International Journal of Heat and Mass Transfer 67 (2013) 260–271

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

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

3D numerical and experimental study of gallium melting in a

rectangular container
O. Ben-David, A. Levy ⇑, B. Mikhailovich, A. Azulay
Pearlstone Center for Aeronautical Engineering Studies, Department of Mechanical Engineering, Ben-Gurion University of the Negev, Israel

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

Article history: Gallium melting in a rectangular container with a heated side face has been investigated. The focus of the
Received 18 December 2012 research is the advancement of the numerical model to 3D status taking into consideration thermal vari-
Received in revised form 6 July 2013 ations of medium properties and the presence of mushy zone, and the model experimental verification
Accepted 19 July 2013
using known data and new data obtained on a specially developed experimental setup. This setup is ori-
ented to trace the liquid–solid interface position and profile and melt velocity using ultrasonic Doppler
velocity measurements in the liquid phase.
Keywords:
The numerical model was based on COMSOL Multiphysics software. 2D and 3D versions were built to
Gallium melting
Convective flow
calculate the melting regime and the flow process characteristics. Outputs of both versions were com-
Liquid–solid interface pared with known experimental data and new data obtained in the present study.
Numerical model The results revealed a better agreement between 3D computational and experimental data indicating
3D effects to a profound effect of the container boundaries on the problem under study.
Ultrasonic velocimetry Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction
Metal melting in a rectangular container has been studied for
many years [1–18], but the respective analytical solutions are
known only for the simplest cases. Meanwhile, heat convection
and metal properties dependence on temperature play an impor-
tant role in real melting processes. They complicate the analysis
and make it necessary to perform computer simulations and vali-
dation by low-temperature physical experiments.
Both kinds of investigations complement each other, since the
assessment of computer simulation efficiency is realized by com-
parison with available experimental data, and in a number of cases
a certain discrepancy between computed and experimental results
is observed [3,18,20]. The latter can be due to either experiment
imperfections or computation methods deficiency.
The known numerical methods make it possible to solve melt-
ing and solidification problems taking into account both the heat
conduction mechanism, which is essential at the initial melting
stage, and the convective heat transfer playing an important role
in the dynamics of the melt and liquid–solid interface [17,18].
In this research, computer simulations and physical low-tem-
perature experiments have been carried out for gallium melt in a
rectangular container with a heated side face (Fig. 1). To improve
the comprehension of such a process, it is necessary to estimate
⇑ Corresponding author. Address: P.O. Box 653, Beer-Sheva 84105, Israel. Tel.:
+972 8 6477092; fax: +972 8 6472814.
E-mail address: avi@bgu.ac.il (A. Levy).
the influence of the numerical model dimension (2D and 3D), tem-
perature dependence of the metal physical properties and bound-
ary conditions on the problem solution. Therefore, it is advisable to
specify the necessary aspects and to modify computational ap-
proach and adapt experimental methods as required.
Industrial-scale hot melting experiments are very expensive
and complicated. Moreover, as known, until now there are no
workable technical means for velocity measurements in high-tem-
perature liquid metals. Therefore, melting experiments at low tem-
perature using gallium or PCM (phase change material) are
performed in various forms in order to understand the melting
process and heat transfer and flow mechanisms. Gau and Viskanta
[1,2], used pure gallium and performed vertical and horizontal
melting, measuring the solid–liquid interface temperature and po-
sition. For this purpose, thermocouples and special L-shaped
probes for draining the container filled with melt were used to ex-
pose the solid–liquid interface. These measuring methods are inva-
sive, disturb the melting process and do not provide information
about the melt flow.
A noninvasive method was applied by Campbel and Koster [3]
who used gallium and measured the solid–liquid interface shape
and position by real-time radioscopy method. X-ray observation
showed a sharp contrast between molten and solid phases with a
3% density difference. Another method can be found in Menon
et al. [4], Pal and Joshi [5] and Katsman et al. [6]. All of them used
different kinds of PCM and performed melting experiments in glass
tubes, which allowed the registration of the solid–liquid interface
shape and position by a high-contrast camera. These methods are

0017-9310/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2013.07.058
 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271 261

Nomenclature

B computational constant (kg/ms) Pr Prandtl number ml/al

cp specific heat (J/kg °C) Ste Stefan number cpl(TH  TC)/L
C Carman–Kozeny constant (kg/m3 s) TH TC Tint Temperatures of hot and cold walls and initial temper-
Fl liquid–solid volume fraction ature, accordingly
~
F force vector (N)
H enthalpy (J/kg) Greek letters
DH modified latent heat (J/kg) l dynamic viscosity (kg/ms)
L latent heat (J/kg) D difference
f(T) enthalpy temperature function (J/kg) q density (kg/m3)
k thermal conductivity (W/m °C) e computational constant
T temperature (°C or K) m kinematic viscosity (m2/s)
P pressure (Pa) K aspect ratio
t time (s) a thermal diffusivity (m2/s)
g gravity acceleration (m/s2)
d diameter (m)
Subscripts
V volume (m3) m melting
~
u velocity vector (m/s) s solid
x,y,z coordinate (m)
l liquid
Q heat quantity (J) p particle
I unit matrix
h container height
Ra Rayleigh number, (q  qm)gh3/qmalml

