You are on page 1of 11

International Journal of Heat and Mass Transfer 173 (2021) 121260

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer


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

Influence of density change during melting inside a cavity: Theoretical


scaling laws and numerical analysis
Moritz Faden a,∗, Andreas König-Haagen a,b, Erwin Franquet c, Dieter Brüggemann a
a
Chair of Thermodynamics and Transport Processes (LTTT), Centre of Energy Technology (ZET), University of Bayreuth, Universitätsstraße 30, Bayreuth,
95440 Germany
b
ENEDI Research Group, Faculty of Engineering of Bilbao, University of the Basque Country UPV/EHU, 48013 Bilbao, Spain
c
Université Côte d’Azur, Polytech’lab, France

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

Article history: An exact prediction of melting processes is important for the design of latent heat thermal energy stor-
Received 29 January 2021 ages. However, in phase change models based on continuum theory, the effect of density change during
Revised 9 March 2021
melting is often not considered, although there is no clear estimate of the error caused by this assump-
Accepted 21 March 2021
tion. To overcome this problem, we use scaling analysis to predict the difference between two macro-
Available online 13 April 2021
scopic models that handle the density change during melting differently. As a test case, we choose the
Keywords: melting in a two-dimensional cavity heated from one side. Numerical simulations are used to confirm
Scaling the scaling analysis. Hence, two different models are built and implemented into OpenFoam. In the first
OpenFOAM model, the density is constant and equal in both phases and the natural convection is taken into ac-
Volume expansion count by the Boussinesq approximation. In the second model, the density is different for both phases
Density change and varies with the temperature in the liquid phase. The scaling analysis reveals that for small Stefan
PCM
numbers, the theoretical maximum deviation in the average phase front position at the same time be-
Phase change material
tween these two models is solely controlled by the liquid/solid density ratio. Furthermore, this maximum
deviation is different for the different regimes of the melting process. The results of the numerical simu-
lations confirm these results and show that the maximum deviation occurs at the end of the convection
dominated regime. Our analysis further reveals that while the results strongly differ in time (≈ 11% for
Octadecane), the flow field and the shape of the phase front is nearly the same if the time is shifted
or the latent heat is scaled accordingly. Taking into account the uncertainty prevailing in the literature
regarding the latent heat of fusion and other material properties of PCM, this may explain why many
researchers found excellent agreement between experiments and their simulations despite neglecting the
density change.
© 2021 Elsevier Ltd. All rights reserved.

1. Introduction phase change materials (PCM) used in these storages can be found
in [15–21] and general information about LHTES in [22–24].
Latent heat thermal energy storage (LHTES) is an attractive pos- Numerical methods are an important tool to access the
sibility to store thermal energy with a low temperature spread and time-dependent thermal power of LHTES. Compared to exper-
a high energy density [1,2]. Therefore, this storage technology has iments, they are cheaper and allow a detailed look into the
attracted the attention of many researchers, leading to dozens of complex temporo-spatial dynamics of the phase change. Multi-
review papers and hundreds of regular articles. Possible applica- ple approaches exist to solve solid-liquid phase change numeri-
tions of LHTES are domestic hot water production [3–7], mobi- cally [25,26]. The most common ones are the so-called fixed-grid
lized thermal energy storage [8], energy storage for batteries ther- or one-fluid methods. They describe both phases (solid and liquid)
mal management [9–11], waste heat recovery in heat-intensive in- as an equivalent homogeneous fluid governed by a unique set of
dustrial plants [12], and in the future Carnot batteries as power- partial differential equations corresponding to the classic balance
heat-power storage units [13,14]. Detailed information about the equations (mass, linear momentum, energy). Therefore, the corre-
sponding system includes average variables that must be defined,
as well as a specific term to cancel the velocity field in the solid

phase. The first numerical methods for solving solid/liquid phase
Corresponding author.
E-mail address: Moritz.Faden@uni-bayreuth.de (M. Faden).
change on fixed grids were developed decades ago. Due to constant

https://doi.org/10.1016/j.ijheatmasstransfer.2021.121260
0017-9310/© 2021 Elsevier Ltd. All rights reserved.
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

some of these methods can be found in König-Haagen et al. [28].


Nomenclature With the basics of these methods well understood, modern re-
search on numerical methods for solid/liquid phase change focuses


A mushy term, kgm−2 s−2 on more efficient methods [29,30], conjugate heat transfer prob-
C constant,- lems [31,32], and unfixed or specifically close contact melting [33–
c specific heat capacity, JK −1 kg−1 36]. A difficulty that receives less attention is the influence of the
D Darcy constant, kgm−3 s−1 temperature-dependent density of the PCM. The vast majority of
D thermal diffusivity, m2 s−1 studies use the same type of model, where the Boussinesq ap-
E Identity matrix,- proximation is applied to model natural convection in the liquid
Fo Fourier number,- phase [29,32–34,37–39]. In other words, except for its effects on
Fo = HDt2 the buoyancy source term of the momentum equation, the den-


g gravitational acceleration, ms−2 sity is assumed constant and the density change accompanying the
L length, m phase change is neglected, although the volume expansion of com-
L latent heat, Jkg−1 mon PCM ranges from -8.3% to 22% [40]. The main reason why this
H height, m assumption is so widespread in the numerical community is the
h specific enthalpy, Jkg−1 increased numerical effort and various difficulties that arise when
Nu Nusselt number,- the density change is taken into account. Either an open boundary,
˙ an elastic respectively moving wall, or a gaseous phase has to be
Nu = λ QT
l included in the simulation domain. The latter is possible by com-
p pressure, Pa
bining the volume-of-fluid technique with the enthalpy-porosity
Ra Rayleigh number,-
ρl2 gβ T H 3 cl model [41,42]. More ’exotic’ volume-of-fluid approaches were also
Ra = ηλ tested, for example by Wang [43] et al., who used a mass trans-
Q˙ heat flux per depth, W m−1 fer mechanism (and then an associated energy transfer) instead of
Ste Stefan number,- a melting/solidification model. Dallaire and Gosselin [44] classified
Ste = cL
T and explained the pros and cons of the different methods to in-
T temperature, K or ◦ clude density change into phase-change simulations. Furthermore,
t time, s they conclude that significant differences can be expected if the
U characteristic velocity scale, ms−1 density variations during phase change are not taken into account.


