Numerical study on heat transfer of turbulent channel ow over periodic grooves

Smith Eiamsa-ard
a
, Pongjet Promvonge
b,

a
Department of Mechanical Engineering, Faculty of Engineering, Mahanakorn University of Technology, Bangkok 10530, Thailand
b
Department of Mechanical Engineering, Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang, Bangkok 10520, Thailand
A B S T R A C T A R T I C L E I N F O
Available online 25 April 2008
Keywords:
Thermal enhancement
Heat transfer
Friction
Turbulent ow
Channel
Groove
A numerical investigation of turbulent forced convection in a two-dimensional channel with periodic
transverse grooves on the lower channel wall is conducted. The lower wall is subjected to a uniform heat ux
condition while the upper wall is insulated. To investigate turbulence model effects, computations based on a
nite volume method, are carried out by utilizing four turbulence models: the standard k, the Renormalized
Group (RNG) k, the standard k, and the shear stress transport (SST) k turbulence models. Parametric
runs are made for Reynolds numbers ranging from 6000 to 18,000 with the groove-width to channel-height
ratio (B/H) of 0.5 to 1.75 while the groove pitch ratio of 2 and the depth ratio of 0.5 are xed throughout. The
predicted results from using several turbulence models reveal that the RNG and the k turbulence models
generally provide better agreement with available measurements than others. Therefore, the k model is
selected to use inprediction of this complex ow. In addition, the results of the heat transfer coefcient, friction
factor, skin friction coefcient and thermal enhancement factor are also examined. It is found that the grooved
channel provides a considerable increase in heat transfer at about 158% over the smooth channel and a
maximum gain of 1.33 on thermal performance factor is obtained for the case of B/H=0.75. This indicates that
the reverse/re-circulation ow in a channel with transverse grooves can improve the heat transfer rate.
2008 Elsevier Ltd. All rights reserved.
1. Introduction
The improvement of convective heat transfer in thermal systems is
needed in many engineering applications with a view to reducing the
size, weight and cost of heat exchangers. Attempts have been made to
enhance heat transfer in the heat exchangers by using roughen
surfaces or turbulators such as rib, groove and helical rib in disturbing
the ow and in providing transverse/longitudinal vortices or three-
dimensional mixing. Comprehensive reviews on heat transfer
enhancement by periodic surfaces mounted obstacles have rarely
been found because of the wide variability of geometric parameters. A
review of measurements of global heat transfer coefcient and
pressure drop for various rib congurations was given by Dalle
Donne and Meyer [1]. Martin and Bates [2] presented velocity eld and
the turbulence structure in an asymmetrically ribbed rectangular
channel with one rib conguration of square ribs and varying channel
height. Hong and Hsieh [3] investigated the inuence of rib alignment
on forced convection in a channel at different rib alignments, either
staggered or in-line, on the internal surfaces of rectangular or square
channels. Jaurker et al. [4] reported the heat transfer and friction
characteristics of ribgrooved articial roughness on one broad heated
wall of a large aspect ratio duct. Distributions of the heat transfer
coefcient and the pressure drop along the wall inside an asymme-
trically ribbed channel measured for thermally developing and
turbulent owover periodic grooves were studied by Lorenz et al. [5].
Numerical investigations on turbulent ow friction and heat
transfer enhancement in ducts or channels with rib, groove or rib
groove turbulators have been carried out extensively. Chaube et al. [6]
investigated the ow and heat transfer characteristics of a two-
dimensional rib roughened (rectangular/chamfered rib) rectangular
duct with only one principal heating wall by using the shear stress
transport k turbulence model. Saidi and Sunden [7] studied the
thermal characteristics in a duct with rib turbulators by using a simple
eddy viscosity model and an explicit algebraic stress model and
reported that the algebraic stress model has some superiority over the
eddy viscosity model for only velocity eld structure prediction.
Tatsumi et al. [8] investigated numerically the ow around a discrete
rib attached obliquely to the ow direction onto the bottom wall of a
square duct and found that noticeable heat transfer augmentationwas