indeed noninvasive, but they cannot provide any quantitative
information about the melt flow in the liquid phase. Moreover,
they are not suitable for opaque media such as metals. All experi-
mental results show a great influence of convective flow on the
melting process and the solid–liquid interface shape.
One of noninvasive measurement methods in phase-change
experiments is the use of ultrasound instruments. Beginning with
medical applications, this method of measuring the velocity of
fluid flows found its use in scientific and engineering research,
especially when dealing with opaque and aggressive liquids,
including liquid metals [7] and [8].
As an example, one can cite the study [9], where mean veloci-
ties of the melt flow and liquid–solid interface have been found
experimentally during the study of the magnetic field effect on
the alloy solidification.
Unlike known experiments using UDV (ultrasonic Doppler velo-
cimeter) technique, in our study we define the liquid–solid inter-
face position and profile, its displacement and longitudinal mean
velocity during the metal melting process.
There are two main approaches to the calculation of moving so-
lid–liquid interface. The first can be found in Ramachandran et al.
[10], Gadgil and Gobin [11], Albert and O’Neill [12] and Vynnycky
and Kimura [13] who have developed deformable grids which
Fig. 1. Melting process scheme.
follow the moving solid–liquid interface. The moving boundary
approach may be appropriate when tracking a sharp boundary
interface is required. However, specific models describing the
interface motion are needed. Since these models use several sim-
plifying assumptions, the accuracy of interface movement and
shape is questionable. Following [13], the analytical solution
describing the moving solid–liquid boundary ‘‘is not uniformly va-
lid for all time, and tends to overestimate the final thickness of the
solid layer’’. Besides, the moving boundary approach does not con-
sider the mushy zone and is more complicated numerically, and
therefore the calculation time for the 3D problem can be much
longer. An alternative approach is the use of a fixed grid, which
simplifies the numerical modeling requirements. Examples of such
approach can be found in Morgan [14], Gartling [15] and Voller
et al. [16–18]. In this approach, conservation equations are solved
for the entire volume, and therefore the velocity of the solid phase
should be zero. Gartling [15] employs a simple way of making the
viscosity of the metal a function of enthalpy: by decreasing extra
enthalpy DH from the latent heat value to zero, he drives the
viscosity to a high value.
2. Present study
In the present study, the enthalpy-porosity method, which al-
lows a fixed-grid solution of the momentum and energy equations,
was adopted. This method was developed by Voller and Prakash
[17]. It is based on solid–liquid volume fraction, while the phase
change is assumed to occur within a narrow temperature range.
In addition, the Darcy source approach is used to simulate motion
in the mushy zone, which consists of solid particles dispersed in a
liquid. Because of lack of information about the solid particles
diameter, the Carman–Kozeny equation [19] for modeling drag
forces in porous media is used. The momentum equation contains
a source term, which depends on the local liquid fraction Fl and the
Carman–Kozeny constant C. When the local liquid fraction equals
zero, there is only solid at this location. The Carman–Kozeny con-
stant is mostly influenced by the morphology and viscosity of the
mushy zone. Voller and Prakash [17] and Shmueli et al. [20] inves-
tigated the influence of this constant on the solution using a
262 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271

