International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
EXPERIMENTAL INVESTIGATION AND
NEURAL MODELING OF
WATER-BUTANOL SYSTEM IN A
SPIRAL PLATE HEAT EXCHANGER
A.Mohamed Shabiulla and S.Sivaprakasam
Department of Mechanical Engineering,
Annamalai University,
Tamil Nadu ,India.
ABSTRACT
In this paper, an experimental investigation of Water-Butanol system in a Spiral plate Heat Exchanger (SHE) is presented.
Experiments have been conducted by varying the mass flow rate of cold fluid (Butanol) and inlet temperature of the hot fluid
(Water), by keeping the mass flow rate of hot fluid constant. The effects of relevant parameters on the performance of spiral
plate heat exchanger are studied.
Also, in this paper, an attempt is made to propose Artificial Neural Network (ANN) models for the analysis of SHE. The data
required to train the ANN models are obtained from the experimental data in the form of polynomial models. The ANN models
are developed using Back Propagation Network (BPN) algorithm incorporating Levenberg-Marquardt (L-M) training method.
The accuracy of the trained networks is verified according to their ability to predict unseen data by minimizing mean square error
(MSE) value. The prediction of the parameters can be obtained without using charts and complicated equations.
The data obtained from ANN and polynomial models for U, W and Nu are compared with those of experimental data. It is
observed that the accuracy between the NN’s predictions, polynomial model’s predictions and experimental values are achieved
with minimum mean absolute relative error less than or equal to ± 7 % for the training and testing data sets respectively,
thereby suggesting the reliability of the Neural Networks (NNs) as a Modeling Tool for Engineers in preliminary design and
analysis of Spiral plate Heat Exchangers (SHEs).
Keywords: Spiral plate Heat Exchanger (SHE), Reynolds Number (Re), Nusselt Number (Nu), Overall heat transfer
coefficient (U), Prandtl Number (Pr) and Artificial Neural Networks (ANNs).
1. INTRODUCTION
Heat exchangers are devices which are used to enhance or facilitate the flow of heat. Their application has a wide
coverage from industry to commerce. The smallest heat exchanger (<1W) is employed in miniature cryogenic coolers
used for infra red thermal imaging. The largest heat exchanger (> I GW) is employed in boilers and condensers. The wide
variety of requirements necessitated various types, shapes and flow arrangements in the design of heat exchangers. They
are widely used in space heating, refrigeration, air conditioning, power plants, chemical plants, petrochemical plants,
petroleum refineries, and natural gas processing. One common example of a heat exchanger is the radiator in a car, in
which a hot engine-cooling fluid, like antifreeze, transfers heat to air flowing through the radiator [1].
Compared to shell-and-tube exchangers, spiral plate heat exchangers are characterized by their large heat transfer surface
area per unit volume , resulting in reduced space, weight, support structure, energy requirements and cost, as well as
improved process design, plant layout and processing. Spiral plate heat exchangers have the distinct advantages over
other plate type heat exchangers in the aspect that they are self cleaning equipments with low fouling tendencies, easily
accessible for inspection or mechanical cleaning. They are best suited to handle slurries and viscous liquids. The use of
spiral heat exchangers is not limited to liquid-liquid services. Variations in the basic design of SHE makes it suitable for
Liquid-Vapour or Liquid-Gas services .
Few literatures have reported about Spiral Heat Exchanger [1 to 4]. Rajavel et al [3 and 4] have conducted experiments on
different process fluids to study the performance of a spiral heat exchanger and developed correlations for different fluid
systems. Martin [1] numerically studied the heat transfer and pressure drop characteristics of spiral plate heat exchanger.
The apparatus used in the investigations had a cross section of 5  300 mm2, number of turns n=8.5, core diameter of 250
mm, outer diameter of 495 mm. Martin developed the following correlation for the particular set up with water as a
medium .
(For 400 < Re < 30000)
(1)
Nu  0.04 Re 0.74 Pr 0.4
More number of experimental data are required to develop correlations to predict the parameters such as Re, Nu etc.
