International Communications in Heat and Mass Transfer 59 (2014) 143147

Contents lists available at ScienceDirect

International Communications in Heat and Mass Transfer

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

Neuro-genetic optimization of laminar slot jets impinging on a

moving surface
Phani Krishna Kadiyala a,, Himadri Chattopadhyay b
a
b

Mechanical Engg. Section, MRAGR Govt. Polytechnic, Vizianagaram 535002, India

Dept. of Mechanical Engg., Jadavpur University, Kolkata 700032, India

a r t i c l e

i n f o

Available online 22 October 2014

Keywords:
Impingement of jets
Laminar slot jets
Optimization
Moving surface
GA and ANN

a b s t r a c t
In this work, heat transfer from a moving surface due to series of impinging slot jets under laminar conditions has
been optimized. For this study numerical investigations were carried out initially using Ansys Fluent 14 and these
results were used to train an articial neural network (ANN). This trained network was integrated into Micro-Genetic Algorithm to get the optimum parameters for better heat transfer from the surface, an optimization procedure proposed by Madadi and Balaji. Pitch of the jets (P), height of the jets (H) and the non-dimensional surface
velocity (Vs) were chosen as dependent variables for optimum heat transfer. 99 simulations were performed by
changing above parameters for each Reynolds number, Re of 100 and 200 were used for case study. Imposition of
surface velocity strongly affects the heat transfer magnitude and distribution following a change in ow structure. The performance of Micro-Genetic Algorithm (GA) was also compared with standard Genetic Algorithm
(GA); it shows that GA reaches optimum in less than half the time of standard GA. The optimum results show
that the pitch of the jets, height of the jets and surface velocity should be as low as possible.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Heat transfer from a moving surface due to jet impingement is of
considerable interest due to high heat transfer rates, which makes it applicable for cooling turbine blades, metal processing and many other
cooling and drying applications [18]. In recent years, turbulent heat
transfer studies were reported on effect of moving surface on slot jets
[28], while these studies on heat transfer from moving surface impinged with laminar slot jets were also reported [1].
Chattopadhyay and Saha [1] have investigated heat transfer from slot
jets on a moving surface under laminar conditions. Axial and knife jet
types were considered and have found that axial jets are always preferred over knife jets. Chattopadhyay et al. [2] studied heat transfer
from a moving surface due to impinging slot jets using the large eddy
simulation technique in the range of Reynolds between 500 and 3000.
The surface velocity has been varied from 0 up to twice the jet velocity.
By increasing surface velocity Nu distribution becomes more uniform
and it also reduces heat transfer. Different types of jets namely annular
jet [3] and circular jet [4] were investigated by Chattopadhyay. It was
found that heat transfer from annular jet was about 20% less than that

Communicated by W.J. Minkowycz.

Corresponding author at: Lecturer in Mechanical Engg., MRAGR Govt. Polytechnic,
Vizianagaram 535002, India.
E-mail address: kadiyala.phani@gmail.com (P.K. Kadiyala).

http://dx.doi.org/10.1016/j.icheatmasstransfer.2014.10.023
0735-1933/ 2014 Elsevier Ltd. All rights reserved.

of a circular jet [3]. The surface velocity largely inuences the heat transfer from the moving surface in the case of circular jet [4]. The large Eddy
Simulation technique was extensively used for analyzing heat transfer
from moving surface due to slot jets under turbulent conditions [2,6,8].
Whereas realizable K- model was used effectively by Chattopadhyay
and Benim [4,7] for high Reynolds number.
By combining optimization techniques with Computational Fluid Dynamics, long computer time can be saved compared to extensive search
[911]. Articial Neural Networks (ANNs) were integrated into Genetic
Algorithms (GAs) to obtain more efcient results [9,10]. Micro-GA is an
alternative to standard GA which is adopted to achieve optimum quickly
with population size less than 10 [11]. Madadi and Balaji [9] used a back
propagation ANN as an objective function which was fed to Micro-GA. A
similar approach was employed with standard GA applied to a natural
convection problem by Kadiyala and Chattopadhyay [10]. Lee et al. [12]
have optimized effectiveness of micro-heat exchanger using the standard GA combined with ANN and named as neuro-genetic optimization.
In the present work the computational domain is adopted from [1] as
shown in Fig. 1. Periodic interfaces have been taken on the left and right
sides of the domain so as to consider a series of jets. Pressure outlets were
imposed at the front and back faces for escaping of incoming uid. After
getting sufcient amount of results from CFD simulations an ANN is
trained to integrate with GA to get optimum results. The variables
that are optimized are pitch of the jets (P), height of the jet (H) and horizontal velocity of moving surface in x direction (Vs). Performance of GA
is also compared with standard GA for the present problem.