number of constant values. It was found that there exist more suit- which means in this case that Tl  Ts is within the range of 0.3–
able values for a specific experiment. It was noted that the constant 0.5 °C. In the present research, Tl  Ts = 0.5 °C was chosen.
C should be investigated for different geometries, materials and The mathematical model of metal melting by a heated vertical
boundary conditions. In our research, the constant C was also wall was examined, for example, by Webb and Viskanta [27]. Here
examined and used as a gauge constant in order to calibrate the we describe the problem considering mass, momentum and energy
numerical model and the experiments results. balance equations under the assumption of a compressible laminar
In our research, the energy equation is expressed by enthalpy flow of a Newtonian liquid. In this study we adopted a simplifying
formulation. Voller and Prakash [17] presented the enthalpy of assumption of Prakash and Voller [17] with some modifications:
the material as a sum of sensible heat and latent heat, H = cpT + DH, continuity equation:
where the latter term is a function of temperature DH = f(T). This
@q
function allows estimating a mushy zone region from the phase- þ r  ðq~
uÞ ¼ 0; ð3Þ
@t
change equilibrium diagram of the material.
In the present research, the enthalpy-porosity method was momentum equation:
adopted, as noted above, but unlike the studies mentioned, the en-  
@~
u   2
thalpy-porosity method was used both in 2D and 3D calculations q þ qð~ u ¼ rP þ r  l r~
u  rÞ~ uT  lðr  ~
u þ r~ uÞI
@t 3
for taking into account boundaries and their effect on the melting
process. Moreover, in the present research, temperature depen- ~ ~
þ F buoyancy þ F drag ; ð4Þ
dence of metal physical properties was used unlike other numeri-
energy equation:
cal studies, which assume a small change and neglect temperature
dependence. @H
At the same time, some details of the melting process remain
q þ qr  ðH~
uÞ ¼ r  ðkrT Þ: ð5Þ
@t
unclear, such as the number of rolls in the liquid metal flow [21],
The ~
F buoyancy term in the momentum equation is the force stimulat-
particularly at the early stage of melting. Various authors obtained
ing natural convection in the container, which is described by the
different melt flow structures in their computations [18,22,23].
temperature dependence of density:
Moreover, scanty particulars are known about 3D simulations of
this problem [24], and experimental data about liquid phase flow ~
F buoyancy ¼ qðTÞ~
g: ð6Þ
have not practically been described (principles of liquid metal flow
measurements are well-known and employed in alloys solidifica- The flow is calculated as compressible in order to take into account
tion process [9,25], but they are not used in melting problems). small changes of density due to temperature differences. Here, ow-
ing to the use of temperature dependence of gallium physical prop-
erties, the expansion coefficient is not necessary.
3. Mathematical model and numerical analysis The term ~ F drag in the momentum equation describes drag forces
between the liquid and solid phases in the mushy zone. Therefore,
3.1. Theoretical model and setup it becomes zero outside the mushy domain. The pressure drop,
which is related to the drag force between the liquid and solid
A model is built for describing gallium melting in a rectangular phases in a fluidized bed, can be calculated by the well-known
container (see Fig. 1). The initial temperature TInt < Tmelt, and all the empirical Ergun expression [28]:
metal is in the solid state. The process starts when the left vertical !
side wall temperature abruptly becomes higher than the melting 150ð1  F l Þ2 ll 1:75ql ð1  F l Þ
temperature, i.e. THot > Tmelt.
rP ¼ 2
þ juj ~u: ð7Þ
F 3l dp dp
When considering a phase change in an alloy (or non-pure me-
tal), three regions are present: a solid region, a liquid region and a The first term in the right-hand part of the last equation represents
porous media region which consists of solid particles dispersed in the Darcy term, and the second is the Forchheimer term. Here ~ u is
the liquid (see Fig. 1). the superficial velocity of the liquid metal, and dp is solid particles
A local liquid–solid volume fraction can be written for a pure diameter in the mushy zone.
material melting at a specific temperature: Due to the low fluid velocity in the mushy zone, the Forchhei-
 mer term can be neglected [26], and the Darcy term can be re-
Vl 0 T < Tm; placed with Carman–Kozeny equation [20], which deals with
F l ðTÞ ¼ ¼ ð1Þ
V 1 T > Tm: drag forces in porous media:
2
For alloys melting, which occurs over a certain temperature range, Cð1  F l Þ
~
F drag ¼  ~
u: ð8Þ
the liquid volume fraction can be calculated by [17]: e þ F 3l
8
> 0 T < T s; Here e is a computational constant used to avoid the division by
V l < TT s
F l ðTÞ ¼ ¼ T l T s Ts 6 T 6 T l; ð2Þ zero when the fluid volume fraction becomes vanishingly small
V >
:
1 T > T l: and tends to zero [20]. When the cell is completely liquid Fl = 1,
the source acquires a zero value, and the momentum equation be-
When the temperature is lower than the solidus temperature, the comes suitable for an actual one-phase fluid flow. When Fl de-
entire cell is solid. At the temperature higher than the liquidus tem- creases in the mushy zone, the value of the source increases, and
perature, the cell is liquid, and at the temperature between the sol- the momentum equation corresponds to the Darcy law. When the
idus and liquidus it is porous. local liquid fraction equals zero, the source term becomes very
Tl  Ts was defined as a small interval around the melting tem- large, and the velocity of the liquid diminishes to zero. In such an
perature. Beckermann and Viskanta [26] noted that an unrealisti- approach, the velocity of the solid should be equal to zero. C is a
cally large Tl  Ts (i.e.>2 °C) would introduce too much artificial constant which depends on the morphology of the mushy zone.
‘smearing’ of the predicted interface, while a too small interval will Voller and Parkash [17] noted that the constant C can reduce or in-
complicate numerical calculation and require a very fine mesh. crease the flow intensity at the mushy zone. This aspect will be dis-
Beckermann and Viskanta [26] used the value Tl  Ts/TH  TC = 0.02, cussed later in the conclusions.
 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271 263

Table 1 2D model was 90,000. For the 3D model, two grids and discretiza-
The investigated cases. tion were examined:
Case DT(°C) Thot(°C) Tint(°C)
1 15 40 25 (a) 32,144 elements with discretization of the first order for the
2 18 45 27 pressure and heat transfer equations and of the second order
3 25 50 25 for the flow equation with 176,610 degrees of freedom,
(b) 10,469 elements with discretization of the first order for all
the conservation equations with 60,000 degrees of freedom.
In order to specify zero velocity in solid cells, Gartling’s method
[15] has been used. The dynamic viscosity in solid cells should ac- The comparison between the results was examined (see Fig. 2),
quire high values in comparison with other terms in the momen- and only minor differences can be seen at the solid–liquid inter-
tum equation, which is achieved following this method [14] by face profile and liquid fraction volume. Therefore, method (b)
using a large computational constant B. The values of the constant was chosen for decreasing calculation time and computer re-
B were examined in the range from 10 to 10,000, and B = 1000 kg/ sources. For method (b) with Tl  Ts = 0.5 °C, energy balance was
ms was found to be the lowest order of magnitude which dimin- calculated in order to verify the numerical scheme. Fig. 3 shows
ishes the velocity magnitude correctly in the entire solid region. a comparison between the heat required for heating and melting
the gallium, Qmelt, and the heat difference between the inlet and
l ¼ ll þ B½1  F l ðTÞ: ð9Þ outlet, Qin  Qout. The agreement is very good, the mean error is
The enthalpy of gallium was calculated by: about 3%, and the energy balance in the numerical scheme is
observed.
H ¼ cp T þ DH; ð10Þ Time discretization was computed by backward differentiation
where cp is the average specific heat of the solid and liquid metal formula (BDF) algorithm [30] which estimates the next time step
according to the volume fraction value Fl, and DH is the modified la- size for the convergence. BDF algorithm always aims at an increase
tent heat added to the enthalpy in cells with melting metal, which is in the next time step size for shortening the calculation time with-
expressed by a normal Gaussian formula: out damaging the convergence of the solution. Segregation was
computed by the conservation equations, the flow equations
 2
L ðT  T m Þ (~
u; P) were solved by PARDISO [31] solver (parallel sparse direct
DH ¼ pffiffiffiffi exp  : ð11Þ
pðT l  T s Þ ðT l  T s Þ linear solver), and the energy equation (T) was solved by MUMPS
[32] solver (multifrontal massively sparse direct solver).
Here L is the latent heat of gallium, Tm is the mean temperature va- The iterative calculation was performed with an under-relaxa-
lue, and Tl  Ts is temperature variance. The boundary condition for tion factor allowing the convergence in high-gradient problems.
the momentum equation is a non-slip condition at the container For smaller under-relaxation factors, the convergence becomes
walls. At the upper boundary, a pressure outlet to the atmosphere slower, and the number of iterations grows. After examining a
is taken into account. In the energy equation, THot and TInt corre- number of under-relaxation factors, optimal values were found.
spond to the two vertical walls, and the remaining walls are For the flow equation, 0.5 factors were used, and for the energy
insulated. equation, an automatic algorithm embedded in the COMSOL Mul-
tiphysics software is found to be optimal.
3.2. Accomplishment of the computations