Volume 2, Issue 9, September 2013
Page 125
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
using mathematical models like regression analysis. The Artificial Neural Network (ANN) modeling technique has the
capability to predict the unseen data by using a reasonable set of experimental data resulting in more speed and accuracy.
Recently many researchers have addressed about ANN modeling to heat exchangers [6 and 7]. However,
not
many
addressed about ANN modeling for Spiral plate Heat Exchangers (SHEs). Hence, an attempt is made in this work to
study the experimental behaviour of SHEs and model the SHE’s characteristics using ANN technique.
2. EXPERIMENTAL SET-UP
The experimental heat exchanger set-up is shown in Fig.1. The heat exchanger was constructed using 316 stainless steel
plates. The spiral plate heat exchanger had a width of 304 mm and a plate thickness of 1 mm. The total heat transfer area
is 2.24 m2. The end connections are shown in Figure 1. The radius of curvature (measured from the centre) of the plate
is177.4 mm and the gap between the plates is 5 mm. The dimensions of the spiral plate heat exchanger are tabulated in
Table 1.
Figure 1 Schematic diagram of the SHE experimental set-up
Table 1: Dimensions of the Spiral Plate Heat Exchanger
Sl.No.
1
2
3
4
5
6
7
8
Parameter
Total heat transfer area (m2)
Plate width (mm)
Plate thickness (mm)
Plate material
Plate conductivity (W/m° C)
Core diameter (mm)
Outer diameter (mm)
Channel spacing (mm)
Dimensions
2.24
304
1
316 Stainless steel
15.364
273
350
5
The heat transfer and flow characteristics of water- butanol system were tested in a spiral plate heat exchanger as shown
in Figure 1. Water was used as the hot fluid. The hot fluid inlet pipe was connected to the central core of the spiral heat
exchanger and the outlet pipe was taken from the periphery of the heat exchanger. Hot fluid was heated by pumping
steam from the boiler to a temperature of about 60°C to 90°C. The hot fluid was pumped to the heat exchanger by using a
fractional horse power (0.367 kW) pump. Butanol was used as the cold fluid. The cold fluid inlet pipe was connected to
the periphery of the exchanger and the outlet was taken from the centre of the heat exchanger.
The cold fluid was supplied at room temperature from a tank and was pumped to the heat exchanger using a fractional
horse power (0.367 kW) pump.The inlet hot fluid flow rate was kept constant and the inlet cold fluid flow rate was varied
using a control valve. The flow of cold and hot fluid was varied using the control valves, V1and V2 respectively. Hot and
cold fluid flow paths of heat exchanger are as shown in Figure 1. Thermocouples T4 and T2 were used to measure the
outlet temperature of hot and cold fluids respectively and T3 and T1 were used to measure the inlet temperature of hot
and cold fluids respectively. For different cold fluid flow rates the temperatures at the inlet and outlet of the hot and cold
fluids were recorded. Temperature data was recorded in the span of ten seconds. The data used in the calculations were
obtained after the system attained steady state. Temperature reading fluctuations were within +/- 0.15°C. Though the
type-K thermocouples had limits of error of 2.2°C or 0.75% when placed in a common water solution the readings at
steady state were all within +/- 0.1°C. All the thermocouples were constructed from the same roll of thermocouple wire,
and hence the repeatability of the temperature readings was high. The operating range of different variables are given in
Volume 2, Issue 9, September 2013
Page 126
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
Table 2.The experiments were conducted by keeping the hot fluid flow rate of 0.5 kg/s and cold fluid inlet temperature at
26oC. The experimental data were obtained by varying cold fluid flow rate from 0.1 to 0.9 kg/s for three different hot fluid
inlet temperatures 60 oC , 70 oC and 80 oC. The experimental results and the corresponding Re, Pr, U, ∆P, W and Nu are
tabulated in Tables 3 to 5.
Table 2: Experimental conditions
Sl.No.
1
2
3
4
Variables
Hot fluid temperature
Cold fluid temperature
Mass flow rate of hot fluid
Mass flow rate of cold fluid
Range
60-80 o C
26 oC
0.5 kg/s
0.1- 0.9 kg/s
2.1 Variables to be Considered for Investigation
2.1.1 Reynolds Number (Re)
It represents the ratio of the inertia force to the viscous force. Therefore, for small Reynolds numbers, the viscous forces
are dominant whereas for larger Reynolds numbers, inertia forces are dominant. Reynolds number is used as the criterion
to determine the change of flow from laminar to turbulent.