u velocity, ms−1 Their subsequent article investigates solid/liquid phase change in a
x axial position, m hermetically sealed capsule [45]. Here, the density change leads to
y vertical position, m a considerable pressure rise and thus requires thermo-mechanical
α volume fraction,- coupling between the PCM and the container. Hassab et al. [46] fo-
β thermal expansion, K −1 cus on the melting process of Eicosan in a vertical cylinder and
δ position of the interface, m find that the difference in the melting time between a constant
ε deviation,- and a variable density model is up to 15%. According to Galione
κ small number,- et al. [47], Octadecane in a spherical capsule melts 8% slower if
λ thermal conductivity, W K −1 m−1 the density and the heat capacity are set to their phase-specific
η dynamic viscosity, kgm−1 s−1 values instead of a single value for both phases. The same authors
ν kinematic viscosity, m2 s−1 introduce an approximate method that considers most of the ener-
ρ density, kgm−3 getic effects of the density change. The basic idea of their method
τ dimensionless time,- is to scale the latent heat with the solid/liquid density ratio and
τ = Fo Ste therefore compensate for the error made in the energy balance
φ auxiliary variable,- when the solid and liquid densities are assumed to be equal in-
side a phase changing control volume. However, since only the la-
α relative to the volume fraction tent part of the thermal energy is considered, their approach as-
δ relative to the position of the phase front sumes small Stefan numbers. In experiments, the density change is,
τ relative to the dimensionless time of course, always present and mostly accommodated by an open-
avg average ing [48,49] or an air layer [50,51]. Therefore, knowledge about the
B constant density case deviation caused by the assumption that the solid and liquid den-
D variable density case sity are equal and constant in both phases will significantly reduce
G Galione’s approach the uncertainty in validating numerical methods for solid/liquid
h hot phase change. Moreover, even for groups of PCM, like organic PCM,
high high Rayleigh number the value of the volume expansion can not be generalized to a spe-
k count variable cific value [52]. Thus, there is a strong need to link the deviation
l liquid caused by a constant density assumption to the density ratio.
low low Rayleigh number Given the aforementioned difficulties and unknowns, the main
M melting research question of the present proposal is to better explain and
med medium Rayleigh quantify the differences between a constant density approach (liq-
m count variable uid value for both phases) and a more physical model including
o old time step density change. The paper has three contributions. To demonstrate
s solid the deviation that results from the assumption of an equal density
∗ not corrected in the liquid and the solid phase, we first carry out a theoretical
scale analysis of the melting of pure substances in a quasi-two-
dimensional rectangular cavity heated from one side. In this scale
further development, especially by Voller and his co-workers [27], analysis, the density change of the material is taken into account.
many different methods are available today. A comparison between Thus, we are able to express the above mentioned deviation as a

2
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

2.1. Conduction asymptote

At the beginning, the PCM is at melting temperature. Thus,


there is no need to preheat the solid and the heat flowing through
the hot wall is solely used to advance the phase front and to heat
the newly created liquid. The energy required for this is propor-
tional to the solid mass per depth, which melts per time step t,
i.e. ms = ρs H δ :
H T dδ dδ
λl ∼ ρs HL + C ρs c l H T , (1)
 δ   dt  dt 
heat flux through the hot wall heat flux needed to advance the phase front
and to heat up the liquid

where δ is the phase front position and C is a constant which de-


pends on the mean temperature of the liquid phase and cannot be
determined by scaling. In equation 1, we assume that the tempera-
ture profile in the liquid is still linear despite the velocity directed
away from the phase front, which is induced through the volume
expansion. We justify this assumption by requiring the magnitude
Fig. 1. State of the domain at t > 0s.
of the diffusive term in the y-direction to be greater than the mag-
nitude of the advective term:
T T
function of the solid/liquid density ratio. Secondly, the scale anal- λl  ρl c l U , (2)
ysis is verified by numerical simulations with and without den- δ2 δ
sity change during melting for octadecane and magnesium chlo- where U is the velocity induced by the volume expansion. Rear-
ride hexahydrate (MCHH). Finally, we perform an additional anal- ranging this equation leads to a kind of Peclet number:
ysis to check whether the scaled latent heat approach by Galione Dl
et al. [47] is useful and compare it with both the constant density  1, (3)
δU
model and the variable density model.
λ
where Dl = ρ cl is the thermal diffusivity. Then, we calculate the
l l
velocity U based on mass conservation across the interface:
 ρs  d δ
2. Theoretical analysis U = 1− . (4)
ρl dt
The problem at hand is the melting of a PCM in a rectangu- If the velocity of the phase front is approximated by the one
lar capsule with a density decrease during the solid-liquid phase without density change, i.e. the fastest possible phase front speed
transition. Initially, the PCM is solid and its temperature is equal and therefore the worst-case scenario, we arrive at the relation:
to the melting temperature TM . At t > 0s, the temperature of the  ρl 
left capsule wall is suddenly raised to Th = TM + T . Due to this 1 1− Ste. (5)
ρs
temperature increase, the PCM begins to melt. During the melt-
ing process, the right wall is kept at melting temperature TM . Both This relation is fulfilled for small Stefan numbers.
the top and the bottom walls are adiabatic. Excess fluid, which re- The differential equation 1 is easily solved by separating the
sults from the volume expansion during melting, leaves the cap- variables:

sule through a small outlet at the upper left corner. A sketch of
λl T
the setup is shown in Fig. 1. δ∼ t. (6)
In the late 1980s, Jany and Bejan [53] developed a scaling the- ρs L ( 1 + C c l 
L
T
)
ory to describe melting in a two-dimensional rectangular capsule
Recasting this equation with dimensionless quantities results
with equal liquid and solid densities. In their article, the melt-
in:
ing process is divided into four regimes: pure conduction, mixed
conduction plus convection, pure convection and shrinking solid. ρl 1 ρ 1
Moreover, they derive a correlation for the progression of the δ ∼ H FoSte =H τ l , (7)
ρs 1 + C Ste ρs 1 + C Ste
phase front based on the pure conduction and the pure convection
asymptote. The correlation is valid until the phase front is in con- where τ is the dimensionless time. Obviously, ρs > ρl leads to a
tact with the right wall, i.e. the start of the shrinking solid regime. slower progression of the phase front. For small Stefan numbers,
In the next section, we will add volume expansion during melting equation 7 reduces to:
to the scaling theory developed by Jany and Bejan [53]. In addi-
ρl
tion, we use the results to predict the deviations between melting δ∼H τ , (8)
with volume expansion and without. The scale analysis relies on ρs
the following assumptions: or to:
ρs δ
• the flow is two-dimensional, τ∼ 2
. (9)
ρl H
• the Prandtl number is greater than one,
• the outlet is small, We assume that the proportionality factor in equation 9 is inde-
• the solid density is greater than the liquid density, pendent of the density ratio and derive the relative deviation be-
• the kinetic energy is neglected because its change is several or- tween the equal density assumption, corresponding to the Boussi-
ders of magnitude smaller than the change in latent heat. nesq approximation, and the variable density case with respect to

