Professional Documents
Culture Documents
Voller
An implicit enthalpy
solution for phase change
problems: with application
to a binary alloy
solidification
V. R. Voller
The purpose of this paper is twofold: called node-jumping scheme in which the space grid
remains fixed but the time step changes ensuring that
(1) to introduced an implicit enthalpy scheme for fast the phase change boundary is always on a node point.
and accurate solution to one-dimensional phase change This scheme requires the application of an implicit time
problems representation of the enthalpy formulation. Recently,
(2) to modify this scheme and apply it to the solution Voller,’ developed an implicit enthalpy method which
of a binary alloy solidification problem dovetails very neatly into the node-jumping technique.
It is this combination which is presented and developed
Enthalpy methods have been a popular means of in this paper.
numerically solving phase change problems for some In the first instance, the new scheme will be demon-
time.1’2 A major reason for this is that an enthalpy for- strated on a simple test problem. After this, the scheme
mulation removes the need to satisfy conditions on the will be modified to deal with a problem of a binary
moving boundary, which means that fixed domain solu- alloy solidification, a problem which involves a phase
tions can be used. Direct applications of numerical change along with coupled heat and mass transfer. The
enthalpy methods, however, may lead to results which major advantage of the node-jumping scheme when
have non-physical oscillations. Such behaviour has been applied to this problem is that in the numerical scheme
noted and analysed by a number of authors.3-6 In order there is no need to account for coupling between
to overcome, or bypass, these oscillations remedial enthalpy and concentration as is the case with previous
schemes have to be applied.2.7~s One of these is the so- numerical schemes. l&l2
0307-904X/87/02110-07/$03.00
110 Appl. Math. Modelling, 1987, Vol. 11, April 0 1987 Butterworth & Co. (Publishers) Ltd
Implicit enthalpy solution for phase change problems: V. R. Voller
Figure 7 Control volumes is a latent heat source term. The term AH in equation
(4) is the latent heat evolved in the phase change. The
functional relationship of AH with temperature, T,
depends on the nature of the phase change. In the case
Sol id Liquid Liquid of a single temperature phase change, refer to equation
(2), then:
0 T< T,,,
AH=
0 N+l L T> T,,,
Solid Liquid
Node jumping scheme
The scheme
In the majority of physical situations, it is necessary
to solve equation (3) via a numerical method. For a
one-dimensional problem an appropriate method is that
of finite difference control volumes. The region of inter-
est is subdivided into a set of discrete nodes. With each
node there is associated a volume such that values of
variables at the node are representative of the values
b AHN= 0 AHN+~ = L/2
in the volume. Three adjacent control volumes are
Figure2 One node jump: (a) time t; (b) time (t + ST) shown in Figure 1. From this figure, the following finite
difference representation of equation (3) may be de-
rived:
The enthalpy formulation
Enthalpy, H, is defined as the sum of sensible and latent
heats. For a melting/freezing problem under conduc-
tion, the enthalpy formulation is usually written as:
where:
half will be liquid with a latent heat value of L. Hence, Example problem
the representative nodal latent heat will be: A material in a liquid state and uniform temperature
Ti = 2 > T,,, = 0 (the phase change temperature) fills
AH; = L/2
the half space x > 0. At time t = 0, the temperature
Now, assume that the numerical time step t is such of the surface at x = 0 is lowered to T, = -10 so that,
that at time t + 6t the phase front is located on the as time advances, a solid layer builds up adjacent to
N + lth node (see Figure2(b)). Hence: the wall. The material has the following thermal proper-
ties, conductivity K = 2, specific heat C = 2.5 x lo6
AH,,,+, = L/2 and latent heat L = 1O8. The solution of this problem
via the node-jumping algorithm is compared with the
and the change in latent heats in the time step in control known analytical solution in Table 1. Note that the step
volumes N and N + 1 will be: size used in these results was 6x = 0.0625 and for the
first jump from x = 0 to x = 0.0625 a fully implicit
S, = AH; - AH, = L/2 mode (f3 = 1) was used, thereafter a Crank-Nicolson
and: mode (0 = f) was adopted.