Re  (ρ V Lc)/μ
(2)
where V is the upstream velocity (m/s) and LC is the characteristic length (m) of the geometry and µ/ρ is the kinematic
viscosity (m2/s) of the fluid .
2.1.2 Prandtl Number (Pr)
It is the ratio between the molecular diffusivity of momentum to the molecular diffusivity of heat. Therefore, it represents
the relative importance of momentum and energy transport by the diffusion process. The relative thickness of velocity and
thermal boundary layers are dependent on the magnitude of the Prandtl number.
Pr  (C P μ)/k
(3)
where C P is the constant pressure specific heat measured in kJ/kg K , µ is the dynamic viscosity measured in Ns/m2
and k is the thermal conductivity measured in W/m K .
2.1.3 Nusselt Number (Nu)
It is the ratio between the heat transfer by convection to the heat transfer by conduction across the fluid layer of thickness
L.
Nu  (hL)/k
(4)
where h is the convection heat transfer coefficient ( W/m2 K) , L is the characteristic length and k is the thermal
conductivity (W/m K) .
2.1.4 Overall Heat Transfer Coefficient (U)
In the case of heat exchangers, various thermal resistances in the path of heat flow from the hot fluid to the cold fluid are
combined and represented as the overall heat transfer coefficient (U).The overall heat transfer coefficient is obtained from
the relation
(5)
U  Q/(A Δ T )
lm
where U is the overall heat transfer coefficient (W/m2 K) , A is the heat transfer area (m 2) and (  T)
lm
is
the
log
mean temperature difference ( K).
2.1.5 Pumping Power (W)
It is the power required to pump the fluid through the flow channel against the pressure drop. WPOWER is measured in
Watt (W) .
Volume flow rate (kg/s) x Pressure drop ( P) (kPa)
(6)
WPOWER 
Density of the fluid (kg/m 3 )
Table 3: Experimental values obtained for hot fluid inlet temperature of 60oC
Sl.
No
Cold
fluid
flow
rate
(kg/s)
Hot fluid
outlet
temp.
( oC)
Cold
fluid
outlet
temp.
(oC)
1
2
0.1
0.3
57.11
53.38
48.15
42.91
Volume 2, Issue 9, September 2013
Re
Prandtl
Number
(Pr)
322.68
881.81
40.03
43.06
Overall
Heat
Transfer
Coeff.
U
(W/m2K)
159.13
330.32
Pressure
drop
∆P
(kPa)
Pumping
power
(W)
0.218
1.247
0.02
0.38
Nusselt
Number
(Nu)
12.5749
27.2438
Page 127
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
3
4
5
0.5
0.7
0.9
50.86
48.98
47.63
40.03
38.09
36.55
1401.2
1902.2
2389.1
44.74
45.88
46.77
446.87
536.72
609.32
2.849
4.934
7.456
1.4
3.5
6.8
38.9716
49.3582
58.8774
Table 4: Experimental values obtained for hot fluid inlet temperature of 70oC
Sl.
No
Cold
fluid
flow
rate
(kg/s)
Hot
fluid
outlet
temp.
(oC)
Cold
fluid
outlet
temp.
(oC)
1
2
3
4
5
0.1
0.3
0.5
0.7
0.9
66.16
61.20
57.84
55.35
53.43
55.45
48.51
44.66
42.06
40.14
Sl.
No
Cold
fluid
flow
rate
(kg/s)
Hot
fluid
outlet
temp.
( oC)
Cold
fluid
outlet
temp.
(oC)
1
2
3
4
5
0.1
0.3
0.5
0.7
0.9
75.17
68.92
64.72
61.62
59.21
63.01
54.33
49.46
46.16
43.74
Re
Prandtl
Number
(Pr)
359.7
954.5
1493.3
2006.78
2504.73
37.04
40.58
42.59
43.98
45.03
Overall
Heat
Transfer
Coeff.
