Applied Energy 229 (2018) 700–714

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Comparison of direct numerical simulation with volume-averaged method T

on composite phase change materials for thermal energy storage
⁎
Xiaohu Yanga,b,c, Qingsong Baia, Zengxu Guoa, Zhaoyang Niua, Chun Yangd, Liwen Jina, ,
⁎
Tian Jian Lue, , Jinyue Yanb,c
a
Institute of the Building Environment & Sustainability Technology, School of Human Settlements and Civil Engineering, Xi’an Jiaotong University, Xi’an 710049, PR China
b
Department of Chemical Engineering and Technology/Energy Processes, Royal Institute of Technology (KTH), 100 44 Stockholm, Sweden
c
School of Sustainable Development of Society and Technology, Mälardalen University, 721 23 Västerås, Sweden
d
School of Mechanical and Aerospace Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798, Singapore
e
State Key Laboratory of Mechanics and Control of Mechanical Structures, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, PR China

H I GH L IG H T S

• Volume-averaged method is compared with direct numerical simulation in phase change.

• DNS results reports significant temperature difference between ligaments and PCM.
• Natural convection strongly affects the melting front shape and temperature distribution.
• VAM with two-temperature model predicts well, although pore-scale features are lost.

A R T I C LE I N FO A B S T R A C T

Keywords: Melting heat transfer in open-cell metal foams embedded in phase-change materials (PCMS) predicted by the
Open-cell metal foam volume-averaged method (VAM) was systematically compared with that calculated using direct numerical si-
Volume averaged method mulation (DNS), with particular attention placed upon the contribution of natural convection in the melt region
Direct numerical simulation to overall phase change heat transfer. The two-temperature model based on the assumption of local thermal non-
Phase change
equilibrium was employed to account for the large difference of thermal conductivity between metallic liga-
Natural convection
ments and PCM (paraffin). The Forchheimer extended Darcy model was employed to describe the additional
flow resistance induced by metal foam. For the DNS, a geometric model of metal foam based on tetra-
kaidehedron cells was reconstructed. The DNS results demonstrated significant temperature difference between
ligament surface and PCM, thus confirming the feasibility of local thermal non-equilibrium employed in VAM
simulations. Relative to the DNS results, the VAM combined with the two-temperature model could satisfactorily
predict transient solid-liquid interface evolution and local temperature distribution, although pore-scale features
of phase change were lost. The presence of natural convection affected significantly the melting front shape,
temperature distribution and full melting. The contribution of natural convection to overall phase change heat
transfer should be qualitatively and quantitatively given sufficient consideration from both macroscopic (VAM)
and microscopic (DNS) point of views. Besides, practical significance and economic prospective using metal
foam in TES unit for WHR system to provide residential heating or hot water is discussed and analyzed.

1. Introduction
The energy demand has increased sharply in the world in recent
decades according to a summary of energy consumption for 69 coun-
tries, especially in developing countries that are accelerating urbani-
zation process [1]. It is indicated that the demand for oil will increase
by 30% while coal and natural gas will increase by 50% from 2007 to
2035. China is the biggest country for energy consumption worldwide
with an increase rate of 18% in 2011, while the US, EU, India, Russia
and Brazil follow behind. Research demonstrates that the energy con-
sumption in China will increase to be more than 15 times by 2050
compared with that in 1970 [2]. The rapid consumption of

⁎
Corresponding authors.
E-mail addresses: lwjin@xjtu.edu.cn (L. Jin), tjlu@xjtu.edu.cn (T.J. Lu).

https://doi.org/10.1016/j.apenergy.2018.08.012
Received 9 October 2017; Received in revised form 28 July 2018; Accepted 3 August 2018
0306-2619/ © 2018 Elsevier Ltd. All rights reserved.
X. Yang et al. Applied Energy 229 (2018) 700–714

Nomenclature u Component velocity in x axis (m s−1)

→
U Velocity vector (m s−1)
Abbreviation v Component velocity in y axis (m s−1)
V Void volume for a TES unit (m3)
DNS Direct numerical simulation Vc Fully charged volume for a TES unit (m3)
Expt Experimental measurement w Component velocity in z axis (m s−1)
MF Metal foam 〈〉 Extrinsic average of a quantity over a control volume
NS Numerical simulation || Magnitude of a vector
TES Thermal energy storage
PCM Phase change material Greek symbols
PF Plate fin
PPI Pore per inch α Cross-sectional area ratio of node to solid ligament
VAM Volume-averaged method αsf Specific area (m−1)
WHR Waste heat recovery β Thermal expansion coefficient (K−1)
δ Numerical constant
Symbols ε Porosity
η Percentage of welding to materials cost
Am Numerical coefficient for damping velocity ρ Density (kg m−3)
CE Inertia coefficient (m−1) μ Dynamic viscosity (N s m−2)
Cp Specific heat (J kg−1 K−1) Π Yearly earned profit ($·yr−1)
Ch Price for residential heat ($·GJ−1) σ Liquid fraction in the porous medium
Cut Specific material cost for metal foam ($·m−3) χ Flow tortuosity
dp Pore diameter (m)
e Thickness ratio of node to solid ligament Subscript
fl Melting fraction
Fo Fourier number e Effective parameter
G Shape function for metallic ligaments f Phase change material
hsf Interstitial heat transfer coefficient (W m−2 K−1) full Full melting rate
k Thermal conductivity (W m−1 K−1) i Initial state
L Latent heat of fusion of PCM (J kg−1) m Melting point
N Number of operating hours (h) s Solid ligaments
Nu Nusselt number td Thermal dispersion
Nu Integral mean Nusselt number w Wall
P Pressure (Pa) 1 Solid state
Qc Total stored energy (J) 2 Liquid state
T Temperature (K)
t Time (s)

conventional fossil fuels has therefore caused a lot of serious energy and
environmental issues, such as energy shortages, global warming, en-
vironmental pollution and etc. [3–5]. For the time being, convectional
fossil fuels have still played the vital role in global energy supply. To
mitigate the environment stress, on one hand, efforts have been paid to
transferring the fossil fuel energy sources towards renewable energy,
such as solar, wind, tidal and biomass and so on; on the other hand,
turning “waste heat” to “useful heat” can also improve the overall en-
ergy efficiency of the existed energy systems, since about 20–50% of
energy used in industry is rejected as waste heat [6,7].
Waste heat recovery (WHR) system is capable of collecting in-
dustrial “waste heat” and converting it to “useful heat” for incoming
reactants preheating, electricity generation, hot water and central
heating for residences, and so on. Thermal energy storage (TES) via
absorbing/releasing latent heat by phase change materials (PCMs) is
one of the proved approaches for WHR. However, the charging/dis-
charging rate suffers significantly from the low thermal conductivity of
engineering-utilized PCMs (e.g. paraffin ∼ 0.2 Wm−1 K−1) [8], thus
limiting its practical applications. To this end, various techniques have
been developed to enhance phase change process. To be summarized,
there are two main approaches according to the thermal enhancement
spreader adopted: movable – micro/nano particles, or non-movable –
porous matrix. Compared with the attainable enhancement by directly
adding micro/nano particles into the PCMs [9], the impregnation of
PCMs into highly porous metal foams with open cells is particularly
promising for structural controllability and satisfactory thermal
enhancement [10–13]. A series of studies have indicated that metal
foams possess a superior performance in enhancing the thermal con-
ductivity of PCMs during phase change process. To model the phase
change process in a PCM-foam composite, two approaches are com-
monly employed according to the scale considered in the simulations:
the volume-averaged method (VAM) and the direct numerical simula-
tion (DNS).
The VAM typically treats the PCM-foam composite as an equivalent
fluid with identical thermophysical properties, with pore-scale thermal
fluid and geometric features lumped into the equivalent fluid. The
predicted results demonstrate global fluid flow and heat transfer char-
acteristics. Assuming local thermal equilibrium or non-equilibrium, one
may employ the one- or two-temperature model. Pioneering work on
thermal equilibrium assumption (one-temperature model) can be traced
back to Beckermann and Viskanta [14], who applied volume-averaged
transport equations to model the process of phase change in porous
media. The topology of solid-liquid interface was found to be greatly
influenced by the onset of natural convection in the melt phase. Feng
et al. [15] and Yang et al. [16] applied the one-temperature model to
simulate the phase change behavior of PCM saturated in open-cell metal
foam, and found that the model may be suitable for solidification
process.
As for melting heat transfer in porous media, the two-temperature
model that considers the temperature difference between solid ligament
surface and PCM may be more appropriate [17]. Assuming local
thermal non-equilibrium, Mesalhy et al. [18] considered the Darcy-