144

P.K. Kadiyala, H. Chattopadhyay / International Communications in Heat and Mass Transfer 59 (2014) 143147

2.2. Governing equations

Nomenclature
B
H
p
P
T
u,v,w
V
Vs
Lx,Ly,Lz
Re
Nu
MRE
n

The simplied form of continuity, momentum and energy equations

for a three dimensional, steady state, incompressible ow and laminar
forced convection heat transfer without viscous dissipation are given
below.
Continuity equation:

Width of the slot jet, m

Height of the jet from moving surface, m
Pressure, N/m2
Pitch of the slot jets, m
Temperature, k
Components of velocity, m/s
Velocity of the uid from jet, m/s
Nondimensional surface velocity in x-direction
Dimensions of computational domain in x, y, z directions, m
Reynolds number of jet based on 2B
Nusselt number of moving surface
mean relative error
number of nodes

u v w

0:
x y z

Momentum equations:

Greek symbols

Thermal diffusivity, m2/s

Dynamic viscosity, kg/m-s

Density, kg/m3

!



2
2
2
u
u
u
p
u u u

w
u v
x
x
x
x
x2 y2 z2

!



v
v
v
p
2 v 2 v 2 v

:
u v w

y
y
y
y
x2 y2 z2

!



2
2
2
w
w
w
p
w w w

u
:

v
w
z
z
z
z
x2
y2
z2

Energy equation:
!



2
2
2
T
T
T
T T T
u

v
w
x
y
z
x2 y2 z2

2. Model and governing equations

2.1. Model
The ow is assumed to be incompressible with constant properties
and without viscous dissipation. The width of the moving surface is 10
times the jet width. The heights of the jet considered are B, 2B and 3B,
whereas the pitch of the jets are 6B, 8B and 10B. The jet delivers the
uid with same properties as the ambient conditions. The transport
phenomena is considered to be in laminar regime. Uniform wall temperature boundary condition is considered for the moving surface. Uniform velocity prole was imposed at the jet entrance.

3. Numerical solution procedure

Structured mesh was generated with ne mesh at the exit of the jet
and near the moving surface, to generate accurate results with less
number of nodes. Different grid sizes were tested as part of grid independence study. By following Richardson extrapolation, error for different grid sizes was calculated. By compromising between computational

Periodic Boundary on sides

B
Lz=8B

H = B, 2B & 3B

Lx = P = 6B, 8B & 10B

Fig. 1. Computational domain.

Ly=10B

P.K. Kadiyala, H. Chattopadhyay / International Communications in Heat and Mass Transfer 59 (2014) 143147

145

Table 1
Validation of initial results with [1].
Re

Vs

Nu (Ref)

Nu

% error

100

0
0.1
0.5
1
2
0
0.1
0.5
1
2

3.74
3.71
3.69
3.29
2.63
5.45
5.39
5.34
4.44
3.42

3.737
3.735
3.703
3.332
2.644
5.506
5.509
5.450
4.656
3.575

0.09
0.68
0.36
1.27
0.54
1.03
2.20
2.06
4.87
4.54

200

time and percentage error, grid with 667,711 nodes was selected. The
adopted grid gave an error of 4.53% compared with grid independent
value. Initially the results for the conguration with P = 8B, H = 2B
were validated with those of Chattopadhyay and Saha [1]. The deviation
was within 25% as shown in Table 1. The trends reporting effect of surface velocity on Nusselt number along the direction of motion are compared with Chattopadhyay and Saha [1], shown in Fig. 2.
Three dimensional governing equations shown above are
discretized on a non-uniform structured grid using the nite volume
method. The velocities and pressures were calculated using semi-implicit pressure linked equation (SIMPLE) scheme [13]. The interpolation
of the gradients of velocities and temperature used the third-order accurate MUSCL scheme [14]. The discretized equations are then linearized using an implicit technique and solved using Ansys Fluent 14, 3d
double precision solver. The convergence criteria for energy equation
and momentum equations are 106 and 104 respectively.
4. Optimization
Optimization was adopted to get best values of pitch of jets (P),
height of jet (H) and moving surface velocity (Vs) for both Reynolds
numbers 100 and 200 by maximizing Nusselt number.
4.1. Articial Neural Networks
Articial Neural Networks (ANNs) are numerical modeling tools that
have found tremendous acceptance in many research areas for modeling real world problems [15]. The ANN used here is a feed forward
back propagation articial neural network. The transfer functions used
are tan-sigmoid transfer function for the rst three layers and pure

