Applied Thermal Engineering 54 (2013) 65e77

Contents lists available at SciVerse ScienceDirect

Applied Thermal Engineering

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

Thermal optimization of PCM based pin fin heat sinks:

An experimental study
Rajesh Baby, C. Balaji*
Heat Transfer and Thermal Power Laboratory, Department of Mechanical Engineering, Indian Institute of Technology Madras, Chennai 600036, India

h i g h l i g h t s

< Experimental investigations on phase change material based pin fin heat sinks.
< PCM materials: n-eicosane, paraffin wax; Heat sink material e aluminium.
< Baseline comparisons with a heat sink filled with PCM, without any fin.
< ANN e GA algorithm to maximize operating time for the n-eicosane heat sink.
< The resulting optimum valid even for paraffin wax based PCM too.

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

Article history: This paper reports the results of an experimental investigation carried out to characterize the thermal
Received 22 August 2012 performance of different configurations of phase change material (PCM) based pin fin heat sinks. Paraffin
Accepted 29 October 2012 wax and n-eicosane are used as the PCMs. Aluminium is used to make the heat sinks and the volume
Available online 31 January 2013
fraction is varied by changing the number of pin fins. Baseline comparisons are done with a heat sink
filled with PCM, without any fin. The effect of PCM volume fraction on the heat transfer performance is
Keywords:
also studied. The results showed that there exists sufficient scope to optimize the thermal design of the
Phase change materials (PCMs)
heat sink. An Artificial Neural Network e Genetic Algorithm hybrid algorithm is then developed to
Electronic equipment
n-Eicosane
determine the optimum configuration of the pin fin heat sink that maximizes the operating time for the
Artificial neural networks n-eicosane based heat sink. The resulting optima was found to be valid even for the paraffin wax based
Genetic algorithm PCM.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction propelling the development of new and innovative thermal man-

agement solutions.
Effective thermal management holds the key to the rapid Walsh and Grimes [1] analyzed the potential of low profile in-
growth and the continued miniaturization of microelectronics. tegrated fan and heat sink solutions in the thermal management of
Optimal thermal control is critical in the design and operation of electronics. Their work stressed the need for designing an inte-
electronic equipment. In several applications like laptops, cellular grated system for thermal control rather than the design of a fan or
phones, digital cameras and control systems in missiles, the ther- heat sink in isolation. Ng et al. [2] proposed a figure of merit and
mal management solution needs to be reliable, safe, inexpensive, a universal performance chart for CPU cooling devices. Data points
light, durable and more importantly energy efficient. The reliability taken from thermosyphons, heat pipes, liquid cooling, and fin-fan
of an equipment is a quantifier of the expected frequency of failure heat sinks were considered for proposing the performance chart.
as a function of time. Out of the many factors affecting the reli- Jaworski [3] introduced a kind of pin fin radiator, in which the pipes
ability of electronic equipment, temperature is one of the most are filled with the PCM. The results showed significant benefits
important. The need to carry out diverse functions and the de- associated with the use of PCM in the structure of heat spreaders.
mands for higher performance of the electronic equipment are Yu and Joshi [4] developed a porous medium model to simulate pin
fin heat sinks in vertical enclosures. Of the various heat transfer
modes, they found that the contribution of radiation is very sig-
* Corresponding author. Tel.: þ91 44 2257 4689; fax: þ91 44 2257 4652. nificant and vital even for lower power levels. By modelling the
E-mail address: [email protected] (C. Balaji). fluid flow and the heat transfer characteristics in a copper heat sink,

1359-4311/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.applthermaleng.2012.10.056
66 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77

Nomenclature 
Tset asymptotic set point temperature,  C
TN ambient temperature,  C
Cp specific heat of PCM, J/kg K
Fo Fourier number, Fo ¼ aPCM tset =L2 Greek symbols
H height of the heat sink, m aPCM thermal diffusivity of the PCM, m2/s
kPCM thermal conductivity of the PCM, W/m K Q modified dimensionless temperature
L length of the heat sink, m Q dimensionless temperature
Q heat input, W rPCM density of the PCM, kg/m3
tset time to reach set point temperature, s f volume fraction of the PCM
tANN time to reach set point temperature predicted by the j volume fraction of the TCE
artificial neural network, s
texpt device operating time from the experiments, s Subscripts
Tbase average heat sink base temperature,  C ANN artificial neural network
Tref reference temperature,  C GA genetic algorithm
Tset set point temperature,  C TCE thermal conductivity enhancer