3
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

the dimensionless time needed for the phase front to be at the for the relative deviation at the same average phase front position
same position: and at:
δ 2 ρs − δ 2 δavg,B − δavg,D ρ 
τD − τB ρl εconvection,δ = · 100% =
s
− 1 · 100%, (19)
εconduction,τ = · 100% =
H H
δavg,D ρl
τD δ 2 ρs
H ρl
ρs  ρ
for the same dimensionless time. Of particular note in Eq. 19 is the
ρ −1
= l ρs · 100% = 1 − l · 100%, (10) absence of the square root in the density ratio. Furthermore, it is
ρl ρs important to point out that εconvection,δ is the deviation for the
where the index D stands for variable density and B for the Boussi- pure convection asymptote. This value is only reached if τ is big
nesq approximation with constant and equal densities in both enough, but the fluid has not yet reached the right wall. It is cer-
phases. Similarly, we calculate the relative deviation in the phase tainly possible that the fluid is in contact with the right wall before
front position for the same dimensionless time: the maximum deviation εconvection,δ occurs. However, with equa-
√  tions 11 and 19, we expect the deviation in the average phase front
δB − δD H τ − H τ ρρsl position to increase after the heat transfer mechanism changes
εconduction,δ = · 100% =  · 100% from conduction to convection. In contrast, the relative deviation
δD H τ ρρsl in the dimensionless time at the same liquid fraction is the same
 for the conduction and the convection asymptote. As simulations
ρl
1− will show, this is also true between these two asymptotes.
ρs ρs
=  = − 1 · 100%. (11)
ρl ρl
ρs 3. Numerical analysis

2.2. Convection asymptote 3.1. Mathematical modeling

In the convection dominated regime, the Nusselt number Nu = For both approaches, a homogeneous model is considered, each

λl T is approximately proportional to the fourth root of the phase being distinguishable by its volumetric phase fraction α . The
ρl2 gβ T H 3 cl flow is supposed laminar and incompressible, while the fluid is as-
Rayleigh number Ra = ηλl formed with the height of the sumed Newtonian. In addition, the velocity in the solid phase is
capsule H as the characteristic length [54]. The reason is that zero by definition, so only liquid properties are transported by con-
the characteristic length should resemble the length of travel in vection. The vanishing of the solid velocity is achieved by adding a
the boundary layer [55]. Rearranging the equation for the Nusselt Darcy source term to the momentum equations:
number and using the above-mentioned proportionality, the heat

→ ( 1 − α )2 −

flux per depth through the hot wall is: A (α ) = D 3 u, (20)
α +κ
Q˙ ∼ Ra0.25 λl T . (12)
where D is the Darcy constant and κ a small number to avoid di-
Again, this heat flux is used to advance the phase front and to
viding by zero. Since we model the melting of pure substances, the
heat up the newly created liquid. In the convection regime, the
Darcy term can be seen as a continuous switch-off method.
phase front is deformed and it is more appropriate to use an aver-
aged phase front position [53]:
3.1.1. Constant density Boussinesq approximation model
dδavg dδavg
Ra0.25 λl T ∼ ρs HL + C ρs c l H T , (13) For the simulations with an identical density in both phases
dt dt and the Boussinesq approximation in the liquid phase, the govern-
where the index avg is short for average. Integrated and recast ing equations are simplified considerably and are known to be:
with dimensionless quantities this yields:
ρ 1
∇ ·−

u = 0, (21)
δavg ∼ H τ l Ra0.25 . (14)
ρs 1 + C Ste
Here, too, the higher solid density leads to a slower progres- ∂−

u −
→− →
sion of the phase front, but the influence is greater. The convec- ρ + ρ∇ · ( u u )
∂t  
tion asymptote lasts until the phase front reaches the right wall −



(δavg ≈ L). Inserted into equation 14 this time is: = −∇ p + ∇ · (η∇ u ) + 1 − β (T − TM ) ρ−

g − A, (22)
L ρs
τ∼ (1 + C Ste )Ra−0.25 . (15)
H ρl ∂h −

ρ + ρ∇ · (hl u ) = ∇ · (λ∇ T ), (23)
Equation 15 implicates that the right wall is reached later if ∂t
there is a volume expansion during melting. Moreover, for small where β is the volumetric coefficient of thermal expansion. In ad-
Stefan numbers, the average melt front position is: dition, we choose the value of the liquid density to be the constant
ρl 0.25 density for both phases.
δavg ∼ H τ Ra , (16)
ρs
while the associated dimensionless time is as follows: 3.1.2. Variable density model
ρ δavg −0.25 Now, the density is no longer assumed to be constant. Instead,
τ∼ s Ra . (17)
it varies depending on the temperature (ρ = ρ (T )). Consequently,
ρl H
the balance equations for the mass, momentum, and energy mix-
Again, we assume that the proportionality factor is independent
ture are:
of the density ratio and arrive at:
τD − τB  ρ ∂ρ −

εconvection,τ = · 100% = 1 − l · 100%, (18) + ∇ · ( ρl u ) = 0 , (24)
τD ρs ∂t
4
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

∂ρ −

u −
→− → Equation 31 is linear in the temperature T k+1,∗ and easily
+ ∇ · ( ρl u u )
∂t solvable. Note that despite the fictitious slope, we still solve an
   isothermal phase change problem. This is evident at convergence

→ −
→ 2 −

= −∇ p + ∇ · η (∇ u + ∇ ( u )T ) − (∇ · u )E (T k+1,∗ = T k ), where the additional term is zero and so the value
3
→ −
− → of the fictitious slope becomes irrelevant. After the linearized dis-
+ρ g − A , (25) crete energy equation is solved, the phase fraction is updated via:
⎧ 
∂ρ h −