Both the results for the movement of the boundary
S ,,,+, = AH;+I - AHN+, = L/2 and temperature, given in Table 1, show that the method
is accurate. In addition, the technique is very efficient,
If the time step was always chosen to ensure that the
requiring somewhere in the order of two to five iter-
phase front always ‘jumped’ from one node to the next
ations at each time step. This means that accuracy can
in a time step the form of the source term in equation
be taken to a fairly high level simply by using more
(6) would always be known in advance. For example,
nodes.
if the phase front was on node N at the start of the
It can be argued that the problem solved in Table
time step and node N + 1 at the end:
I does not present a very severe test on the method.
ifP= NorP= N+ 1 In the next section, however, the same node-jumping
Sp = L/2 principal will be used and the technique modified to
[0 otherwise solve a binary alloy problem, a more complex problem
This fact ensures that the system of equations resulting of coupled heat and mass transfer.
from equation 6 will be linear and tridiagonal and hence
can be easily and efficiently solved by the tridiagonal
matrix algorithm. The only unknown in the system is Binary alloy problem
the time step required to jump from node to node. The The problem
value of the time step can be approximated using an
iterative procedure, with the time step modified from The binary alloy problem is that of solidification of a
one iteration to the next. The point of convergence for metal containing a small concentration of impurity
such a procedure being reached when the predicted (solute). The model of the system used here is identical
value of h at node N + 1 is: to that proposed by Crowley and Ockendon.” Modifica-
tion and application of the node-jumping scheme, how-
h N+l = CT, ever, will result in a more accurate and efficient solution
than those previously presented. The developed solu-
The node-jumping procedure may be summarized as
tion will also be, to the author’s knowledge, the first
follows:
implicit technique proposed for general treatment of
(1) phase front on node N the binary alloy problem (note the implicit scheme pro-
(2) initial guess for time step (at), = previous value posed by Whitei requires that the phase change temper-
(3) set SN and SN+i = L/2, all other Si to zero ature T, remains fixed; in general this will not be the
(4) solve equation (6) for nodal distribution of sensible case).
heat An equilibrium phase diagram for a dilute binary alloy
(5) update value of time step (via a discretized Newton is shown in Figure 3 where the liquidus and solidus lines
method on the function hN+I - CT,,, = 0, for example) are assumed to be straight. Consider a small volume
(6) repeat steps (4) and (5) until convergence, e.g. until element of binary alloy in thermodynamic equilibrium
[CT,,, - hN+,l < TOL. at uniform temperature T and overall average concen-
tration of solute c. If conditions are such that c + kzT In a similar manner, after the jump when the front is
> 0 (k, = (liquidus slope))‘) then the volume is in the onnodeN+ 1:
stable liquid state. If c + k,T< 0 (k, = (solidus slope)-‘)
then the element is in a stable solid state. When the CN+I = (k, + M/2 Vv+,
values of c and T are such that they correspond to a These observations can be utilized in Crank-Nicolson
point between the solidus and liquidus lines, the volume finite difference representations of equations (10) and
is undergoing the phase change and both solid and liquid (11) to give for heat:
phases, at different concentrations of solute, will be
present separated by a plane front.
If equilibrium is maintained, a discontinuity in con- hp-h;=;F;+;Fp+Sp (14)
centration at the phase front will exist such that:
E,V,-E;V,“=;Gb+;G, (15)
k,
Cl = -c-2 (9)
k2
where GP = [D,.,( VW - VP) - D,(V, - V,)]
This discontinuity makes the concentration similar to Under a node-jumping regime:
the enthalpy function H( 7) previously introduced. The
1
enthalpy is a function of temperature T which is conti- ifP= NorP= N-t 1
Sp= L/2
nuous at the boundary. Concentration can be made a (16)
[0 otherwise
k2
function of the chemical activity, V,” defined by:
ci = k;V; ifP>N+l
0.125 -
0.025 -
0 I I I
0 0.1 0.2 0.3 0.4
X
figure3 Equilibrium diagram Figure 6 Concentration profile at r = 0.02 (key as for Figure
4)
2.0L
1.0 -
I I
3 0 0.1 0.2 0.3 0.4
X
t
Figure 7 Enthalpy profile at time r = 0.02 (key as for Figure
Figure 4 Temperature history at x = 0.1; (- 1analytic; (0) 4)
node jumping
o 0.075 - l
0.05 -
0.025 -
0 1 I 1
0 0.01 0.02 0.03
t
Figure 5 Cancellation history at X = 0.1 (key as for Figure 4) Figure 8 Phase front movement (key as for Figure41
final iterative time step is such that the sum of the pre- the surrounding control volume, a volume which is half
dicted values in the control volume changing phase are solid and half liquid, hence:
close to zero (i.e. TN+, + V,,, = 0) which is the require-
ment of the heat mass coupling. c/v = (k, + k&v/2 (18)
Another feature of the node-jumping solution is that The value cN is known, therefore V,,, can be calculated
it is implicit. Previous techniques have, in the main, and the liquid and solid concentrations on either side
been explicit so that the coupling between temperature of the front calculated as:
and concentration can be easily incorporated. The only
c, = k,V, cZ = k2V,\z (19)
other implicit technique known to the author, is that
due to White.‘” This technique, however, appears to respectively. Both of these values, as derived from the
assume that the phase change occurs at a fixed value numerical concentration when the front is on x = 0.1,
of temperature T,,, and chemical activity V,,,. This are shown in Figure 5. The sharp manner in which the
removes the need for a coupling relationship because jump discontinuity is modelled by the node-jumping
the nature of a control volume, i.e. solid or liquid can results is a notable feature. In the Crowley and Ocken-
be determined purely on the temperature value. Assum- don results (Figure 2(b) of reference 11) the numerical
ing a fixed temperature may not always be sound and concentration history indicates a smearing of the discon-
some problems may exhibit variations in the phase tinuity.
