International Journal of Heat and Mass Transfer 78 (2014) 1145–1154

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

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

Nanofluid PCMs for thermal energy storage: Latent heat reduction

mechanisms and a numerical study of effective thermal storage
performance
Aitor Zabalegui, Dhananjay Lokapur, Hohyun Lee ⇑
Mechanical Engineering, Santa Clara University, CA, 500 El Camino Real, Santa Clara, CA 95053, United States

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

Article history: The latent heat of fusion of paraffin-based nanofluids has been examined to investigate the use of
Received 7 February 2014 enhanced phase change materials (PCMs) for thermal energy storage (TES) applications. The nanofluid
Received in revised form 13 June 2014 approach has often been exploited to enhance thermal conductivity of PCMs, but the effects of particle
Accepted 17 July 2014
addition on other thermal properties affecting TES are relatively ignored. An experimental study of par-
Available online 15 August 2014
affin-based nanofluids containing various particle sizes of multi-walled carbon nanotubes has been con-
ducted to investigate the effect of nanoparticles on latent heat of fusion. Results demonstrated that the
Keywords:
magnitude of nanofluid latent heat reduction increases for smaller diameter particles in suspension.
Nanofluid
Phase change materials
Three possible mechanisms – interfacial liquid layering, Brownian motion, and particle clustering – were
Thermal energy storage examined to explain further reduction in latent heat, through the weakening of molecular bond struc-
tures. Although additional research is required to explore detailed mechanisms, experimental evidence
suggests that interfacial liquid layering and Brownian motion cannot explain the degree of latent heat
reduction observed. A finite element model is also presented as a method of quantifying nanofluid
PCM energy storage performance. Thermal properties based on modified effective medium theory and
an empirical relation for latent heat of fusion were applied as model parameters to determine energy
stored and extracted over a given period of time. The model results show that while micro-scale particle
inclusions exhibit some performance enhancement, nanoparticles in PCMs provide no significant
improvement in TES performance. With smaller particles, the enhancement in thermal conductivity is
not significant enough to overcome the reduction in latent heat of fusion, and less energy is stored over
the PCM charge period. Therefore, the nanofluid approach may not be justifiable for energy storage appli-
cations. However, since the model parameters are dependent on the material properties of the system
observed, storage performance may vary for differing nanofluid materials.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction Zhang et al. [2] suggested micro and nano-encapsulation of PCMs

Thermal energy storage (TES) based on phase change phenom-
ena is promising for sustainable thermal power generation and res-
idential heating, due to the high specific storage capacity of phase
change materials (PCMs). However, the characteristically low ther-
mal conductivity of PCMs limits the rate at which they can
exchange thermal energy with other heat transfer media. Due to
this low charge/discharge rate, PCMs in TES applications may not
meet the energy demand over given periods of time, and thus, will
require larger heat exchangers. Several methods have been pro-
posed to enhance heat transfer rates of PCMs. Wu et al. [1] and
⇑ Corresponding author. Tel.: +1 408 554 5283.
E-mail address: hlee@scu.edu (H. Lee).
to increase the surface area to volume ratio, while Fukai et al. [3]
and Velraj et al. [4] suggested structural modifications to thermal
storage systems, such as the addition of fin configurations and
the insertion of a metal matrix. The nanofluid approach, involving
the dispersion of high conductivity particles into a base fluid, has
also been demonstrated as an effective way of enhancing thermal
conductivity, although the mechanisms behind it are controversial
[5,6]. An advantage of the nanofluid approach is its ability to be
combined with the previously mentioned structural modifications
to further augment heat transfer efficiency. Theoretical and exper-
imental studies by Prasher et al. [7] and Gao et al. [8] have identi-
fied nanoparticle clustering effects as the determining factor for
nanofluid conductivity enhancement. Furthermore, modified effec-
tive medium theory (EMT) by Nan et al. [9,10] considers high

