. No attempt was
made to optimize or vectorize the code. Such steps may be expected to produce
significant reductions in CPU times. In particular, code vectorization would result
in substantially faster execution on the Clray 2/1. However, these CPU times com-
pare very favorably with those quoted in the literature for comparable investigations
[26, 271. These requirements may still be considered modest for the complex prob-
lem investigated, and indicate that two-dimensional fine-grid solutions based on the
enthalpy-porosity technique are entirely feasible using a microcomputer. For three-
dimensional phase-change analyses, however, it is anticipated the application of vec-
torized code on a supercomputer would be required to yield acceptable execution
times.

Resutts and Discussion

The position of the melt front and streamlines at 3, 6, LO, and 19 min are
shown in Figs. 3a-6a. The melt front is virtually planar after 3 min as the natural
 A. D. BRENT ET AL.
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

Fig. 3 Flow and temperature fields at 3 min for the

two-dimensional melting of gallium. Panel (a) depicts
the streamlines and melt front; panel (b) the isotherms.

convection field has just begun to develop (Fig. 3a). Figure 4a shows the developing
morphology after 6 min when the natural convection has intensified and is beginning
to have a pronounced influence on the overall energy transport from the heated wall
to the melt face. The morphology of the melt front is as expected; fluid rising at the
heated wall travels across the cavity and impinges on the upper section of the solid
front, thereby resulting in this area melting back beyond the mean position of the
front. The influence of the convection currents intensifies with time, and the mor-
phology of the melt front becomes more acute as melting progresses. Figures 5a and
6a show the melt front and streamlines after 10 and 19 min, respectively. From Fig.
6a it is evident that after 19 min the shape of the melt front is governed primarily
by convection effects, with conduction exerting very little influence. Although the
upper section of the melt front advances rapidly due to the impingement of warm
fluid, the lower section moves considerably slower, thereby resulting in the acute
morphology illustrated in Fig. 6a.Development of the intensity of the free convection
field may be traced by examining Fig. 7, in which the maximum absolute value of
the stream function is plotted against time. In the early stages of melting the max-
 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE 311

imum absolute value of the stream function increases rapidly as the temperature gra-
dients driving the convection currents begin to develop. As the volume of the molten
fraction in the cavity increases with time, the influence of viscous stresses against
the cavity walls decrease, thereby allowing the fluid velocity (and consequently the
absolute value of the stream function) to rise. The fluid motion continues to intensify
until it levels off after approximately 12 min, by which time the natural convection
field in the cavity is well developed.
The regions of maximum heat flux may be identified by examining the isotherm
plots given in Figs. 3b-6b. The steepest temperature gradients are present in two
areas, viz., at the upper section of the melt face where warm fluid impinges on the
solid, and at the lower section of the heated wall where cool fluid returning from
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

the melt front impinges on the heated wall. The 'insulatingn nature of the melt front
in the cavity is also evident from the isotherm plots. Almost the entire temperature
drop across the cavity occurs in the molten region, leaving the temperature of the
residual solid essentially uniform near the initial temperature. The reason for this is
that virtually all the energy entering the cavity at the heated wall is absorbed by the

Fig. 4 Flow and temperature fields at 6 min for the

two-dimensional melting of gallium. Panel (a) depicts
the streamlines and melt front; panel ( b ) the iso-
themls.
 A. D. BRENT ET AL.
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

Fig. 5 Flow and temperature fields at 10 min for the

two-dimensional melting of gallium. Panel (a) depicts
the streamlines and melt front; panel ( b ) the iso-
therms.

phase change at the melt front, allowing only a small fraction to penetrate the slightly
supercooled solid. As the major temperature gradients occur in the liquid phase, it
is the thermal conductivity of the liquid rather than the solid that controls heat trans-
fer, and hence the problems associated with the theoretical treatment of anisotropic
solid conductivity d o not arise.
The melt fronts calculated in this study and those determined experimentally
by Gau and Viskanta [21] at various times are plotted in Fig. 8. From Fig. 8 it is
evident that both the qualitative behavior and acute morphology of the experimental
melt fronts have been realistically duplicated in the numerical study. An appreciation
of the accuracy of the numerical predictions may be obtained by studying Fig. 9, in
which the computed fronts at 2, 6, 10, and 17 min are compared directly to their
experimental counterparts. Excellent agreement exists between the computed and
experimental melt front positions, and small discrepancies between the measured and
calculated results may be attributed to factors such as three-dimensional effects in
the experimental apparatus used to determine the front position, experimental error,
and variations in fluid properties.
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

Fig. 6 Flow and temperature fields at 19 min for the

two-dimensional melting of gallium. Panel (a) depicts
the streamlines and melt front; panel (b) the iso-
therms.