701
X. Yang et al.
Brinkman-Forchiemer effect on free convection in the melt region. The
phase change behavior was analyzed for different Fourier numbers and
the predictions agreed well with those in literature. Since heat transfer
between ligament surface and the interstitial fluid was difficult to
measure, the heat transfer coefficient across a bank of cylinders was
commonly employed. By varying the interstitial heat transfer coeffi-
cient, Krishnan et al. [19] discussed the effect of interstitial Nusselt
number on the propagation of melting front. Besides, the influence of
Rayleigh and Stefan numbers on transient phase change heat transfer
was also documented and discussed. Chen et al. [20] investigated
thermal energy storage in a solar flat-plate collector with metal foam as
heat transfer enhancer for phase change. It was demonstrated that, in
comparison with the one-temperature model, the two-temperature
model can better predict phase change heat transfer in the solar thermal
collector. Tian and Zhao [21] solved the problem of coupled heat
conduction and natural convection in phase transition and liquid zones
for PCM-metal-foam composites. They confirmed the applicability of
local thermal non-equilibrium model to simulate melting heat transfer
in porous materials that have much higher thermal conductivity than
PCMs. Li et al. [22] conducted both experimental and numerical in-
vestigations on melting phase change in copper foams, and argued that
the enhanced heat conduction compensated the suppressed natural
convection, yielding significant enhancement in phase change rate.
Zhang et al. [23] conducted numerical simulations on similar scenario
using a more realistic natural convection thermal boundary, with local
temperature difference between ligament surface and interstitial PCM
measured at one selected point. Qu et al. [24] simulated the phase
change performance of a metal foam heat sink for power battery
thermal management. A two-dimensional (2D) model with local
thermal non-equilibrium was adopted. Finally, it should be mentioned
that the newly developed Lattice Boltzmann Method (LBM) was also
employed to investigate phase change heat transfer in porous materials
[25–28].
Although VAM is capable of predicting phase change behavior of
porous media, it only provides an equivalent description of both fluid
flow and heat transfer characteristics in the complex foam structures, as
pore-scale features are lost under the basic assumption of this method.
For comparison, DNS has competitive preponderance in providing
physical insight into pore-scale phase change characteristics, such as
local solid-liquid interface shape and propagation, temperature
Fig. 1. Image of open-cell metal (copper) foam and reconstructed sphere-cut tetrakaidecahedron cells.
702
Applied Energy 229 (2018) 700–714
distribution and temperature difference between metallic ligament
surface and PCM. However, thus far, pore-scale numerical simulations
are scarce [29–34]. Notably, Hu et al. [29,32] conducted direct nu-
merical simulation of the melting heat transfer in open-cell metal foams
having idealized cellular topologies: a body-centered-cubic (BCC) unit
cell was constructed to model the phase change process with constant
thermal boundary condition. With temperature distribution and phase
interface evolution illustrated and the dominant role of heat conduction
identified, they confirmed that the DNS approach acquired affordable
computation resources and time. Sundarram and Li [30] and Yao et al.
[33] developed a 3D (three-dimensional) finite element (FE) model to
calculate the effective thermal conductivities of PCM-metal-foam
composites. The effects of pore size and porosity upon effective con-
ductivity as well as melting heat transfer characteristics were quanti-
fied. Feng et al. [31] and Zhang et al. [34] also conducted pore-scale
numerical simulations on phase change heat transfer in metal foams,
and suggested that the melting heat transfer could be further enhanced
with extra fin inserts and gradient packaging. Besides, PCM was also
impregnated into graphite foam to form graphite-foam-PCM composites
[35–37]. With graphite foam involved, the effective thermal con-
ductivity of the composite was significantly increased and the phase
change process was greatly enhanced.
The above literature survey demonstrates that the influence of open-
cell metal foam as heat spreader on phase change heat transfer has been
mainly simulated using the VAM approach, and only heat conduction-
dominated phase change was considered at pore-scale by the DNS ap-
proach. The contribution of local natural convection at pore level to
local phase interface evolution as well as global phase change behavior
is yet justified, nor fully understood. Besides, systematical comparison
between the two approaches has not been performed. In the present
study, melting heat transfer in open-cell metal foams was simulated
using both the VAM and DNS approaches. Phase interface, temperature
and velocity fields obtained using the two approaches were system-
atically compared. Particular attention was placed upon exploring the
physical mechanisms underlying pore-scale phase change and justifying
the contribution of local natural convection to both microscopic (pore-
scale) and macroscopic (global) phase change behaviors. While the
DNS-based results were used as benchmark of comparison, existing
experimental results were also used for better validation.
X. Yang et al. Applied Energy 229 (2018) 700–714

2. Direct numerical simulation (DNS) Table 1

Thermo-physical properties of PCM (Paraffin) [40] and aluminum [39].
2.1. Reconstruction of metal foam with open cells Material Parameter Value

High porosity open-cell metal foams typically have stochastic cel- Paraffin Melting temperature (K) 307 ∼ 309
Latent heat (J kg−1) 250,000
lular morphology, as shown in Fig. 1. To directly model the melting
Density (kg m−3) (solid/liquid) 880/760
heat transfer of PCM saturated in a metal foam block, the full size of the Special heat capacity (J kg−1 K−1) (solid/liquid) 2000
block should be considered. To simplify the problem, two fundamental Thermal conductivity (W m−1 K−1) 0.4/0.2
assumptions need to be made: one is to assume the stochastic metallic Kinematic viscosity (m2 s−1) 2.508 × 10−3
matrix to be periodic and the other is to employ tetrakaidecahedron Thermal expansion coefficient 0.001
Aluminum Density (kg m−3) 2719
unit cell (UC) to represent the real foam topology. Fig. 1 presents the
Special heat capacity (J kg−1 K−1) 871
reconstructed tetrakaidecahedron cells, which was centrally cut by a Thermal conductivity (W m−1 K−1) 202.4
sphere of 5.08 mm in diameter, resulting in a pore density of 5 PPI
(pores per inch). Correspondingly, the volume fraction (porosity) of the
UC (or foam) was 0.90. The UC of Fig. 1 was validated by our previous 2.4. Governing equations
DNS results on solidification of PCM saturated in metal foam [38].
For direct numerical simulation, heat transfer in metallic matrix was
2.2. Computational domain considered from heat transfer during phase change. The metallic matrix
was motionless, thus only energy equation was needed:
The periodicity of the problem posed above suggested that only one ∂Ts
piece of the whole structure was needed for direct numerical simula- ρs cps = ks ∇2 Ts
∂t (1)
tion, as illustrated in Fig. 2, with a total of 18 cells (3 × 6 × 1 for
x × y × z direction) considered to save computational cost. While all Conjugate heat transfer was considered at the interface between
tetrahedral elements were applied according to the geometric feature of PCM and ligament surface, as:
tetrakaidehedron cell, fine tetrahedral cells were applied on matrix
∂Tf ∂T
surface (Fig. 2) to catch the details of solid ligament surfaces. Tf = Ts, kf → = ks →s
∂n ∂n (2)

2.3. Initial and boundary conditions
The front and rear faces of the computational domain were treated
as symmetry, and the top, the bottom as well as the right faces were
thermally insulated. The temperatures of the PCM and the metallic
matrix were initially assumed to be the same as Ti (< Tm1) that equaled
to ambient air temperature. A constant temperature wall boundary was
applied on the left face, enabling phase change in the immediate vici-
nity of the heating wall. The initial temperature was set at 300 K and
the heating wall temperature at 368 K.
where ρ, cp and k were the density, specific heat and thermal con-
ductivity; the subscripts f and s stood for phase change material and
solid ligament, respectively.
Upon hating, the PCM in solid phase was going to melt, followed by
natural convection in the melt. The thermo-physical properties of the
PCM were assumed to be independent of temperature, except for the
density of PCM in liquid phase (Table 1) [39,40]. The Boussinesq model
was applied to the momentum equations to estimate the contribution of
local free convection to the melting process. The governing equations
were given by:

Fig. 2. Computational domain and representative mesh for DNS (direct numerical simulation).