obtained downstream of the rib, produced by a strong secondary ow
motion. Yang and Hwang [9] conducted numerically to examine the
heat transfer enhancement in rectangular ducts with slit and solid ribs
mounted on one wall using the k turbulence model.
Luo et al. [10] reported the turbulent convection behavior in a
horizontal parallel-plate channel with periodic transverse ribs using
the standard k model and a Reynolds stress model. They found that
International Communications in Heat and Mass Transfer 35 (2008) 844-852
Communicated by W.J. Minkowycz.
Corresponding author.
E-mail address: kppongje@kmitl.ac.th (P. Promvonge).
0735-1933/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.icheatmasstransfer.2008.03.008
Contents lists available at ScienceDirect
International Communications in Heat and Mass Transfer
j our nal homepage: www. el sevi er. com/ l ocat e/ i chmt
the anticlockwise vortex was observed in the downstream region of a
rib while the length and relative strength of the vortex predicted by
these two models were signicantly different. Kim and Lee [11]
presented an optimal design of a square channel with V-shaped
square ribs extruded on both walls to enhance turbulent heat transfer
by using the shear stress transport (SST) model.
Apart fromusing different turbulence models as mentioned earlier,
large eddy simulation (LES) was also introduced to investigate
turbulent ows in periodically grooved channel by Yang [12] and in
a ribbed tube by Vijiapurapu and Cui [13] and found that the large-
scale turbulent structures are signicantly affected by the geometry of
the groove or the rib.
Several recent studies cited above reveal that the grooved channel
ows have been examined only for heat transfer coefcients or the
ow eld and turbulence structure from different groove pitches/
heights with rarely varying groove sizes. In the present work, heat
transfer (Nu), friction factor (f), skin friction coefcient (C
f
Re) and
thermal enhancement factor ((Nu/Nu
s
) / (f/f
s
)
1/3
) characteristics in a
perodic grooved channel with different groove widths are studied
numerically. The main goals of the present work are (1) to examine the
performance of the four turbulence models (standard k turbulence,
Renormalized Group (RNG) k turbulence, standard k turbulence
and SST k turbulence models) in prediction of heat transfer in
turbulent grooved channel ows and (2) to investigate the effect of
groove-width ratio (B/H) on heat transfer, ow friction and thermal
enhancement characteristics.
2. Geometry of the present problem
The system of interest is a horizontal plane channel with nine
periodic grooves (eight ribs) along the lower channel wall as shown in
Fig. 1. The channel height is set to H=40 mmwhile the channel length,
rib land (s) and groove width (B) were set to 47H, H and H,
respectively. To ensure a fully developed ow, the rst groove was
located at the distance of 20H downstream of the entrance while the
last groove was set to 10H upstream of the exit. The grooved channel
having eight ribs with a test section length of L=680 mm (17H),
20 mm rib height (e=0.5H), 40 mm rib length (s=H) and 40 mm
channel clear height (H) as depicted in Fig. 1. The dimension of the
constant heat ux test section mentioned above is the same as the
grooved channel of Lorenz et al. [5] used for validation. To examine an
effect of the groove size on heat transfer and friction loss in the
channel, the groove width ratio, (B/H), is varied to be B/H=0.5, 0.75,
1.0, 1.25, 1.5 and 1.75, for each test run of calculations.
A uniform rectangular mesh with grid adoption for y
+
2 at an
adjacent wall region is used to resolve the laminar sub-layer and is
shown in Fig. 2. To obtain grid independence solution, the number of
cells is varied between 50,900 and 155,992 meshes. The mean inlet
velocity between 1.19 and 3.58 m/s based on Reynolds number cited
above, zero pressure gradients at the exit and no slip wall boundary
conditions are taken for the present computation. In previous
research, Chaube et al. [6] suggested that the calculation with 2-
dimensional ow model yields the results closer to measurements as
compared that with 3-dimensional ow. In this work, the 2D ow is
therefore carried out for saving computer memory and computational
time.
3. Mathematical modeling
3.1. Flow governing equations
The phenomenon under consideration is governed by the steady 2-
Dimensional formof the continuity, the time-averaged incompressible
Fig. 1. Grooved channel ow conguration.
Nomenclature
B distance between grooves, m
C
f
friction coefcient
C