Fig. 3. Nu Vs X at Z = 0.5 for H = 2B and Re = 100.

linear transfer for the last output layer. The weights and bias values
are updated in the ANN using LevenbergMarquardt algorithm [9,10].
A total of 99 simulations were calculated for both Reynolds numbers
100 and 200. Out of these, 81 were used to train an ANN and the remaining were used to test it. The number of neurons in hidden layers is selected depending on mean relative error (MRE) and correlation
coefcient. ANN of two hidden layers with 5 neurons each is used for
both cases, input layer has three neurons and output layer has one
neuron.
MRE
where
a
t

Fig. 2. Comparison of Nu variation along X with surface velocity of present results with
Chattopadhyay and Saha [1].

n
1X
jai t i j
n i1 jt i j

ANN predicted value

target or actual value

Fig. 4. Variation of Nu W.R.T Vs for various pitches when Re = 100, H = 2B.

146

P.K. Kadiyala, H. Chattopadhyay / International Communications in Heat and Mass Transfer 59 (2014) 143147

Genetic Algorithm is an evolutionary optimization technique which

was derived from the concepts of evolution of species and survival of
the ttest [16]. Application of Genetic Algorithms in the area of heat
transfer has been extensively increased from the past 15 years due to
its advantages [11]. Here the performance of GA in achieving optimum
value has been compared with GA. The ow chart for optimization
was given by Kadiyala and Chattopadhyay [10]. For single generation
population size is 20 and the crossover fraction is 0.8.

reported. Symmetric distribution of Nu is observed at all pitches for stationary condition which takes an inclined S-shaped at a surface velocity
of 1.0, a similar trend being reported by Chattopadhyay and Saha [1].
Fig. 4 shows the variation of heat transfer with surface velocity. It can
be observed that as pitch increases, the value of Nu reduces at all levels
of surface velocity. However the change is more prominent for the case
of stationary jet and the difference reduces with increasing surface velocity. At Re = 100, 200 change in pitch from 6B to 10B reduces heat
transfer by 20 and 18% respectively. But the reduction is only 9% for
Vs = 2.
For the jet with height of B, the effect of surface velocity on heat
transfer is almost negligible. Fig. 5 shows the variation of Nu for different jet heights. Height of the jet largely affects the heat transfer with
positive surface velocity, compared to stationary condition. Uniform
heat transfer distribution can be achieved with lesser surface velocity
for H = 3B.
The Nusselt number always decreases with increase in velocity. The
surface velocity effects the ow regime strongly, as it has been observed
that strong asymmetry is discerned with increasing surface velocity. The
ow eld for such cases was previously discussed by Chattopadhyay et
al. [1,2,6]. It has been observed that jet height effects the distribution of
Nusselt number considerably. For H = B more heat is transferred nearer
the impingement zone, whereas it is distributed evenly for higher jet
heights.
From the above observations it is clear that heat transfer depends on
pitch of jets (P), height of the jet (H) and surface velocity (Vs). Data were
generated for different combinations of Re, P, H and Vs. This output is
further used for optimization purpose as described below.

4.3. Micro-Genetic Algorithms

5.2. Results from GA integrated with ANN optimization

The Micro-Genetic Algorithm (GA) is a small population Genetic Algorithm that operates on same principle as above. Due to small population, the computational time required to achieve best t solution is less.
Maximizing Nusselt number is the objective taken for optimization. The
population size is 5 for each generation. The Nusselt number value will
be generated by the ANN, whenever required by the algorithm. The ow
diagram was given by Madadi and Balaji [9].

By using Micro-GA it has been observed that the result may be obtained in less than half the time taken by standard GA, after 51 generations of population size 5. The optimum congurations for Re equal to
100 and 200 are shown in Table 2. The range of variables taken is
P = 6B10B, H = B3B and Vs = 02. GA reaches global optimum
after 51 generations.
It may be noted here that as the optimization parameter, surface averaged Nu is considered in the work. However, the optimization parameter may be dened by the user and the same strategy can be applied to
nd the new sets of optimum values.

Fig. 5. Variation of Nu W.R.T Vs for various heights of jets when Re = 100, P = 8B.