+ ∇ · (ρl hl u ) = ∇ · (λ∇ T ), (26)
⎨α k + ∂α  (T k+1,∗ − T k ) for 0 < α < 1
∂T
∂t α k+1,∗ = k (32)
where ρ , λ and ρ h are defined as convex combinations of their ⎩ k c k+1,∗
α + L (T − TM ) otherwise
liquid and solid values [56]:
and corrected to stay bounded between 0 and 1:
φ = αφl + (1 − α )φs . (27)    
We assume a constant heat capacity in both phases, set the ori- α k+1 = min max 0; α k+1,∗ ; 1 . (33)
gin of the enthalpy to the melting temperature and split the en-
thalpy into a sensible and a latent part. Hence, the product of den- The next step is to set the temperature inside cells with 0 <
sity and enthalpy is: α < 1 to the melting temperature to prevent cells from skipping
the phase change. Lastly, the thermophysical-properties are up-
ρ h = (1 − α )ρs cs (T − TM ) + αρl L + αρl cl (T − TM ). (28) dated and the residual is calculated. The whole procedure is re-
peated until convergence.
3.2. Numerical solution process
3.2.2. Numerical settings
The mathematical model is implemented into the open source
The results shown in the next section are calculated with sec-
CFD toolbox OpenFOAM (v1812). OpenFoam is written in C++ and
ond order spatial schemes and the implicit Euler time scheme. The
is based on the cell-centered finite volume method. The solution
time step is calculated by a maximum CFL number of 1 or 1.5, de-
of the energy equation is part of a merged PISO-SIMPLE (PIMPLE)
pending on the case. For numerical stability, the initial tempera-
algorithm. For a detailed description of the pressure-velocity cou-
ture and the right wall temperature are 0.05 K below the melting
pling solution process, see the original paper by Issa [57]. Here,
temperature. Furthermore, we assume that the phase boundary is
we focus on the energy equation and assume that the mass fluxes
located at α = 0.5. In all simulations, the length is equal to the
are known. The computational method for solving the nonlinear
height of the capsule. The grid is orthogonal and equidistant with
energy equation is explained for different solid and liquid densi-
62500 cells.
ties. The method for identical densities follows by setting the liq-
uid density to a constant value and then taking ρs = ρl .
4. Results and discussion
3.2.1. Temperature phase fraction coupling
Inserting the equation of state (Eq. 28) into the energy equation In this section, we compare the numerical results obtained
(Eq. 26) results in an equation for the temperature and the phase with the constant density Boussinesq approximation model and
fraction: the variable density model for the melting of pure substances with
each other as well as with the correlation given by Jany and Be-
∂ρ cT −
→ ∂αρl −

+ cl ∇ · (ρl u T ) = ∇ · (λ∇ T ) − L + ∇ · ( ρl u ) jan [53]. In addition, we check how well the theoretical analysis
∂t ∂t conducted in Section 2 predicts the deviation between two meth-
∂ρ c −
→ ods. For the simulations, we use octadecane and magnesiumchlo-
+ TM + c l ∇ · ( ρl u ) , (29) ride hexahydrate (MCHH) as PCM. The thermophysical properties
∂t
of these two PCM are given in Table 1.
where ρ c = αρl cl + (1 − α )ρs cs . Additionally, the condition applies If not stated otherwise, the Stefan number of the simulations
that α can only be different from 0 or 1 if the temperature of the is below 0.01 to match the small Stefan number assumption of the
cell is the melting temperature. Equation 29 could be solved by analytical model. This results in T = 1 K for octadecane and T =
the classical fictitious heat source method. However, this method 0.5 K for MCHH. For both materials, two Rayleigh numbers are in-
converges very slowly [30]. To accelerate the convergence of the vestigated by adjusting the gravitational acceleration: 1.3 · 106 and
method, we introduce a new iterative scheme in the spirit of 1.3 · 107 for octadecane and 6.1 · 104 and 6.1 · 105 for MCHH. The
Swaminathan and Voller [58,59]. First, the phase fraction inside an cases with the lower Rayleigh number are termed Ralow and the
iteration step k is Taylor-expanded:

∂α  k+1,∗
α k+1 = α k + (T − T k ),
∂ T k
(30) Table 1
Thermophysical properties of octadecane [60] and magnesiumchloride hexahydrate
(MCHH) [61,62]
where the derivative is a very steep fictitious slope e.g. 5 · 104 K1 .
Next, the transient terms of equation 29 are discretized and equa- Property Unit Value
tion 30 is inserted into this semi-discretized energy equation. The PCM - Octadecane MCHH
superscript o denotes the old time step and m the mass flux from Melting temperature K 301.13 388.25
−1
the PISO step. This yields: Latent heat of fusion kJkg 236.98 157.43
−3
Density (solid) kgm 867.9 1595.5
(ρ c )k T k+1,∗ − (ρ cT )o −
→ m Density (liquid) kgm
−3
775.5 1455.7
+ cl ∇ · ( (ρl u ) T k+1,∗ )
kt k+1,∗ Volumetric thermal expansion coefficient K− 1 8.69 · 10−4 3.76 · 10−4
= ∇ · (λ ∇ T ) Specific heat capacity (solid) kJK −1 kg−1 1.941 1.830
 Specific heat capacity (liquid) kJK −1 kg−1 2.234 2.570
(αρl )k −(αρl )o k ∂α  T k+1,∗ −T k −
→ (31)
−L t + ρ l ∂T  t + ∇ · ( ρl u ) m Thermal conductivity (solid) W K −1 m−1 0.334 0.70
k Thermal conductivity (liquid) W K −1 m−1 0.151 0.63
(ρ c )k − (ρ c )o −

Dynamic viscosity mPas 3.78 16.70
+TM + c l ∇ · ( ρl u ) m . Darcy constant kgm−3 s−1 1013 1013
t

5
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

Fig. 2. Liquid fraction as a function of the dimensionless time for the cases with Ste < 0.01.

Fig. 3. Evolution of the phase front for Octadecane (Ralow ). The upper row corresponds to the constant density Boussinesq approximation model. The lower one to the
variable density model. The snapshots are taken at the same time.

ones with the higher Rayleigh number Rahigh . Besides, there are tween the two models increases throughout the convection domi-
cases where we increase the temperature of the left capsule wall nated regime (linear increase in α in Fig. 2). As soon as the shrink-
to increase the Stefan number. This also increases the Rayleigh ing solid regime begins, i.e. the moment when the fluid reaches
number and results in a third Rayleigh number termed Ramed . the right wall, the slope of the curves begins to decrease. As a re-
At the end of the section, we discuss the approach by Galione sult, the difference between the two models also decreases. This
et al. [47] to scale the latent heat with the solid/liquid density ra- is due to the fact that the net heat flux (hot wall heat flux minus
tio. cold wall heat flux) is declining in the last regime with increasing
First, it is noticeable that at the same dimensionless time, the global liquid fraction. The liquid fraction of the simulations with
liquid fraction obtained with the constant density Boussinesq ap- the variable density model is always smaller than the one with-
proximation model is greater than the one obtained with the vari- out. Hence, the net heat flux and the melting rate are higher. At
able density model. The reason is that for the cases with a con- the end of the melting process, which is not simulated due to ex-
stant density, less energy is needed to melt the PCM, simply be- cessive time investment, the solid is completely molten. Thus, the
cause there is less PCM mass (ρl < ρs ). Consequently, the cases difference in the liquid fraction between the two methods must be
with the assumption of a constant density melt faster (Fig. 2). It is zero. It is also visible that during the first instants, where conduc-
striking that for all cases, the difference in the liquid fraction be- tion prevails, the Rayleigh number has no influence and the evolu-

6
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

Fig. 4. Relative deviation of the liquid fraction as a function of the dimensionless time for different Rayleigh and Stefan numbers.