http://dx.doi.org/10.1016/j.ijheatmasstransfer.2014.07.051
0017-9310/Ó 2014 Elsevier Ltd. All rights reserved.
1146 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154
nanoparticle aspect ratio and Kapitza resistance, and shows good
agreement with reported values.
Although considerable research efforts have focused on nano-
fluid thermal conductivity, other thermal properties affecting
TES, such as specific heat capacity or latent heat of fusion, do not
receive as much attention. Specific heat capacity is commonly
believed to exhibit no nanoscale effects [11], but recent experi-
mental findings by Shin and Banerjee [12,13] and Wang et al.
[14] show significant heat capacity enhancement in nanostruc-
tures and nanofluids. Shin and Banerjee proposed that the
observed nanofluid heat capacity enhancement was due to
improved thermal properties of semi-solid layers at particle inter-
faces, formed by liquid layering effects. On the other hand, nano-
fluid latent heat of fusion is expected to linearly decrease as
particles not contributing to phase change are added to the base
fluid [15,16]. However, several experimental studies on nanofluids
have reported additional reduction beyond latent heat EMT. Wu
et al. [15] reported a nearly 10% drop from the expected latent heat
of Cu/paraffin nanofluids with 25 nm diameter particles at 1% vol-
ume fraction. Zeng et al. [17] observed a similar reduction at 1%
volume fraction with copper nanowires in tetradecanol. Ho and
Gao [18] reported a less significant reduction of approximately
3% for alumina-in-paraffin emulsions, with 177.8 nm diameter par-
ticles at 2% volume fraction. Despite the similarity of these obser-
vations and their inconsistency with theory, latent heat
characterization was not the intended focus of these studies.
Although no possible reduction mechanisms were suggested, Wu
et al. [15] proposed that a new model for latent heat EMT of
solid–liquid mixtures is needed. From the reported findings, it is
apparent that nanofluids of smaller particle diameter exhibit
greater reduction in latent heat of fusion. Studies to investigate this
relationship have yet to be conducted.
Since most TES applications require energy to be stored or
released in a given amount of time, storage performance can be
evaluated as how much energy is transferred over a given duration.
Both latent heat of fusion and thermal conductivity affect energy
storage performance. While enhanced thermal conductivity
improves the rate of heat transfer, reduced latent heat decreases
specific energy storage capacity. Although reduced latent heat
can also increase energy charge/discharge rates due to shortened
melt time, greater nanofluid PCM volume will be required to
account for reduced storage capacity. Without reliably quantifiable
thermal properties, the nanofluid approach may not be effective for
PCM thermal energy storage.
In this work, a comprehensive characterization study of nano-
fluid latent heat of fusion has been conducted. Possible reduction
mechanisms are explored with respect to measured results. In
addition, a numerical model of nanofluid phase change is pre-
sented as a predictive tool to determine the amount of energy
stored/extracted in a given amount of time. Applying nanofluid
thermal conductivity enhancement approximated by Nan’s EMT,
the proposed numerical model serves as a predictive tool to deter-
mine the effects of the diameter-dependent latent heat reduction
observed.
2. Sample preparation and experimental method
Multi-Walled Nanotubes (MWNTs) of 15.5, 40, 65, and 400 nm
outer diameter (Cheap Tubes Inc.) were dispersed in liquid paraffin
(126 MP Wax – 3032, Candlewic) at 0.2%, 0.5%, and 1.0% particle
volume fractions, using high frequency pulse sonication (VCX750
Ultrasonic Processor, Sonics & Materials, Inc.). Once mixed, stock
nanosuspensions were individually poured into an acrylic mold,
solidified, and cut into 12  12 mm samples. Sample latent heat
of fusion was measured using differential scanning calorimetry
(STA 449 F3 Jupiter, NETZSCH). More details about the sample
preparation and experimental method were presented in our pre-
vious work [19].
An optical microscope (SMZ1500, Nikon) was utilized to exam-
ine suspension stability, since the surface charge accumulation
caused by non-conductive paraffin hinders imaging with electron
microscopy. No significant change in the degree of nanoparticle
clustering was observed before and after melt cycling. In addition,
no observable sedimentation occurred for stock left in liquid state
on a hot plate over a five-day period. Another vial of stock under-
went phase transition cycling over the same five-day period. The
stock was melted twice per day and left in liquid state for at least
two hours at a time. No visible sedimentation was detected.
3. Characterization results and discussion
Differential Scanning Calorimeter (DSC) measurements show
that at each volume fraction tested, sample latent heat of fusion
reduces for nanofluids of smaller particle diameter. At 1% particle
volume fraction, nanofluids with 15.5 nm diameter MWNTs exhi-
bit an additional 10% reduction below the mass loss prediction.
Shown in Fig. 1, all samples exhibit linear latent heat reduction
with particle loading, below that expected by traditional EMT.
Looking at a single volume fraction, latent heat is observed to
reduce with smaller diameter particles in suspension. Therefore,
additional reduction below the mass loss prediction is shown to
be independent of particle volume fraction. The apparent diame-
ter-dependence suggests that interface effects may have a role in
nanofluid latent heat reduction. Defining r as latent heat reduction
rate with respect to particle volume fraction, Fig. 1 shows that the
magnitude of reduction rates increases as particle diameter
decreases. Several experimental studies also appear to follow this
trend [15,17,18]. However, the nanofluids investigated by both
Wu et al. and Ho and Gao contain spherical particles, which may
explain observed differences from the measured results. An aver-
age value was estimated from provided SEM images of the work
by Zeng et al., which may not serve as an accurate representation.
Nevertheless, the strong dependence on particle diameter sug-
gested by these results serves as a basis for assessing the contribu-
tions of the proposed latent heat reduction mechanisms.
Theoretically, the latent heat of fusion of composite materials is
expected to linearly decrease with particle addition,
Fig. 1. Normalized nanofluid latent heat of fusion versus particle volume fraction
measured by DSC, along with experimental works by other groups. At constant
particle volume fractions, nanofluid latent heat is shown to reduce with decreasing
particle size.
 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154 1147