Fig. 7 Relationship between the maximum absolute value of

the stream function and time for the two-dimensional melting of
gallium.
 A. D. BRENT ET AL.

CONCLUSIONS
The applicability of the enthalpy-porosity technique for modeling an isothermal
phase change in a metallurgical system has been verified by examining the two-
dimensional melting of pure gallium under the influence of natural convection in a
rectangular cavity. Results obtained from this numerical study were compared to
experimental data available in the literature. The method converges rapidly and is
capable of accurately predicting both the position and morphology of the melt front
at various times with relatively modest computational requirements. The results of
this work may be taken to be a sound validation of the enthalpy-porosity technique
for simulating isothermal phase changes in metallurgical systems.
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

Time (min)

Fig. 8 Calculated and experimental melt fronts at various times. Panel

(a) shows melt front positions determined experimentally by Gau and
Viskanta 1211; panel (b) depicts the theoretical melt front positions cal-
culated in this study.
 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

Numerical
- -- Experimental (Gau and Viskanta (211)
Fig. 9 Comparison of calculated and experimental melt fronts for the two-dimensional melting
of gallium.

Validation of this approach for mushy phase-change systems is to be considered

next. A detailed study of the quantitative influence of the morphology constant C in
a mushy phase-change situation is also required as the criteria for assigning a value
to this parameter need to be established. The application of the enthalpy-porosity
method to the modeling of "freeze-linings" or "skulls" in metallurgical processes
will also be undertaken. The ability to predict both the morphology and position of
these protective solid layers that isolate a melt from the structural components of the
vessel in which it is contained is of particular interest in the field of ultra-pure metals
processing.

APPENDIX
Conservation Equations and Boundary Conditions for the
Two-Dimensional Melting of Gallium
1. Conservation equations
1.1 Continuity
 A. D. BRENT ET AL.

1.2 Momentum equations

x direction:

y direction:
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

where

1.3 Energy equation

where

2. Boundary conditions
2.1 Left wall (x = 0):

u=v=o
T = Tho,

2.2 Top and bottom walls O, = 0 and y = Y):

2.3 Right wall (x = X):

 ENTHALPY-POROSITY TECHNIQUE FOR PHASE CHANGE

REFERENCES

1. J. Crank, Free and Moving Boundary Problems. Clarendon Press, Oxford, 1984.
2. E. M. Sparrow, S. V. Patankar, and S. Ramadhyani, Analysis of Melting in the Presence
of Natural Convection in the Melt Region, J. Heat Transfer, vol. 99, pp. 520-526, 1977.
3. N. Ramachandran, I . R. Gupta, and T. Jalunu, Thermal and Fluid Flow Effects During
Solidification in a Rectangular Cavity, Int. J . Heat Mass Transfer, vol. 25, pp. 187-
194, 1982.
4. A . Gadgil and D. Gobin, Analysis of Two-Dimensional Melting in Rectangular Enclo-
sures in the Presence of Convection, J. Heat Transfer, vol. 106, pp. 20-26, 1984.
5. M. R. Albert and K. O'Neill, Transient Two-Dimensional Phase Change with Convec-
tion Using Deforming Finite Elements, in R. W. Lewis et al. (eds), Computer Techniques
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

in Heat Transfer. vol. I , Pineridge Press, Swansea, 1985.