Mahmoudi [5] showed that heat sink made of copper with louvered effective in stretching the duration of operation of the electronic
fins has an optimum design configuration. Li et al. [6] analyzed the equipment in comparison with plate fin heat sinks. This study takes
thermal and fluid behaviour of impingement jet cooling in plate fin this forward and evaluates the performance of pin fin based heat
heat sinks. Schmidt [7] listed the following design considerations in sinks, by way of conducting exhaustive experimental studies over
selecting a cooling method for commercial computers: perfor- a range of the pertinent parameters.
mance, leakage current, functionality, reliability, cost, space and Zhang et al. [21] conducted extensive studies on preparation of
power required for operating the cooling hardware. From these paraffin/expanded graphite composite phase change material and
design considerations, it was concluded that air cooling capabilities its thermal energy storage properties and observed a significant
may be nearing an end. Today conventional cooling methods are reduction in the composite PCM charging duration as compared to
being replaced by the advent of new and energy efficient tech- paraffin alone. Experimental investigations were done by Zhao
niques. In cases where the application is transient [8e10] or et al. [22] with paraffin wax RT 58 as the PCM under constant heat
intermittent, phase change materials based heat sinks offer con- flux conditions. They also conducted experiments for solidification
siderable promise. By considering the specific case of Samsung SPH under natural and forced convection modes of heat transfer and
E2500 mobile phone, Luo et al. [11] carried out a system level concluded that the presence of metal foams can make the sample
thermal analysis. They also suggested the viability of using PCM solidification much faster than that with pure PCM samples. So-
based heat sinks in similar applications. lidification time reduced to more than half with the use of metal
A phase change material is selected mainly based on the phase forms. Sanusi et al. [23] used paraffin PCM embedded with graphite
change temperature and the enthalpy of fusion. The melting tem- nanofibers for energy storage and solidification. They studied the
perature of the PCM is kept below the maximum operating tem- performance of graphite nanofibers impregnated into the PCM for
perature of the equipment. Fins made of high thermal conductivity different aspect ratios and power densities and found a significant
material like copper, aluminium or porous matrix made up of these reduction in the solidification time in the case of PCM embedded
materials are known to be thermal conductivity enhancers (TCEs) with graphite nanofibers. Fok et al. [24] conducted experimental
and are used to compensate for the very low thermal conductivity investigations on PCM based finned heat sinks for the cooling of
of the PCM when the latter is used alone. Alawadhi and Amon [12] portable electronic devices.
carried out experimental and numerical studies at constant and Different optimization techniques have been explored in liter-
varying power levels on PCM based heat sinks for portable elec- ature to arrive at optimal designs of PCM based heat sinks. Yusup
tronic devices. Kandasamy et al. [13] experimentally investigated et al. [25] conducted an excellent review of the use of evolutionary
the application of phase change materials in thermal management optimization techniques with special focus on machining param-
of electronic equipment. The heat flux levels used in this study eters mainly done between 2007 and 2011. Gosselin et al. [26]
were 2.4, 3.6 and 4.8 kW/m2. Saha et al. [14] employed two values present a critical review of the use of Genetic Algorithms (GA) in
of the heat fluxes 3.09 and 6.17 kW/m2 for determining the opti- heat transfer problems. Ermis et al. [27] carried out a heat transfer
mum distribution of fins in PCM based heat sinks. analysis of the phase change process using artificial neural net-
Faraji and Qarnia [15] studied the effect of PCM in the cooling of works (ANN). They considered a finned tube thermal energy stor-
protruding electronic components. They used n-eicosane as the age system for experimental and numerical investigations. They
PCM. Akhilesh et al. [16] conducted numerical investigations to found that the ANN results to be more accurate in predicting the
improve the heat transfer performance of PCM based heat sinks. total amount of heat stored than the numerical model results.
They developed a thermal design procedure for PCM based heat Optimization based on either ANN or GA or a combination of the
sinks with fins made of high thermal conductivity material. Nayak two is becoming increasingly popular and is finding applications in
et al. [17] considered a porous matrix, plate and pin fins as the TCEs many areas like environmental sciences and biotechnology [28e
for the numerical investigation. Xiang et al. [18] developed a PCM 30], fuel production [31,32], nuclear engineering [33] and so on.
based heat sink for cooling high power light emitting diode. Ye et al. Madadi and Balaji [34], used this technique for optimizing the
[19] analyzed the fluid flow patterns and the heat transfer in a plate location of multiple discrete heat sources in a ventilated cavity.
fin unit. Paraffin was used as the PCM. Baby and Balaji [20] com- The preceding review shows that eventhough a considerable
pared the heat transfer performance of a plate fin heat sink with amount of literature is available on PCM based heat sinks, literature
that of a pin fin heat sink, both filled with n-eicosane by conducting on the optimization of the heat sinks is scarce. In view of this, in the
experimental investigations. They found that the pin fin was highly present study a hybrid optimization technique, completely based
 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77 67

Fig. 1. Isometric views of various heat sink assemblies used in the present study.

on experimental results that combines ANN and GA for maximizing focuses on the effect of varying power levels, volume fraction of the
the time taken by a PCM based fin heat sink to reach a set point PCM and volume fraction of the TCE on the heat transfer perfor-
temperature is attempted. The ANN driven by the experimental mance of pin fin heat sinks and proposes a strategy to design an
data is a key feature in the present study. The present study mainly optimized heat sink using the ANNeGA route.

Table 1
Properties of materials employed in the present study.

Material Thermal conductivity Specific heat Latent Melting Density

(W/m K) (kJ/kg K) heat (kJ/kg) point ( C) (kg/m3)
n-Eicosane 0.39 (solid) 1.9 (solid) 237.4 36.5 810 (solid)
0.157 (liquid) 2.2 (liquid) 770 (liquid)
Paraffin wax 0.21 (solid) 2.89 173.4 53e57 870 (solid)
0.12 (liquid) 840 (liquid)
Aluminium 202.4 0.87 e 660.4 2719
Cork 0.045 0.350 e e 120
68 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77

This weight penalty makes copper not quite suitable for the type of
applications, especially the thermal management of portable elec-
tronic equipment discussed in this paper. The phase change ma-
terials used are n-eicosane and paraffin wax. The details regarding
the properties of the materials used in the present investigation are
given in Table 1. A photograph of the pin fin heat sink assembly is
given in Fig. 2. The horizontality of the heat sink assembly is
ensured with the help of a spirit level and four adjusting screws as
shown in the photograph. It is seen that the top surface is covered
with a perspex sheet. The pin fins have a size 2  2 mm2, are of
20 mm high and are made by Electrical Discharge Machining
(EDM). A 60  42 mm2 plate heater of 2 mm thickness, made up of
a standard coil-type nichrome wire wound over a mica sheet is
used to mimic heat generation in electronic chips. In order to avoid
contact resistance, a thermally conducting paste is employed at the
recess provided in the heat sink. The power levels used in the ex-
periments vary from 4 to 10 W in steps of 0.5 W. For a power level of
4 W, the heat flux is 1.587 kW/m2, and for 10 W, the heat flux is
3.968 kW/m2. A total of thirteen calibrated K - type thermocouples
are used at different locations of the heat sink to map the tem-
perature. AralditeÔ epoxy is used to fix the thermocouples in po-
Fig. 2. A photograph of the heat sink assembly. sition. The thermocouple locations are detailed in Fig. 3. In this
figure, T1 to T10 denote the thermocouples one to ten. Thermo-
couples H1 and H2 are kept at the heat sink base and are used to
2. Experimental set up measure the base temperature.
The thermocouples are connected to a PC e based data acquis-
Experimental investigations were done with three different pin ition unit. The power input to the plate heater is supplied from an
fin heat sink geometries with 33, 72 and 120 pin fins, filled with independently controlled DC power unit which has a voltage range
PCM, and giving rise to volume fractions of 4%, 9% and 15% of 0e30 V and a current range of 0e2 A. A pictorial representation
respectively. Baseline comparisons are done with a heat sink filled of the experimental set up is given in Fig. 4. The uncertainties in the
with PCM, but without any fin. The test section considered in all voltage and the current measurements are 0.1 V and 0.01 A
cases is an 80  62 mm2 base, with a TCE height of 25 mm. Fig. 1 respectively. The uncertainty in the measured temperature was
shows the isometric views of no fin, 33 pin fin, 72 pin fin and 120 found to be 0.2  C. The uncertainty in the power measurement is
pin fin heat sink configurations respectively. All the sides of the obtained using the procedure given in Ref. [35] and is found to be
heat sink, except the top, are insulated with cork. The aluminium 3.2%. To ensure repeatability, the experiments are carried out
pin fins act as the thermal conductivity enhancer in the present several times and it was seen that identical temperatureetime
case. Low density, corrosion resistance and high thermal conduc- distributions are obtained with a maximum deviation of 0.8  C in
tivity make aluminium the material of choice for the TCE. Metal the temperature excess, after factoring in the variation in ambient
joining process like soldering, brazing and welding can be easily
done with aluminium. Eventhough the thermal conductivity of
copper is double than that of aluminium, copper has a much higher
density, which is well over three times the density of aluminium.