tion is similar for octadecane and MCCH. The agreement between is shown. In the diffusive regime, i.e. for τ → 0,the relative devi-
the constant density model and the correlation by Jany and Be- ation between both methods is very close to ( ρs /ρl − 1 ) · 100%
jan [53] is good (depicted as squares in Fig. 2). Significant devia- as predicted by the theoretical analysis. As soon as the convec-
tions are only visible for MCHH with the low Rayleigh number at tion regime begins, the deviation increases. For the cases with
a large dimensionless time. Note that the correlation is only valid Ste < 0.01, the deviation approaches the asymptotic value of the
until the convection regime ends. pure convection asymptote (ρs /ρl − 1 ) · 100%. However, this value
A graphical representation of the melting process can be seen in is never reached, because the condition that τ must be large is not
Fig. 3 for Octadecane (Ralow ). The sequence of images shows that yet fulfilled when the convection regime ends and the shrinking
with the variable density model the PCM melts slower. This is par- solid regime begins. In the shrinking solid regime, the deviation
ticularly noticeable when the constant density Boussinesq approx- declines. Although the Rayleigh number has no influence on the
imation model reaches the right wall (upper row Fig. 3) and the theoretical maximum deviation, a higher Rayleigh number leads to
variable density model still has some PCM left for melting (lower a greater maximum deviation observed between simulations per-
row Fig. 3). Nevertheless, the phase front looks very similar for formed with the two models. At first, this seems to be obvious,
both models. since a higher Rayleigh number increases convection and the pure
A quantitative evaluation of the difference between the two convection asymptote is reached earlier, i.e. a smaller τ fulfills the
methods is given in Fig. 4a for Octadecane and in Fig. 4b for condition of being large enough. However, a high Rayleigh number
MCHH. In these figures, the relative deviation in the liquid fraction deforms the phase front more strongly. Thus, the shrinking solid
(the same as the average phase front position) regimes begins earlier. A mechanism that decreases the maximum
observable deviation.
Increasing the Stefan number also leads to a higher deviation
between the two models. We suspect that the fluid leaving the
αD (τ ) − αB (τ )
εα = · 100%, (34) capsule in the variable density model is at least partly responsible
αD (τ )
7
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

For small Stefan numbers, the theoretical analysis predicts this


difference to be (1 − ρl /ρs ) · 100%. With the density values of Oc-
tadecane and MCHH, the theoretical deviation for these two PCM
is 10.65% respectively 8.76%. The numerical results confirm the the-
oretical prediction (Fig. 5), even in the shrinking solid regime. This
means that the results of a simulation with the constant den-
sity Boussinesq approximation model and with the variable den-
sity model strongly differ in time. However, the flow field and the
shape of the phase front are nearly the same if the time is adjusted
accordingly (Fig. 6). We conjecture that the difference between a
simulation with a constant density plus the Boussinesq approxi-
mation and a simulation with different solid/liquid densities plus
a variable density in the liquid phase is only in time because the
speed of the phase front is extremely low in the melting of PCM.
So there are two different time scales: one for the buoyancy-driven
flow and one for the progression of the phase front, whereas the
former one is far shorter than the latter one. The secondary flow,
which occurs as a result of the volume expansion near the phase
front, does therefore not significantly alter the primary flow i.e. the
vortex which results from natural convection and the scaling law
1
Nu ∼ Ra 4 remains valid.
Fig. 7 depicts the Nusselt number based on the net heat flux
flowing into the capsule. The Nusselt number shows the typical
behavior for melting from a vertical wall. Initially, when the heat
Fig. 5. Relative deviation of the dimensionless time plotted over the liquid fraction transfer is dominated by diffusion, it is very high. Then, the heat
for MCHH and octadecane. The black lines are the results obtained by scaling (Eq. flux, and hence the Nusselt number, decreases with an increas-
10 and Eq. 18). The symbols represent numerical results. ing distance between the hot wall and the cold PCM. This de-
crease continues until the starting convection stabilizes the heat
flux. In the shrinking solid regime, the net heat flux and the Nus-
for this. The fluid leaves the capsule with high temperature and
selt number decline because some of the energy flowing through
energy. This energy is not available for melting and the melting
the hot wall leaves the capsule through the cold wall without be-
rate is lower compared to the constant density model. Neverthe-
ing available for melting. The time shift between the two methods
less, the influence of the Stefan number is rather small. Thus, even
is clearly visible in the Nusselt number. Moreover, during the con-
for non-negligible Stefan numbers, the value of the pure convec-
vection regime the Nusselt number is nearly constant with time
tion asymptote (ρs /ρl − 1 ) · 100% is very useful to predict the de-
and independent of the density change. Compared to our numer-
viation in the liquid fraction during the convective regime while
ical results, the correlation by Jany and Bejan [53] gives slightly
simulating melting from a vertical wall without including density
higher values for the Nusselt number in the convection dominated
change.
regime.
Now, we change our point of view and look at the relative de-
Finally, we check if the approach by Galione et al. [47] to scale
viation in the dimensionless time at the same liquid fraction:
the latent heat with the solid/liquid density ratio works. For the
τD ( α ) − τB ( α ) cases with Ste → 0, the maximum deviation in the liquid fraction
ετ = · 100%. (35)
τD ( α ) between the scaled latent heat and the variable density method

Fig. 6. Velocity Field during the melting of Octadecane (Ralow ). The left picture is with density change and the right one is without. The right picture was taken ≈ 10.65%
earlier than the left one.

8
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

Fig. 7. Nusselt number as a function of the dimensionless time for all the cases with Ste < 0.01. Also plotted is the Correlation given by Jany and Bejan [53] (squares).

Fig. 8. Liquid fraction as a function of the dimensionless time: comparison with results obtained with the approach of Galione et al. [47] (index G)

is below 0.1% regardless of the PCM (Fig. 8). Hence, the symbols control volume and neglects the unimportant secondary flow in-
representing the results obtained with the Galione et al. approach duced by the density change. For higher Stefan numbers, the dif-
lie exactly on the lines representing the simulations with density ference in the liquid fraction increases as the amount of sensi-
change. The dimensionless time needed to melt 75% of the MCHH ble heating depends on the density change. So if someone wants
(Ralow ) is 0.15872 for the scaled latent heat approach, 0.15873 for to use this method, they should also scale the heat capacity with
the simulations with density change and 0.14485 for the constant the solid/liquid density ratio. However, it is easy to find situations
density case. For the Octadecane case with the low Rayleigh num- where this approach will perform considerably worse, e.g. close
ber, the values are 0.083768, 0.083775 and 0.074879. Obviously, the contact melting where a volume expansion upon melting results in
values of the scaled latent heat approach and the simulations with a thicker melt gap, melting inside a partially filled macro capsule
the variable density model are almost the same. where the volume expansion leads to a higher heat transfer area
The explanation why scaling the latent heat works surprisingly or a cycle of melting and solidification during which the value of L
well is that it accounts for the important energy balance inside a has to be augmented with ρs /ρl for melting and ρl /ρs for solidifi-