qbf ;s hsl;bf ð1  /p Þ
hsl;nf ¼ ; ð1Þ
qnf
where qbf,s denotes the base fluid density in solid phase, qnf is the
theoretical nanofluid density, and hsl,bf is the solid–liquid latent heat
of the base fluid. As mentioned previously, nanofluid latent heat
reduction is defined by the mass – or volume – of particles within
the nanofluid that does not contribute to phase change. In accor-
dance with experimental findings, latent heat reduction beyond
the mass loss prediction suggests that aside from nanoparticle vol-
ume, there is additional volume not contributing to latent heat. It is
proposed that this additional volume is represented by effective
volumes of strained base fluid molecular structure, which require
less energy to break down during melting. Molecular strain may
be attributed to the following interfacial phenomena: interfacial Fig. 2. Cross-section of MWNT with diameter dp, surrounded by a densely packed
liquid layering, Brownian motion, and particle clustering. Each of layer and strained layer of base fluid molecules, with a collective interface phase
these effects is diameter-dependent, and has a greater impact with width w.
reduced particle size. Therefore, these phenomena are considered as
mechanisms for the diameter-dependent nanofluid latent heat
reduction observed. The contribution of each reduction mechanism
width is thin (on the order of molecular spacings), demonstrating
is analyzed by approximating the respective strained region volume
that the effect of van der Waals forces is relatively weak. Therefore,
generated, and comparing it to the strained volume required to
base molecules beyond the DPL do not experience significant
explain observed reduction.
movement, and the total interface phase width should scale on
the same order as the DPL width. Consequently, interface volume
3.1. Interfacial liquid layering
fractions should also scale similarly to effective interface volume
fractions consisting of only the DPL.
The first mechanism, interfacial liquid layering, facilitates latent
Interface volume fractions for each particle size tested can be
heat reduction through the weakening of base fluid molecular
evaluated from measured nanofluid latent heat, using a modified
structure. At the interface, van der Waals forces attract nearby base
mass loss prediction for a ternary system. The nanofluid ternary
fluid molecules, forming a more ordered and densely packed layer.
system consists of the base fluid, nanoparticles, and interface
During this process, molecular bonds between surrounding base
phase. As a minimum estimate of interface volume fraction, inter-
fluid molecules are strained, and require less energy to break down
face phase structure is assumed to be completely broken down and
during melting. The effects of strain propagate normal to the inter-
not contributing to latent heat:
face, as the inverse of distance from the particle surface. Since the
number density of layered molecules increases with interface den-
qbf ;s hsl;bf ð1  /p  /i Þ
sity, smaller diameter particles generate greater volumes of hsl;nf ;tern ¼ ð5Þ
strained regions. To consider this effect theoretically, interface
qnf
volume fraction (/i) is defined as the volume of interface phase Using Eq. (5), interface volume fractions required to fit mea-
(Vi) – including both the densely packed layer at the interface and sured nanofluid latent heat are calculated and summarized in
surrounding strained layer – over the total nanofluid volume (Vnf). Table 1, for all particle diameters tested. DPL volume fractions
Vi are also calculated, using Eq. (4), assuming a DPL width of
/i ¼ ð2Þ w = 2 nm. In Table 1, all interface volume fractions are shown nor-
V nf
malized by particle volume fraction, and thus, are represented as
Representing total nanofluid volume as a function of particle ratios of either interface phase or DPL volume to individual particle
volume and particle volume fraction, Eq. (2) can be expressed as: volume. Also included are MWNT geometries and the required
  width of the interface phase to fit measured reduction, calculated
Vi
/i ¼ /p ð3Þ from Eq. (4).
Vp
The resulting required interface volume fractions significantly
where / is volume fraction, with i and p subscripts denoting inter- overestimate interface volume fractions consisting of only the
face and particle, respectively. DPL. Corresponding required interface phase widths are on the
Interface and particle volumes can be expressed through geo- order of particle diameter, which is highly inconsistent with
metric functions of particle diameter (dp), length (L), and interface approximations in literature [22]. Since strained regions within
phase width (w), as shown in Fig. 2. the interface phase are unlikely to occupy volumes two orders of
0  2  2 1 magnitude greater than respective DPL volumes, interfacial layer-
pL dp d
þ w  2p ! ing effects cannot solely explain the degree of latent heat reduction
B 2 C 4w 4w2
/i ¼ /p B
@  2 C
A ¼ /p dp þ 2 ð4Þ observed.
pL d2p dp A correlation between interfacial liquid layering and measured
latent heat can be made, considering the inverse proportionality to
After simplification, Eq. (4) demonstrates that interface volume particle diameter in Eq. (4). As shown in Fig. 3(a), each particle
fraction is inversely proportional to particle diameter when w/dp is diameter independently shows a linear relation to latent heat
much less than unity. reduction, but a weaker fit is demonstrated (r = 0.8625) when
It has been established from both experimental studies and considering all particle diameters tested. Alternatively, a very
molecular dynamics simulations that the width of the densely strong correlation (r = 0.9729) among all particle diameters is
packed layer (DPL) is no more than 1–2 nm [20,21]. Since attractive seen in Fig. 3(b), considering proportionality to the inverse square
forces dissipate normal to the particle surface, base molecules fur- root of particle diameter. Proportionality to the inverse square root
ther away from the interface migrate shorter distances. The DPL of particle diameter was originally considered because it describes
1148 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154

Table 1
Interface volume fractions required to fit measured nanofluid latent heat for all particle sizes tested (Vi,req/Vp), calculated with Eqs. (3) and (5). Required interface volume fractions
are compared to effective interface volume fractions consisting of a 2 nm thick densely packed layer (Vi,DPL/Vp). Interface volume fractions are normalized by particle volume
fraction (/i//p), and represented as ratios of interface volume to particle volume.

MWNT diameter [nm] MWNT length [lm] MWNT aspect ratio w required to fit hsl,nf [nm] Vi,req/Vp Vi,DPL/Vp
15.5 6.5 419.4 17.5 9.62 0.58
40.0 15.0 375.0 39.0 7.69 0.21
65.0 15.0 230.8 50.0 5.44 0.13
400.0 27.5 68.9 125.0 1.65 0.02

Fig. 3. Normalized nanofluid latent heat versus functions of particle diameter. (a) Plotted versus inverse particle diameter, representing interfacial liquid layering, with a
correlation coefficient, r = 0.8625. (b) Plotted versus inverse square root particle diameter, a proportionality of Brownian diffusion, with r = 0.9729.