703
X. Yang et al. Applied Energy 229 (2018) 700–714

Continuity equation:
→
∇· U = 0 (3)

Momentum equation:
→
∂U → → → (1−f )2 →
ρf + ρf (U ·∇) U = −∇P + μf ∇2 U + ρf →
g β (Tf −Tm) + 3 l Am U
∂t fl + δ
(4)
Energy equation:

∂Tf → ∂f
ρf cpf + ρf cpf U ·∇Tf = ∇ ·(kf ∇Tf )−ρf L l
∂t ∂t (5)

where ρf →
g β (Tf −Tm) was introduced to account for the buoyancy flow in
(1 − f )2 →
the melt phase, 3 l Am U was introduced to damp the velocity in
fl + δ
Momentum equation within the solid PCM domain ( fl = 0 ), two nu-
merical constants δ and Am were the damping coefficient: one relatively
small (δ = 10−10 ) and the other sufficiently large ( Am = 1015 ); the last
term in Energy equation described the latent heat absorption during
phase change process and fl was the melting fraction. The melting
fraction was determined depending upon the melting temperature
range (Tm1 ∼ Tm2), as:

⎧ 0 at T ⩽ Tm1 solid
⎪ Tf − Tm1
fl = at Tm1 < T < Tm2 mushy
⎨ Tm2 − Tm1
⎪ 1 at T ⩾ Tm2 liquid (6) Fig. 3. Computational domain and representative mesh for simulation based on
⎩
volume averaging.

2.5. Numerical procedure

The commercially available software ANSYS-FLUENT 17.1 was

employed to solve the governing equations, with the finite volume
method (FVM) with tetrahedral elements adopted for the modeling.
High resolution scheme was selected for better computation accuracy
and the solution was thought to be converged when the residuals of all
the governing equations were less than 10−6. Three sets of meshes with
2,675,390, 5,341,442 and 8,851,256 elements and three time steps of
0.1 s, 0.05 s and 0.01 s were tested for the case of Ti = 300 K and
Tw = 368 K. It was established that the predicted values from the last
two sets of meshes exhibited a deviation less than 1.3%. Hence, the set
of mesh having the order of magnitude of 5,341,442 elements and a
time step of 0.05 s was used in subsequent simulations. Validation of Fig. 4. Dependence of temperature at fixed position on time: comparison of
the numerical simulation was discussed in Section 4. VAM-based simulation results with experimental measurements [23].

Table 2
3. Volume averaged method (VAM)
Comparison of numerical prediction and experimental measurement [23] at a
given points.
3.1. Treatment of PCM-foam composite layer
Time Texpt at TNS at Relative Texpt at TNS at Relative
elapsed/s ligament/K ligament/K error/% PCM/ PCM/ error/%
Porous media saturated with fluid have been commonly treated as
K K
equivalent fluid with effective thermophysical properties. The bulk
porous medium is divided into small control volumes consisted of one 1000 307.3 305.8 0.49 306 305.3 0.23
entire pore structure. Thermal transport at pore scale is then lumped 2000 315.8 313.7 0.67 314.1 313.1 0.32
3000 322.2 319.7 0.78 320.5 319.1 0.44
into each control volume where volume-averaged integral calculation is
4000 327.8 325.3 0.76 326.1 324.9 0.37
performed. With VAM, the heat and mass transfer in porous media can 5000 332.8 329.4 1.02 331.1 329.2 0.57
be solved by adding source terms to the original Navier-stokes and 6000 337.8 335.6 0.65 336.2 335.4 0.24
energy equations. 7000 342.2 340 0.64 341.2 339.9 0.38
The governing equations for melting phase change heat transfer in a 8000 345.5 342.8 0.78 344.7 342.8 0.55
9000 348.2 345.4 0.80 347.8 345.3 0.72
PCM-metal-foam composite layer can be summarized as follows:
10,000 351 349.6 0.40 351.2 349.6 0.46
Continuity equation:
Note: Expt and NS stand for experimental measurement and numerical simu-
∂ρf →
+ ∇ (ρf 〈U 〉) = 0 lation, in respective.
∂t (7)

Momentum equations:

704
X. Yang et al. Applied Energy 229 (2018) 700–714

Fig. 5. Comparison between solid-liquid interface evolutions obtained separately with VAM and DNS: (a) with local natural convection considered; (b) conduction
dominated (natural convection neglected).

705
X. Yang et al. Applied Energy 229 (2018) 700–714

density difference between solid and liquid phases was neglected. The
open-cell metal foam of Fig. 1 was taken as homogeneous and isotropic
with respect to its effective thermophysical properties.

