Natural convection heat transfer from inclined plate-n heat sinks

Ilker Tari

, Mehdi Mehrtash
Mechanical Engineering Department, Middle East Technical University, 06800 Ankara, Turkey
a r t i c l e i n f o
Article history:
Received 12 April 2012
Received in revised form 23 August 2012
Accepted 26 August 2012
Available online 27 October 2012
Keywords:
Plate n array
Vertical heat sink
Inclined heat sink
Natural convection
Electronics cooling
a b s t r a c t
The steady-state natural convection from heat sinks with parallel arrangement of rectangular cross sec-
tion vertical plate ns on a vertical base are numerically investigated in order to obtain a validated model
that is used for investigating inclined orientations of a heat sink. Taking a previous experimental study as
a basis, aluminum heat sinks with two different practical lengths are modeled. The models and the sim-
ulation approach are validated by comparing the at plate heat sink results with the available correla-
tions, and by comparing the nned heat sink results with the experimental data. Natural convection
and radiation heat transfer rates from the fronts of the heat sinks heated from the back with a heater
are obtained from nite volume computational uid dynamics simulations. The sensitivities of the heat
transfer rates to the geometric parameters are determined. A set of dimensionless correlations for the
convective heat transfer rate is suggested. The validated model is used for several upward and downward
inclination angles by varying the direction of gravitational acceleration. At small inclinations, it is
observed that convection heat transfer rate stays almost the same, even increases slightly for the down-
ward inclinations. At larger angles, the phenomenon is investigated for the purpose of determining the
ow structures forming around the heat sink. For the inclination angles of 4, 10, 20, 30, 45,
60, 75, +80, 85, 90 from the vertical, the extent of validity of the obtained vertical case correla-
tion is investigated by modifying the Grashof number with the cosine of the inclination angle. It is
observed that the correlation is valid in a very wide range, from 60 (upward) to +80 (downward). It
is also observed that the ow separation inside the n channels of the heat sink is an important phenom-
enon and determines the validity range of the modied correlation. It is further shown that the correla-
tions are also applicable to all available inclined case data in the literature, verifying both our results and
correlations. Since the investigated ranges of parameters are suitable for electronic device cooling, the
suggested correlations have a practical use in electronics cooling applications.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
Heat sinks with extended surfaces have been widely used in
various engineering applications especially in electronics cooling.
Due to ease of manufacturing, parallel arrangement of rectangular
cross section plate ns on a at base is the most common heat sink
geometry. Heat sinks with this geometry have been used for both
forced and natural convection. In the case of forced convection,
the geometric parameters of a heat sink highly depend on the
remaining components of the cooling system, such as the fan and
the enclosure; therefore, the optimal values of these parameters
depend on the considered application. In contrast, in natural con-
vection, it is possible to optimize the parameters of the heat sink
geometry in an application independent manner. Consequently,
in literature, there are several attempts for optimizing parameters
in the case of natural convection for vertical and horizontal orien-
tations of the heat sink. However, the suggested correlations for
the optimal values of the geometric parameters and for possible
natural convection heat transfer rates considerably vary among
different studies; these attempts do not converge to a consistent
set of correlations.
The objective of the present study is to obtain a consistent set of
correlations for all orientations of plate-n heat sinks, including
the vertical. At the end of the study, our efforts converge to a single
correlation covering a wide range of angles between vertical and
horizontal orientations in both upward and downward facing
directions of the heat sink. The suggested correlation agrees very
well with all available experimental data in literature for the in-
clined case ([1,2], with %5.8 and %8.8 mean relative errors,
respectively).
Heat sinks with parallel arrangement of rectangular cross sec-
tion plate ns on a at base are used preferably in vertical or up-
ward facing horizontal orientations in order to obtain higher
natural convection rates [2]. In certain scenarios, however, these
two orientations may not be suitable due to several system related
constraints, e.g., lack of available space on the side and top surfaces
of the electronic box or lack of any vertical or horizontal surface in
0017-9310/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2012.08.050

Corresponding author. Tel.: +90 312 2102551; fax: +90 312 2102536.
E-mail address: [email protected] (I. Tari).
International Journal of Heat and Mass Transfer 56 (2013) 574593
Contents lists available at SciVerse ScienceDirect
International Journal of Heat and Mass Transfer
j our nal homepage: www. el sevi er . com/ l ocat e/ i j hmt
the design. Moreover, one may be forced to use plate-n heat sinks
in inclined orientations when a vertical or horizontal heat sink is
inclined due to a rotation of the device. These possible scenarios
motivate this study.