the diameter-dependency of particle diffusion due to Brownian maximum when solely in the radial direction. The range of Brown-
motion. The strong correlation between this proportionality and ian sweep volumes can be expressed as a function of average
observed latent heat reduction suggests that Brownian motion Brownian diffusion length, assuming a 2 nm thick interface phase
may have a significant role as a reduction mechanism. width:
Another important parameter to note in Table 1 is the aspect " 2  2 #  2
ratio of the samples tested. Aspect ratio decreases from approxi- dp dp dp
pL þw  þ pk þ w 6 V BrS
mately 420 for the 15.5 nm diameter nanoparticles to 69 for the 2 2 2
400 nm nanoparticles. Consequently, these aspect ratios also show " 2  2 #
an inverse proportionality to the square root of particle diameter. dp dp
6 pL þw  þ kLðdp þ 2wÞ ð6Þ
The mutual proportionality of particle aspect ratio and observed 2 2
latent heat reduction implies not only a direct correlation, but also
For Brownian motion in the diffusive regime, average Brownian
that potential reduction mechanisms should be highly dependent
diffusion length, k [m], is inversely proportional to the square root
on aspect ratio.
of particle diameter,
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3.2. Brownian motion pffiffi 2kB T
k¼ t ð7Þ
3pldp
Brownian movement facilitates latent heat reduction by caus-
ing a disruption of base fluid molecular structure as nanofluids where l is dynamic viscosity, t is the time scale, kB is the Boltzmann
undergo phase change. The random movement of nanoparticles constant, and T is the absolute temperature. The time scale for
in the medium results in an effective sweep volume of weakened Brownian diffusion is of the same order of magnitude as momen-
bond structure that requires less energy to breakdown during tum relaxation time [23]. For a particle of mass m, momentum
phase change, and may account for the additional volume not con- relaxation time is given as:
tributing to latent heat. By calculating average Brownian diffusion mp
length, k, the sweep volume generated by a single particle in sus- sp ¼ ð8Þ
6pldp
pension can be estimated. The magnitude of this Brownian sweep
volume can be compared to the equivalent interface phase volume, Instead of approximating Brownian sweep volumes, required
Vi,req, required to fit measured nanofluid latent heat reduction – Brownian diffusion times to explain observed latent heat reduction
found using Eqs. (3) and (5). If both volumes are on the same order can be estimated and compared to respective momentum relaxa-
of magnitude, Brownian movement can theoretically account for tion times. Axial and radial Brownian sweep volumes were fit to
latent heat reduction not explained by liquid layering effects. The required interface phase volumes, Vi,req, to calculate required aver-
range of Brownian sweep volumes for cylindrical particles is at a age diffusion length (Eq. (6)). Required diffusion lengths were
minimum when diffusion is solely in the axial direction, and applied to Eq. (7) to calculate ranges of required diffusion times.
 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154
Fig. 4. Approximate time scales required for axial and radial Brownian sweep volumes of various diameter MWNTs to explain observed latent heat reduction, compared to
respective momentum relaxation time scales.
A comparison of the resulting time scales is shown in Fig. 4, dem-
onstrating that required diffusion time scales are two orders of
magnitude larger than the momentum relaxation time. Thus, the
time scale for Brownian motion is too slow to explain the addi-
tional reduction in nanofluid latent heat of fusion.
3.3. Particle clustering
Particle aggregation into high aspect ratio clusters has been
described as the key mechanism for nanofluid thermal conductiv-
ity enhancement [7,22,24]. Percolation effects due to direct contact
between aggregated particles explain thermal conductivity
enhancement beyond traditional Maxwell theory. The effective
volume of an aggregate cluster is larger than the volume of nano-
particles within the cluster, and exhibits higher thermal conductiv-
ity than the base fluid. Keblinski et al. state that even for densely
packed aggregates, roughly 25% of cluster volume is occupied by
base fluid filling the voids between particles. Aggregates are
formed as a result of inter-particle attraction due to van der Waals
forces. It is proposed that base fluid structure is strained as parti-
cles migrate towards each other. Similar to Brownian motion, the
clustering or movement of particles in the medium weakens bond
structure, leading to a reduction in latent heat. Strained base fluid
volume within aggregate clusters can account for the interface vol-
ume required to explain observed latent heat reduction.
As described by Prasher et al. [7,24], aggregates are character-
ized by their radius of gyration (Ra). Radius of gyration is defined
as the root mean square of the average radius from the cluster’s
center of mass. Prasher et al. define /int as the volume fraction of
particles within aggregate clusters, and /a as the volume fraction
of aggregates in the nanofluid. Hence, particle volume fraction
can be defined as
/p ¼ /int /a ð9Þ
For a completely dispersed nanofluid, /int = 1 and /p = /a, since
each aggregate is composed of a single particle. On the other hand,
/a = 1 and /int = /p for a completely aggregated nanofluid, since the
entire nanofluid volume is composed of a single aggregate cluster.
The volume fraction of particles within formed clusters, /int, is
given by Potanin and Russel [25]:
/int ¼ ð2Ra =dp Þ
df 3
ð10Þ
1149
where df is the fractal dimension of the aggregates, ranging from
1.75 to 2.5 [7]. Prasher et al. [7] assume df = 1.8, based on observa-
tions by Wang et al. [26] showing that nanofluids exhibit diffusion-
limited cluster–cluster aggregation (DLCCA). Low range fractal
dimensions represent a weak repulsive barrier, which is character-
istic of DLCCA.
To investigate particle clustering’s maximum potential contri-
bution to latent heat reduction, ratios of maximum effective clus-
ter volume to particle volume can be estimated, assuming a
completely aggregated nanofluid. Since cluster volumes include
the volume of particles within the cluster, maximum cluster vol-
ume ratios should be compared to the following ratio:
V iþp;req V i;req þ V p
¼ ð11Þ
Vp Vp
Eq. (11) is a more appropriate comparison than the required
interface volume fractions in Table 1, which do not include particle
volume. Since /a = 1 for a fully aggregated nanofluid, Eq. (9) can be
reduced to /int = /p. Thus, the maximum cluster radius of gyration
is given by:
ðRa Þmax ¼ ðdp =2Þð/p Þ1=ðdf 3Þ ð12Þ
where /int in Eq. (10) has been replaced with particle volume frac-
tion, /p. Assuming spherical clusters, calculated values of (Ra)max at
various particle volume fractions allow for the estimation of maxi-
mum cluster volumes, (Va)max.
As seen in Fig. 5, resulting maximum cluster volumes, normal-
ized by particle volume, are orders of magnitude larger than
required interface volume fractions (Eq. (11)). Required interface
volume ratios from Table 1 are normalized by particle volume;
and thus, are well-represented by a power function trend line of
the form Vi,req/Vp = C(dp)1/2. On the other hand, maximum cluster
volume ratios are shown for several particle volume fractions.
Cluster volume ratios demonstrate linear fit with respect to diam-
eter, as they describe a spherical to cylindrical volume ratio. Trend
line approximations show that smaller particle volume fractions
produce greater cluster volume fractions. Cluster volume fractions
are shown to increase with larger particle size, producing a greater
divergence from required interface volume fraction. Despite these
trends, larger particles at smaller volume fractions are less likely to
aggregate.
1150 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154
Fig. 5. Ratios of maximum aggregate volume to particle volume for various particle volume fractions, with respect to particle diameter. Maximum aggregate volume is
calculated assuming a spherical cluster of radius (Ra)max (Eq. (12)). Volume ratios are compared to interface volume fractions required to explain latent heat reduction,
normalized by particle volume fraction (Eq. (11)).
Although the effect of strain within aggregate clusters is not
considered in Fig. 5, the approximated scale of cluster volume frac-
tions suggest that clustering is the principal mechanism for nano-
fluid latent heat reduction. Exact cluster volume fractions required
to fit required interface volume fractions can also be calculated,
providing an estimate of the necessary degree of aggregation
within the nanofluid. Firstly, the effective radius of volume Vi+p,req
is calculated and substituted into Eq. (10), in place of the cluster
radius of gyration. Lastly, resulting values of /int are used in Eq.
(9) to calculate required cluster volume fraction. These values are
provided in Fig. 6, normalized by particle volume fraction, and
are also well-described by the inverse square root of particle diam-
eter. Fig. 6 shows that cluster volume fraction must be at least an
order of magnitude larger than particle volume fraction to account
for the latent heat reduction observed.
Fig. 6. Minimum cluster volume fraction to particle volume fraction ratios required
for particle clustering to explain observed latent heat reduction, with respect to
particle diameter. Approximated ratios scale proportionally with particle volume
fraction, and are well-described by a power function trend line of the form
/a,req//p = C(dp)0.5.
Ultimately, additional research efforts are needed to further
investigate particle clustering as a latent heat reduction mecha-
nism. Direct measurement of nanofluid thermal conductivity, for
example, will allow for the estimation of cluster volume fraction
using Prasher’s method [7,27]. Approximations of cluster volume
fraction can be compared to ratios provided in Fig. 6. In addition,
molecular dynamics simulation should be conducted to provide a
clearer understanding of the proposed reduction phenomena.
According to the strong correlation demonstrated in Fig. 3(b),
observed latent heat reduction is well-described by a dependency
on the inverse square root of particle diameter. This relationship
was the basis for considering Brownian motion as a reduction
mechanism. As shown in previous work [19], the volume ratios
Vi,req/Vp in Table 1, defined as required interface volume fraction
normalized by particle volume fraction, can be described by a
power function trend line of the form Vi,req/Vp = C(dp)1/2. Thus,
strained phase volume over particle volume may be considered
approximately proportional to the inverse square root of particle
diameter. By using this power function to calculate required inter-
face volume fraction, Eq. (5) may be used to approximate nanofluid
latent heat of fusion. Regardless of the underlying mechanisms for
further reduction in latent heat of fusion, the empirical relation
presented can be used to identify the validity of the nanofluid
approach for TES. Since the developed nanofluids are intended
for use in TES applications, it is important to consider the effects
of change in thermal properties on TES performance.
3.4. Numerical study of nanofluid PCM thermal storage performance
In our previous work [19], nanoparticle size is shown to have a
much greater impact on latent heat than thermal conductivity. In
short, a smaller particle diameter leads to a slight increase in ther-
mal conductivity and a significant reduction in latent heat, sug-
gesting that greater volumes of nanofluid PCM can be melted in
a given period of time by reducing particle size. However, reduc-
tion of latent heat will decrease the amount of energy stored per
unit mass. These counteracting effects complicate the quantitative
assessment of nanofluid energy storage performance.
A numerical model has been developed to determine the opti-
mum particle diameter and volume fraction to maximize the
 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154 1151