The computations were performed for a rectangular container 4. Experimental setup

of 0.06  0.06  0.09 m. Three cases were examined (Table 1), for
gallium with the properties listed in Table 2.
The numerical solution was built using COMSOL Multiphysics
software which uses the finite element method (FEM). Both 3D
and 2D models were numerically explored to describe the liquid–
solid interface by using the volume-of-fluid technique [20].
The grid was a built-in rectangular mesh. The grid size was cho-
sen after careful examination of the solution convergence, the total
cells number being 8932 for the 2D. The discretization of the con-
servation equations is done by Galerkin’s method [29] with La-
grange’s elements of the first order for the pressure and heat
transfer equations and of the second order for the flow equation
in the 2D model. The number of degrees of freedom (DOF) in the
In order to validate our theoretical model and numerical simu-
lation, an experimental system was designed and constructed.
Melting experiments were performed in a Perspex
0.06  0.06  0.09 m rectangular container (Fig. 4). Two vertical
walls serving as a multipass heat exchanger were made of brass
(5 in Fig. 4). The rest of the container walls were made of 0.01 m
thick Perspex. Additional Perspex walls (3 in Fig. 4) were placed
for better insulation. 99% metallic gallium was used, whose prop-
erties allow measurements in the temperature region close to
room temperature.
Temperature regimes were specified by Haake F3 laboratory
thermostats (1 in Fig. 4) with the precision to ±0.5 °C. For temper-

Table 2
Gallium properties [33,34].

Property Value
Gallium (solid) qs 5905 kg/m3
298K < T < 303K ks (60.66  0.183T + 6.03  104T2  7.136  107T3)W/m  K
cps (9858.85 + 64.57T  0.1014T2)J/kg  K
Gallium (liquid) Tm 303K
303K < T < 323K ql (6273.98  0.605T + 4.82  105T2)kg/m3
kl (15.70 + 0.031T + 2.97  105T2)W/m  K
cpl (828.32  3.056T + 0.0079T2  9.13  106T3)J/kg  K
ll (0.0156  1.053  104T + 3.198  107T2  5.005  1010T3)kg/m  s
L 80,200 J/kg  K
264 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271
Fig. 2. Comparison of solid–liquid interface profiles at different times and liquid
fraction volume with time calculated by methods (a) and (b).
ature distribution measurements in the liquid and solid phases of
the metal and for determining the liquid–solid interface position,
eighteen calibrated T-type thermocouples (1/16-inch diameter
stainless steel tubes) with ±0.3 °C accuracy were uniformly spaced
along the length and height of the working container. Thermocou-
ples readings were registered by the Agilent 34970A Data Logger
Fig. 3. Comparison between the heat required for heating and melting the gallium, Qmelt, and the heat difference between the inlet and outlet, Qin  Qout, for two DT values
(normalized to Qin  Qout maximal value).
connected to PC. LabVIEW 2012 software of National Instruments
was used for data processing.
The interface position, its displacement with time, and one of
the melt velocity components between thermally stabilized oppo-
site vertical walls were recorded by ultrasonic Doppler velocimeter
(UDV) DOP2000, Model 2150, Signal Processing, Lausanne, Swit-
zerland [7]. Ultrasound Doppler velocimetry is based on the Dopp-
ler shift theory. The device sends an ultrasonic pulse into the fluid
and receives an echo from particles of the fluid. The delay of the
echo provides the distance between the transducer and a particle,
and the Doppler shift reveals the particle velocity. It takes a short
time to obtain a fluid velocity profile along the ultrasonic beam
(Fig. 5). In spite of specific problems connected with such measure-
ments (acoustic coupling, wetting conditions, medium reflecting
capabilities etc.), they usually provide sufficient information on
the mean flow velocity [25].
Five ultrasonic transducers TR0405RS (4 MHz) and TR1005RS
(10 MHz) were placed into special slots in a hot plate 7 (Fig. 4) with
a possibility to switch the transducers according to a specified or-
der by a multi-channel device multiplexer. It was taken into ac-
count that transducers with higher emission frequency allowed
to obtain a higher resolution, but with a concurrent growth of sig-
nal attenuation.
Ultrasonic Doppler velocimetry provides a full profile of the
flow velocity along the beam line (internal plot in Fig. 5). In our
experimental situation, the divergence of the ultrasonic beam
(1.8–2.5°) allows to estimate the lateral size of the measuring re-
gion close to the far cold wall depending on the distance from the
transducers 2.8 mm to 3.8 mm.
Before the experiment, the solid gallium is melted and poured
into the experimental container through tube 8 (Fig. 4) at the top
of the box. A tube 9 serves to let the air out. Both tubes also allow
changing the volume during melting. Gallium solidifies by keeping
a uniform initial temperature below melting temperature for sev-
eral hours at the container using two heat exchangers. Melting is
initiated by switching the left bath to a proper temperature
exceeding the melting point. The left container wall has a constant
temperature during the experiment. The temperature of the oppo-
site wall is kept below the melting point, while other walls are
thermally insulated. Reading and recording of the thermocouples
and UDV transducers starts in parallel.
The mentioned temporal advancement of the experimental li-
quid–solid interface is illustrated in Fig. 4. For its plotting, both
the registered time-varied velocities profiles and peak positions
of UDV power display were used. A great number of distinct points
 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271 265