6. E. M. Sparrow and Y. Ohkubo, Numerical Analysis of Two-Dimensional Transient Freezing
Including Solid-Phase and Tube-Wall Conduction and Liquid-Phase Natural Convection,
Numer. Heat Transfer, vol. 9, pp. 59-77, 1986.
7. K. Morgan, A Numerical Analysis of Freezing and Melting with Convection, Comp.
Methods Appl. Eng., vol. 28, pp. 275-284, 1981.
8. D. K. Gartling, Finite Element Analysis of Convective Heat Transfer Problems with
Phase Change, in K. Morgan et al. (eds.), Computer Methods in Fluids, Pentech, Lon-
don, 1980.
9. V. R. Voller, N. C. Markatos, and M. Cross, Solidification in Convection and Diffusion,
in N. C. Markatos et al. (eds.), Numerical Simulations of Fluid Flow and HeallMass
Transfer Processes, Springer-Verlag, Berlin, 1986.
10. V. R. Voller, N. C. Markatos, and M. Cross, Techniques for Accounting for the Moving
Interface in Convection/Diffusion Phase Change, in R. W. Lewis and K. Morgan (eds.),
Numerical Methods in Thermal Problenis, vol. 4 , Pineridge Press, Swansea, 1985.
I I. V. R. Voller, M. Cross, and N. C. Markatos, An Enthalpy Method for Convection/
Diffusion Phase Change, Int. J . Numer. Methods Eng., vol. 24, pp. 271-284, 1987.
12. V. R. Voller and C. Prakash, A Fixed Grid Numerical Modelling Methodology for Con-
vection/Diffusion Mushy Region Phase Change Problems, Int. J . Heat Mass Transfer,
vol. 30, pp. 1709-1719, 1987.
13. C. Gau and R. Viskanta, Melting and Solidifcation of a Metal System in a Rectangular
Cavity, Int. J . Heat Mass Transfer, vol. 27, pp. 113-123, 1984.
14. J. Szekely and P. S. Chhabra, The Effects of Natural Convection on the Shape and
Movement of the Melt-Solid Interface in the Contmlled Solidification, Met. Trans. B ,
vol. 1, pp. 1195-1203, 1970.
15. F. M. Chiesa and R. I. L. Guthrie, Natural Convection Heat Transfer Rate During the
Solidification and Melting of Metals and Alloy Systems, J . Hear Transfer, vol. 99, pp.
520-526, 1977.
16. W. K u n and D. J. Fisher, Fundamenrals of Solidifcation, Trans. Tech., Switzerland,
1986.
17. D. R. Poirier, Permeability for Flow of lnterdendritic Liquid in Columnar-Dendritic Al-
loys, Met. Trans. B . . vol. 18B, pp. 245-256, 1987.
18. S. V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere, Washington,
D.C.. 1980.
19. C. Benard, D. Gobin, and F. Maninez, Melting in Rectangular Enclosures: Experiments
and Numerical Simulations, J . Heat Transfer, vol. 107, pp. 794-802, 1985.
20. B. W. Webb and R. Viskanta, Natural-Convection-Dominated Melting Heat Transfer in
an Inclined Rectangular Enclosure, Int. J . Heat Mass Transfer, vol. 29(2), pp. 183-192,
1986.
 318 A. D. BRENT ET AL.

21. C. Gau and R. Viskanta, Melting and Solidification of a Pure Metal on a Vertical Wall,
J . Heat Transfer, vol. 108, pp. 174-181, 1986.
22. W. H. Cubberley (ed.), Metals Handbook-Properties and Selection: Nonferrous Alloys
and Pure Metals, 9th ed.. American Society of Metals, Metals Park, Ohio, pp. 736-
737, 1979.
23. Gallium and Gallium Compounds, in Kirk-Othmer Encyclopedia of Chemical Technoi-
ogy, 3d ed., vol. l I, p. 606, Wiley, New York.
24. M . J. Duggin, The Thermal Conductivity of Liquid Gallium, Phys. Lett.. vol. 29A(8),
pp. 470-47 1, 1969.
25. A . 0 . Ukanwa, Diffusion in Liquid Metal Systems, NASA Contractor Technical Report
NASA-CR-144016, p. 9, June 1975.
26. C. J . Ho and R. Viskanta, Heat Transfer During Melting from an Isothermal Vertical
Downloaded by [University of Sydney Library] at 12:44 15 August 2017

Wall, J. Heat Transfer, vol. 106(c), pp. 12- 19, 1984.

27. G . E. Schneider, Computation of Heat Transfer with Solid/Liquid Phase Change In-
cluding Free Convection, J. Thermophys. Heat Transfer, vol. 1(2), pp. 136-145, 1987.

Received September 2 1 . 1987

Accepted November 3. 1987

Requests for reprints should be sent to V. R. Voller.