turbulence model constant

D
h
hydraulic diameter of channel, (=2H)
e rib height or groove depth, m
f friction factor
H channel height, m
h convective heat transfer coefcient, W m
2
K
1
k thermal conductivity, W m
1
K
1
L channel length of test section, m
Nu Nusselt number
p static pressure, Pa
Re Reynolds number
s rib land/length, m
T temperature, K
u mean velocity, m s
1
u
i
uctuation velocity components, m s
1
Greek letter
kinematic viscosity, kg s
1
m
1

t
eddy viscosity, kg s
1
m
1

ij
Reynolds stress, m
2
s
2

w
wall shear stress, Pa
thermal enhancement factor, (=(Nu/Nu
s
) / (f/f
s
)
1/3
)
turbulent specic dissipation rate, s
1
turbulent dissipation rate, m
2
s
3
density, kg m
3
Subscript
ave average
s smooth
845 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852
NavierStokes equations and the energy equation. In the Cartesian
tensor system these equations can be written as:
Continuity equation:
A
Ax
i
u
i
0 1
Momentum equation:
A u
i
u
j

Ax
j

Ap
Ax
i

A
Ax
j

Au
i
Ax
j

Au
j
Ax
i


A
Ax
j
uV
i
uV
j
P

2
Energy equation:
A
Ax
i
u
i
T
A
Ax
j
C C
t

AT
Ax
j

3
where and
t
are molecular thermal diffusivity and turbulent
thermal diffusivity, respectively and are given by
C

Pr
; and C
t


t
Pr
t
: 4
The Reynolds-averaged approach to turbulence modeling requires
that the Reynolds stresses, uV
i
uV
j
in Eq. (2) needs to be modeled. For
closure of the equations, the k turbulence model is chosen. A
common method employs the Boussinesq hypothesis to relate the
Reynolds stresses to the mean velocity gradients:
uV
i
uV
j

t
Au
i
Ax
j

Au
j
Ax
i

5
The turbulent viscosity term
t
is to be computed from an
appropriate turbulence model. The expression for the turbulent
viscosity is given as

t
C

k
2
e
6
The modeled equation of the TKE, k is written as:
A
Ax
i
ku
i

A
Ax
j


t

k

Ak
Ax
j

G
k
e 7
Similarly the dissipation rate of TKE, is given by the following
equation:
A
Ax
i
eu
i

A
Ax
j


t

e

Ae
Ax
j

C
1e
e
k
G
k
C
2e

e
2
k
8
where G
k
is the rate of generation of the TKE while is its destruction
rate. G
k
is written as:
G
k
uV
i
uV
j
Au
j
Ax
i
9
Fig. 3. Effect of grid densities on Nu
x
prediction at Re=12,000 and B/H=1.0.
Fig. 2. Grid arrangement for grooved channel ow.
846 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852
The boundary values for the turbulent quantities near the wall are
specied with the enhanced wall treatment method. C

=0.09,
C
1
=1.44, C
2
=1.92,
k
=1.0,

=1.3 and Pr
t
=0.9 are chosen to be
empirical constants [14] in the turbulence transport equations.
Thegoverningequations aresolvedusinganitevolumeapproachand
the SIMPLE algorithm [15]. The solutions are considered to be converged
when the normalized residual values reach 10
5
for all variables.
3.2. Turbulence models
In contrast to the molecular viscosity, the eddy viscosity depends
strongly on the ow property. Therefore, selecting the turbulence
model, to accomodate the ow behavior of each application, is very
important. To attainthe accurate aerodynamic predictioninthe grooved
channel, the predictive ability of four different turbulence models,
including, the standard k turbulence model, the Renormalized Group
(RNG) k turbulence model, the standard k turbulence model, and
theShear Stress Transport (SST) kturbulence model, are investigated.
More details of each model can be found in Ref. [16].
3.3. Solution procedure
The time-independent incompressible NavierStokes equations and
the turbulence model were discretized using the nite volume method.
Fig. 5. Axial distribution of Nu
x
on the top of different rib locations at Re=12,000 and B/H=1.0.
Fig. 4. Effect of turbulence models on Nu
x
in a single pitch at Re=12,000 and B/H=1.0.
847 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852
QUICK (Quadratic upstream interpolation for convective kinetics
differencing scheme) and central differencing ow numerical schemes
were applied for convective and diffusive terms, respectively. The
discretized nonlinear equations were implemented implicitly. To
evaluate the pressure eld, the pressurevelocity coupling algorithm
SIMPLE (Semi Implicit Method for Pressure-Linked Equations) was
selected. At the inlet, uniform velocity prole has been imposed.
Impermeable boundary condition has been implemented over the
channel wall while constant heat ux condition is applied to the lower
wall of test section. The turbulence intensity was kept at 10%at the inlet.
Four parameters of interest for the present case are: (1) friction
factor, (2) skin friction coefcient, (3) Nusselt number and (4) thermal
enhancement factor. The friction factor, f is computed by pressure
drop, p across the length of test section, L, having the hydraulic
diameter, D
h
=2H as
f
p=L D
h
1
2
u
2
7
The skin friction coefcient, C
f
is given by
C
f