Fig. 4. Schematic diagram of the experimental system: 1 – constant temperature bath, 2 – plexiglas walls, 3 – plexiglas walls for insulated condition, 4 – thermocouples
inside the container, 5 – heat exchanger, 6 – heat exchanger thermocouples, 7 – ultrasonic transducers, 8,9 – pouring and compensating tubes.

Fig. 5. Liquid–solid interface temporal progress for specific UDV transducer at the level y = 53 mm. Inset: Typical UDV display where the horizontal axis is the flow velocity
[mm/s], and the vertical axis is the length of the container [mm].
266 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271

Table 3
Dimensionless parameters at the temperature TH for the melting process (K = 0.67 is
aspect ratio, qm is the density at melting temperature 303 K, and ql, ml, al are liquid
gallium properties at TH).

Case DT(oC) PrT H SteT H RaT H

1 15 0.0247 0.07444 3.712  105

2 18 0.0244 0.08933 6.017  105
3 25 0.0242 0.12406 8.376  105

Fig. 6. Comparison of liquid–solid interface position at different times in GV

experiment and in numerical 2D and 3D solutions at Thot = 38 °C and DT = 10 °C.

of liquid–solid interface were sampled, and a polynomial fitting

was used to illustrate its temporal motion and profile. The maximal
error of the points processing is described by R2 = 0.99.

5. Results and discussion

The numerical modeling of the melting problem is extremely

complicated, since besides the presence of two phases, there is also
a moving liquid–solid interface and a mushy zone. Moreover, metal
properties change with temperature during the process. The same
reasons cause many impediments to experimental measurements
in low-temperature metal.
Note that our problem is solved in a dimensional form, but to
make it possible to unify the obtained results and make general
conclusions, the values of main dimensionless numbers are given
in Table 3.

5.1. Comparison with experimental data

5.1.1. Solid–liquid interface position and profile

Fig. 6 shows a comparison of liquid–solid interface position at
different times in Gau and Viskanta experiments [2] and in our
numerical models results for mid vertical x–y section.
For the 2D model, the agreement is reasonable at the moment
120 s, but since the moment 360 s, the profile shape and its varia-
tion with time is different from the 3D case and the experimental
results. For the subsequent process development, the error is larger
in the upper part of the container (for example, at t = 1020 s the
mean error is about 13% and the maximal error is about 23%).
For the 3D model, the results agreement is better (the mean error
is about 9% and the maximal error is about 20%).
Fig. 7. Comparison of liquid–solid interface position at different moments of time
obtained from: – UDV experimental data, 2D and 3D computations for three
cases: (a) DT = 15 °C, (b) DT = 18 °C, (c) DT = 25 °C.
A set of experiments was carried out to validate the developed
numerical model. As noted above, liquid melt dynamics was stud-
ied by means of ultrasonic measurements.
It is established that the lower boundary of the measured veloc-
ity is within the range down to 1 mm/s [7] and [9], and the estima-
tion of measurements accuracy varies within the range of about
 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271 267

Fig. 8. Temporal development of melting metal velocities at the point (45, 53),
DT = 15 °C o – experimental points, – fitting of experimental data, 2D and 3D
computations.