Fig. 3. Schematic diagram showing the thermocouple positions. Fig. 4. A pictorial representation of the experimental set up.
 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77 69

temperature, when experiments are conducted at different times Table 2

and/or different days. For a fuller discussion on the fabrication of Results of the neuron independence study.

the pin fin heat sink, the heat sink assembly, characterization of SI no No of hidden MRE MSE R2
the PCM and TCE see Ref. [20]. At this juncture, it is instructive to neurons
mention that there are critical differences between this study and 1 1 0.08797 0.00023 0.92486
[20]. These are (i) In Ref. [20] only n-eicosane was considered. Here 2 2 0.04833 0.00007 0.97802
3 5 0.04615 0.00005 0.98244
both paraffin and n-eicosane are considered, (ii) thermal perfor-
4 7 0.03944 0.00005 0.98384
mance of heat sinks with same TCE volume fraction of the pin fin 5 8 0.03987 0.00006 0.98059
and plate fin geometries were compared in Ref. [20], but here heat 6 9 0.04046 0.00006 0.98027
sinks with pin fin geometry and different volume fractions of the 7 10 0.03817 0.00005 0.98324
TCE are considered, as the pin fin geometry was found to give su- 8 14 0.03651 0.00006 0.98158
9 15 0.03562 0.00005 0.98430
perior heat transfer performance when compared with plate fin 10 16 0.03272 0.00005 0.98309
geometry [20], (iii) volume fraction of the PCM in Ref. [20] was fixed 11 19 0.04096 0.00005 0.98448
at f ¼ 1, while in this study it is a variable. In view of the above, the 12 23 0.3852 0.00005 0.98438
present study represents a much more comprehensive exploration 13 27 0.03696 0.00005 0.98256
14 29 0.04072 0.00009 0.97195
of the parameter space.
15 40 0.03588 0.00005 0.98548
16 50 0.04017 0.00007 0.97835
3. Definitions used in the present study

The dimensionless time is given by the Fourier number ðFoÞ parameters which affect the heat transfer characteristics of the pin
defined as Fo ¼ aPCM tset =L2 , where aPCM is the thermal diffusivity fin heat sink, it is intuitively seen that a functional relationship
of the PCM, m2/s, L is the length of the heat sink, m and tset is the exists between Q, f, j and Fo. This can be generally represented as
time required to reach the set point temperature. The modified Fo ¼ f ðQ; f; 1 þ jÞ. Once the function form ‘f ’ is known, we can
dimensionless temperature, Q is defined as follows determine the time required for reaching the set point tempera-
ture, tset . In an ANN model, though an expression for ‘f’ is not
Q ¼ ðTset  TN Þ=Tref (1) explicitly developed, the network is so developed that it returns
a unique value for Fo for every combination of Q, f and ð1 þ jÞ. The
Levenberg Marquardt back propagation is used for training the
Tref ¼ Q =ðHkPCM Þ (2)
neural network. An artificial neural network typically has three
In the above equation, Tset is the set point temperature (say, 50 or layers (i) input layer (ii) hidden layer and (iii) output layer. The
60  C), which is measured at the heat sink base. The volume frac- number of hidden layers and the number of neurons in each hidden
tion of the PCM, f is given by the ratio of the volume of the PCM to layer are invariably fixed on the basis of detailed neuron inde-
the difference between the total empty volume of the heat sink and pendence studies. Preliminary investigations showed that an
the volume occupied by the fins. The volume fraction of the TCE is increase in the number of hidden layers did not affect the perfor-
denoted by, j which is the ratio of the volume of the fins to the total mance of the network significantly and in view of this all studies
volume of the heat sink. were carried out for networks with one hidden layer. The details of
The dimensionless temperature, Q can be defined as the ratio of the neuron independence study are given in Table 2. The perfor-
difference between the asymptotic set point ½Tset  , taken to be
mance indices chosen to arrive at these results are the Mean

105 C in the present study (maximum temperature recorded in the
experiments corresponding to a power level of 10 W) and the heat
sink base temperature to the difference in the set point and the
ambient temperature. It is mathematically represented as


 
Q ¼ Tset

 Tbase Tset  TN (3)

where, Tbase is the average heat sink base temperature

Finally, the enhancement ratio is defined as the ratio of the time
taken by the heat sink with fins to reach a set point temperature to
that taken by a heat sink without fins.

4. Artificial neural network

Artificial neural network (ANN) is a simplified model of the

structure of a biological neural network. Interconnected processing
units are the heart of the neural network architecture. Generally,
the processing unit consists of a summing part and an output part.
When the summing part receives different input values, it assigns
weights to each value and computes a weighted sum. The weighted
sum is called the activation value. From this activation value, the
output part produces a signal. The determination of whether
a weight is positive or negative is decided by the sign of the weight
for each input. The input and output can be deterministic, sto-
chastic or fuzzy [36].
A back propagation feed forward neural network has been
considered in the present study. By carefully examining the Fig. 5. Neuron architecture employed in this study.
70 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77

Fig. 6. Melting phase of the heat sink with 72 pin fin, filled with different PCMs. Fig. 8. Spatial variation of temperature within the PCM for an n-eicosane heat sink at
a power level of 6 W.