9
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

cation. Therefore the approach has to be handled with care. Taking Methodology, Software, Formal analysis, Funding acquisition, Writ-
into account the uncertainty in the value of the latent heat of fu- ing - review & editing. Erwin Franquet: Methodology, Formal anal-
sion of most PCM, it is likely that some researchers using the con- ysis, Writing - original draft, Writing - review & editing. Dieter
stant density Boussinesq approximation model applied the scaled Brüggemann: Funding acquisition, Supervision.
latent heat approach by accident. To be completely clear, for ma-
terials with different solid and liquid densities, a simulation with Acknowledgments
the constant density Boussinesq approximation model must differ
from the experimental results to be correct within its assumptions. The authors are grateful for the financial support of the German
Research Foundation (DFG) under grant no. BR 1713/20-1 and BR
1713/20-2.
5. Conclusion
References
The combined application of scaling analysis and numerical
simulations proved to be a useful tool for studying the influence [1] H. Mehling, L.H. Cabeza, Heat and cold storage with PCM. An up-to-date intro-
of density change when a PCM melts inside a one-side heated cav- duction into basics and application, Heat and Mass Transfer, Springer, 2008.
[2] L. Cabeza, A. Castell, C. Barreneche, A. de Gracia, A. Fernández, Materials used
ity. For small Stefan numbers, the velocity induced by the den- as PCM in thermal energy storage in buildings: A review, Renewable and Sus-
sity change has no significant impact on the melting process, and tainable Energy Reviews 15 (3) (2011) 1675–1695, doi:10.1016/j.rser.2010.11.
for this reason, the effect of the density change is almost entirely 018.
[3] Z. Wang, F. Qiu, W. Yang, X. Zhao, Applications of solar water heating system
energetic. Comparing the constant density Boussinesq model (liq- with phase change material, Renewable and Sustainable Energy Reviews 52
uid density for both phases) with the variable density model for (2015) 645–652, doi:10.1016/j.rser.2015.07.184.
small Stefan numbers and taking into account the results which [4] M.H. Abokersh, M. Osman, O. El-Baz, M. El-Morsi, O. Sharaf, Review of the
phase change material (PCM) usage for solar domestic water heating sys-
are obtained by the scaling theory, the following conclusions can tems (SDWHS), International Journal of Energy Research 42 (2) (2017) 329–
be drawn: 357, doi:10.1002/er.3765.
[5] S.F. Ahmed, M. Khalid, W. Rashmi, A. Chan, K. Shahbaz, Recent progress in
1. The shape of the phase front and the velocity field is not solar thermal energy storage using nanomaterials, Renewable and Sustainable
Energy Reviews 67 (2017) 450–460, doi:10.1016/j.rser.2016.09.034.
changed for the same mean phase front position and therefore
[6] G. Alva, L. Liu, X. Huang, G. Fang, Thermal energy storage materials and sys-
the results are only shifted in time. tems for solar energy applications, Renewable and Sustainable Energy Reviews
2. The deviation resulting from this time shift is accurately pre- 68 (2017) 693–706, doi:10.1016/j.rser.2016.10.021.
[7] S. Kunkel, P. Schütz, F. Wunder, S. Krimmel, J. Worlitschek, J.U. Repke, M. Rädle,
dicted by the scaling theory, i.e. the relative deviation equals
Channel formation and visualization of melting and crystallization behaviors
(1 − ρl /ρs ) · 100%. in direct-contact latent heat storage systems, International Journal of Energy
3. The deviation in the mean phase front position at the same di- Research 44 (6) (2020) 5017–5025, doi:10.1002/er.5202.
mensionless time is non-linear and the maximum deviation is [8] S. Guo, Q. Liu, J. Zhao, G. Jin, W. Wu, J. Yan, H. Li, H. Jin, Mobilized thermal
energy storage: Materials, containers and economic evaluation, Energy Con-
in the convection dominated regime. version and Management 177 (2018) 315–329, doi:10.1016/j.enconman.2018.09.
4. The solid/liquid density ratio is the only parameter that affects 070.
the theoretical maximum deviation in the mean phase front [9] M. Malik, I. Dincer, M.A. Rosen, Review on use of phase change materials in
battery thermal management for electric and hybrid electric vehicles, Interna-
position. This maximum deviation occurs for infinite time and tional Journal of Energy Research 40 (8) (2016) 1011–1031, doi:10.1002/er.3496.
without the interaction of the cold right wall. [10] L. Ianniciello, P.H. Biwolé, P. Achard, Electric vehicles batteries thermal man-
5. The maximum theoretical deviation in the mean phase front agement systems employing phase change materials, Journal of Power Sources
378 (2018) 383–403, doi:10.1016/j.jpowsour.2017.12.071.
position at the same dimensionless time obtained by scaling [11] A.R.M. Siddique, S. Mahmud, B.V. Heyst, A comprehensive review on a passive
is not necessarily reached in the simulation because when (phase change materials) and an active (thermoelectric cooler) battery thermal
the convection dominated regime ends and the shrinking solid management system and their limitations, Journal of Power Sources 401 (2018)
224–237, doi:10.1016/j.jpowsour.2018.08.094.
regime begins, the deviation declines. The higher the Rayleigh
[12] T. Nomura, N. Okinaka, T. Akiyama, Waste heat transportation system, using
number of the investigated case the closer is the maximum de- phase change material (PCM) from steelworks to chemical plant, Resources,
viation to the theoretical maximum. Conservation and Recycling 54 (11) (2010) 10 0 0–10 06, doi:10.1016/j.resconrec.
2010.02.007.
[13] A. Thess, Thermodynamic efficiency of pumped heat electricity storage, Physi-
Furthermore, the scaled latent heat approach of Galione
cal Review Letters 111 (11) (2013) 110602, doi:10.1103/PhysRevLett.111.110602.
et al. [47] works well for the investigated geometry. But it is easy [14] R.B. Laughlin, Pumped thermal grid storage with heat exchange, Journal of Re-
to find situations where their approach will perform considerably newable and Sustainable Energy 9 (4) (2017), doi:10.1063/1.4994054.
worse. We therefore recommend using it with caution. [15] M.K. Rathod, J. Banerjee, Thermal stability of phase change materials used in
latent heat energy storage systems: A review, Renewable and Sustainable En-
Neglecting volume change during melting is only one of many ergy Reviews 18 (2013) 246–258, doi:10.1016/j.rser.2012.10.022.
sources of uncertainty in the simulation of melting processes. In [16] M.M. Kenisarin, Thermophysical properties of some organic phase change ma-
future work, we will investigate the effect of uncertain thermo- terials for latent heat storage. A review, Solar Energy 107 (2014) 553–575,
doi:10.1016/j.solener.2014.05.001.
physical properties and experimental boundary condition on the [17] C. Barreneche, M.E. Navarro, L.F. Cabeza, A.I. Fernández, New database to select
evolution of melting processes. phase change materials: Chemical nature, properties, and applications, Journal
of Energy Storage 3 (2015) 18–24, doi:10.1016/j.est.2015.08.003.
[18] A. Kylili, P.A. Fokaides, Life cycle assessment (LCA) of phase change materials
Declaration of Competing Interest (PCMs) for building applications: A review, Journal of Building Engineering 6
(2016) 133–143, doi:10.1016/j.jobe.2016.02.008.
[19] S.N. Gunasekara, V. Martin, J.N. Chiu, Phase equilibrium in the design of phase
The authors declare that they have no known competing finan- change materials for thermal energy storage: State-of-the-art, Renewable and
cial interests or personal relationships that could have appeared to Sustainable Energy Reviews 73 (2017) 558–581, doi:10.1016/j.rser.2017.01.108.
[20] K. Reddy, V. Mudgal, T. Mallick, Review of latent heat thermal energy storage
influence the work reported in this paper.
for improved material stability and effective load management, Journal of En-
ergy Storage 15 (2018) 205–227, doi:10.1016/j.est.2017.11.005.
[21] M. Walczak, F. Pineda, Á.G. Fernández, C. Mata-Torres, R.A. Escobar, Materi-
CRediT authorship contribution statement als corrosion for thermal energy storage systems in concentrated solar power
plants, Renewable and Sustainable Energy Reviews 86 (2018) 22–44, doi:10.
Moritz Faden: Conceptualization, Methodology, Software, Val- 1016/j.rser.2018.01.010.
[22] A. de Gracia, L.F. Cabeza, Phase change materials and thermal energy storage
idation, Formal analysis, Investigation, Writing - original draft, for buildings, Energy and Buildings 103 (2015) 414–419, doi:10.1016/j.enbuild.
Writing - review & editing, Visualization. Andreas König-Haagen: 2015.06.007.