2–8% and even more. At the same time, it is noteworthy that the
measurements accuracy depends on the experimental conditions.
For its unbiased assessment, one should have the results of alterna-
tive measurements performed in a different way, or a standard
case for which the velocity distribution is known (in time, along
the coordinates, etc.). In the absence of such comparison tests,
we have to be based on the described findings.
We have also to note some problems with UDV measurement
caused by melting peculiarities – transient character of the process
under study, including thermal variations of medium properties,
possible presence of the mushy zone with inconstant sound speed,
gallium oxides appearance [25,33]. Moreover, varying conditions
may be a cause of ultrasound pulses advancing failure and their
breakdown, etc.
Therefore, velocity profiles obtained using UDV require some
kind of processing, and they were averaged, for the sake of compar-
ison, by several approximation methods. The most convenient and
simple among them is the polynomial fitting for spatial averaging.
Time-averaged profiles were obtained from long-term recorded
velocity profiles by choosing close-in-time values and accomplish-
ing standard statistical procedures.
For our experimental situation, most results were obtained
using 128 emissions per profile. The sensitivity of 10 MHz trans-
ducers was acceptable at pulse repetition frequency (PRF) in the
range from 160 Hz to 1 kHz to obtain velocities in the range from
0.012 m/s to 0.072 m/s. For 4 MHz transducers, the same PRF inter-
val allows to get velocities from 0.028 m/s to 0.18 m/s.
Fig. 7 displays a comparison of liquid–solid interface position at
different moments of time between the ultrasonic experimental
results and 2D and 3D numerical solutions for a vertical x–y section
at a distance of 20 mm from the center.
Experimental velocity values were processed using OriginPro
7.5 statistics tools, where the standard error of the mean value is Fig. 9. Longitudinal velocity profile temporal increase at the height y = 53 mm (600,
characterized by R2 = 0.99. 700, 800 s one after another); o – experimental points, parabolic approximation,
3D computations.
One can see a good agreement for the 3D solution and a less ex-
act one for the 2D solution, meaning that the 3-rd dimension ef-
fects cannot be neglected, and a flow through the depth of the 5.1.2. Liquid metal velocity
container has an essential effect on the heat transfer and the melt- Temporal behavior of the melting metal longitudinal velocity
ing rate. The mean error for the 2D solution is 15–20.5% depending component at a fixed point (x = 45 mm, y = 53 mm) is illustrated in
on the temperature difference DT, whereas the maximal one is Fig. 8. As indicated, 3D method reveals a better matching with the
about 40%. For the 3D solution the mean error is from 8.8% to velocity development. Here the experimental points were approxi-
11.3%, and it also depends on the temperature difference, while mated using OriginPro 7.5 software by an exponential dependence
the maximal error is about 14%. of the 2-nd order with a reduced chi-squared value 1.58 and R2 = 0.82.
268 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271
Fig. 10. Temporal temperature change at the distance x = 40 mm for different
heights: 15 mm, 30 mm, 40 mm; DT = 15 °C o – experimental points, 2D and
3D computations.
Fig. 9 shows velocity profiles comparison at the height
y = 53 mm along the length of the container at different times in
a vertical x–y section 20 mm from the center. Parabolic fitting of
experimental data with R2 = 0.99 and SD = 0.25 demonstrates a
good agreement with 3D computations.
Fig. 11. Temporal development of liquid–solid interface position, 2D model. Case 1:
DT = 15 °C, maximal velocities vary from 8 to 15 mm/s.
5.1.3. Temperature distribution
Temperature distribution can be a direct indication of the li-
quid–solid interface position. Fig. 10 shows a comparison of the
temporal temperature change between the numerical models
and the experiment in mid x–y section. A considerable temperature
increase occurs both in computations and measurements when the
liquid–solid interface passes over the point of interest. As expected,
because of weak convective flow, one can see that temperature val-
ues fall with the level decrease. Here a better agreement exists be-
tween the measurements and 3D computations, too, which proves
that 3D effects have a great influence on the melting process.
 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271 269

Figs. 11 and 12 illustrate the solid–liquid interface temporal

extension according to 2D numerical results in different cases,
where solid phase is shown in blue, the liquid in red and the mushy
zone in a transition range between blue and red. As expected, a
convective flow is developing in time. First, the flow is negligibly
weak, and the main heat transfer mechanism is thermal conduc-
tion, and thus the solid–liquid interface profile is nearly a straight
line. Later, the convective flow becomes more significant [17]. Con-
vection becomes the main heat transfer mechanism and has a great
influence on the liquid–solid interface morphology. The liquid
phase heats up at the hot wall, changes its density and flows up-
ward. Then it flows toward the liquid–solid interface, cools and
goes down. A comparison between Figs. 11 and 12 shows that
the convection becomes more significant much earlier, and the vol-
ume flow is higher in case 2 because of a higher temperature dif-
ference. Accordingly, the liquid–solid interface profile becomes
deformed and moves much faster.
Melting conditions determine an interesting flow regime effect
related to small rolls nucleation at the early stages of liquid phase
flow (Fig. 13). These rolls are observed in 2D flows, and their
dynamics can be analyzed within the framework of this model,
while in 3D model (Fig. 14) a single roll dominates through the en-
tire melting process.
One can see in Fig. 14 the liquid–solid interface position at dif-
ferent times according to 3D numerical results, where the third
dimension influence is considered. On the right side of Fig. 13 a
top view of the melting process at the height y = 45 mm can be
seen. The liquid–solid profile is curved due to the flow regime
developing along the third dimension. The presence of side walls

Fig. 12. Temporal development of liquid–solid interface position, 2D model. Case 3:

DT = 25 °C. Maximal velocities vary from 15 to 22 mm/s.

Comparative analysis of the obtained computational data with

experimental ones (previous and described herein) reveals their
acceptable fit. The developed numerical 3D model is suitable for
the prediction of both the temporal dynamics of liquid–solid inter-
face motion inside the container during melting, and melts flow in
liquid phase including early stages of its formation.