U
(W/m2K)
167.43
345.4
465.95
558.36
632.6
Pressure
drop
∆P
(kPa)
Pumping
power
(W)
Nusselt
Number
(Nu)
0.207
1.199
2.757
4.795
7.27
0.02
0.366
1.4
3.41
6.65
13.2107
28.2110
40.0546
50.4908
60.0543
Table 5: Experimental values obtained for hot fluid inlet temperature of 80oC
Re
Prandtl
Number
(Pr)
401.6
1035.3
1593.88
2119.05
2625.56
34.41
38.30
40.6
42.21
43.42
Overall
Heat
Transfer
Coeff.
U
(W/m2K)
175.65
359.95
483.73
578.2
654.59
Pressure
drop
∆P
(kPa)
Pumping
power
(W)
Nusselt
Number
(Nu)
0.198
1.156
2.673
4.668
7.097
0.02
0.35
1.36
3.32
6.5
13.9169
29.2744
41.2370
51.7095
61.2858
3. MODELING OF SPIRAL HEAT EXCHANGER
In this paper, an attempt is made to develop ANN models for U, Nu and W. The data required to train the ANN models
were obtained from the corresponding polynomial models for U, Nu and W. The polynomial model has the capacity to
generate more number of data from few experimental data as given in Table 3.The following sections present the results
for polynomial models and ANN models for hot fluid inlet temperature of 60oC.
3.1 Polynomial Models
In this paper, the second order polynomial models are developed for Reynolds Number Vs Overall Heat transfer
Coefficient and Prandtl Number Vs Overall Heat transfer Coefficient based on the experimental data as given in Table 3
for 60oC. The corresponding plots and the polynomial equations obtained are shown in Figs 2 and 3, respectively. It is
observed from both the graphs that the polynomial model output exactly matches with the experimental data for Re Vs U
and Pr Vs U.
Figure 2 Comparison of experimental data with polynomial model output (Re Vs U)
Volume 2, Issue 9, September 2013
Page 128
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
Figure 3 Comparison of experimental data with polynomial model output (Pr Vs U)
A third order polynomial model is developed for Reynolds Number Vs Pumping Power based on experimental data as
given in Table 3 for 60oC.The corresponding plot and the polynomial equation obtained are shown in Fig. 4. It is
inferred from the graph that the polynomial model output exactly matches with the experimental data for Re Vs W.
Figure 4 Comparison of experimental data with polynomial model output (Re Vs W)
3.2 ANN Models
In order to find the relationship between the input and output data derived from experimental work, a more powerful
method than the traditional ones are necessary. ANN is an especially efficient algorithm to approximate any function with
finite number of discontinuities by learning the relationships between input and output vectors. These algorithms can
learn from the experiments and also are fault tolerant in the sense that they are able to handle noisy and incomplete data.
The ANNs are able to deal with non-linear problems and once trained can perform estimation and generalization rapidly.
They have been used to solve complex problems that are difficult to be solved if not impossible by the conventional
approaches, such as control, optimization, pattern recognition, classification and so on, specially it is desired to have the
minimum difference between the predicted and observed outputs. ANNs are biological inspirations based on the various
brain functionality characteristics. They are composed of many simple elements called neurons that are interconnected by
links and act like axons to determine an empirical relationship between the inputs and outputs of a given system. Multiple
layer arrangement of a typical interconnected neural network is shown in Fig.6. It consists of an input layer, an output
layer and one hidden layer with different roles. Each connecting line has an associated weight. ANNs are trained by
adjusting these input weights (connection weights), so that the calculated outputs may be approximated by the desired
values.
The output from a given neuron is calculated by applying a transfer function to a weighted summation of its input to give
an output, which can serve as input to other neurons as follows.
 k 1

(7)
α
 F    w α
 β 
ijk
i(k

1)
jk
jk
k
 i 1

Volume 2, Issue 9, September 2013
Page 129
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
The model fitting parameters wijk are the connection weights. The nonlinear activation transfer functions Fk may have
many different forms. The classical ones are threshold, sigmoid, Gaussian and linear function etc. The training process
requires a proper set of data i.e., input (I1 ) and target output (t i). During training the weights and biases of the network are
iteratively adjusted to minimize the network performance function. The typical performance function that is used for
training feed forward neural networks is the network Mean Squares Errors (MSE).