3.2. Thermophysical properties

To solve the governing equations for melting heat transfer in PCM-

saturated open-cell metal foams, the structural parameter (asf ), the in-
terstitial heat transfer coefficient (hsf ) as well as the key thermophysical
properties (ke, kfe, kse, ktd, K , CE ) needed to be determined. As pre-
viously mentioned, for the VAM, pore-scale structural features were
lumped into the macroscopic parameters, e.g., the porosity and pore
density. Calmidi et al. [41] proposed a model to link pore-scale struc-
Fig. 6. Melting fraction plotted as a function of time: comparison between VAM tural parameters to the macroscopic ones, and obtained the specific
and DNS. area as:
1.18
ρf ∂〈u〉 ρf → ∂〈p〉 μf μf ρf CE → ⎞ asf = 3π (1−ε )
+ 2 (〈U 〉·∇) 〈u〉 = − + ∇2 〈u〉−⎜⎛ + |〈U 〉|⎟ 〈u〉 dp (13)
σ ∂t σ ∂x σ ⎝ K K ⎠
(1−fl )2 where dp denoted the average pore size of open-cell foam and could be
− Am 〈u〉 estimated by:
fl3 +δ (8)
0.0224
ρf ∂〈v〉 ρf → μf μf ρf CE → ⎞ dp =
∂〈p〉 ω (14)
+ 2 (〈U 〉·∇) 〈v〉 = − + ∇2 〈v〉−⎜⎛ + |〈U 〉|⎟ 〈v〉
σ ∂t σ ∂y σ ⎝K K ⎠ The pore density ω was commonly provided by the manufacturer.
(1−fl )2 The interstitial heat transfer coefficient could in general not be di-
− Am 〈v〉 + ρf gβ (〈Tf 〉−Tm1)
fl3 + δ (9) rectly measured by experiments due to the complex porous structure
and the complicated fluid flow condition within a pore. Consequently,
ρf ∂〈w〉 ρf → ∂〈p〉 μf μf ρf CE → ⎞ the empirical model of Žukauskas [42] was often employed to simplify
+ 2 (〈U 〉·∇) 〈w〉 = − + ∇2 〈w〉−⎜⎛ + |〈U 〉|⎟ 〈w〉
σ ∂t σ ∂z σ ⎝ K K ⎠ the heat transfer scenario in open-cell foams and satisfactory agreement
(1−fl )2 was achieved. In the present study, this model was also employed, as
− Am 〈w〉 [42]:
fl3 + δ (10)
0.4 0.37
→ ⎧ 0.76Re Pr kf /d, 0 < Re⩽40
where the Forchheimer extended Darcy equation ( μf
K
+
ρf CE
K )
|〈U 〉| 〈u〉 ⎪
hsf = 0.52Re0.5Pr 0.37kf / d, 40 < Re⩽1000
was employed to describe the additional flow resistance induced by ⎨
(1 − fl )2 ⎪ 0.26Re0.6Pr 0.37kf / d, 1000 < Re⩽20000 (15)
metal foam, Am 〈u〉 denoted the source term for damping the ve- ⎩
fl3 + δ
locity before melting occurred (with numerical constants taking same where d was the average ligament diameter of the open-cell foam. In
values as those for DNS). The free motion driven by temperature dif- practice, d could be estimated by examining scanning electron micro-
ference in the melt phase was assumed to be Boussinesq free convec- scope (SEM) images.
tion, governed by ρf gβ (〈Tf 〉−Tm1) in Momentum equation. With regard to thermophysical properties, the following models
For the energy equations, due to large difference in thermophysical were employed.
properties between metal ligaments and PCM, thermal non-equilibrium Effective thermal conductivity [43]:
was considered to model transient melting heat transfer between liga-
ke (1−ε ) kf
ments and PCM. Therefore, the two-temperature model was adopted, = + ε
as: ks
(
1−e +
3e
2α ) ⎡⎣3(1−e) + 3
2
αe⎤
⎦
ks
(16)
Energy equation for PCM:
Thermal dispersion conductivity [44]:
df ∂〈Tf 〉 →
ερf ⎛cpf + L l ⎞
⎜ ⎟ + ρf cpf 〈U 〉·∇〈Tf 〉 = ∇ ·((kfe + ktd ) ∇〈Tf 〉) 0.36
⎝ dt ⎠ ∂t ktd = ρ cpf d u2 + v 2 + w 2
1−ε f (17)
−hsf asf (〈Tf 〉−〈Ts 〉) (11)
Permeability and inertial coefficient [45]:
Energy equation for metal foam:
ε [1−(1−ε )1/3]
∂〈Ts 〉 K= dp2
(1−ε ) ρs cps = ∇ ·(kse ∇〈Tf 〉)−hsf asf (〈Ts 〉−〈Tf 〉) 108[(1−ε )1/3−(1−ε )] (18)
∂t (12)
where ∇ ·((kfe + ktd ) ∇〈Tf 〉) in Energy equation for PCM (Eq. (11)) de- −1
cd ε ⎛1.18 1−ε 1 ⎞
scribed the thermal diffusivity term and here the thermal dispersion CE = 0.095 ⎜ ⎟
12 3(χ −1) ⎝ 3π G ⎠ (19)
conductivity ktd was also accounted for in the melt phase; while in
Energy equation for metal foam (Eq. (12)) only thermal conductivity of In the above equations, ke, kf, ks denoted the (effective) thermal
metal foam was considered; for both two Energy equations (Eqs. (11) conductivity of PCM-metal foam composite, saturating fluid and solid
and (12)), hsf asf (〈Ts 〉−〈Tf 〉) denoted the interstitial heat transfer be- ligaments; e and α were the thickness and cross-sectional area ratio of
tween solid ligaments and the saturating PCM. node to solid ligament (e = 0.3, α = 1.5); cd was the drag coefficient
To numerically simulate the transient phase change behavior of which was suggested to have a mean value of 1.56 by Bhattacharya
PCM saturated in open-cell metal foam, a few assumptions were also et al. [46] in the porosity range of 0.8991–0.978; χ and G represented
made for simplification. PCM in liquid phase was considered to be in- the flow tortuosity and shape function for metallic ligaments, which
compressible Newtonian fluid and the volume expansion of PCM due to could be calculated by:

706
X. Yang et al. Applied Energy 229 (2018) 700–714

ε 3.3. Computational domain and initial/boundary conditions

χ=
1−(1−ε )1/3 (20)
For VAM-based numerical simulations, the PCM-metal foam com-
G = 1−e−(1 − ε )/0.04 (21) posite was treated as an equivalent fluid having effective thermo-
physical properties. The metallic ligaments and interstitial fluid were

Fig. 7. Temperature distribution of PCM on the mid-plane (z/t = 1/2) at Fo = 0.0057, 0.0113, 0.0171, 0.0227: comparison between VAM and DNS.

707
X. Yang et al. Applied Energy 229 (2018) 700–714

not distinguished physically in the computational domain, which had Technique were employed for numerical computation. During each
the size of L × t × H (length × width × height) as shown in Fig. 3. The iteration, the convergence was obtained only if the residual was less
initial conditions were: than 10−6.
The solution accuracy for a transient problem was significantly in-
〈Tf 〉 = 〈Ts 〉 = Ti, at t = 0 (22)
fluenced by mesh quality and time step. Both the mesh and time step
The boundary conditions were: sensitivities were tested using three sets of meshes with 80 × 100 × 15,
110 × 140 × 30 and of 130 × 150 × 45 elements and three sets of
〈Tf 〉|x = 0 = 〈Ts 〉|x = 0 = Tw (23)
time steps with 0.1 s, 0.05 s and 0.01 s. The initial and boundary con-
∂〈Tf 〉 ∂〈Ts 〉 ditions were previously presented: Ti = 300 K, Tw = 368 K, the right,
= =0 top and bottom faces were thermally insulated and the other two walls
∂x ∂x x=L (24)
x=L
were symmetry planes. The test sequence was as follows: first, with a
∂〈Tf 〉 ∂〈Ts 〉 ∂〈Tf 〉 ∂〈Ts 〉 time step of 0.5 s, three sets of meshes of 80 × 100 × 15,
= = = =0 110 × 140 × 30 and of 130 × 150 × 45 were tested; then, with a fixed
∂y y=0
∂y y=0
∂y y=H
∂y y=H (25) mesh density of 110 × 140 × 30, three time steps of 0.5 s, 0.1 s and
∂〈Tf 〉 ∂〈Tf 〉 0.05 s were tested. By comparing the numerical results obtained with
∂〈Ts 〉 ∂〈Ts 〉
= = = =0 different meshes and time steps, it was established that the mesh with a
∂z z=0
∂z z=t ∂z z=0
∂z z=t (26) total element size of 110 × 140 × 30 and a time step of 0.05 s was
sufficient to achieve high accuracy, as the predicted values from the last
3.4. Sensitivity test two sets of meshes/time steps exhibited a deviation less than 1.7%.

To solve the transient heat conduction with latent heat released and
the local natural convection in melt phase, the finite volume method
(FVM) was employed. The Pressure-Linked Equations algorithms
(known as SIMPLE) was employed for pressure and velocity coupling.
The staggered grid system with structural meshes was applied to dis-
cretize the continuity, momentum and energy equations. The Tri
Diagonal Matrix Algorithm (TDMA) and the Gauss-Seidel Iteration
4. Validation
Both the DNS and VAM approaches were used to characterize
transient melting heat transfer in PCM saturated in open-cell metal
foam.
For the DNS approach, melting phase change occurred in the PCM
trapped in individual pores and heat transfer occurred between metallic

Fig. 8. Flow streamlines at selected time at Fo = 0.0057, 0.0113, 0.0171 and 0.0227: comparison between VAM and DNS.