5.2. 2D and 3D numerical simulation-flow analysis

Computations give a possibility to distinguish the melt flow de-

tails from the early stages including rolls birth and its further Fig. 13. Rolls nucleation and transformation at early flow stages, time from 15 to
development. 40 s, case 3: DT = 25 °C.
270 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271

Fig. 14. Temporal development of liquid–solid interface position, 3D model. Case 3: DT = 25 °C. Left side: midplane of vertical cross-section with maximal velocities from12
to 14 mm/s; right side: midplane of horizontal cross-section with maximal velocities from 1.2 to 6.1 mm/s.

and non-slip boundary conditions influences the flow regime along
the third dimension, which is absent in the 2D model.
The conditions of our computations, temperature regimes and
container dimensions were chosen in compliance with those
described by Gau and Viskanta [2]. But even in this particular case,
the influence of the side walls on the process is confirmed by calcula-
tions and indicates that 3D models are necessary for similar problems.
A comparison between Figs. 12 and 13 displays that in the 3D
numerical model liquid–solid interface advance is less dynamic,
and its volume flow is less intensive. Another difference appears
at the early stages of flow formation. In the 2D model, flow origin
is accompanied with small rolls development. Their quantity defi-
nitely depends on the process parameters and changes during the
melting progress. On the contrary, in the 3D model a single roll is
formed in the beginning, enlarging during the melting process.
Alexiades et al. [21] noted that several studies using finite ele-
ment approach in order to predict gallium melting in a rectangular
container got different solutions. One of them found that during a
rather long time, the flow structure consists of a number of small
rolls, while another found only a single roll. It seems that in the
present study the 2D and 3D numerical models provide both
solutions, but a physical explanation has to be developed.
 O. Ben-David et al. / International Journal of Heat and Mass Transfer 67 (2013) 260–271
For example, in the 2D model case 1 (DT = 15 °C) one can see
four convective rolls in the liquid phase at the early periods (until
100 s).Throughout 300 s, the rolls quantity is reduced to three,
later (at 500 s) there are two rolls left, and after that (1020 s) a sin-
gle big roll remains. In case 3 (DT = 25 °C) the rolls confluence pro-
cess is much faster. At the beginning only three rolls are present in
the flow, after that a reduction to a pair of rolls can be seen (140 s),
and one big roll remains during the further process development
(about 500 s).
Beyond any doubt, the temperature difference between the hot
and cold walls has a great influence on the flow regime and its inten-
sity, including the rolls quantity. Detailed analysis of this depen-
dence is not executed here, but we emphasize the numerical
model capabilities for the study of the effect of geometrical ratios,
boundary conditions and temperature gradient on the rolls quantity.
A certain transition region (mushy zone) can be also seen in
both 2D and 3D computations. The width of this region at the
lower side of the container is much wider due to a lower temper-
ature gradient.
6. Conclusions
A numerical and experimental study of solid–liquid phase tran-
sition in a rectangle container is performed. An advanced numeri-
cal model was developed for the computations of metal melting in
a rectangular container with a heated vertical wall. The thermal
dependences of all material properties were taken into account
at the calculation.
The Carman–Kozeny constant values were examined, and their
optimal value was found for the prediction of the liquid–solid
interface behavior. The error of solid liquid interface position
determination grows with the temperature difference, which can
be explained by the way we calibrate the numerical models
according to experimental results. Several different values of the
constant C were examined for such calibration of the first case,
DT = 15 °C, and C = 1011 was found to be the most accurate. In
other cases, the same constant was used. The growing error may
indicate to a certain dependence between the constant C and the
temperature difference.
According to 2D computations, the melt flow origin is accompa-
nied by small rolls nucleation. Their quantity depends on the pro-
cess parameters and reduces during the melting progress in such a
way that one big roll remains in the further process development.
Calculations reveal the mushy zone appearance close to the li-
quid–solid interface. As expected, this zone is wider in the lower
part of the container because of smaller temperature gradients.
A set of experimental data in liquid phase was obtained using
the ultrasonic Doppler velocimeter. These results, together with
collateral temperature measurements, describe liquid–solid inter-
face dynamics and temporal behavior of longitudinal mean flow
velocity component.
The local velocities obtained by ultrasonic measurements were
compared with those predicted by the numerical simulations, both
2D and 3D. The comparison demonstrates a better agreement be-
tween the experimental data and the results of the 3D numerical
simulation, which confirms a great influence of the third
dimension.
It was shown that 3D computations reveal an acceptable fit