w
1
2
u
2
8
The heat transfer is measured by local Nusselt number which can
be obtained by
Nu x
h x D
h
k
9
The average Nusselt number can be obtained by;
Nu
1
L
Nu x Ax 10
The thermal enhancement factor () is given by
Nu=Nu
s
= f =f
s

1=3
11
Fig. 6. Contour plot of stream function of groove-8: (a) B/H=0.75, (b) B/H=1.25 and (c) B/H=1.75.
848 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852
where Nu
s
and f
s
stand for Nusselt number and friction factor for the
smooth channel, respectively.
4. Numerical results
4.1. Grid independent
Predicted results are reported for fully developed ow through the
channel with transverse grooves at Reynolds number, Re=12,000 based
onthe entrance velocity, u=1.19 m/s. The local Nusselt numbers withthe
RNG k turbulence model for ve different grid densities of 50,960
cells, 75,950 cells, 102,140 cells, 129,069 cells, and 155,992 cells and a
near wall elements spacing of y
+
2, are presented in Fig. 3. Numerical
results with grids having y
+
b2 only slightly differs from the results
presented here. For grid independence test, the number of cells is varied
from 50,960 to 155,992 in various steps. It is found that after 129,069
cells, further increase in cells is less than 3% variation in local Nusselt
number value which is taken as criterion for grid independence.
4.2. Effect of turbulence models
Fig. 4 compares the axial variation of surface Nusselt number (Nu
x
)
along a single groove of one pitch of previous measured data [5] with
the present work using various turbulence models (the standard k
model, the RNG k model, the standard k model and the SST k
model) at Re=12,000 and B/H=1.0. The average top surface Nusselt
number of the channel with a groove is also shown in the gure. The
numerical predicted result with the RNG k turbulence model or the
standard k turbulence model agrees well with the experimental
data [5] than that with the standard kmodel or the SST kmodel,
except at l/H=1.5 to 2.0. The predicted and measured results are in a
similar trend for using both the RNG k model and the standard k
model. Thus, any of the RNG k model or the standard k model
appears to be suitable in prediction of this ow. The highest heat
transfer rate is found on the top surface of rib or at l/H=0.5 to 1.5. The
differences between the calculated Nusselt number and the experi-
mental data are generally within 6%. In addition, the predictions with
Fig. 7. Contour plot of TKE of groove-8: (a) B/H=0.75, (b) B/H=1.25 and (c) B/H=1.75.
849 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852
the standard k model and the SST k model are under-predicted
in comparison with measurements, especially at the top front of rib (l/
H=0.5). To verify thermal periodicity, the local Nusselt number
distributions on the top surface of all ribs (0.5l/H1.5) are presented
in Fig. 5. Numerical results showthat the periodic thermal condition is
reached after the fth period at any Reynolds number.
4.3. Flow structure
The time-averaged channel ow structure in the presence of
grooves could be easily discerned by considering the stream function
plots in Fig. 6(ac). The gures present stream functions of turbulent
channel ow through grooves using the k model for B/H=0.5, 0.75,
and 1.75, respectively. Here the streamlines in the grooved channel are
presented. It can be visible that the largest re-circulation zone can be
found in the groove region for B/H=1.75. The larger the groove, the
longer the re-circulation zone becomes.
The contour plots of turbulence kinetic energy are displayed in Fig. 7
(ac) for different width ratios (B/H). The peak turbulence intensity
values, predicted by the k turbulence model, are seen on the topof rib
front regions, while the turbulence intensity is observed to be very low
at groove bottom wall area. A large turbulence kinetic energy zone is
found between the adjacent ribs close to the main owwhichyields the