708
X. Yang et al.
Fig. 9. DNS-predicted temperature distributions of PCM on selected planes (z/t = 0, 1/6, 1/3 and 1/2) at Fo = 0.0171.
ligaments surfaces and the trapped PCM. If the pore-scale structural
features of the foam were precisely captured, the phase change beha-
vior of the PCM-foam composite could be accurately characterized. In
the present study, inspired by our previous studies on heat conduction
in open-cell foams [38], Tetrakaidehedron cells were again used to
reconstruct the topological features of the foam.
As for the VAM-based numerical model, significant assumption was
made: the metal-foam-PCM composite was approximated as an
equivalent fluid that had the same effective thermophysical properties
as the composite. The sequence for model validation was as follows:
first, the VAM-based simulation was validated by comparing the pre-
dicted temperature variation history with existing experimental mea-
surements [23] under the same initial and boundary conditions ad-
dressed in Ref. [23]; subsequently, the DNS-based simulation was
further validated by quantitatively comparing with the VAM-based si-
mulation results. Zhang et al. [23] conducted experimental measure-
ment on the transient melting heat transfer in paraffin embedded in
open-cell copper foam under the constant heat flux thermal boundary.
The rectangular enclosure with the inner size of
100.0 × 100.0 × 10.0 mm was used to house the metal foam saturated
by paraffin (composite PCM). The center of the composite PCM was
selected as the point for monitoring temperature variation. Two ther-
mocouples were employed to measure the temperatures for metallic
ligament surface and the PCM inside the pore space, in respective. One
was directly inserted in the pore space filled with PCM and the other
was soldered on the ligament surface. Fig. 4 compared the temperature
variation history at selected points obtained from the VAM approach
with the experimental measurements [23]. Good agreement was
achieved between numerical and experimental data (maximum devia-
tion 1.02%, see Table 2), confirming the feasibility of the
709
Applied Energy 229 (2018) 700–714
computational models developed in the present study. Further, by
employing the non-equilibrium thermal model, slight difference be-
tween the ligament surface and the saturating paraffin was also ob-
served in Fig. 4.
5. Results and discussion
5.1. Solid-liquid interface
Fig. 5 compared the evolution of solid-liquid interface predicted
using the VAM with that using the DNS at four selected times
Fo = 0.0057, 0.0113, 0.0171 and 0.0227. Particular effort was made to
justify the contribution of local natural convection to the evolution. Due
to the symmetry nature in VAM simulations, only the 2D results were
depicted. For the DNS-based results, the grey and white colors denoted
separately the solid and liquid paraffin, and the tetrakaidehedron cells
in brown color could be clearly seen. The blue and red colors in the
VAM-based results set the two melting fraction limits, 0 and 1.0, which
represented the paraffin in solid phase and liquid phase respectively;
and the other colors turning from blue to red denoted the mushy zone
(mixture of solid and liquid phases). Due to the narrow temperature
range of phase change (307 ∼ 309 K) in the present PCM, the thickness
of the mushy zone was not large. Further, given the fundamental as-
sumption underlying the VAM approach, the metallic ligaments could
not be seen in the simulation results.
Fig. 5(a) demonstrated the specific features of solid-liquid interface
evolution. Initially, once the solid paraffin adjacent the heating
boundary started to melt, the interface gradually moved away from the
boundary. At this stage, the volume of the melt phase was so small that
natural convection was not strong enough to enhance local heat
X. Yang et al.
Fig. 10. Comparison between VAM- and DNS-predicted temperature profiles at
select points: (a) natural convection dominated case; (b) heat conduction
dominated case; temperature variations at selected points on metallic ligaments
and interstitial PCM with natural convection (c) and without natural convection
(d).
710
Applied Energy 229 (2018) 700–714
transfer. Hence, the solid-liquid interface was globally flat (see
Fig. 5(a), DNS result at Fo = 0.0057) and parallel to the heating
boundary. Subsequently, more solid paraffin was melt and the melt
phase flew upward, leading to local acceleration of interface evolution
(see Fig. 5(a), DNS results at Fo = 0.0171 and 0.0227). This contributed
to a macroscopically “slope” shape of the phase interface. From pore-
scale point of view, it could be seen that the melting interface was
microscopically uneven in contrast to the smooth one observed in VAM
simulations (Fig. 5(a)). A large portion of the PCM adjacent the metallic
ligaments was able to be melt at a faster speed, and the PCM located in
the bigger window of the tetrakaidehedron cell was still in solid phase
while the solid PCM next to the ligaments had already melt (Fig. 5(a),
DNS results). In contrast, with VAM simulations, no metallic ligaments
could be observed and a continuous melting interface separated the
solid and liquid paraffin. However, broadly speaking, the two numer-
ical models yielded qualitatively similar predictions of transient solid-
liquid interface evolution.
The DNS results of Fig. 5(a) and (b) also quantified the contribution
of local natural convection to melting interface propagation. In com-
parison, the solid-liquid phase interface predicted by the VAM was
globally flat and parallel to the heating boundary due to the conduc-
tion-dominated heat transfer mechanism. Initially, the volume of melt
phase was tiny so that the phase interface was similar, with or without
natural convection considered (see Fig. 5(a) and (b), the DNS results).
With time elapsed, local natural convection was gradually enhanced
which accelerated interface evolution in the top domain (see Fig. 5,
DNS results at Fo = 0.0227). However, the conduction-dominated heat
transfer mode resulted in a globally flat solid-liquid phase interface.
This phenomenon was also observed in the VAM-based results without
natural convection considered.
The contribution of natural convection to the evolution of melting
interface was addressed qualitatively in Fig. 5. To quantitatively com-
pare the two models, the melting fraction curve was plotted in Fig. 6.
The DNS approach predicted an increasing trend of the melting fraction
as a function of dimensionless time (Fo number), quantitatively similar
to that predicted by the VAM approach. This provided an indirect va-
lidation of the present pore-scale numerical model (i.e., the DNS ap-
proach). According to the development of local natural convection, the
history of interface evolution could be divided into three stages: the
pre-developed stage, the developed stage, and the post-developed stage.
At the pre-developed stage, the curve with natural convection con-
sidered overlapped with the conduction-dominated one; when more
solid paraffin was melt (melting fraction exceeding 0.3), the difference
between the two cases gradually increased, which was consistent with
the qualitative observation in Fig. 5. This could be explained as follows:
solid paraffin adjacent the heating boundary initially started to melt but
the melting phase had such a small volume that strong natural con-
vection was not developed, thus limiting the heat transfer enhancement
by natural convection. At the developed stage, more solid paraffin was
melt and the gradually-increased melting volume enabled convection
development. This could be validated by the interface evolution results
of Fig. 5 where the interface morphology gradually changed from plane
to slope. At this stage, natural convection played a vital role in en-
hancing the phase change heat transfer. Beyond this stage, the solid
paraffin formed a curve-edged triangle located in the right corner of the
studied domain, resulting in the post-developed stage. The solid PCM in
this corner region took about 30% of the full melting time to melt.
The present results demonstrated that while the VAM approach was
capable of predicting the overall phase change behavior both qualita-
tively and quantitatively, the pore-scale melting features were lumped
into the homogenization treatment of the PCM-foam composite. In