amount of heat transferred over a given period of time. In accor- The total enthalpy in solid and liquid state is defined as:
dance with our experimental findings, the numerical model incor- 8R
porates values for nanofluid latent heat of fusion that consider < T melt qcs dT; for T < T melt
T
additional volume not contributing to phase change. Specifically, E ¼ R Tinit R ð14Þ
: melt qc dT þ T
T melt qc l dT þ qhsl ;
final
s for T > T melt
the empirical relation for nanofluid latent heat given in Fig. 6 is T init

applied to Eq. (5) to provide approximate values for the model.

The numerical model presented in this paper simulates tran- where subscripts init and final denote initial and final states, and
sient nanofluid phase change in an annular storage container. Tmelt signifies melting temperature. In discretized form, solid and
PCM in the outer annulus is heated convectively by a working fluid liquid state enthalpies are defined as:
in the inner annulus, while the outside of the container is insulated 8  
>
< qcs T ij  T i ; for T ij < T melt
and cooled by the surrounding air. Conduction heat transfer within
the annulus is assumed to be axially symmetric, and thus, can be Eij ¼     ð15Þ
>
: qcs T ij  T i þ qcl T ij  T m þ qhsl ; for T ij > T melt
modeled as one-dimensional. Convection within the PCM is
neglected. The model incorporates a finite element approach,
A discretized equation for the enthalpy at inner nodes [j = 2 to
based on the enthalpy method [27]. The enthalpy method solves
j = (N  1)] is derived from the first law, considering the control
for stored internal energy at individual nodes at each time step,
volume (CV) in Fig. 7(b). Applying an energy balance, conduction
determines the phase of the material accordingly. Fig. 7 shows
into the CV is equal to the sum of internal energy and conduction
the configuration of nodes within the annulus, along with the con-
out of the CV. The discretized form of this energy balance is given
trol volumes used to derive discretized equations for both the
by:
inner nodes and boundary conditions. Nodes are spaced at Dr

steps, and range from ro (j = 1) at the heated interface to rN (j = N) q00cond r ð2pr j1=2 LÞ  q00cond rþDr ð2prjþ1=2 LÞ
at the container wall. !
h  i T iþ1  T i
All discretized equations are derived from the first law of ther- i 2 2 j j
¼ qcj pL r jþ1=2  r j1=2 ð16Þ
modynamics, neglecting work done due to PCM expansion during Dt
phase change.
  where subscripts r and j signify radial and nodal position, respec-
1 @ @T dT
kr ¼ qc ð13Þ tively, and superscript i is the iterative time step, initially at i = 1.
r @r @r dt
Since the surface area at each node increases moving outward from

Fig. 7. (a) Configuration of nodes within annular PCM storage container, at initial temperature TPCM,init, heated convectively by a working fluid at constant temperature T1,wf.
Nodes range from ro (j = 1) at the heated interface to rN (j = N) at the container wall, spaced apart by Dr steps. (b) Control volume defined to derive the discretized governing
equation for the enthalpy at inner nodes [j = 2 to j = (N  1)]. (c & d) Control volumes defined to derive discretized boundary conditions for the enthalpy at (c) the heated
interface (j = 1) (d) storage container wall (j = N).
1152 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154