strong inuence of turbulence intensity on heat transfer enhancement.
4.4. Heat transfer
Fig. 8 presents the heat transfer results for air ow in the channel
for six different groove-width ratios (B/H=0.5, 0.75, 1.0, 1.25, 1.5, and
1.75). In the gure, the mean Nusselt numbers are related as a function
of Reynolds number. The result obtained from smooth channel is also
plotted for comparison. The variation of average Nusselt number is
also presented in the gure which shows that the Nusselt number
increases with the increase of Reynolds number. In all cases, the
grooved channel ows give higher values of Nusselt number than that
for smooth channel ow due to the induction of high re-circulation/
reverse ow and thin boundary layer in the grooved channel, leading
to higher temperature gradients. The maximum Nusselt number is
obtained for B/H=0.75 while the lowest one is for B/H=0.5. In
addition, the Nusselt number values for B/H=0.75 and 1.0 are found to
be nearly the same. The increases in Nusselt number values for using
B/H=0.5, 0.75, 1.0, 1.25, 1.5, and 1.75, respectively, are about 145%,
158%, 157%, 156%, 154%, and 150% over the smooth channel.
4.5. Friction loss
Fig. 9 shows the predicted results of friction factor using the k
turbulence model for different width ratios (B/H) in the range of
Reynolds number investigated for xed values of other parameters. It
is seen that the friction factor decreases with increasing Reynolds
number in all cases due to the suppression of viscous sub-layer with
the increase of Reynolds number. The grooved channel is seen to yield
Fig. 8. Variation of mean Nusselt number with Reynolds number.
Fig. 9. Variation of friction factor with Reynolds number.
850 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852
higher friction factor than the smooth channel. The friction factor for
air owing in the grooved channel with larger width ratio is found to
be higher than that with smaller width ratio. The friction factor
increases with the increase of width ratio (B/H) and thus, the B/
H=1.75 provides maximum friction factor.
The distribution of local skin friction coefcient (C
f
) in a groove for
three different width ratios (B/H=0.5, 1.0 and 1.75) at Re=12,000, is
presented in Fig. 10. In the gure, it is evident that the presence of the
groove involves the increased value of surface friction on the channel
wall. The peak skin friction coefcient (C
f
) can be observed at the top
front of rib while the lowest of the skin friction coefcient is found at
the groove surface for all width ratios (B/H). However, the skin friction
coefcients are increased again at the locations corresponding to the
zones of counter rotating owas seen in the gure. It indicates that the
highest skin friction coefcient (C
f
) can be observed at the area of high
turbulent intensity especially at the front of the rib and groove area.
4.6. Performance evaluation
Fig. 11 presents the thermal enhancement factor (Nu/Nu
s
)(f/f
s
)
(-1/
3)
for air owing in the grooved channel with different width ratios (B/
H). In the gure, the thermal enhancement factor tends to decrease
with the rise of Reynolds number. The channel with B/H=0.75
provides the highest value of thermal enhancement factor while
that with B/H=1.75 yields the lowest one. Though the channel with
transverse grooves causes the energy loss of the uid ow, the loss
Fig. 10. Distribution of local skin friction coefcient on groove-8 wall.
Fig. 11. Variation of thermal enhancement factor with Reynolds number.
851 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852
value is low especially at lower Reynolds number. It is found that the
thermal enhancement factor values vary between 1.00 and 1.33 for B/
H=0.75. For the results investigated, the grooved channel with B/
H=0.75 gives the best overall thermal performance and one with B/
H=1 yields nearly the same value.
5. Conclusions
In this present investigation, a numerical prediction has been