In practice, there is certainly a need for investigating inclined
plate-n heat sinks. For example, when natural convection with
vertical plate-n heat sinks was suggested as a viable solution
for cooling of at panel displays with high power components [3]
and of laptop computers [4], the major concern was how to handle
the situation if the device is operated when the heat sink is inclined
due to the inclination of the screen.
Even though there are several studies investigating natural con-
vection heat transfer from vertical plate ns protruding from a ver-
tical base [514] or from a horizontal base [1521], Mittelmann
et al. [1] and Starner and McManus [2] are the only ones investigat-
ing inclined orientations of plate-n heat sinks. The ranges of the
parameters investigated in the previous works are summarized
in Table 1, together with the ones for the present work (the last
row). The last column in Table 1 shows the investigated inclination
angles from the vertical, where the zero corresponds to the vertical
orientation of the base while upward inclinations are negative. No-
tice that Starner and McManus [2] deal only with upward 45 incli-
nation, and Mittelman et al. [1] with downward inclination angles
between 60 and 90. The present study, by investigating wide
range of upward and downward angles, lls a gap in literature.
Firstly, we consider the vertical case, for which extensive exper-
imental data is available for numerical model validation. Secondly,
upon validating our numerical model and suggesting a set of corre-
lations for the vertical case, we use both the model and the corre-
lations to investigate the inclined case. Since there is practical
motivation for studying inclined orientations for electronics cool-
ing applications, the size of the simulated heat sink is selected
accordingly.
2. Numerical model and method
A recent experimental study, Yazicioglu and Ync [13,14], is
taken as the base case for the validation of our vertical model.
The experimental set-up is numerically modeled and simulations
are performed for the same set of parameters, using the informa-
tion presented in [14]. The simulation results are compared with
the experimental heat sink temperatures in order to verify that
the simulations are representing the conditions of the experi-
ments. The data in [14] together with the data from literature
([2,79] for the vertical and [1,2] for the inclined cases) are used
for comparison and verication.
The experimental set-up in [14] consists of a heat sink installed
on the front of a heater plate with layers of insulation in the back.
The experiments have been performed in a small room with nearly
stagnant air and nearly constant temperature. In our model, the
backside insulation of the heater plate is simplied and replaced
with an equivalent aerated concrete block (see Fig. 1). The assem-
bly in Fig. 1 is placed in an air lled cubical room of 3 m sides with
walls that are kept at uniform 20 C, as shown in Fig. 2. The prop-
erties of the model components are given in Table 2. The technical
drawings of two investigated heat sinks with their dimensions are
shown in Fig. 3. The six locations marked on each heat sink base
are the locations of the thermocouples in the experiments. These
locations are used in our simulations for convergence monitoring
and comparisons with the experimental temperature data.
In order to simulate experimental cases as closely as possible,
the dimensions of the aerated concrete insulation are determined
after several trials of matching the heat sink surface temperatures
to the experimental results at the thermocouple locations.
Steady state solutions are obtained by using the zero-equation-
turbulence model with initial ambient air temperature of 20 C. Air
is taken as an ideal gas at atmospheric pressure. No slip boundary
condition is used for all surfaces. There is no contact resistance be-
tween solid surfaces.
A non-conformal mesh structure with a very ne grid around
the cooling assembly and a coarse grid for the rest of the room is
employed. Grid independence is achieved by examining three dif-
ferent grid densities with 1,685,832, 2,834,264 and 4,077,608 cells,
and then selecting the medium density mesh, i.e., the one with
2,834,264 cells, as it yields results matching to those of the ne
mesh.
ANSYS Fluent solver [22] is used for solving the continuity,
momentum and thermal energy equations for air and the heat con-
duction equation within the solids. To handle the radiative heat
transfer, the surface-to-surface radiation model is used.