Relative Error (MRE), Mean Square Error (MSE) and Absolute frac-
tion of variance (R2), The expressions for the above are, final network used in the present study with three input variables
  and one output variable. The output variable is the time to reach
n 
1X tANN  texpt  a set point temperature in dimensionless form.
MRE ¼ (4)
n i¼1 texpt
5. Genetic algorithms
1X n
2
MSE ¼ t  texpt (5)
n i ¼ 1 ANN Genetic algorithms (GAs) are procedures based on mechanics of
2 3 natural selection and genetics, developed by Holland [37]. GAs are
6Pn
t 2 7 particularly suited to handle complex optimization problems. The
2 6 i ¼ 1 ANN  texpt 7 major difference between the GA and the traditional optimization
R ¼ 16 Pn
2 7 (6)
4 texpt 5 techniques as described by Goldberg [38] is that (i) GA work with
i¼1
a coding of the set of parameters and not directly the parameters.
(ii) GA conducts search from a population of points and does not
It can be seen from Table 2, that for a network with 16 hidden
work with a solution at every iteration. (iii) GA uses just the in-
neurons, the MRE is lowest at 0.03272 and R2 for this network is
formation on the objective function and does not use information
0.98309 (indicated with bold face in Table 2) and hence subsequent
on derivatives or other auxiliary knowledge and (iv) GA use sto-
studies were done with a network having one hidden layer with 16
chastic and not deterministic rules in conducting the search.
neurons contained in it. It is instructive to mention that in so far as
neuron independence studies are concerned, the Fourier numbers
are converted to time in seconds to evaluate the performance of 6. Results and discussion
various architectures.
MATLAB software is used for training and developing the net- 6.1. Melting and solidification patterns of different PCMs
work. In the present study, a total of 516 data points are available
from the experiments. About 80% of the randomly selected data are A comparison of the melting and solidification patterns for the
used for training the network and the remaining 20% are used to heat sink with 72 pin fins is presented in Figs. 6 and 7 respectively
test the network performance. Fig. 5 shows the architecture of the for the two PCM materials (n-eicosane and paraffin wax from Sigma

Fig. 7. Solidification phase of the heat sink with 72 pin fin, filled with different PCMs. Fig. 9. Spatial variation of temperature within the PCM, paraffin wax at 6 W.
 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77 71

Fig. 10. A comparison of the base temperature for f ¼ 1 at 8 W for an n-eicosane based Fig. 12. A comparison of the base temperature for f ¼ 0.3 at 8 W for an n-eicosane
heat sink. based heat sink.

Aldrich) with f ¼ 1. The average value of the temperature recorded the heat sink. Sensible heating in the pre-melting region, latent
by thermocouples T1 and T2 is considered here. These are located heating and sensible heating in the post-melting case constitute
at a height of 5 mm from the inside of the heat sink, i.e., at a height the complete melting phase. In the pre-melting region, conduction
of 10 mm from the heat sink base and protruding 10 mm into the is the only heat transfer mechanism and an almost linear rela-
PCM from the side. n-Eicosane has a melting point of 36.5  C and is tionship exists between temperature and time. When the temper-
ideally suited for applications in portable electronic equipment, as ature of the PCM reaches the melting point, the melting begins and
many studies have shown that when the temperature reaches the temperature remains within the melting range. This is the
45  C, users of electronics devices feel discomfort [39]. Paraffin wax, latent heating phase. Conduction and free convection within the
on the other hand, does not have a sharp melting point like n- molten PCM are present in the post melting case. All the regions
eicosane. The melting point of paraffin wax ranges between 53  C listed above for the melting phase are present in the solidification
and 57  C, and is ideally suited for applications at higher temper- phase too. In the solidification phase, the temperature reaches
atures, for which n-eicosane, is not suitable. A comparison of the a value below the melting temperature, before the phase change
melting phase for the two materials is given in Fig. 6 for a power material begins to solidify and release the heat. This is known as
level of 8 W. The properties of the two materials along with those of subcooling. A close examination of Fig. 7 shows that there is a slight
aluminium and cork (used to make the heat sink assembly) are subcooling from 100 to 115 min for the n-eicosane based heat sink.
given in Table 1. From Table 1, it is seen that the latent heat for n-
eicosane is 237.4 kJ/kg while that of paraffin wax is 173.4 kJ/kg. The
6.2. Spatial variation of temperature within the PCM
high latent heat of n-eicosane is also responsible for the increased
time duration for the latent heating as shown in Fig. 6. The costs of
The spatial variation of temperatures within the PCMs is rep-
100 g of n-eicosane and paraffin wax are $30 and $5 respectively
resented in Figs. 8 and 9 for n-eicosane and paraffin wax respec-
(2012 prices).
tively for the 72 pin fin heat sink at a power level of 6 W. The
The solidification phase represented in Fig. 7 is for a power level
abscissas of these figures represent the distance of the melt front up
of 7 W. The time required for the solidification phase is quite high
to 15 mm in the vertical direction. The average reading of the
when compared to the melting phase with the present design of
thermocouples H1 and H2 is used to record the base temperature

Fig. 11. A comparison of the base temperature for f ¼ 0.6 at 8 W for an n-eicosane Fig. 13. Temperature distribution with different volume fraction of the PCM at 8 W for
based heat sink. heat sink with 72 pin fin heat sink with n-eicosane as PCM.
72 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77

Fig. 14. Temperature distribution with different volume fractions of the PCM at 8 W for
heat sink for the 72 pin fin heat sink with paraffin wax as PCM.
Fig. 15. Enhancement in operating time due to presence of fins for a base temperature
of 47  C, f ¼ 1.
and this is considered to be the initial condition for the problem
under investigation. The average of the thermocouple readings T1 here is 3.175 kW/m2. In order to determine the dimensionless

and T2, T3 and T6, T4 and T5 which are at a height of 5 mm, 10 mm temperature, Q , the asymptotic set point temperature Tset , taken is
and 15 mm respectively (see Fig. 3) is used to map the temperatures same for both the PCMs. The Fourier number, Fo which represents
inside the PCM at these positions. Though temperature readings are the dimensionless time is used in the abscissa to denote time. To
recorded at every 5 s, for better readability and clarity, tempera- determine Fo, in the present case, the average values of the corre-
tures after 1000 s, 5000 s and 9000 s alone are shown in these sponding properties are used, whenever properties in the solid and
figures. As the plate heater is kept at the heat sink base, in all cases liquid phases are found to vary. A change of 0.01 in the Fourier
the starting point shows a higher temperature. After this point, the number, Fo corresponds to 450.7 s in the case of n-eicosane as the
spatial temperature variation is not significant for both the PCMs PCM. For a dimensionless temperature, Q ¼ 0.6, the Fo is compared
considered for the present study. Natural convection within the for different volume fractions of the PCM, for the no fin, 33, 72 and
melt plays a key role here in maintaining a uniform temperature 120 pin fin configurations. For the case of f ¼ 1, it is seen that the
with the PCM in the finned heat sink. This input is significant as Fourier number, Fo is 0.03 and 0.13 respectively for the heat sink
many researchers are currently working to develop the numerical with no fin and 72 pin fin heat sink. Among the other two config-
models of phase change phenomenon in finned heat sinks. Solving urations of the heat sink assembly, the heat transfer performance of
for melt convection will significantly improve the quality of the the 33 pin fin heat sink assembly is better as the Fo ¼ 0.095 is
simulations and bring them closer to experiments. higher in comparison to 0.085 for the 120 pin fin case. These are
shown in Fig. 10. For other values of the volumetric fraction, f,
6.3. Comparison of heat sinks with different volume fractions of the namely 0.6 and 0.3 too, trends are the same. The difference be-
PCM for n-eicosane based heat sinks tween the performance of 33 and 120 pin fin heat sinks narrows
down, once f changes from 1 to 0.6 and then to 0.3 for the same
A comparison of the heat transfer performance of all the heat value of Q ¼ 0.6 as seen from Figs. 11 and 12. The 72 pin fin heat
sinks with n-eicosane for different volume fractions of the PCM is sink is found to perform much better among the heat sink config-
given in Figs. 10e12. For all the cases considered here, the tem- urations considered in the present study.
perature taken is the average of the heat sink base temperature, H1 The performance of the n-eicosane heat sink with 72 pin fins for
and H2. The heat flux corresponding to the power level of 8 W used different volume fractions of the PCM (f ¼ 0.3, 0.6 and 1) is shown