conducted to study heat transfer and owfriction behaviors in turbulent
channel ows over periodic grooves. Verication of the heat transfer of
the groove channel is performed by comparing with the previous
research under similar conditions. Comparison of predicted results using
the standard k turbulence model, the RNG k turbulence model, the
standardkturbulence model andthe SSTkturbulence model with
measurements shows that the RNG k turbulence model or the k
turbulence model is more suitable in predicting heat transfer and uid
channel ow over periodic grooves on one wall. The heat transfer in the
channel withperiodic grooves canbeenhancedalthoughhigher pressure
loss of the uid ow is obtained. Thermal enhancement factor of about
1.33 is found at lower Reynolds number for B/H=0.75.
Acknowledgement
The rst author would like to gratefully acknowledge Mahanakorn
University of Technology (MUT) for the nancial support of this
research.
References
[1] M. Dalle Donne, L. Meyer, Turbulent convective heat transfer from rough surfaces
with two-dimensional rectangular ribs, International Journal Heat and Mass
Transfer 20 (1977) 583620.
[2] S.R. Martin, C.J. Bates, Small-probe-volume Laser Doppler Anemometry measure-
ments of turbulent ow near the wall of a rib-roughened channel, Flow
Measurement Instrumentation 3 (1992) 8188.
[3] Y.M. Hong, S.S. Hsieh, Heat transfer and friction factor measurement in ducts with
staggered and inline ribs, Transaction ASME Journal Heat Transfer 115 (1993)
5865.
[4] A.R. Jaurker, J.S. Saini, B.K. Gandhi, Heat transfer and friction characteristics of
rectangular solar air heater duct using ribgrooved articial roughness, Solar
Energy 80 (2006) 895907.
[5] S. Lorenz, D. Mukomilow, W. Leiner, Distribution of the heat transfer coefcient in
a channel with periodic transverse grooves, Experimental Thermal and Fluid
Science 11 (1995) 234242.
[6] A. Chaube, P.K. Sahoo, S.C. Solanki, Analysis of heat transfer augmentation and ow
characteristics due to rib roughness over absorber plate of a solar air heater,
Renewable Energy 31 (2006) 317331.
[7] A. Saidi, B. Sunden, Numerical simulation of turbulent convective heat transfer in
square ribbed ducts, Numerical Heat Transfer, Part A: Applications 38 (1) (2000)
6788.
[8] K. Tatsumi, H. Iwai, K. Inaoka, K. Suzuki, Numerical analysis for heat transfer
characteristics of an oblique discrete rib mounted in a square duct, Numerical Heat
Transfer, Part A: Applications 44 (8) (2003) 811831.
[9] Y.T. Yang, C.W. Hwang, Numerical calculations of heat transfer and friction
characteristics in rectangular ducts with slit and solid ribs mounted on one wall,
Numerical Heat Transfer, Part A: Applications 45 (4) (2004) 363375.
[10] D.D. Luo, C.W. Leung, T.L. Chan, W.O. Wong, Flow and forced-convection
characteristics of turbulent ow through parallel plates with periodic transverse
ribs, Numerical Heat Transfer, Part A: Applications 48 (1) (2005) 4358.
[11] K.Y. Kim, Y.M. Lee, Design optimization of internal cooling passage with V-shaped
ribs, Numerical Heat Transfer, Part A: Applications 51 (11) (2007) 11031118.
[12] K.S. Yang, Large eddy simulation of turbulent ows in periodically grooved
channel, Journal of Wind Engineering and Industrial Aerodynamics 84 (2000)
4764.
[13] S. Vijiapurapu, J. Cui, Simulation of turbulent owin a ribbed pipe using large eddy
simulation, Numerical Heat Transfer, Part A: Applications 51 (12) (2007)
11371165.
[14] B.E. Launder, D.B. Spalding, Lectures in Mathematical Models of Turbulence,
Academic Press, London, 1972.
[15] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere, Washington
DC, 1980.
[16] A. Ridluan, S. Eiamsa-ard, P. Promvonge, Numerical simulation of 3D turbulent
isothermal ow in a vortex combustor, International Communications in Heat and
Mass Transfer 34 (2007) 860869.
852 S. Eiamsa-ard, P. Promvonge / International Communications in Heat and Mass Transfer 35 (2008) 844-852