MSE 
1 N
1 N
2
2
 (e i ) 
 (t i  α i )
N i 1
N i1
(8)
There are many different types of neural networks, differing by their network topology and/or learning algorithm. In this
(3)
paper the back propagation learning algorithm, which is one of the most commonly used algorithms is designed to predict
the SHE performances. Back propagation is a multilayer feed forward network with hidden layer between the input and
output. The simplest implementation of back propagation learning is the network weights and biases updates in the
direction of the negative gradient that the performance function decreases most rapidly.There are various back
propagation algorithms such as Scaled Conjugate Gradient (SCG), Levenberg-Marquardt (LM) and Resilient back
Propagation (RP). LM is the fastest training algorithm for networks of moderate size and it has the memory reduction
feature to be used when the training set is large.
The neural nets learn to recognize the patterns of the data sets during the training process. Neural nets teach themselves
the patterns of data set letting the analyst to perform more interesting flexible work in a changing environment. Although
neural network may take some time to learn a sudden drastic change, it shows its excellence in adapting constantly
changing information. However the programmed systems are constrained by the designed situation and they are not valid
otherwise. Neural networks build informative models whereas the more conventional models fail to do so. Because of
handling very complex interactions, the neural network can easily model data, which are too difficult to model
traditionally (statistics or programming logic).Performance of neural networks is at least as good as classical statistical
modeling and even better in most cases. The neural networks built models are more reflective of the data structure and are
significantly faster.
3.2.1
Architecture of Neural Model
Fig.5 illustrates the Simulink representation of the generation of input-output data from polynomial model for Re Vs U,
where the random Re inputs are given and the corresponding U values are recorded. Same procedures are followed to
record the values for Pr Vs U and Re Vs W.
Figure 5 Simulink diagram for generating input-output data using polynomial model (Re Vs U)
The neural network is trained using delayed output and current input obtained from the polynomial models. The network
is built with two inputs, one output with one hidden layer that contains ten neurons, so the design is 2-10-1 as shown in
Fig.6.The activation function for the hidden layer is Tansigmoidal, while for the output layer linear function is selected
and they are bipolar in nature. The Levenberg Marquardt (LM) learning algorithm does the correct choice of the weight.
The parameters used for training Re Vs U are given below:
Input vectors
Output vector
Momentum factor
Learning rate
Training parameter goal
Volume 2, Issue 9, September 2013
:
:
:
:
:
[ U(k-1) Re(k)]

U(k)
0.5
0.001
1e-4
Page 130
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
3.2.2 Training and Model Validation of Neural Models
The data set used for training is sufficiently rich to ensure the stable operation, since no additional learning takes place
after training. During training the NN learns by fitting the input-output data pairs for Reynolds number (Re) Vs Overall
heat transfer coefficient (U), Prandtl number (Pr) Vs Overall heat transfer coefficient (U) and Reynolds number (Re) Vs
Pumping power (W). This is achieved by using the LM algorithm.
Figure 6 Neural network architecture of a neural model for SHE
The random input data Re and the corresponding output data U obtained from the polynomial model for 1000 samples are
shown in Figs.7 and 8, respectively. The training is taking place until the error goal is achieved at 1e-4 as shown in
Fig.9. Fig.10 demonstrates the relationship between the training data and the predicted network output after training. As
can be seen from Fig.10 the error between the training data (target) and the NN model output is small and the points
almost fit. This may be improved by several factors such as increasing the number of epochs, increasing learning rate,
decreasing goal, etc. The training pattern of Mean Squared Error (MSE) value is shown in Fig.11 and this error goal is
attained within 119 epochs. Figs. 12 and 13 show the comparison of experimental data with polynomial and ANN model
output for experimental Re values. Table 6 shows the relative error between the actual values with polynomial model and
NN model output for overall heat transfer coefficient (U). Hence, it is proved that the neural network has the ability to
predict U as a function of Re.