10
M. Faden, A. König-Haagen, E. Franquet et al. International Journal of Heat and Mass Transfer 173 (2021) 121260

[23] C. Veerakumar, A. Sreekumar, Phase change material based cold thermal en- [43] W. Wang, H. Li, S. Guo, S. He, J. Ding, J. Yan, J. Yang, Numerical simulation
ergy storage: Materials, techniques and applications–A review, International study on discharging process of the direct-contact phase change energy stor-
Journal of Refrigeration 67 (2016) 271–289, doi:10.1016/j.ijrefrig.2015.12.005. age system, Applied Energy 150 (2015) 61–68, doi:10.1016/j.apenergy.2015.03.
[24] G. Alva, Y. Lin, G. Fang, An overview of thermal energy storage systems, Energy 108.
144 (2018) 341–378, doi:10.1016/j.energy.2017.12.037. [44] J. Dallaire, L. Gosselin, Various ways to take into account density change in
[25] P. Verma, Varun, S. Singal, Review of mathematical modeling on latent heat solid–liquid phase change models: Formulation and consequences, Interna-
thermal energy storage systems using phase-change material, Renewable and tional Journal of Heat and Mass Transfer 103 (2016) 672–683, doi:10.1016/j.
Sustainable Energy Reviews 12 (4) (2008) 999–1031, doi:10.1016/j.rser.2006.11. ijheatmasstransfer.2016.07.045.
002. [45] J. Dallaire, L. Gosselin, Numerical modeling of solid-liquid phase change in
[26] Y. Dutil, D. Rousse, S. Lassue, L. Zalewski, A. Joulin, J. Virgone, F. Kuznik, K. Jo- a closed 2D cavity with density change, elastic wall and natural convec-
hannes, J.-P. Dumas, J.-P. Bédécarrats, A. Castell, L.F. Cabeza, Modeling phase tion, International Journal of Heat and Mass Transfer 114 (2017) 903–914,
change materials behavior in building applications: Comments on material doi:10.1016/j.ijheatmasstransfer.2017.06.104.
characterization and model validation, Renewable Energy 61 (2014) 132–135, [46] M.A. Hassab, M.M. Sorour, M.K. Mansour, M.M. Zaytoun, Effect of volume ex-
doi:10.1016/j.renene.2012.10.027. pansion on the melting process’s thermal behavior, Applied Thermal Engineer-
[27] V.R. Voller, An overview of numerical methods for solving phase change prob- ing 115 (2017) 350–362, doi:10.1016/j.applthermaleng.2016.12.006.
lems, Advances in Numerical Heat Transfer 1 (9) (1996) 341–380. [47] P.A. Galione, O. Lehmkuhl, J. Rigola, A. Oliva, Fixed-grid numerical modeling
[28] A. König-Haagen, E. Franquet, M. Faden, D. Brüggemann, Influence of the con- of melting and solidification using variable thermo-physical properties - Ap-
vective energy formulation for melting problems with enthalpy methods, In- plication to the melting of n-Octadecane inside a spherical capsule, Interna-
ternational Journal of Thermal Sciences (May) (2020) 106477, doi:10.1016/j. tional Journal of Heat and Mass Transfer 86 (2015) 721–743, doi:10.1016/j.
ijthermalsci.2020.106477. ijheatmasstransfer.2015.03.033.
[29] I. Danaila, R. Moglan, F. Hecht, S. Le Masson, A Newton method with adap- [48] M. Faden, C. Linhardt, S. Höhlein, A. König-Haagen, D. Brüggemann, Velocity
tive finite elements for solving phase-change problems with natural convec- field and phase boundary measurements during melting of n-octadecane in a
tion, Journal of Computational Physics 274 (2014) 826–840, doi:10.1016/j.jcp. cubical test cell, International Journal of Heat and Mass Transfer 135 (2019)
2014.06.036. 104–114, doi:10.1016/j.ijheatmasstransfer.2019.01.056.
[30] M. Faden, A. König-Haagen, D. Brüggemann, An Optimum Enthalpy Approach [49] F.L. Tan, Constrained and unconstrained melting inside a sphere, Interna-
for Melting and Solidification with Volume Change, Energies 12 (5) (2019) 868, tional Communications in Heat and Mass Transfer 35 (4) (2008) 466–475,
doi:10.3390/en12050868. doi:10.1016/j.icheatmasstransfer.20 07.09.0 08.
[31] W. Ogoh, D. Groulx, Effects of the number and distribution of fins on the stor- [50] C.J. Ho, J.Y. Gao, An experimental study on melting heat transfer of paraf-
age characteristics of a cylindrical latent heat energy storage system: A nu- fin dispersed with Al2O3 nanoparticles in a vertical enclosure, Interna-
merical study, Heat and Mass Transfer 48 (10) (2012) 1825–1833, doi:10.1007/ tional Journal of Heat and Mass Transfer 62 (1) (2013) 2–8, doi:10.1016/j.
s00231-012-1029-3. ijheatmasstransfer.2013.02.065.
[32] D. Hummel, S. Beer, A. Hornung, A conjugate heat transfer model for un- [51] J. Vogel, D. Bauer, Phase state and velocity measurements with high temporal
constrained melting of macroencapsulated phase change materials subjected and spatial resolution during melting of n-octadecane in a rectangular enclo-
to external convection, International Journal of Heat and Mass Transfer 149 sure with two heated vertical sides, International Journal of Heat and Mass