" i #
the center of the annulus, the radius at each node must be 2kN1=2 r N1=2 Dt
considered. Eiþ1
1 ¼ Ei1 þ T iN1
Drðr2N  r2N1=2 Þ
The conduction terms in Eq. (16) may be represented by " !#
i
Fourier’s law for one-dimensional conduction: 2Dt kN1=2 r N1=2
 2 þ U o rN T iN
i
kj1=2 ð2pr j1=2 LÞðT j1  T j Þ
i
kjþ1=2 ð2pr jþ1=2 LÞðT j  T jþ1 Þ r N  r 2N1=2 Dr
þ " #
Dr Dr 2U o r N Dt
h
 i T iþ1  T i
! þ 2 Ti ð24Þ
j j r N  r 2N1=2 1;air
¼ qcij pL r 2jþ1=2  r 2j1=2 ð17Þ
Dt
Solved enthalpy values at each node and iterative time step
Distributing qc within the time derivative, the term on the right are used to solve for temperatures, liquid volume fraction k,
hand side can be expanded in order to combine qcT terms as dis- and thermal conductivity at the following time step. Table 2 pro-
cretized enthalpy terms, as defined in Eq. (15). vides a summary of the appropriate equations to estimate these
    properties, given different ranges of solved enthalpy values.
qcj T iþ1
j  qcj T i þ qcj T i  qcj T ij Liquid volume fraction is zero when the material is in solid state,
¼ Ejiþ1  Eij ð18Þ and 1 when in liquid state. When the temperature of the material
Dt
has reached melting point, the material enters a ‘‘partial mush’’
Squared radius terms at half nodes can also be expanded:
state. In this mush state, the liquid volume fraction is defined
   
Dr 2
 
Dr 2 as the fraction of stored enthalpy over the enthalpy required for
r 2jþ1=2  r 2j1=2 ¼ r j þ  rj  ¼ 2r j Dr ð19Þ the material to fully melt. Therefore, liquid volume fraction is
2 2
an indicator of the material’s state, and its progress towards melt-
and shown to reduce to the simple expression, 2rjDr. By incorporat- ing during phase change. Equations for temperature were derived
ing the simplified expressions of Eqs. (18) and (19) into Eq. (17), from Eq. (15). Thermal conductivity in the particle mush state is
dividing out by pL and 2rjDr, and rearranging terms, the final form taken from Alexiades and Solomon [27], signifying a ‘‘sharp front’’
of the governing discretized equation for the enthalpy at inner melting interface, with layers of solid and liquid in a serial
nodes is given as: arrangement.
" # Model parameters were chosen by considering a nanofluid PCM
Dt  
i
Eiþ1
j ¼ Eij þ 2
kj1=2 r j1=2 T ij1 of CNT particles dispersed in paraffin wax. The physical and ther-
rj ðDrÞ mal properties of multi-walled carbon nanotubes and paraffin
" #
Dt   wax are shown in Table 3. These values are used to approximate
i i
 2
kjþ1=2 r jþ1=2 þ kj1=2 r j1=2 T ij effective nanofluid thermal conductivity using Nan et al.’s model.
r j ðDrÞ
" # An empirical relation of aspect ratio to particle diameter, taken
Dt  i  from the nanoparticles tested, was applied for other particle sizes
þ k r
jþ1=2 jþ1=2 T ijþ1 ð20Þ modeled.
r j ðDrÞ2
Initially, the temperature of the wax is assumed to be uniform
Discretized enthalpy equations at the heated interface j = 1 and throughout the container, at 293 K. The temperature of the heat
container wall j = N are derived from boundary conditions match- source (the working fluid) is 358 K, with an overall heat transfer
ing conduction and convection. An energy balance is applied to coefficient (Ui) of 3000 W/m2 K, and the surrounding air tempera-
each of the CVs in Fig. 7(c) and (d). Starting with Fig. 7(c), the ture is 293 K, with an overall heat transfer coefficient (Uo) of
appropriate energy balance is: 1 W/m2 K. The inner and outer radii of the annulus are 1 and
  5 cm, respectively.
  ki1þ1=2 ð2pr 1þ1=2 LÞ T i1  T i2 Using the above parameters, a storage container of pure paraffin
U i ð2pr1 LÞ T 1;wf  T i1 
Dr undergoing an 8 h charge period was modeled at 0.1 s time inter-
2   3
vals. The temperature distribution along the radius of the cylinder
pL r21þ1=2  r21 T nþ1
1  T n
1
¼ qc 1 4 5 ð21Þ at 2, 4, 6, and 8 h is shown in Fig. 8(a). The horizontal line on the
Dt plot represents the melting temperature of pure paraffin (326 K).
A distinct change in the temperature profile is visible at the melt-
By using the same methods of simplification in Eqs. (18) and ing interface throughout the charge cycle. At 8 h, the wax in the
(19), Eq. (21) can be expressed in its final discretized form: container is almost completely melted. Therefore, 8 h was chosen
" i
!# as the benchmark charge time to compare nanofluid PCM storage
2 Dt k1þ1=2 r1þ1=2
Eiþ1
1 Ei1
¼  2 U i r1 þ T i1 performance. The total enthalpy stored after 8 h is calculated
r1þ1=2  r 21 Dr through numerical integration of the nodal enthalpy values along
2 3 " #
i the radius of the annulus. Total enthalpy was calculated for nano-
2k1þ1=2 r 1þ1=2 Dt i 2U i r 1 Dt
þ 4  5
 T2 þ 2 Ti ð22Þ fluid PCMs with particle diameters ranging from 10 nm to 1 lm, at
Dr r 21þ1=2  r21 r 1þ1=2  r 21 1;wf 0% to 2% volume fractions. As shown in Fig. 8(b), total stored
enthalpy values are normalized relative to pure paraffin, and plot-
To derive the discretized boundary condition at the container ted with respect to particle volume fraction.
wall, the energy balance for the control volume in Fig. 7(d) is given Fig. 8(a) shows the temperature profile and progression of the
as: melting front along the radius of the container. The horizontal line
  represents the melting temperature of pure paraffin. Fig. 8(b) com-
i
kN1=2 ð2prN1=2 LÞ T iN1  T iN  
pares the energy stored in the nanofluid system at the end of 8 h
 U o ð2pr N LÞ T iN  T 1;air
Dr for different particle concentrations and sizes. The results show
  
qcN pL r2N  r2N1=2 T Niþ1  T iN that the addition of highly conductive, micro-scale particles
¼ ð23Þ increases storage capacity relative to pure paraffin. However, for
Dt nanoscale inclusions, the enhancement in thermal conductivity is
and the final discretized equation takes the form: less significant than the reduction in latent heat of fusion, and
 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154 1153

Table 2
Equations to calculate temperature, liquid volume fraction k, and thermal conductivity, based on ranges of nodal enthalpy values. The three ranges of enthalpies signify material
in solid, partial mush, and liquid state.