comparison, the DNS approach provided not only good prediction of
the overall phase change behavior but also captured pore-scale phase
change features. Nevertheless, the computational time needed by the
DNS approach was 29.4 times longer than that needed by the VAM
approach for the same phase change process.
X. Yang et al.
Fig. 11. (a) Comparison between DNS- and VAM-predicted boundary wall heat
flux profiles; (b) DNS-predicted average Nu number, both with and without
natural convection considered.
5.2. Temperature distribution
Fig. 7 presented snapshots of temperature distribution on the mid-
plane (z/t = 1/2) at selected times (Fo = 0.0057, 0.0113, 0.0171,
0.0227). Again, two different approaches (DNS and VAM) contribution
were employed. The temperature distribution was consistent with the
melting interface evolution results of Fig. 5. For instance, abundant
“slope” isothermal lines were observed: for pore-scale simulation, the
Fig. 12. Schematic illustration of TES unit assemblies.
711
Applied Energy 229 (2018) 700–714
presence of metallic ligaments resulted in locally tortuous isothermal
lines; for volume-averaged simulation, the isothermal lines resembled
the melting interface shown in Fig. 5. With natural convection con-
sidered, the uniformity of temperature in the gravity direction (y axis)
vanished. In contrast, when conduction was dominant, the temperature
was more uniform (isothermal lines more parallel to y-axis). On the left
surface, the PCM adjacent the heating boundary as well as the top re-
gion exhibited a higher temperature than PCM in other regions, due
mainly to the significant contribution of local natural convection to
melting heat transfer. This could be further validated by streamlines in
the melt phase shown in Fig. 8. A maximum flow velocity of 0.01 m/s
was found near the heating boundary and natural convection became
increasingly notable as more PCM was melt. Local natural convective
circulation was qualitatively predicted by the two numerical ap-
proaches. The fluid (melt) washed the solid PCM from the top to the
bottom, leading to the “slop” shape of the solid-liquid phase interface.
At pore scale, due to the presence of metal ligaments, the flow lines
were considerably tortuous, which might explain the uneven melting
interface observed in the vicinity of these ligaments.
For better demonstration of the pore-scale features of local tem-
perature distribution, comparisons were made among different planes
(z/t = 0, 1/6, 1/3 and 1/2) at Fo = 0.0171. As could be seen from
Fig. 9, natural convection played a significant role in shifting the
temperature distribution within the top region of the computation do-
main. The temperature fields globally resembled each other at different
planes but differed locally near the ligaments. The ligaments helped
increase the PCM temperature nearby. To quantitatively illustrate the
temperature variation history, two points located in the PCM domain
were selected to demonstrate the cases with/without natural convec-
tion, as shown in Fig. 10(a) and (b). Both approaches gave quantitative
predictions of temperature variation. The temperature at P1 behaved
similarly in trend as that at P2 but differed in values for the case with
natural convection. Comparatively, both the trend and value for tem-
perature at P1 and P2 were the same when natural convection was
neglected. This might serve as a quantitative evidence for the uniform
temperature distribution along gravity direction (y axis) in Figs. 7 and
9.
To show local thermal non-equilibrium, temperature variation his-
tories of the ligament surface as well as the saturating PCM were illu-
strated in Fig. 10(c) and (d). Another five points, P3 to P7, were
X. Yang et al.
Table 3
Item used for profit estimation.
Cut/$·m−3 V/m−3 η /% N/h tfull/s ε Ch/$·GJ−1
750 0.03 5 10 146 (MF) 0.97 6.67
200
(PF)
Note: MF and PF stand for metal foam and plate fin, in respective.
selected on the x−z plane at y/H = 0.5: P3 was located in the PCM
domain in the middle of the square cross-section of the tetra-
kaidehedron cell; the other four points, P4 to P7, were located in the
metallic ligament domain. Again, natural convection was considered in
Fig. 10(c) while heat conduction dominated the process in Fig. 10(d).
As could be seen, there did exist temperature difference between liga-
ment surface and interstitial PCM. The temperature in the PCM domain
was initially lower than that on the ligament surface. However, after
melting occurred on y/H = 0.5, i.e., Fo > 0.227, the PCM temperature
became equal to that on ligament surface. The initially lower tem-
perature in the PCM domain was due mainly to the fact that the PCM
absorbed heat and changed phase but maintained a smaller tempera-
ture rise during phase change. Upon phase changing, the temperature
difference gradually decreased, eventually approaching local thermal
equilibrium at pore scale.
5.3. Heat transfer performance
Numerical simulation can provide plenty of results in temperature
and velocity distribution within the whole computational domain
during the full melting process. To characterize the heat transfer per-
formance of phase change process, some advanced post-processing will
be carried out in the following part.
To quantify the effect of natural convection on melting heat
transfer, a modified Nusselt number based on the heat transfer coeffi-
cient of the heating wall was introduced, as:
h (t ) D q (t ) D
Nu (t ) = = w
kf (Tw−Ti ) kf
where h (t ) was the transient heat transfer coefficient, D was the char-
acteristic length of metal foam (thickness L), kf was the thermal con-
ductivity of the liquid PCM, qw(t) was the surface heat flux on the
heating boundary (which was the area-averaged value for DNS), Ti and
Tw were the temperatures of the initial state and heating wall, respec-
tively. With both DNS and VAM, the predicted Nusselt number was
plotted as a function of time in Fig. 11(a). There was good agreement
between the two approaches, with or without considering natural
convection. Thus, from the heat transfer point of view, these results
clearly demonstrated the feasibility and flexibility of the present pore-
scale simulations. Consideration of natural convection enhanced sig-
nificantly the Nusselt number. For example, the Nu number with nat-
ural convection was 11.4 times higher than that without natural con-
vection at Fo = 0.02.
To compare transient quantities during phase change, Yang et al.
[38] recommended an integral averaged method over the entire phase
change process. With ϕ (t ) and tfull denoting the transient quantity and
the full phase change time, the integration was performed as:
t
∫0 full ϕ (t ) dt
ϕ =
t full
The, upon integrating upon time, the average Nu number (Nu )
could be obtained as:
tfull qw (t ) D
Nu = ∫0 (Tw−Ti ) kf t full
dt
The corresponding results were presented in Fig. 11(b). The average
Applied Energy 229 (2018) 700–714
Nu number with natural convection considered was 7.7 times higher
than that without. This indicated, from the heat transfer point of view,
Π/$·yr−1 that natural convection should be given sufficient justification when
dealing with melting phase change process. However, although the Nu
3440 (MF) ratio (7.7) of the convection-dominated case to the conduction-domi-
2580 nated one was large, the full melting time was only reduced by 45.5%.
(PF)
This might be explained as follows: initially there was insufficient vo-
lume of the melt phase so that natural convection occurred locally near
the heating boundary, which was gradually enhanced after the melting
fraction reached 0.40 (the two curves of Fig. 6 were coincident when
the melting fraction was lower than 0.40). Further, in view of the fluid
motion features associated with natural convection, the top region al-
ways had fresh fluid with a higher temperature, thus enjoying a higher
local heat transfer rate (see temperature distribution in Fig. 7, VAM
results with natural convection). On the other hand, the corner region
that was at the bottom and far from the heating boundary suffered
significantly from low heat transfer rate. This was in consistence with
the results of melting interface propagation in Figs. 5 and 9. As a result,
for the solid PCM in the corner region to melt, about 30% of the full
melting time was consumed. During this period of time (i.e., the post-
developed stage), the contribution of natural convection to melting heat