Table 3
Time (s) taken by the heat sink to reach different set point temperatures (with n-eicosane as the PCM).

Type of heat sink Time to reach 42  C Time to reach 47  C Time to reach 52  C

Power level, W 5 6 7 8 5 6 7 8 5 6 7 8
Volume fraction of the PCM, f ¼ 1
No fin 510 275 140 112 1280 600 340 255 3830 1765 960 550
33 Pin fin 1825 950 465 385 5395 4075 2515 1745 6220 4830 3985 3465
72 Pin fin 4745 3760 3415 2385 6075 4425 3975 3085 9595 5455 4680 3725
120 Pin fin 4060 3355 2710 2270 4725 3845 3135 2660 5660 4500 3620 3050

Volume fraction of the PCM, f ¼ 0.6

No fin 450 240 160 125 1160 535 365 270 3120 1560 920 565
33 Pin fin 1520 895 400 365 3745 2930 1750 1595 4940 3475 2700 2300
72 Pin fin 3365 2565 2065 1600 4290 3140 2515 2050 5835 4050 3185 2555
120 Pin fin 2625 2145 1675 1495 3200 2565 2015 1790 4010 3130 2425 2130

Volume fraction of the PCM, f ¼ 0.3

No fin 375 200 180 114 985 455 370 230 2330 1210 875 490
33 Pin fin 1405 855 460 320 2410 1975 1580 1015 3575 2560 2040 1610
72 Pin fin 2090 1425 1100 750 2970 2025 1600 1155 4345 2875 2245 1595
120 Pin fin 2330 1650 1360 1095 2970 2060 1680 1365 3910 2615 2095 1690
 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77 73

Fig. 16. Enhancement in operating time due to presence of fins for a base temperature Fig. 18. Time to reach a set point temperature of 60  C at different power levels for the
of 52  C, f ¼ 1. 72 pin fin heat sink. PCM used is paraffin wax.

in Fig. 13. From the figure, it is seen that Fo is quite high for f ¼ 1, as
opposed to that for f 0.6 or 0.3. The masses of the PCM n-eicosane From Table 3, it is seen that for the same power level (5 W), the
with f ¼ 1, for the no fin, 33, 72 and 120 pin fin heat sinks are 49.32, operating time can change from 510 to 9595 s, depending on the
47.46, 43.87 and 39.75 g respectively. From this study, it can be choice of the control variables. This clearly indicates that there
inferred that there is a strong influence of both the PCM volume exists a tremendous scope to optimize the thermal design of the
fraction and the volume fraction of the TCE on the heat transfer heat sink. Table 3 shows the results for three values of 4, namely
performance of pin fin heat sinks. These experimental findings 0.3, 0.6 and 1 with the power changing from 5 to 8 W in steps of
strongly suggest the need for optimization of the phase change 1 W. An enhancement in the operation time due to the presence of
material based finned heat sinks from the view point of enhancing fins for heat sink base temperatures of 47  C and 52  C for 4 ¼ 1 is
the thermal performance. represented in Figs. 15 and 16. The enhancement ratio is defined as
the ratio of the time taken by the heat sink with fins to reach a set
6.4. Comparison of different volume fractions of the PCM for point temperature to that of a heat sink without fins. The power
paraffin wax based 72 pin fin heat sink level used vary from 5 to 8 W. These correspond to heat flux levels
of 1.984 and 3.175 kW/m2 respectively.
The performance of the paraffin wax based heat sink for dif- In Fig. 15, it is seen that for a set point temperature of 47  C, the
ferent volume fractions of the PCM (f ¼ 0.3, 0.6 and 1) with 72 pin enhancement is very significant for a power level of 7 and 8 W for
fins is given in Fig. 14. In this figure, a change of 0.01 in the Fourier the different heat sink geometries. For all the power levels, the heat
number, Fo, corresponds to 958.4 s. From Fig. 14, it is seen that Fo is transfer performance of the 72 pin fin heat sink is superior followed
quite high for f ¼ 1 as opposed to that for f ¼ 0.3 or 0.6. by the 120 pin fin heat sink.
When heat is transferred from the plate heater to the PCM, the
6.5. Enhancement in the operation time for different heat sinks layer of the PCM which is in the immediate vicinity of the heat sink
using n-eicosane as PCM base will change its phase and behaves like an insulator, con-
sequent upon the low thermal conductivity of the PCM, and this
Table 3 gives a comprehensive summary of the time taken by all
the heat sinks (with n-eicosane as the PCM) used in the present
study to reach different set point temperatures of 42, 47 and 52  C.

Fig. 19. Time to reach a set point temperature of 70  C at different power levels for the
Fig. 17. Enhancement in operating time at different f for a 72 pin fin case at 42  C. 72 pin fin heat sink. PCM used is paraffin wax.
74 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77

Fig. 22. Convergence of the ANNeGA hybrid algorithm showing the best fit and the
Fig. 20. Parity plot showing the adequacy of the ANN in predicting the time to reach mean.
a set point temperature.