Solver condition Ti+1

j ki+1
j ki+1
j

06 Eiþ1 < qcs ðT melt  T init Þ Eiþ1

j 0 ks
j T init þ qcs
Tmelt  1
qcs ðT melt  T init Þ 6 Eiþ1
j < qcs ðT melt  T init Þ þ qhsl Eiþ1
j
qcs ðT melt T init Þ kiþ1
j
1kiþ1
qhsl kl T melt
þ ks T j
melt

qcs ðT melt  T init Þ þ qhsl 6 Eiþ1 Eiþ1

j
qðhsl cs T init cl T melt Þ 1 kl
j
qðcs þcl Þ

Table 3
Physical and thermal properties of multi-walled carbon nanotubes and paraffin wax
applied to the numerical model of a nanofluid PCM storage container undergoing a
charge cycle.
Nanoparticle: Multi-walled carbon nanotubes
Thermal conductivity [28]
Kapitza resistance
Bulk medium: Paraffin wax
Thermal conductivity [3]
Density [3]
Specific heat capacity
Latent heat of fusion
Melting temperature
energy storage capacity is decreased over the PCM charge period.
The reduction in storage performance is more severe as particle
size decreases. With 10 nm particles, the relative energy stored is
less than that of pure paraffin. These results suggest that the nano-
fluid approach is not an effective method for increasing PCM ther-
mal storage performance. Although larger nanofluid particle
diameters exhibit enhanced storage performance, the benefits
may be negligible when considering cost and manufacturability,
increased viscosity, and long term suspension stability. However,
since the model parameters are dependent on material properties
such as aspect ratio and interfacial resistance, storage performance
may differ with the nanofluid materials selected. Additional stud-
ies are needed to fully understand the mechanisms of latent heat
reduction.
4. Conclusions
The latent heat of fusion of paraffin-based nanofluids containing
multi-walled carbon nanotubes (MWNTs) has been investigated,
utilizing differential scanning calorimetry (DSC). The latent heat
21 W/m K of fusion of nanofluid samples containing various diameter
1  108 Km2/W
multi-walled nanotubes was observed to reduce below theoretical
expectations. The rate of nanofluid latent heat reduction, with
0.21 W/m K (solid)
respect to particle volume fraction, was shown to increase in mag-
0.12 W/m K (liquid)
900 kg/m3 (solid) nitude with smaller diameter nanoparticles in suspension. Interfa-
780 kg/m3 (liquid) cial phenomena such as interfacial liquid layering, Brownian
1888 J/kg K (solid) motion, and particle clustering were proposed as reduction mech-
2272 J/kg K (liquid) anisms and explored with respect to measured results. It is con-
1.8  105 J/kg
326 K
cluded that Brownian motion and liquid layering are incapable of
solely accounting for the degree of latent heat reduction observed.
Particle clustering is demonstrated as a potential mechanism for
latent heat reduction, but further investigation is needed.
The effect of latent heat reduction on nanofluid PCM thermal
energy storage performance was also presented. Defining storage
performance as the amount of heat stored or extracted over a given
duration, our previous work showed that nanofluids of smaller par-
ticle diameter can theoretically improve storage performance. There
is a limit, however, to the practicality of reduced specific storage
capacity. Although more heat is transferred at a faster rate, larger
volumes of phase change material (PCM) are needed to account
for the reduced amount of energy stored per unit mass. A one-
dimensional finite element model of nanofluid phase change in an
annular storage container was presented as a predictive tool to
assess diameter-dependent nanofluid storage performance. Based
on the enthalpy method, discretized equations and boundary
conditions were derived from the first law of thermodynamics to
calculate internal energy at iterative time steps. The development