transfer was not as much as that in the “developed” stage. However, by
observing the melting interface evolution history (Fig. 5(b), the VAM
results) in the conduction-dominated case, the interface was uniform in
the gravity direction and the corner area was smaller than that with
natural convection. Therefore, the full melting time was only reduced
by 45.5%, rather than 7.7 times less.
6. Techno-economic assessment
6.1. Energy recovery from waste heat
Waste heat is inevitably produced during an industrial process and
released by two main exhausts: gas or water. Industrial process pro-
duces a large amount of waste heat. For example, the cooling demand of
both industrial sectors and buildings in Europe is about 2.21 × 109 GJ/
y, and 78% of the total cooling demand is generated by industrial
(27) cooling applications (1.72 × 109 GJ/a) [47]. Direct release of waste
heat to environment not only reduces the overall energy efficiency of an
industrial process, but also makes thermal pollution to the environ-
ment. Outdoor thermal comfort level is significantly degraded due
mainly to the increased outdoor air temperature [48]. To both increase
the energy efficiency and avoid polluting the environment, WHR
technologies seem to be an appropriate way. WHR collects, stores and
transforms waste heat to useful thermal energy, either directly to the
thermodynamics process or provide hot water and central heating for
residential buildings. By using WHR, on one hand, the energy efficiency
of an industrial process is improved; on the other hand, the energy
consumption of fossil fuels to provide residential heating can be sig-
nificantly reduced, with less carbon dioxide, oxynitride and oxysulfide
emissions.
6.2. Technical assessment of metal foam composite PCM
Metal foams have promising structural controllability and robust-
ness for they can be cut into pieces and sintered with heat transfer
environment as channel or tube heat exchanger in typical engineering
applications [38]. Metal-foam-cored heat exchanger can be directed
used as the key element for a WHR system, favoring a high heat transfer
(28)
efficiency. Besides, the metal foam/PCM composites can be packed
separately and get charging/discharging like battery to form a flexible
energy supplier. It is very convenient to assemble several composite
PCM in parallel or series, to get either a much larger TES capacity or a
(29)
cascaded utilization of energy, as shown in Fig. 12. The present study
not only puts a physical insight into the thermo-fluidic characteristic
into the transport phenomena in porous media with interstitial phase
712
X. Yang et al.
change occurred, but also provides heat transfer performance evalua-
tion of composite PCMs potentially for WHR. Practical design of WHR
heat exchanger may benefit from the analysis of phase change heat
transfer in composite PCM unit.
6.3. Preliminary economic analysis
Open-cell metal foam can significantly enhance phase change heat
transfer, favoring a higher heat transfer coefficient. A higher heat
transfer coefficient indicates that the temperature difference between
the heat transfer media in two sides (exhaust gas and PCM) is smaller;
i.e. WHR system using metal foam for TES heat exchanger can extract
thermal energy from exhaust gas with a lower temperature. On the
other hand, if the temperature for the exhaust gas is fixed, using metal
foam is capable of tremendously reducing the full melting time for
thermal energy storage. This reveals that more waste energy can be
recovered and stored by the metal-foam cored phase change heat ex-
changer. Let’s compare the lifetime cost of TES unit with plate fin and
using metal foam as heat transfer enhancement media. The capital in-
vestment Cci using metal foam contains two parts: material cost Cm and
the welding cost Cw. Therefore, the capital investment can be expressed
as:
Cci = Cm + Cw (30)
Regarding that metal foam will be inserted and welded into the
space between two plates, the material cost is thus depended on the
void fraction of the TES unit V (volume for PCM). According to the
manufacturing routine, the welding cost is typically estimated to be η
percent of the total material cost. Therefore, the capital investment
turns to be
Cci = Cut V (1 + η)
(31)
−3
where Cut ($·m ) is the specific cost the materials.
The capital cost increases with using metal foam, while it can ex-
tract and store more energy from exhaust gas for a given period of time.
For comparison, two cases for TES with/without metal foam were si-
mulated: one used plate fin style; the other one inserted metal foam into
the spaces between every two plates. The full melting time for the TES
unit is tfull, and the available charging hours is N per day; therefore, the
fully charged volume Vc can be estimated to be
N This work was supported by the National Natural Science
Vc = Vε
t full (32)
Here, ε represents the porosity for metal foam and it equals to be unit
for the case without metal foam. Compared with the amount of latent
heat, sensible heat stored in a TES unit within a medium-low tem-
perature range can be neglected. The energy Qc stored in a TES unit can
be then estimated as
Qc = ρPCM Vc L
(33)
If the stored thermal energy is to provide hot water or central
heating for residential buildings, the profits Π for the case with metal
foam can be estimated to be
N
Π = Ch Qc− Cci
t full (34)
where the prices for residential heating is Ch ($·GJ−1).
Table 3 shows the profit estimation of the practical TES unit with/
without metal foam. The profit for the waste heat to provide residential
hot water is calculated in one-year operation. Although metal foam
material increases the capital cost (about $5,570 for total 7.4 m3)
compared with the plate fin case, more profits can be obtained due to
more energy stored within the same operating hours. A 6.5 year is
found to be the payback time for using metal foam. For a 15-year
lifetime, the metal-foam-cored plate heat exchanger (7.4 m3) can save
about $12,900 in comparison with plate heat exchanger. This indicates
713
Applied Energy 229 (2018) 700–714
the economic applicability of metal foam as heat transfer enhancement
for TES. It is pleasantly expected to obtain much more profits if large
scale of the metal-foam-cored plate heat exchanger was used for TES.
7. Conclusions
The melting behavior of phase change material (PCM) impregnated
into open-cell metal foam was investigated from both macroscopic and
microscopic point of view using two different approaches: the volume-
averaged method and the unit-cell-based direct numerical simulation.
The effects of natural convection on solid-liquid interface propagation,
local and global temperature distributions and wall heat transfer coef-
ficient were justified, both qualitatively and quantitatively. Main con-
clusions were drawn as follows:
(1) Temperature difference between metal ligaments surface and the
saturating PCM was not negligible, thus demonstrating the applic-
ability of the two-temperature model for modelling phase change in
PCM-metal-foam composites.
(2) The volume-averaged method predicted both qualitatively and
quantitatively similar phase change behaviors as the direct nu-
merical simulation from the global point of view. The direct nu-
merical simulation could additionally capture pore-scale features of
solid-liquid interface evolution and temperature distributions in
cellular foam structure as well as interstitial PCM.
(3) The average Nu number calculated with natural convection con-
sidered was 7.7 times higher than that without considering natural
convection. The melting front in the former propagated faster than
the latter, leading to 45.5% reduction in full melting time. These
results indicated the significant role played by buoyancy flow in
melting phase change process.
(4) For the same operating hours (10 h per day), metal-foam-cored TES
unit can store more energy compared with the case of plate fin heat
exchanger. A payback time of 6.47 year is found to use metal foam
for TES and a considerable amount of money (25, 800 $) can be
obtained by using metal foam with a total 7.4 m3 TES tank, after a
30-year operating.
Acknowledgements
Foundation of China (51506160), China Post-doctoral Science
Foundation Funded Project (2015M580845, 2016T90916), Shaanxi
Province Post-Doctoral Science Foundation Project (2016BSHYDZZ54),