the benefit of a more uniform distribution of heat in a heat sink

prevents further transfer of heat. This may cause local overheating
with more number of fins. Because of the fluctuations in the
and even failure of the device. The high thermal conductivity of the
enhancement ratio for various set point temperatures due to the
pin fin heat sink material aluminium in the present case establishes
conflicting requirements of the PCM and the TCE volume fraction, it
an alternate path to transfer the heat effectively, which ensures
is imperative that an optimum design of heat sink be sought to
a more uniform melting of the PCM. This in turn, reduces the
prolong the operation duration of the electronic equipment,
chances of local overheating as a consequence of which the phase
thereby increasing the life span and reliability of the device. Fur-
change temperature remains almost constant. Because of the
thermore, from Table 3, it is seen that while in general, the time to
availability of a large number of alternative paths in the 120 pin fin
reach the set point temperature increases with increasing f, lead-
case, to reach 47  C, the enhancement is shown more than what is
ing one to believe and conclude that the optimum always lies at
seen in the 33 pin fin heat sink at higher power levels. However,
f ¼ 1, a closer look reveals that this is not true always. For the no fin
from Fig. 16, it is seen that for a set point temperature of 52  C, the
case at all power levels it is seen that the time to reach set point
performance of 33 pin fin heat sink is surprisingly better than 120
temperature is not a linear function of f. Hence, a full fledged
pin fin heat sink at higher power levels. This counter intuitive
optimization with all the pertinent parameters taken into consid-
behaviour is largely due to the smaller volume of the PCM in the
eration becomes a necessity. The details of the optimization strat-
120 pin fin case as opposed to 33 pin fin heat sink which outweighs
egy employed in this study, wherein GA coupled with ANN is given
in Section 7.
Of the various configurations of the pin fin heat sink assembly
considered in this study, the 72 pin fin heat sink is seen to perform
better. Fig. 17 presents the variation of the enhancement for the 72
pin fin case with power, for various volume fractions of the PCM.
The set point temperature considered here is 42  C. An enhance-
ment of 24 is obtained for a power level of 7 W and f ¼ 1. From
Fig. 17, it can be observed that enhancement is more at higher
power levels for a given f and at higher f for a given power.

6.6. Enhancement in the operating time for different volumetric

fraction of paraffin wax in the 72 pin fin heat sink

The bar charts given in Figs. 18 and 19 show the time required
for the heat sink to reach a set point temperature of 60  C and 70  C
with 72 pin fins at different volumetric fractions of the PCM (par-
affin wax for f ¼ 0.3. 0.6 and 1). The power levels vary from 7 to
10 W. An average value of the heat sink base temperature is used
here. It is seen that time to reach the set point temperature de-
creases with increasing power levels, as expected. From Fig. 19, it is
seen that when the set point temperature is increased from 60  C to
70  C, the difference in the heat transfer performance for f ¼ 1 and
f ¼ 0.6 narrows down. For the heat sink with f ¼ 0.3, the perfor-
mance is the lowest at all power levels.

7. Optimization of heat sinks with n-eicosane

Fig. 21. Flow chart describing the coupled ANNeGA algorithm used in the present For accomplishing the optimization, first a neural network is
study. developed to predict the dimensionless time ðFoÞ of a heat sink
 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77 75

Table 4 conducted in this study, a minimum power level of 4 W is main-

Results of the GA output for various values of the control parameters. tained and as a consequence of this, the Q corresponding to this
Sl. no Population Crossover Q f 1þj power is given as the upper bound. The bounds for the input var-
size fraction iable 1 þ j, are 1 and 1.2 corresponding to heat sink with no fin and
1 12 0.5 0.031 0.961 1.098 a heat sink with more than 120 pin fins respectively. The value of
0.031 0.952 1.096 ð1 þ jÞ for 120 pin fins corresponds to 1.15. From the experimental
0.031 0.947 1.095
study, it is evident that as the number of pin fins is increased there
2 16 0.5 0.031 0.947 1.095
0.031 0.952 1.097 is a penalty in the PCM volume, causing the performance of the
0.031 0.952 1.097 heat sink to deteriorate. In view of this, a value of 1.2 is given as the
3 20 0.5 0.031 0.951 1.096 upper bound for 1 þ j.
0.031 0.953 1.096 As already stated, GA works with a population or a set of can-
0.031 0.959 1.098
4 24 0.5 0.031 0.953 1.096
didate solutions and tries to improve the average quality of these
0.031 0.952 1.096 solutions with iterations. The GA first initializes the population. The
0.031 0.956 1.097 variables are coded first as binary strings. Depending on the accu-
5 28 0.5 0.031 0.954 1.097 racy level, the population size can be specified. The fitness function
0.031 0.952 1.096
(objective function) values are then calculated by substituting the
0.031 0.951 1.096
6 20 0.3 0.031 0.962 1.098 variables in the fitness function and in the present study the fitness
0.031 0.96 1.099 function is evaluated from the ANN output. Reproduction, cross-
0.031 0.954 1.097 over and mutation are the important operations done to create
7 20 0.4 0.031 0.943 1.097 a new population. When the individual strings are copied according
0.031 0.959 1.098
0.031 0.955 1.097
to their objective function values, it is known as reproduction.
8 20 0.6 0.031 0.962 1.098 There are many ways in which the reproduction operator can be
0.031 0.948 1.095 implemented in algorithmic form. In the roulette wheel scheme,
0.031 0.956 1.098 the space on the wheel reflects the relative fitness of each member
9 20 0.7 0.031 0.963 1.099
of the population. The selection of parents for reproduction is based
0.031 0.96 1.098
0.031 0.956 1.098 on the rotation of the wheel. Similar to the selection procedure in
10 20 0.8 0.031 0.95 1.096 a tournament match, the tournament selection scheme is operated.
0.031 0.952 1.096 When one crossover point is selected, binary string from beginning
0.031 0.95 1.096 of chromosome to the crossover point is copied from one parent
Average 0.031 0.954 1.097
and the rest are copied from the second parent and this is known as
single point crossover. Two point, uniform and arithmetic are fre-
quently used of crossover strategies. A change of bits done ran-
configuration given the other three quantities namely Q, f and domly or stochastically in order to neutralize the undesirable
1 þ j. The rationale behind developing an ANN network from effects of crossover, known as mutation. The probability of muta-
a limited number of experiments is to device a methodology to tion is kept very close to 0.05 in this study. Once the important
obtain an optimized heat sink configuration. The time, cost and the operations of the GA are completed, the new generation is formed
effort required, if the optimization is attempted from experimental and in the present study, the decoded solution will act as input to
data alone, is enormous and so replacement of the experiments the ANN. Then by using the neural network, new fitness function
with an equally good neural network is inevitable. When ANN is values are evaluated till the desired fitness function tolerance limit
coupled with a powerful global search technique like GA, it is of 106. Otherwise, the procedure is repeated for the next genera-
capable of delivering a fast and inexpensive optimization scheme. It tion. Once the tolerance limit is reached or the specified iterations
is worth mentioning that the ANN in the present case is driven are over, then this scheme terminates. A flow chart for the coupled
purely based on the experimental data. ANNeGA algorithm for estimating the optimal configuration of the
A parity plot of the ANN network for n-eicosane as PCM where input parameters for the problem under consideration, to obtain
Fo is the output and Q, f, and ð1 þ jÞ are the inputs is shown in
Fig. 20. The ANN is developed from the data obtained from exper-
iments using the n-eicosane PCM. Here, the abscissa represents the
experimentally generated Fo and the ordinate represents Fo pre-
dicted by the neural network for a given (Q, f, 1 þ j). As a majority
of the data lie within 15% of the parity line, it can be concluded that
the neural network developed in the present study is adequate to
predict the Fourier number, Fo and can thus serve as a replacement
or “surrogate” to the experiments. To test the performance of the
ANN, nearly 20% of the independent experimental data (i.e., the
data not used for training the network) have been considered. For
this testing set, the RMS error in the Fourier number predicted by
the ANN is 0.008 and correlation coefficient of the ANN fit is 0.93.
The high correlation coefficient and the low value of the RMS error
reinforce our faith in the ability of the neural network to predict the
time to reach set point temperature.
The objective function to be maximized is Fo ¼ f (Q, f, 1 þ j),
subject to 0 < Q < 0.031, 0 < f < 1, 1 < 1 þ j < 1.2. From the
experimental data available for different power levels, it is clear
that, as the power level decreases, the time to reach a set point Fig. 23. Experiments with paraffin wax for different volume fractions, f of the PCM for
temperature and eventually, Fo increases. For the experiments the 72 pin fin heat sink.
76 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77