(2020) 119205, doi:10.1016/j.ijheatmasstransfer.2019.119205. Transfer 127 (2018) 1264–1276, doi:10.1016/j.ijheatmasstransfer.2018.06.084.
[33] Y. Kozak, G. Ziskind, Novel enthalpy method for modeling of PCM melting [52] L.F. Cabeza, G. Zsembinszki, M. Martín, Evaluation of volume change in phase
accompanied by sinking of the solid phase, International Journal of Heat change materials during their phase transition, Journal of Energy Storage 28
and Mass Transfer 112 (2017) 568–586, doi:10.1016/j.ijheatmasstransfer.2017. (September 2019) (2020) 101206, doi:10.1016/j.est.2020.101206.
04.088. [53] P. Jany, A. Bejan, Scaling theory of melting with natural convection in an enclo-
[34] M. Faden, A. König-Haagen, S. Höhlein, D. Brüggemann, An implicit algo- sure, International Journal of Heat and Mass Transfer 31 (6) (1988) 1221–1235,
rithm for melting and settling of phase change material inside macrocap- doi:10.1016/0017- 9310(88)90065- 8.
sules, International Journal of Heat and Mass Transfer 117 (2018) 757–767, [54] A. Bejan, Convection Heat Transfer, John wiley & sons, 2013.
doi:10.1016/j.ijheatmasstransfer.2017.10.033. [55] J.H. Lienhard, On the commonality of equations for natural convection from
[35] K. Schüller, J. Kowalski, Spatially varying heat flux driven close-contact immersed bodies, International Journal of Heat and Mass Transfer 16 (11)
melting–A Lagrangian approach, International Journal of Heat and Mass Trans- (1973) 2121–2123, doi:10.1016/0017-9310(73)90116-6.
fer 115 (2017) 1276–1287, doi:10.1016/j.ijheatmasstransfer.2017.08.092. [56] V. Alexiades, J.B. Drake, A weak formulation for phase-change problems with
[36] F. Rösler, Modellierung und Simulation der Phasenwechselvorgänge in bulk movement due to unequal densities, Free Boundary Problems Involving
makroverkapselten latenten thermischen Speichern, Logos-Verlag, Berlin Solids (1993) 82–87.
(2014), zugleich: Diss. Univ. Bayreuth, 2014, 2013. [57] R.I. Issa, Solution of the implicitly discretised fluid flow equations by operator-
[37] F.L. Tan, S.F. Hosseinizadeh, J.M. Khodadadi, L. Fan, Experimental and compu- splitting, Journal of Computational Physics 62 (1) (1986) 40–65, doi:10.1016/
tational study of constrained melting of phase change materials (PCM) inside 0 021-9991(86)90 099-9.
a spherical capsule, International Journal of Heat and Mass Transfer 52 (15-16) [58] V.R. Voller, C.R. Swaminathan, General source-based method for solidification
(2009) 3464–3472, doi:10.1016/j.ijheatmasstransfer.2009.02.043. phase change, Numerical Heat Transfer, Part B: Fundamentals 19 (2) (1991)
[38] S. Wang, A. Faghri, T.L. Bergman, A comprehensive numerical model for melt- 175–189, doi:10.1080/10407799108944962.
ing with natural convection, International Journal of Heat and Mass Transfer [59] C.R. Swaminathan, V.R. Voller, A general enthalpy method for modeling so-
53 (9-10) (2010) 1986–20 0 0, doi:10.1016/j.ijheatmasstransfer.20 09.12.057. lidification processes, Metallurgical Transactions B 23 (5) (1992) 651–664,
[39] P.A. Galione, O. Lehmkuhl, J. Rigola, A. Oliva, Fixed-grid modeling of solid- doi:10.1007/BF02649725.
liquid phase change in unstructured meshes using explicit time schemes, Nu- [60] M. Faden, S. Höhlein, J. Wanner, A. König-Haagen, D. Brüggemann, Review of
merical Heat Transfer, Part B: Fundamentals 65 (1) (2014) 27–52, doi:10.1080/ Thermophysical Property Data of Octadecane for Phase-Change Studies, Mate-
10407790.2013.836399. rials 12 (18) (2019) 2974, doi:10.3390/ma12182974.
[40] R. Tamme, T. Bauer, E. Hahne, Heat Storage Media, in: Ullmann’s Encyclope- [61] S. Höhlein, A. König-Haagen, D. Brüggemann, Thermophysical characterization
dia of Industrial Chemistry, 17, Wiley-VCH Verlag GmbH & Co. KGaA, 2009, of MgCl2·6H2O, xylitol and erythritol as phase change materials (PCM) for la-
pp. 421–438, doi:10.10 02/143560 07.a12_b30.pub2. tent heat thermal energy storage (LHTES), Materials 10 (4) (2017), doi:10.3390/
[41] V. Shatikian, G. Ziskind, R. Letan, Numerical investigation of a PCM-based heat ma10040444.
sink with internal fins, International Journal of Heat and Mass Transfer 48 (17) [62] S. Ushak, A. Gutierrez, H. Galleguillos, A.G. Fernández, L.F. Cabeza, M. Grágeda,
(2005) 3689–3706, doi:10.1016/j.ijheatmasstransfer.2004.10.042. Thermophysical characterization of a by- product from the non-metallic in-
[42] V. Shatikian, G. Ziskind, R. Letan, Numerical investigation of a PCM-based dustry as inorganic PCM, Solar Energy Material and Solar Cells 132 (2015)
heat sink with internal fins: Constant heat flux, International Journal of Heat 385–391.
and Mass Transfer 51 (5–6) (2008) 1488–1493, doi:10.1016/j.ijheatmasstransfer.
2007.11.036.

11

You might also like