China Northwest Architecture Design and Research Institute CO. Ltd
(TC-2016A-092), the fundamental research funds for central uni-
versities (xjj2016042) and the Beijing Key Lab of Heating, Gas Supply,
Ventilating and Air Conditioning Engineering (NR2016K01). Xiaohu
Yang is financially supported by China Scholarship Council (CSC) to
conduct research in Royal Institute of Technology (KTH) and
Mälardalen University (MDH), Sweden.
References
[1] Sharma SS. Determinants of carbon dioxide emissions: Empirical evidence from 69
countries. Appl Energy 2011;88:376–82.
[2] Berardi U. Building energy consumption in US, EU, and BRIC countries. Proced Eng
2015;118:128–36.
[3] Chua KJ, Chou SK, Yang WM, Yan J. Achieving better energy-efficient air con-
ditioning – A review of technologies and strategies. Appl Energy 2013;104:87–104.
[4] Hu Y, Yan J. Characterization of flue gas in oxy-coal combustion processes for CO2
capture. Appl Energy 2012;90:113–21.
[5] Tian Y, Zhao L, Meng H, Sun L, Yan J. Estimation of un-used land potential for
biofuels development in (the) People’s Republic of China. Appl Energy
2009;86:S77–85.
[6] Ji C, Qin Z, Dubey S, Choo FH, Duan F. Three-dimensional transient numerical study
on latent heat thermal storage for waste heat recovery from a low temperature gas
flow. Appl Energy 2017;205:1–12.
[7] Yang X, Lu Z, Bai Q, Zhang Q, Jin L, Yan J. Thermal performance of a shell-and-tube
X. Yang et al.
latent heat thermal energy storage unit: Role of annular fins. Appl Energy
2017;202:558–70.
[8] Yang X, Lu T, Kim T. Temperature effects on the effective thermal conductivity of
phase change materials with two distinctive phases. Int Comm Heat Mass Transfer
2011;38:1344–8.
[9] Liu YD, Zhou YG, Tong MW, Zhou XS. Experimental study of thermal conductivity
and phase change performance of nanofluids PCMs. Microfluid Nanofluid
2009;7:579–84.
[10] Fan L, Khodadadi J. Thermal conductivity enhancement of phase change materials
for thermal energy storage: a review. Renew Sustain Energy Rev 2011;15:24–46.
[11] Zhao CY, Lu W, Tian Y. Heat transfer enhancement for thermal energy storage using
metal foams embedded within phase change materials (PCMs). Sol Energy
2010;84:1402–12.
[12] Zhao CY, Wu Z. Heat transfer enhancement of high temperature thermal energy
storage using metal foams and expanded graphite. Solar Energy Mater Solar Cells
2011;95:636–43.
[13] Fleming E, Wen SY, Shi L, da Silva AK. Experimental and theoretical analysis of an
aluminum foam enhanced phase change thermal storage unit. Int J Heat Mass
Transf 2015;82:273–81.
[14] Beckermann C, Viskanta R. Natural-convection solid liquid-phase change in porous-
media. Int J Heat Mass Transf 1988;31:35–46.
[15] Feng SS, Zhang Y, Shi M, Wen T, Lu TJ. Unidirectional freezing of phase change
materials saturated in open-cell metal foams. Appl Therm Eng 2015;88:315–21.
[16] Yang X, Wang W, Yang C, Jin L, Lu TJ. Solidification of fluid saturated in open-cell
metallic foams with graded morphologies. Int J Heat Mass Transf 2016;98:60–9.
[17] Harris KT, Haji-Sheikh A, Nnanna AA. Phase-change phenomena in porous media–a
non-local thermal equilibrium model. Int J Heat Mass Transf 2001;44:1619–25.
[18] Mesalhy O, Lafdi K, Elgafy A, Bowman K. Numerical study for enhancing the
thermal conductivity of phase change material (PCM) storage using high thermal
conductivity porous matrix. Energy Convers Manage 2005;46:847–67.
[19] Krishnan S, Murthy JY, Garimella SV. A two-temperature model for solid-liquid
phase change in metal foams. J Heat Trans 2005;127:995–1004.
[20] Chen ZQ, Gu MW, Peng DH. Heat transfer performance analysis of a solar flat-plate
collector with an integrated metal foam porous structure filled with paraffin. Appl
Therm Eng 2010;30:1967–73.
[21] Tian Y, Zhao CY. A numerical investigation of heat transfer in phase change ma-
terials (PCMs) embedded in porous metals. Energy 2011;36:5539–46.
[22] Li WQ, Qu ZG, He YL, Tao WQ. Experimental and numerical studies on melting
phase change heat transfer in open-cell metallic foams filled with paraffin. Appl
Therm Eng 2012;37:1–9.
[23] Zhang P, Meng Z, Zhu H, Wang Y, Peng S. Melting heat transfer characteristics of a
composite phase change material fabricated by paraffin and metal foam. Appl
Energy 2017;185:1971–83.
[24] Qu Z, Li W, Tao W. Numerical model of the passive thermal management system for
high-power lithium ion battery by using porous metal foam saturated with phase
change material. Int J Hydro Energy 2014;39:3904–13.
[25] Huo Y, Rao Z. Investigation of phase change material based battery thermal man-
agement at cold temperature using lattice Boltzmann method. Energy Convers
Manage 2017;133:204–15.
[26] Gao DY, Chen ZQ. Lattice Boltzmann simulation of natural convection dominated
melting in a rectangular cavity filled with porous media. Int J Therm Sci
2011;50:493–501.
[27] Gao DY, Chen ZQ, Chen L. A thermal lattice Boltzmann model for natural
714
Applied Energy 229 (2018) 700–714
convection in porous media under local thermal non-equilibrium conditions. Int J
Heat Mass Transf 2014;70:979–89.
[28] Tao YB, You Y, He YL. Lattice Boltzmann simulation on phase change heat transfer
in metal foams/paraffin composite phase change material. Appl Therm Eng
2016;93:476–85.
[29] Hu X, Patnaik SS. Modeling phase change material in micro-foam under constant
temperature condition. Int J Heat Mass Transf 2014;68:677–82.
[30] Sundarram SS, Li W. The effect of pore size and porosity on thermal management
performance of phase change material infiltrated microcellular metal foams. Appl
Therm Eng 2014;64:147–54.
[31] Feng SS, Shi M, Li YF, Lu TJ. Pore-scale and volume-averaged numerical simula-
tions of melting phase change heat transfer in finned metal foam. Int J Heat Mass
Transf 2015;90:838–47.
[32] Hu X, Wan H, Patnaik SS. Numerical modeling of heat transfer in open-cell micro-
foam with phase change material. Int J Heat Mass Transf 2015;88:617–26.
[33] Yao Y, Wu H, Liu Z. Numerical study of heat conduction of high porosity open-cell
metal foam/paraffin composite at pore scale. ASME 2016 heat transfer summer
conference collocated with the ASME 2016 fluids engineering division summer
meeting and the ASME 2016 14th international conference on nanochannels, mi-
crochannels, and minichannels: American society of mechanical engineers; 2016. p.
V001T02A6-VT02A6.
[34] Zhang Z, He X. Three-dimensional numerical study on solid-liquid phase change
within open-celled aluminum foam with porosity gradient. Appl Therm Eng
2017;113:298–308.
[35] Sedeh MM, Khodadadi J. Thermal conductivity improvement of phase change
materials/graphite foam composites. Carbon 2013;60:117–28.
[36] Sedeh MM, Khodadadi J. Interface behavior and void formation during infiltration
of liquids into porous structures. Int J Multiphase Flow 2013;57:49–65.
[37] Sedeh MM, Khodadadi J. Solidification of phase change materials infiltrated in
porous media in presence of voids. J Heat Transfer 2014;136. 112603.
[38] Yang X, Feng S, Zhang Q, Chai Y, Jin L, Lu TJ. The role of porous metal foam on the
unidirectional solidification of saturating fluid for cold storage. Appl Energy
2017;194:508–21.
[39] Yang S, Tao W. Heat transfer. Beijing: Higher Education Press; 2004.
[40] Rubitherm Technologies GmbH; 2017.
[41] Calmidi VV. Transport phenomena in high porosity fibrous metal foams. Boulder,
CO: University of Colorado; 1998.
[42] Žukauskas A. Heat transfer from tubes in crossflow. Adv Heat Transf
1972;8:93–160.
[43] Yang XH, Bai JX, Yan HB, Kuang JJ, Lu TJ, Kim T. An analytical unit cell model for
the effective thermal conductivity of high porosity open-cell metal foams. Trans
Porous Med 2014;102:403–26.
[44] Georgiadis JG, Catton I. Dispersion in cellular thermal-convection in porous layers.
Int J Heat Mass Transf 1988;31:1081–91.
[45] Yang XH, Lu TJ, Kim T. An analytical model for permeability of isotropic porous
media. Phys Lett A 2014;378:2308–11.
[46] Bhattacharya A, Calmidi VV, Mahajan RL. Thermophysical properties of high por-
osity metal foams. Int J Heat Mass Transf 2002;45:1017–31.
[47] Oró E, Miró L, Farid MM, Martin V, Cabeza LF. Energy management and CO 2
mitigation using phase change materials (PCM) for thermal energy storage (TES) in
cold storage and transport. Int J Refrig 2014;42:26–35.
[48] Huang Q, Meng X, Yang X, Jin L, Liu X, Hu W. The ecological city: considering
outdoor thermal environment. Energy Proce 2016;104:177–82.