The above procedure is repeated for Pr Vs U and Re Vs W. The random input data Pr and Re are generated and applied to
the corresponding polynomial models and the output data U and W are obtained from the polynomial model for 1000
samples. The training is taking place until the error goal is achieved as shown in Fig.9. Fig.14 shows the comparison of
experimental data (U) with polynomial and ANN model output for experimental Pr values. It is observed that the ANN
predicts U and W, which exactly matches with the polynomial model output and the actual value. This is achieved by
minimizing MSE values at 1e-4 within 102 and 64 epochs, respectively with NN architecture of 2-10-1. Fig.15
demonstrates the relationship between the training data and the predicted network output after training for pumping
power. As can be seen from Fig.15 that the error between the training data (target) and the NN model output is small and
the points almost fit. Fig. 16 shows the comparison of experimental data (W) with polynomial and ANN model output for
experimental Re values. Table 7 shows the relative error between the actual values with polynomial model and NN model
output for W. Hence, it is proved that the neural network has the ability to predict U as a function of Pr and W values as
function of Re.
2500
Reynolds Number (Re)
2000
1500
1000
500
0
0
100
200
300
400
500
600
Samples
700
800
900
1000
Figure 7 Random variation in Reynolds Number (Re) applied to polynomial model.
Volume 2, Issue 9, September 2013
Page 131
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
Overall Heat Transfer Co-efficient (U)
700
600
500
400
300
200
100
0
100
200
300
400
500
600
Samples
700
800
900
1000
Figure 8 Overall heat transfer co-efficient for random Reynolds Number (Re)
Figure 9 Block diagram of a training process of neural model for SHE
Overall Heat Transfer Co-efficient (U)
800
Polynomial model output
NN model output
700
600
500
400
300
200
100
0
100
200
300
400
500
600
Samples
700
800
900
1000

Figure 10 Comparison of polynomial model output ( U ) and NN model predicted output ( U )
of spiral heat exchanger for random Re
Figure 11 Reduction in Mean Squared Error (MSE) value during the training process of NN model
Volume 2, Issue 9, September 2013
Page 132
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
Overall Heat Transfer Co-efficient (U)
650
Experimental data
Polynomial model output
NN model output
600
550
500
450
400
350
300
250
200
150
0
500
1000
1500
Reynolds Number (Re)
2000
2500
Simulated Overall Heat Transfer Co-efficient (U)

Figure 12 Comparison of polynomial model output ( U ) and predicted NN model output ( U ) with
actual experimental data( U) for experimental Re values
700
Polynomial model output
NN model output
600
500
400
300
200
100
150
200
250
300
350
400
450
500
550
Experimental Overall Heat Transfer Co-efficient (U)
600
650

Figure 13 Comparison of polynomial model output (U) and predicted NN model output( U ) with
actual experimental data (U).
Table 6 :Relative error between the actual and predicted values of
overall heat transfer coefficient (U)
Overall Heat Transfer Co-efficient
Experimental
Polynomial
NN model
model output
output
%Error
Polynomial
NN model
model
159.1300
330.3200
161.4999
325.1995
162.5806
325.2974
-1.488
1.55
-2.16
1.52
446.8700
536.7200
449.2555
543.3584
449.2394
543.4361
-0.533
-1.23
-0.53
-1.25
609.3200
610.7692
610.8979
-0.237
-0.258
Figure 14 Comparison of polynomial model output and predicted NN model output with
experimental data (U) for various Pr values.
Volume 2, Issue 9, September 2013
Page 133
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
10
Polynomial model output
NN model output
Pumping Power (P)
8
6
4
2
0
-2
0
100
200
300
400
500
600
Time (sec)
700
800
900
1000

Figure 15 Comparison of polynomial model output ( W ) and predicted NN model output ( W )
of spiral heat exchanger for random Re
8
7
Pumping Power (W)
6
5
4
3
2
1
0
0
500
1000
1500
Reynolds Number (Re)
2000
2500
Figure 16 Comparison of polynomial model output and predicted NN model output with
experimental data (W) for various Re.
Table 7: Relative error between the actual and predicted values
of pumping power (W).