Nomenclature
c
p
heat capacity (J/K)
d base plate thickness (mm)
g gravitational acceleration (m/s
2
)
Gr
/
modied Grashof number, Gr
/
= [gbTS
4
]/[m
2
(LH)
0.5
]
h average heat transfer coefcient (W/(m
2
K))
H n height (mm)
k thermal conductivity of uid (air) (W/(m K))
L n length (mm)
Nu
L
average Nusselt number based on L, Nu
L
= (hL)=k
Nu
s
average Nusselt number based on S, Nu
S
= (hS)=k
p pressure (atm)
P non-dimensional pressure
Pr Prandtl number
Q
c
convection heat transfer rate from n array (W)
Q
in
power supplied to heater plate (W)
Q
r
radiative heat transfer rate from n array (W)
Ra Rayleigh number based on L, Ra = gbL
3
(T
w
- T
a
)/(ma)
S n spacing (mm)
S
opt
optimum n spacing (mm)
t n thickness (mm)
T
w
average base wall temperature (C)
T
a
ambient temperature (C)
T temperature (C)
T
f
lm temperature (C)
DT base-to-ambient temperature difference (K)
W heat sink width (mm)
u, v, w velocities in x, y and z directions
v
/
, w
/
characteristic velocities in y and z directions
v, w non-dimensional velocities in y and z directions
x, y, z Cartesian coordinates
X, Y, Z non-dimensional coordinates
Greek symbols
a thermal diffusivity (m
2
/s)
b volumetric thermal expansion coefcient (1/K)
e emissivity
h angle of inclination with respect to vertical position (de-
grees)
l dynamic viscosity (s/m
2
)
q density (kg/m
3
)
m kinematic viscosity (m
2
/s)
/ non-dimensional temperature
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 575
2.1. Vertical model validation
The model, mesh and solution parameters are validated by
comparing the simulation results with the results from available
empirical correlations after replacing the nned heat sink with a
at plate. Considered correlations are
Mc Adam
/
s relation [23],
Nu
L
= 0:59Ra
1=4
(1)
Churchill and Chu
/
s rst relation [24],
Nu
L
= 0:825
0:387Ra
1=6
[1 (0:492=Pr)
9
16
[
8=27
" #
2
(2)
Churchill and Chu
/
s second relation [24],
Nu
L
= 0:68
0:67Ra
1
4
1 (0:492=Pr)
9
16
h i4
9
(3)
In Eqs. (1)(3), the Rayleigh number is dened based on the heat
sink length L as Ra = gbL
3
(T
w
T
a
)=ma in which g is the gravita-
tional acceleration, T
w
and T
a
are respectively the wall and ambient
air temperatures, and b, m, a and Pr are the properties of air.
The comparison is given in Table 3. The agreement of the results
especially with Eq. (3) is very good (with 5.1% average relative er-
ror). This is an indication of sufcient grid structure and suitable
modeling approach. In addition to the at plate case, the numerical
temperature results are compared at every step with the available
experimental results showing a very good agreement.
2.2. Convective heat transfer rate
In [14] Yazicioglu used a calibration that replaces the heat sink
with an assembly of two identical parallel at plates separated by a
very small distance (two-at-plate-case) so that the front heat
Table 1
Ranges of parameters investigated in the literature.
Reference Fin length
L (mm)
Base width
W (mm)
Fin height
H (mm)
Fin
thickness t
(mm)
Fin spacing
S (mm)
Base-to-ambient
temperature difference DT
(C)
Optimum n
spacing S
opt
(mm)
Angle from vertical h (deg)
[1] 200 130 21.534 1.1 717 16.455.6 60, 70, 80, 90
[2] 127 254 6.3525.4 1.02 6.357.95 2590 0, 45, 90
[6] 203 66.3 6.35
19.05
2.3 4.819 3590 0
[7] 150 190 10, 17 3 345 2040 9, 9.5 0, 90
[8] 250 190 60 3 333 2080 911 0, 90
[9] 150, 250,
375, 500
190 30, 60, 90 119 376 2040 9.511 0
[10] 150, 250,
375, 500
190 30, 60, 90 119 376 2040 9.511 0
[11] 100 250 525 3 534 14106 7 0
[12] 2549 2549 13.5 1 311 1522 0
[13,14] 250, 340 180 525 3 585.5 21162 11.2 0
[15] 130390 130 10.534 1.1 7 1676 611 90
[16] 500 190 60 3 577 2040 12 0, 90
[17] 100 250 626 3 6.283 13133 10.520 90
[18] 100 535 520 3696 90
[19] 127381 6.350 438.1 33100 90
[20] 127381 2647 438.1 2070 90
[21] 750 712 37 412 4060 90
Present 250, 340 180 525 3 585.5 14185 11.75 0, 4, 10, 20, 30, 45, 60,
75, +80, 85, 90
Aerated Concrete Block
(340450100 mm)
Heater Plate
(1802505 mm)
Finned Heat Sink
Fig. 1. Schematic view of the model for the heat sink length of 250 mm.
Fig. 2. The 3D view of the computational domain. The domain is inclined with the
angle h by changing the direction of the gravitational acceleration g.
576 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
Table 2
Properties of the components.