Fig. 8. (a) Temperature distribution along the radius of the annulus at different time intervals. (b) Total enthalpy saved after an 8 h charge cycle, normalized relative to pure
paraffin.
1154 A. Zabalegui et al. / International Journal of Heat and Mass Transfer 78 (2014) 1145–1154
of a governing tridiagonal matrix is shown, including equations to
calculate temperature and material phase at each node within the
finite element mesh. The numerical model shows that while the
addition of micro-scale particle increases thermal storage perfor-
mance, nanoscale inclusions are shown to degrade storage perfor-
mance relative to pure paraffin. With smaller particles, the
beneficial effect of thermal conductivity enhancement is negated
by significant reduction in latent heat of fusion, resulting in less
energy stored over a given charge period. These results suggest that
the nanofluid approach is not an effective method for improving
PCM thermal storage performance, and may not be justifiable for
thermal storage applications. However, these results are confined
to the material properties of the system observed, and storage per-
formance may vary depending on the nanofluid materials used.
Conflict of interest
None declared.
Acknowledgments
This work was supported by the School of Engineering,
Sustainability Initiative Grant, and the Center for Science,
Technology, and Society at Santa Clara University. The authors
would like to acknowledge Bernadette Tong for her valuable
comments during revisions.
References
[1] W. Wu, H. Bostanci, L.C. Chow, Y. Hong, C.M. Wang, M. Su, J.P. Kizito, Heat
transfer enhancement of PAO in microchannel heat exchanger using nano-
encapsulated phase change indium particles, Int. J. Heat Mass Transfer 58 (1–
2) (2013) 348–355.
[2] H. Zhang, X. Wang, D. Wu, Silica encapsulation of n-octadecane via sol–gel
process: a novel microencapsulated phase-change material with enhanced
thermal conductivity and performance, J. Colloid Interface Sci. 343 (1) (2010)
246–255.
[3] J. Fukai, Y. Hamada, Y. Morozumi, O. Miyatake, Improvement of thermal
characteristics of latent heat thermal energy storage units using carbon-fiber
brushes: experiments and modeling, Int. J. Heat Mass Transfer 46 (23) (2003)
4513–4525.
[4] R. Velraj, R.V. Seeniraj, B. Hafner, C. Faber, K. Schwarzer, Heat transfer
enhancement in a latent heat storage system, Sol. Energy 65 (3) (1999) 171–
180.
[5] P. Keblinski, J. Eastman, D. Cahill, Nanofluids for thermal transport, Mater.
Today 8 (6) (2005) 36–44.
[6] C. Kleinstreuer, Y. Feng, Experimental and theoretical studies of nanofluid
thermal conductivity enhancement: a review, Nanoscale Res. Lett. 6 (1) (2011)
229.
[7] R. Prasher, P.E. Phelan, P. Bhattacharya, Effect of aggregation kinetics on the
thermal conductivity of nanoscale colloidal solutions (nanofluid), Nano Lett. 6
(7) (2006) 1529–1534.
[8] J.W. Gao, R.T. Zheng, H. Ohtani, G. Chen, Experimental investigation of heat
conduction mechanisms in nanofluids. Clue on clustering, Nano Lett. 9 (12)
(2009) 4128–4132.
[9] C.-W. Nan, R. Birringer, D.R. Clarke, H. Gleiter, Effective thermal conductivity of
particulate composites with interfacial thermal resistance, J. Appl. Phys. 81
(10) (1997) 6692.
[10] C.-W. Nan, G. Liu, Y. Lin, M. Li, Interface effect on thermal conductivity of
carbon nanotube composites, Appl. Phys. Lett. 85 (16) (2004) 3549.
[11] S.-Q. Zhou, R. Ni, Measurement of the specific heat capacity of water-based
Al2O3 nanofluid, Appl. Phys. Lett. 92 (9) (2008) 093123.
[12] D. Shin, D. Banerjee, Enhancement of specific heat capacity of high-
temperature silica-nanofluids synthesized in alkali chloride salt eutectics for
solar thermal-energy storage applications, Int. J. Heat Mass Transfer 54 (5–6)
(2011) 1064–1070.
[13] D. Shin, D. Banerjee, Enhanced specific heat of silica nanofluid, J. Heat Transfer
133 (2) (2011) 024501.
[14] L. Wang, Z. Tan, S. Meng, D. Liang, G. Li, Enhancement of molar heat capacity of
nanostructured Al2O3, J. Nanopart. Res. 3 (2001) 483–487.
[15] S. Wu, D.-S. Zhu, X. Zhang, J. Huang, Preparation and melting/freezing
characteristics of Cu/paraffin nanofluid as phase-change material (pcm),
Energy Fuels 24 (2010) 1894–1898.
[16] X.-j. Wang, D.-s. Zhu, S. yang, Investigation of pH and SDBS on enhancement of
thermal conductivity in nanofluids, Chem. Phys. Lett. 470 (1-3) (2009) 107–
111.
[17] J.-L. Zeng, F.-R. Zhu, S.-B. Yu, L. Zhu, Z. Cao, L.-X. Sun, G.-R. Deng, W.-P. Yan, L.
Zhang, Effects of copper nanowires on the properties of an organic phase
change material, Sol. Energy Mater. Sol. Cells 105 (2012) 174–178.
[18] C.J. Ho, J.Y. Gao, Preparation and thermophysical properties of nanoparticle-in-
paraffin emulsion as phase change material, Int. Commun. Heat Mass Transfer
36 (2009) 467–470.
[19] A. Zabalegui, B. Tong, H. Lee, Investigation of thermal properties in nanofluids
for thermal energy storage applications, in: ASME 2013 Summer Heat Transfer
Conference, ASME, Minneapolis, MN, 2013.
[20] C.J. Yu, A.G. Richter, J. Kmetko, S.W. Dugan, A. Datta, P. Dutta, Structure of
interfacial liquids: X-ray scattering studies, Phys. Rev. E 63 (2) (2001).
[21] L. Xue, P. Keblinski, S.R. Phillpot, S.U.S. Choi, J.A. Eastman, Effect of liquid
layering at the liquid–solid interface on thermal transport, Int. J. Heat Mass
Transfer 47 (19–20) (2004) 4277–4284.
[22] P. Keblinski, R. Prasher, J. Eapen, Thermal conductance of nanofluids: is the
controversy over?, J Nanopart. Res. 10 (7) (2008) 1089–1097.
[23] D.S. Lemons, Paul Langevin’s 1908 paper ‘‘on the theory of Brownian motion’’
[‘‘Sur la théorie du mouvement brownien’’, C. R. Acad. Sci. (Paris) vol. 146
(1908) pp. 530–533], Am. J. Phys. 65 (11) (1997) 1079.
[24] R. Prasher, W. Evans, P. Meakin, J. Fish, P. Phelan, P. Keblinski, Effect of
aggregation on thermal conduction in colloidal nanofluids, Appl. Phys. Lett. 89
(14) (2006) 143119.
[25] A. Potanin, W. Russel, Fractal model of consolidation of weakly aggregated
colloidal dispersions, Phys. Rev. E 53 (4) (1996) 3702–3709.
[26] B.-X. Wang, L.-P. Zhou, X.-F. Peng, A fractal model for predicting the effective
thermal conductivity of liquid with suspension of nanoparticles, Int. J. Heat
Mass Transfer 46 (14) (2003) 2665–2672.
[27] V. Alexiades, A.D. Solomon, Mathematical Modeling of Melting and Freezing
Processes, Hemisphere, Washington, DC, 1993.
[28 D. Yang, Q. Zhang, G. Chen, S. Yoon, J. Ahn, S. Wang, Q. Zhou, Q. Wang, J. Li,
Thermal conductivity of multiwalled carbon nanotubes, Phys Rev B 66 (16)
(2002).