the time to reach a set point temperature is maximum, is given in temperature, f is the volume fraction of the PCM and j is the
Fig. 21 and the convergence of the ANNeGA hybrid algorithm volume fraction of the TCE. The predicted optima were finally
showing the best fit and the mean fit is shown in Fig. 22. It is validated with independent measurements. Finally, the optimal
seen that 51 generations are sufficient to guarantee convergence. solution obtained was found to hold for a paraffin wax based heat
The GA coupled with ANN algorithm was then subjected to several sink as well.
tests to check its robustness for tournament selection criterion with
crossover fraction varying from 0.3 to 0.8 and population size var-
ied from 12 to 28. The details are shown in Table 4. As there is no References
significant variation in the optimal solution, the robustness of the [1] E. Walsh, R. Grimes, Low profile fan and heat sink thermal management so-
algorithm is revalidated. An average of the total 30 optimal solu- lution for portable applications, Int. J. Therm. Sci. 46 (2007) 1182e1190.
tions is taken as the final optimal set and the values (0.031, 0.954, [2] K.C. Ng, C.R. Yap, M.A. Chan, A universal performance chart for cpu cooling
1.097) corresponding to (Q, f, 1 þ j). The value of 1.097 for ð1 þ jÞ devices, Heat Transfer Eng. 29 (2008) 651e656.
[3] M. Jaworski, Thermal performance of heat spreader for electronics cooling
denotes 77 pin fins in the heat sink. The solution was also found to with incorporated phase change material, Appl. Therm. Eng. 35 (2011)
be stable if the selection criterion is changed to either roulette or 212e219.
stochastic uniform. [4] E. Yu, Y. Joshi, Heat transfer enhancement from enclosed discrete components
using pinefin heat sinks, Int. J. Heat Mass Transfer 45 (2002) 4957e4966.
To finally validate the correctness of the optima obtained, fur- [5] J. Mahmoudi, Modeling of fluid flow and heat transfer in a copper based heat
ther experiments are conducted again for the 72 pin fin heat sink sink application, Int. J. Green Energy 3 (2006) 219e237.
with f ¼ 0.95. In view of this, one of the 2 controlling variables [6] H.Y. Li, K.Y. Chen, M.H. Chiang, Thermal-fluid characteristics of plate-fin heat
sinks cooled by impingement jet, Energy Convers. Manage. 50 (2009) 2738e
namely the volume fraction of the PCM is removed from the opti- 2746.
mal solution. By using the 33 pin fin and the 120 pin fin heat sink, [7] R. Schmidt, Challenges in electronic cooling opportunities for enhanced
i.e., the number of fins has been varied from 33 to 120. It is seen that thermal management techniques microprocessor liquid cooled minichannel
heat sink, Heat Transfer Eng. 25 (2004) 3e12.
the 72 pin case gives close but slightly inferior results compared to [8] S. Krishnan, S.V. Garimella, S.S. Kang, A novel hybrid heat sink using phase
what has been predicted. As the 77 pin fin case (optimum value change materials for transient thermal management of electronics, IEEE Trans.
predicted by the ANNeGA algorithm) is geometrically similar to the Compon. Packag. Technol. 28 (2005) 281e289.
[9] X.Q. Wang, A.S. Mujumdar, C. Yap, Effect of orientation for phase change
72 pin fin heat sink, additional experiments are not performed with
material (PCM)-based heat sinks for transient thermal management of electric
the 77 pin fin heat sink to save cost, time and effort. components, Int. Commun. Heat Mass Transfer 34 (2007) 801e808.
[10] R. Kandasamy, X.Q. Wang, A.S. Mujumdar, Transient cooling of electronics
using phase change material (PCM)-based heat sinks, Appl. Therm. Eng. 28
8. Performance of the optimal configuration for paraffin wax
(2008) 1047e1057.
[11] Z. Luo, H. Cho, X. Luo, K. Cho, System thermal analysis for mobile phone, Appl.
Finally, experiments were also done for the 72 pin fin heat sink Therm. Eng. 28 (2008) 1889e1895.
using paraffin wax as the PCM for the optimum configuration [12] E.M. Alawadhi, C.H. Amon, PCM thermal control unit for portable electronic
devices: experimental and numerical studies, IEEE Trans. Compon. Packag.
obtained using n-eicosane PCM. The value of Q ¼ 0.031, is kept Technol. 26 (2003) 116e125.
constant, by suitably choosing the excess temperature [13] R. Kandasamy, X.Q. Wang, A.S. Mujumdar, Application of phase change
ðTset  TN Þand keeping the power level at 6.5 W. Fig. 23 represents materials in thermal management of electronics, Appl. Therm. Eng. 27
(2007) 2822e2832.
the variation in operation duration for various values of f. The heat [14] S.K. Saha, K. Srinivasan, P. Dutta, Studies on optimum distribution of fins in
transfer performance of f ¼ 0.95 is much superior to the other heat sinks filled with phase change materials, ASME J. Heat Transfer 130
volume fractions of the paraffin wax. This reaffirms the correctness (2008) 034505/1e034505/4.
[15] M. Faraji, H. El Qarnia, Passive cooling of protruding electronic components by
of the optimal configuration predicted by the hybrid ANNeGA latent heat of fusion storage, ASME J. Electron. Pack 131 (2009) 021011.
algorithm, and so the optimum arrived at for n-eicosane heat [16] R. Akhilesh, A. Narasimhan, C. Balaji, Method to improve geometry for heat
sink is also valid for paraffin wax, notwithstanding the differences transfer enhancement in PCM composite heat sinks, Int. J. Heat Mass Transfer
48 (2005) 2759e2770.
in their properties and the operating temperature levels of the
[17] K.C. Nayak, S.K. Saha, K. Srinivasan, P. Dutta, A numerical model for heat sinks
system. with phase change materials and thermal conductivity enhancers, Int. J. Heat
Mass Transfer 49 (2006) 1833e1844.
[18] J. Xiang, C. Zhang, F. Jiang, X. Liu, Y. Tang, Fabrication and testing of phase
9. Conclusions
change heat sink for high power led, Trans. Nonferrous Metals Soc. China 21
(9) (2011) 2066e2071.
Experiments have been carried out by changing the number of [19] W.B. Ye, D.S. Zhu, N. Wang, Numerical simulation on phase-change thermal
fins, volume fraction of the PCM and power density to evaluate the storage/release in a plate-fin unit, Appl. Therm. Eng. 31 (2011) 3871e3884.
[20] R. Baby, C. Balaji, Experimental investigations on phase change material based
heat transfer characteristics of PCM based pin fin heat sinks. Two finned heat sinks for electronic equipment cooling, Int. J. Heat Mass Transfer
phase change materials (PCMs) namely n-eicosane and paraffin 55 (2012) 1642e1649.
wax were used in the present study. An enhancement factor of 24 [21] Z. Zhang, N. Zhang, J. Peng, X. Fang, X. Gao, Y. Fang, Preparation and thermal
energy storage properties of paraffin/expanded graphite composite phase
for a power level of 7 W and f ¼ 1, in the operation time for the heat change material, Appl. Energy 91 (2012) 426e431.
sink with 72 pin fins with n-eicosane as opposed to heat sink [22] C.Y. Zhao, W. Lu, Y. Tian, Heat transfer enhancement for thermal energy
without fin was obtained. This establishes, the superior perfor- storage using metal foams embedded within phase change materials (PCMs),
Solar Energy 84 (2010) 1402e1412.
mance of PCM based pin fin heat sinks. A large number of uniformly [23] O. Sanusi, R. Warzoha, A.S. Fleischer, Energy storage and solidification of
distributed fins are largely responsible for this enhancement in paraffin phase change material embedded with graphite nanofibers, Int. J.
heat transfer. An optimized configuration of the heat sink is Heat Mass Transfer 54 (2011) 4429e4436.
[24] S.C. Fok, W. Shen, F.L. Tan, Cooling of portable hand-held electronic devices
obtained using a hybrid optimization technique, wherein Genetic using phase change materials in finned heat sinks, Int. J. Therm. Sci. 49 (2010)
Algorithm (GA), were coupled with a feed forward back propaga- 109e117.
tion artificial neural network (ANN). The highlight of the work is [25] N. Yusup, A.M. Zain, S.Z.M. Hashim, Evolutionary techniques in optimizing
machining parameters: review and recent applications (2007e2011), Expert
the use of actual experimental data to develop the ANN. The trained
Syst. Appl. 39 (2012) 9909e9927.
network, becomes a surrogate to the experiments, and was used to [26] L. Gosselin, M. Tye-Gingras, F. Mathieu-Potvin, Review of utilization of genetic
drive the GA based optimization to maximize the operating time of algorithms in heat transfer problems, Int. J. Heat Mass Transfer 52 (2009)
the device, for an n-eicosane based heat sink. For the problem 2169e2188.
[27] K. Ermis, A. Erek, I. Dincer, Heat transfer analysis of phase change process in
under consideration, the optimum configuration is Q ¼ 0.031, a finned-tube thermal energy storage system using artificial neural network,
f ¼ 0.95 and 1 þ j ¼ 1.097, where Q is the modified dimensionless Int. J. Heat Mass Transfer 50 (2007) 3163e3175.
 R. Baby, C. Balaji / Applied Thermal Engineering 54 (2013) 65e77 77