4.2. Genetic Algorithms

5. Results and discussions

5.1. Effect of different parameters on heat transfer

6. Conclusions
From the simulation data local Nusselt number was calculated as
Nu

T
:
z

The surface averaged value is obtained from the relationship

Z Z
Nu
Lx Ly

Nux; y
dxdy:
Lx Ly

The local variation in heat transfer distribution was reported earlier

by Chattopadhyay and Saha [1]. Our results conrm to this as shown in
Fig. 3. Here the variation in Nu at stationary condition and at Vs = 1 is

Table 2
Optimization results.

GA
GA

In this work CFD has been combined with biologically inspired optimization techniques for analyzing the heat transfer from a moving surface due to impinging slot jets. Simulations were carried out by varying
different parameters like P, H and Vs. The results from the above simulations are used to train a feed forward back propagation ANN which
is fed into Micro-Genetic Algorithm to get optimum heat transfer
(Nu). Micro-GA shows faster convergence to optimum than standard
GA.
Initial simulations show that with lower value of pitch between jets
and minimum height of the jet leads to higher heat transfer at all surface
velocities. Optimization results also conrm this. The surface velocity
should also be minimum, as suggested in previous works. The work establishes a methodology for combining several parameters for optimization of heat transfer performance from impinging jets.
References

Re

Vs

Nu

Time (sec)

100
200
100
200

6
6
6
6

1.126
1
1.122
1

0.016
0.125
0.056
0.194

4.87
7.04
4.87
7.05

4
4
10
10

[1] H. Chattopadhyay, S.K. Saha, Simulation of laminar slot jets impinging on a moving
surface, ASME J. Heat Transf. 124 (2002) 10491055.
[2] H. Chattopadhyay, G. Biswas, N.K. Mitra, Heat transfer from a moving surface due to
impinging jets, ASME J. Heat Transf. 124 (2002) 433440.
[3] H. Chattopadhyay, Numerical investigations of heat transfer from impinging annular
jet, Int. J. Heat Mass Transf. 47 (2004) 31973201.

P.K. Kadiyala, H. Chattopadhyay / International Communications in Heat and Mass Transfer 59 (2014) 143147
[4] H. Chattopadhyay, Effect of surface motion on transport processes due to circular
impinging jetsa numerical study, Drying Technol. 24 (2006) 13471351.
[5] H. Chattopadhyay, S.K. Saha, Numerical investigations of heat transfer over a moving
surface due to impinging knife-jets, Numer. Heat Transf. Part A 39 (2001) 531549.
[6] H. Chattopadhyay, S.K. Saha, Turbulent ow and heat transfer from a slot jet impinging on a moving plate, Int. J. Heat Fluid Flow 24 (2003) 685697.
[7] H. Chattopadhyay, A.C. Benim, Turbulent heat transfer over a moving surface due to
impinging slot jets, ASME J. Heat Transf. 133 (2011) 104502.
[8] T. Cziesla, G. Biswas, H. Chattopadhyay, N.K. Mitra, Large Eddy Simulation of ow
and heat transfer in an impinging slot jet, Int. J. Heat Fluid Flow 22 (2001) 500508.
[9] R.R. Madadi, C. Balaji, Optimization of the location of multiple discrete heat sources
in a ventilated cavity using articial neural networks and micro genetic algorithm,
Int. J. Heat Mass Transf. 51 (2008) 22992312.

147

[10] P.K. Kadiyala, H. Chattopadhyay, Optimal location of three heat sources on the wall
of a square cavity using genetic algorithms integrated with articial neural networks, Int. Commun. Heat Mass Transf. 38 (2011) 620624.
[11] L. Gosselin, M. Tye-Gingras, F. Mathieu-Potvin, Review of utilization of genetic algorithms in heat transfer problems, Int. J. Heat Mass Transf. 52 (2009) 21692188.
[12] H.W. Lee, Y.J. Teng, I.A. Azid, K.N. Seetharamu, Neuro-genetic optimization of micro
compact heat exchanger, Int. J. Numer. Methods Heat Fluid Flow 17 (2007) 2033.
[13] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Taylor and Francis, 1978.
[14] B. van Leer, Towards the ultimate conservative difference scheme. V. A second-order
sequel to Godunov's method, J. Comput. Phys. 32 (1979) 101136.
[15] B. Yegnanarayana, Articial Neural Networks, Prentice-hall of India, 1999.
[16] D.E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning,
Addison Wesley, 1989.