change temperature and concentration as the phase Figures 6 and 7 show the enthalpy and concentration
change proceeds. The node-jumping technique deter- profiles at time t = 0.02. Once again, these results have
mines the nature of a control volume from the sum T high accuracy and discontinuities are well modelled by
+ V, an approach that will allow for variations. similar relationships to those given in equations
(18)-(19).
Additional information which is a natural con-
An application sequence of the node-jumping method is the movement
To illustrate the application of the node-jumping binary of the front. Numerical predictions are compared with
alloy solution the test problem introduced by Crowley analytical ones in Figure 8. As with all other results,
and Ockendon” is used. A binary mixture with an equi- the accuacy is good. Furthermore, note that other
librium diagram as in Figure 3, initially at uniform tem- enthalpy-based methods will not provide a front position
perature Ti = 1 and concentration c, = 0.1 fills the half prediction quite so readily.
space x > 0. At time t = 0 the surface temperature
is lowered to T, = -1, so that phase change can com-
Conclusions
mence. It is assumed that there is no mass flow across
the surface x = 0 and the properties of the alloy are: In this paper, the enthalpy node-jumping technique for
the solution of one-dimensional phase change problems
Kr = K,= 1 k, = 1 kZ = 2
first proposed by Voller and Cross’,’ has been updated.
Specific heat, C = 1 L=l The major modification has been to incorporate the
implicit numerical enthalpy technique reported by
D,=D2=1
Voller.” This technique dovetails into the node-jumping
The latter expression implies that the mass diffusivities concept and results in an efficient solution procedure
take the values 1 for the solid and f for the liquid. for heat conduction problems which can utilize the tri-
Rubinsteini4 has proposed an analytical similarity diagonal matrix algorithm and requires few iterations
solution for the binary alloy problem. In certain cases, per time step.
however,i5 the implicit assumption made by Rubinstein The basic node-jumping technique has been extended
of a plane front breaks down and a mushy region occurs. and applied to a problem of binary alloy solidification.
For the test problem used here, no such breakdown A problem which involves both heat and mass transfer.
occurs and the Rubinstein analytical solution reported Enthalpy techniques have been applied to this problem
by Crowley and Ockendon” can be used for comparison before. In the main, however. these techniques are
purposes. explicit in nature and require an additional relationship
Solutions of the problem using a step of 6x = 0.0125 to account for the coupling between the temperature
are compared in Figures 4-8. Figure 4 shows the temper- and concentration,” or if the technique is implicit not
ature history of the point x = 0.1. The agreement always physical assumptions are required.‘” The node-
between the analytical and numerical solutions is to jumping technique presented here requires no ad-
within three significant figures. Furthermore, the oscilla- ditional relationships in that the numerical equations
tions noticeable in the results of Crowley and Ockendon are solved in an uncoupled form. The heat/mass coupl-
(Figure 2(a) of reference 11) are not present in the node- ing is then accounted for by choosing a time step which
jumping solution. Note, however, that in some later moves the tront exactly one node space. In addition,
work, Crowley,ih employing the remedial scheme of the technique is implicit, yet still applicable to a general
Tacke,” has also removed the oscillations. binary alloy problem in which the phase change temper-
Figure 5 gives the concentration history at x = 0.1. ature and concentration varies as the phase change pro-
Again the numerical results are very close to the analyti- ceeds.
cal solution and free of oscillations. The numerical Results using the node-jumping scheme appear to be
values of the concentration may be manipulated to pre- more accurate than previous methods retaining the
dict the jump discontinuity in the concentration history. expected physical behaviour especially with reference
The jump discontinuity will occur when the front is on to discontinuities and temperature history. The effi-
the node located at x = 0.1 (node N say). At this time ciency of the technique also means that accuracy may
the nodal concentration at x = 0.1 is representative of be achieved on adding more node points to the solution.