[28] M.S. Bhatti, D. Kapoor, R.K. Kalia, A.S. Reddy, A.K. Thukral, RSM and ANN mod- [33] X. Huo, Z. Xie, A novel channel selection method for CANDU refuelling
eling for electrocoagulation of copper from simulated wastewater: multi based on the BPANN and GA techniques, Ann. Nucl. Energy 32 (2005)
objective optimization using genetic algorithm approach, Desalination 274 1081e1099.
(2011) 74e80. [34] R.R. Madadi, C. Balaji, Optimization of the location of multiple discrete heat
[29] C.R. Chen, H.S. Ramaswamy, Modeling and optimization of variable retort sources in a ventilated cavity using artificial neural networks and micro ge-
temperature (VRT) thermal processing using coupled neural networks and netic algorithm, Int. J. Heat Mass Transfer 51 (2008) 2299e2312.
genetic algorithms, J. Food Eng. 53 (2002) 209e220. [35] S.P. Venkateshan, Mechnical Measurements, first ed., Ane Books India, New
[30] T. Nazghelichi, M. Aghbashlo, M.H. Kianmehr, Optimization of an artificial Delhi, 2008.
neural network topology using coupled response surface methodology and [36] B. Yegnanarayana, Artificial Neural Networks, PHI Learning Pvt. Ltd., 2004.
genetic algorithm for fluidized bed drying, Comput. Electron. Agr. 75 (2011) [37] J.H. Holland, Adaptation in Natural and Artificial Systems: an Introductory
84e91. Analysis with Applications to Biology, Control, and Artificial Intelligence, The
[31] M. Rajendra, P.C. Jena, H. Raheman, Prediction of optimized pretreatment MIT Press, Cambridge, 1992.
process parameters for biodiesel production using ANN and GA, Fuel 88 [38] D.E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine
(2009) 868e875. Learning, Addison-Wesley, Boston, 1989.
[32] R.S. Prakasham, T. Sathish, P. Brahmaiah, Imperative role of neural networks [39] A. Patapoutian, A.M. Peier, G.M. Story, V. Viswanath, Thermo TRP channels and
coupled genetic algorithm on optimization of biohydrogen yield, Int. J. beyond: mechanisms of temperature sensation, Nat. Rev. Neurosci. 4 (2003)
Hydrogen Energy 36 (2011) 4332e4339. 529e539.