with experimental data and can be exploited as an effective instru-
ment for the prediction of the liquid–solid interface dynamics and
melt flow features, starting from the early stages of flow
development.
The obtained data require a certain expansion. Nevertheless,
they obviously demonstrate the potentialities of both the
271
described numerical 3D model and measuring method as effective
tools for the study of similar problems.
References
[1] C. Gau, R. Viskanta, Effect of natural convection on solidification from above
and melting from below of a pure metal, Int. J. Heat Mass Transfer 28 (1984).
[2] C. Gau, R. Viskanta, Melting and solidification of pure metal on a vertical wall,
Heat Transfer 108 (1986) 174–181.
[3] T.A. Campbel, J.N. Koster, Visualization of liquid–solid interface morphologies
in gallium subject to natural convection, Cryst. Growth 140 (1994) 414–425.
[4] A.S. Menon, M.E. Weber, A.S. Mujumdar, The dynamics of energy storage for
paraffin wax in cylindrical containers, Can. J. Chem. Eng. 61 (1983) 643–653.
[5] D. Pal, Y.K. Joshi, Melting in a side heated tall enclosure by a uniformly
dissipating heat source, Int. J. Heat Transfer 44 (2001) 375–387.
[6] L. Katsman, V. Dubovsky, G. Ziskind, R. Letan, Experimental investigation of
solid–liquid change in cylindrical geometry, in: Thermal Engineering Summer
Heat Transfer Conference, Vancouver, Canada, 2007.
[7] http://signal-processing.com.
[8] P. Oborin, I. Kolesnichenko, Study of liquid metal flow and crystallization in a
plane later with stirring, in: Conference on Theoretical and Applied
Magnetohydrodynamics, Perm, Russia, 2012, p. 75.
[9] A. Cramer et al., Liquid metal model experiments on casting and solidification
processes, J. Mater. Sci. 39 (2004) 7285–7294.
[10] N. Ramachandran, J.R. Gupta, Y. Jalunu, Thermal and fluid flow effects during
solidification in a rectangular cavity, Int. J. Heat Mass Transfer 25 (1982) 187–
194.
[11] A. Gadgil, D. Gobin, Analysis of two dimensional melting in rectangular
enclosures in the presence of convection, J. Heat Transfer 106 (1984) 20–26.
[12] M.R. Albert, K. O’Neill, Transient two-dimensional phase change with
convection using deforming finite elements, Int. Comput. Tech. Heat Transfer
1 (1985).
[13] M. Vynnycky, S. Kimura, An analytical and numerical study of coupled
transient natural convection and solidification in a rectangular enclosure, Int. J.
Heat Mass Transfer 50 (2007) 5204–5214.
[14] K. Morgan, A numerical analysis of freezing and melting with convection,
Comput. Meth. Appl. Eng. 28 (1981) 275–284.
[15] D.K. Gartling, Finite element analysis of convective heat transfer problems
with change of phase, in: K. Morgan et al. (Eds.), Computer Methods in Fluids,
Pentech, London, 1980, pp. 257–284.
[16] V.R. Voller, M. Cross, N.C. Markatos, An enthalpy method for convection/
diffusion phase change, Int. J. Numer. Methods Eng. 24 (1987) 271–284.
[17] C. Prakash, V.R. Voller, A fixed grid numerical modelling methdology for
convection–diffusion mushy region phase-change problem, Int. J. Heat Mass
Transfer 30 (1987) 1709–1719.
[18] A.D. Brent, V.R. Voller, K.J. Reid, The enthalpy-porosity technique for modeling
convection–diffusion phase change: application to the melting of a pure metal,
Numer. Heat Transfer 13 (1988) 297–318.
[19] W.D. Carrier III, Goodbye, Hazen: Hello, Kozeny–Carman, J. Geotech.
Geoenviron. Eng. (2003) 1054–1056.
[20] H. Shmueli, G. Ziskind, R. Letan, Melting in a vertical cylindrical tube:
numerical investigation and comparison with experiments, Int. J. Heat Mass
Transfer 53 (2010) 4082–4091.
[21] V. Alexiades, N. Hannoun, T.Z. Mai, Resolving the controversy over Tin and
Gallium melting, Numer. Heat Transfer 44 (Part B) (2003) 253–276.
[22] J.A. Dantzig, Modeling liquid–solid phase changes with melt convection, Int. J.
Numer. Methods Eng. 28 (1989) 1769–1785.
[23] F. Stella, M. Ginagi, Melting of a pure metal on a vertical wall: numerical
simulation, Numer. Heat Transfer Appl. 38 (Part A) (2000) 193–208.
[24] Y. Takeda, Ultrasonic Doppler method for velocity profile measurement in fluid
dynamics and fluid engineering, Exp. Fluids 26 (1999) 177–178.
[25] D. Brito, H.C. Nataf, P. Cardin, J. Aubert, J.P. Masson, Ultrasonic Doppler
velocimetry in liquid gallium, Exp. Fluid 31 (2001) 653–663.
[26] C. Beckermann, R. Viskanta, Natural convection solid/liquid phase change in
porous media, Int. J. Heat Transfer 31 (1987) 35–46.
[27] B.W. Webb, R. Viskanta, Analysis of heat transfer during melting of a pure
metal from an isothermal vertical wall, Numer. Heat Transfer 9 (1986) 539–
558.
[28] N.S. Cheng, Application of Ergun equation to computation of critical shear
velocity subject to seepage, J. Irr. Drain. Eng. ASCE 129 (4) (2003) 278–283.
[29] O.C. Zienliewicz, P. Nithiarasu, R.L. Taylor, Finite Elem. Methods Fluid Dyn. 6
(2005).
[30] P.N. Brown, K.E. Grant, S.L. Lee, R. Serban, D. EShmaker, A.C. Hindmarsh, C.S.
Woodward, Sundials: suite of nonlinear and differential/algebraic equation
solver, ACNT Math. Softw. 31 (2005) 363.
[31] O. Schenk, K. Gartner, PARDISO User Guide, 2011, http://www.pardiso-
project.org.
[32] P. Amestoy et al., 2011. http://www.graal.ens-lyon.fr/MUMPS.
[33] Y. Tasaka, Y. Takeda, T. Yanagisawa, Ultrasonic visualization of thermal
convective motion in a liquid gallium layer, Flow Meas. Instrum. 19 (2008)
131–137.
[34] D.R. Lide (Ed.), CRC Handbook of Chemistry and Physics, 84th ed., CRC Press,
2003.