Pumping power(W)
Experimental
Polynomial
NN model
model
output
output
0.02
0.0206
0.0196
0.38
0.3749
0.3711
1.4
1.4898
1.4949
3.5
3.68
3.6745
6.8
7.2
7.1599
%Error
Polynomial
NN model
model
-3
1.342
-6.4
-5.143
-5.882
2
2.342
-6.779
-4.986
-5.293
In order to further evaluate the performance of ANN models, Root Mean Square Error (RMSE) value is calculated
between the actual value and the predicted output value from the ANN models, the RMSE value calculation for U is given
by
1 N
2
RMSE 
(9)
 (U (k ) - Û(k))
N k1
where U (k ) is the actual value and Û(k) is the predicted value of overall heat transfer coefficient for various Re values.
The RMSE values calculated for Re Vs U, Pr Vs U and Re Vs W are tabulated in Table 8.
Table 8: Root Mean Square Error values for NN models
NN Model
Volume 2, Issue 9, September 2013
RMSE Value
Re Vs U
Pr Vs U
0.1962
0.0703
Re Vs W
0.0025
Page 134
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 9, September 2013
ISSN 2319 - 4847
4. CONCLUSIONS
Experimental studies of Water-Butanol system in a Spiral plate Heat Exchanger has revealed the following results.
1. The Overall Heat Transfer Coefficient increases almost linearly as the Reynolds Number increases.
2. The Overall Heat Transfer Coefficient increases almost linearly as the Prandtl Number increases.
3. The Pumping Power Requirements increases exponentially as the Reynolds Number increases.
4. The ANN has the capability to predict the unseen data with minimum relative error of ± 7% and RSME less than
0.1 as seen from Tables 6,7 and 8.
References
[1] Holger Martin, Heat Exchangers, Hemisphere Publishing Corporation, London, 1992.
[2] Rangasamy Rajavel and Kaliannagounder Saravanan, “Heat Transfer Studies on Spiral Plate Heat Exchanger”,
Journal of Thermal Science, Vol.12, No.3, pp. 85-90, 2008.
[3] R.Rajavel and K. Saravanan, “An Experimental Study on Spiral Plate Heat Exchanger for Electrolytes”, Journal of
the University of Chemical Technology and Metallurgy, Vol.43, No.2, pp. 255-260, 2008.
[4] K.S.Manoharan and K. Saravanan, “Experimental Studies in a Spiral Plate Heat Exchanger”, European Journal of
Scientific Research, Vol.75, No.2, pp. 157-162, 2012.
[5] S.A.Mandavgane and S.L.Pandharipande, “Application of Optimum ANN Architecture for Heat Exchanger
Modeling”, Indian Journal of Chemical Technology, Vol.13, pp. 634-639, 2006.
[6] A.R.Moghadassi, S.M.Hosseini, F.Parvizian, F.Mohamadiyon, A.Behzadi Moghadam and A. Sanaeirad, “An Expert
Model for the Shell and Tube Heat Exchangers Analysis by Artificial Neural Networks”, ARPN Journal of
Engineering and Applied Sciences, Vol.6,No.9, pp.78-93,2011
AUTHOR
A.MOHAMED SHABIULLA received the B.E. in Mechanical Engineering and M.E. degree in Internal
Combustion Engineering from College of Engineering, Anna University, Chennai, Tamil Nadu in 2003 and
2007, respectively. He is working as an Assistant Professor in the Department of Mechanical Engineering,
Annamalai University, Chidambaram, Tamil Nadu. He is currently carrying out his research in Heat Exchangers.
Dr. S. SIVAPRAKASAM received the B.E. in Mechanical Engineering and M.E. degree in Thermal Power
Engineering from Annamalai University, Chidambaram, Tamil Nadu in 1995 and 1997, respectively. He received
his Ph.D. Degree in alternative Fuels from Annamalai University, Chidambaram, Tamil Nadu in 2006. He is
working as an Associate Professor in the Department of Mechanical Engineering, Annamalai University, Chidambaram,
Tamil Nadu. His areas of interests are IC Engines, Alternative Fuels and Heat Exchangers.
Volume 2, Issue 9, September 2013
Page 135