Extension of the application of the technique to 2 Voller, V.R. and Cross, M. Application of control volume enthalpy
problems which involve properties changing across the methods in the solution of Stefan problems, in Computational techni-
ques in heat transfer, (eds. Lewis, R. W., et al.), Pineridge, Swansea,
phase front should not present many problems. The 1985
major modification would be in the development of the 3 Price, P.H. and Slack, M.R. The effect of latent heat on numerical
finite difference equations, equations (6), (14)) and (15), solutions of the heat flow equation, Br. J. Appl. Phys., 1954, 5,
and not in the solution technique itself. 245-2 75
4 Bonacina, C., Comini, G., Fasano, A. and Primicerio, M. Numeri-
Problems in more dimensions or non-planar (i.e.
cal solution of phase change problems, In?. J. Heat Mass Transfer,
mushy region) problems cannot currently be solved by 1973, I&1825-I832
the node-jumping approach. If a multi-dimensional 5 Voller, V.R., Cross, M. and Walton, P.G. Assessment of weak
problem is to be solved, then the shape of the phase solution numerical techniques for solving Stefan problems, Numerical
change surface needs to be known a priori, so that the methods in thermal problems (eds. Lewis, R. W. and Morgan, K.),
Pineridae Press. Swansea. 1979. pp. 172-181
appropriate solid and liquid fractions in a control 6 Belly G.E. On the performance of the enthalpy method, Int. J.
volume can be calculated. In the case of a mushy region, Heat Mass Transfer, 1982, 25,587-589
phase change difficulties arise because a control volume 7 Voller, V.R. and Cross, M. Accurate solutions of moving boundary
cannot be separated into two distinct phases (i.e. solid problems using the enthalpy method, ht. J. Heat Mass Transfer, 1981,
24,545-556
and liquid) which is a current reqmrement of node jump-
8 Tacke, K. Discretization of the enthalpy method for planar phase
ing.In fact, it should be noted that the node-jumping change, in Free boundary problems: application and theory, Vol. IV
technique is only suitable for the binary alloy problem (eds. Bossavit, Damlamian and Fremond), Pitman, London, 1985
presented here in the circumstances that it retains a 9 Voller, V.R. Implicit finite-difference solutions of the enthalpy
plane front and does not exhibit the mushy region beha- formulation of Stefan problems, ZMA .I. Numer. Anal., 1985, 5,
201-214
viour noted by Wilson et al.15 Clearly more work is 10 Fix, G.J. Numerical methods for alloy solidification problems,
required to approach multi-dimensional and mush in Moving boundary problems (eds. Wilson, D.G., et al.), Academic
region problems via an enthalpy node-jumping or simi- Press, New York, 1978
lar technique. 11 Crowley, A.B. and Ockendon, J.R. On the numerical solution
of an alloy solidification problem, In?. J. Heat Mass Transfer, 1979,
In conclusion, the node-jumping enthalpy method 22,941-947
presented in this paper should provide a tool for efficient 12 Wilson, D.G., Solomon, A.D. andAlexiades, V. Amodel of binary
analysis of one-dimensional heat conduction phase alloy solidification, Int. J. Numer. Meth. Engin., 1984, 20,1067-1084
change problems. Furthermore, the success of the modi- 13 White, R.E. The binary alloy problem: existence, uniqueness
and numerical approximation, SIAM J. Numer. Anal., 1985, 22,
fication and application to the problem of binary alloy
205-244
solidification indicates that the principle of node-jump- 14 Rubinstein, L.I. The Stefan problem, Trans. of Math. Monogr.,
ing can cope with fairly sophisticated phase change 21,Am. Math. Sot., 1971
problems. 15 Wilson, D.G., Solomon, A.D. and Alexiades, V. A shortcoming
of the explicit solution for the binary alloy problem, Left. in Heat
References Mass Transfer, 1982, 9,421-428
16 Crowley, A.B. Numerical solution of alloy solidification problem
1 Crank, J. Free and moving boundary problems, Clarendon Press, revisited, in Free Boundary Problems: application and theory, Vol.
Oxford, 1984, pp. IV (eds. Bossavit, Damlamian and Fremond), Pitman, London, 1985