Component Material type Specic heat (J/kg K) Conductivity (W/m K) Emissivity Roughness (mm)
Concrete Block Aerated Concrete 1000 0.15 0.9 2
Heater Base Plate Aluminum 900 130 0.2 0.02
Fin Array Aluminum 900 130 0.2 0.02
Fig. 3. Locations of the six temperature monitoring points (thermocouple positions) together with the dimensions of the heat sinks.
Table 3
Comparison of Nusselt numbers for vertical at plate case with the correlations from literature.
Q
in
(W) Ra Average Nu
L
McAdam
/
s relation Churchill and Chu
/
s rst relation Churchill and Chu
/
s second relation Present study
20 4.86E+07 49.26 49.26 43.58 45.59
30 6.08E+07 52.11 52.63 46.05 48.33
40 6.85E+07 53.68 54.51 47.42 49.92
50 7.36E+07 54.66 55.67 48.25 50.83
60 7.78E+07 55.41 56.57 48.91 51.53
70 8.01E+07 55.82 57.06 49.25 51.99
80 8.17E+07 56.09 57.37 49.48 52.16
90 8.27E+07 56.26 57.55 49.61 52.48
100 8.29E+07 56.30 57.59 49.63 52.29
110 8.28E+07 56.29 57.57 49.62 52.45
120 8.25E+07 56.23 57.48 49.56 52.28
130 8.20E+07 56.15 57.37 49.48 52.25
140 8.14E+07 56.05 57.23 49.38 52.09
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 577
transfer is only due to the radiative transfer between the plates (by
neglecting the heat conduction through the thin air layer), giving a
conductive heat loss value from the backside using the knowledge
of the input power to the heater. When the actual nned heat sink
is installed, he assumed that the fraction of heat loss from the back
stays the same as the two-at-plate-case.
We believe that his assumption is not correct because a nned
heat sink is expected to transfer higher fractional heat rate from
the front as compared to a at plate, causing a lower fraction of
heat loss from the back. In our opinion, he should have used a
guard heater to compensate for the backside heat losses following
the same path in most of the similar studies in literature [1,2,6
10], or a heat sink structure that is symmetrical on both sides elim-
inating the need for insulation as it was done in [5].
Nevertheless, we believe that the temperature measurements in
[14] are still correct. Therefore, we calibrate our insulation thick-
ness to get temperature results matching the temperature mea-
surements in [14]. By this way, we replicate the experimental
case and at the same time we can estimate the backside conductive
loss and convective heat transfer rates from the front side of the
vertical plate-n heat sink directly from our simulations. Notice
that due to the aforementioned problem, in our opinion, the con-
vective heat transfer rates calculated in [14] are underestimates,
as shown in Table 4.
2.3. Optimum n spacing
There is an optimum n spacing maximizing the heat transfer
rate from the heat sink for each heater input power, thus, minimiz-
ing the average heat sink temperature. To determine this optimum
n spacing, the approach followed in both [8,14] is adopted. The
optimization approach consists of three steps: trying several n
spacing values sampled from the interval bounded by the small-S
and large-S limits, drawing asymptotes from both left (small-S lim-
it) and right (large S-limit), and identifying the spacing correspond-
ing to the intersection of the asymptotes as the optimum value,
S
opt
.
2.4. Other parameters affecting heat sink performance
In addition to the n spacing (S), the main geometric factors
affecting the performance of a plate-n heat sink are the heat sink
length (L) and the n height (H). The n thickness (t) is taken as
constant throughout the simulations because it only affects the
conduction resistance inside the ns; the selected thickness value,
which is 3 mm, is enough to guarantee that overall surface efcien-
cies are nearly equal to one. Similarly, the base thickness (d) is also
kept constant. Two different lengths (250 and 340 mm), three dif-
ferent n heights (5, 15 and 25 mm) and ve different n spacing
values (5.85, 8.8, 14.7, 32.4 and 85.5 mm) are investigated for ve
heater input powers (25, 50, 75, 100 and 125 W).
2.5. Acquisition of data from literature
The data from older works in literature, of which data is not
available, is obtained by digitizing the gures in each one of them.
Digitization error depends on the gure quality as well as the value
of the data point. Using the ground truth from recent works, we
estimated the digitization error as 1%. For Leung et al. data [79],
for which the ground truth is not available, we obtained an esti-
mate of the digitization error as 2.5% by comparing the digitized
x-axis values obtained from Q/WL versus S gures with the re-
ported values of n spacing (S) given as investigated range.
2.6. Statistical analysis of the data
While statistically analyzing the simulation data we consider
both the classical least squares approach and robust approach
using Huber and Tukey [25] norms for measuring deviations from
the t. Since all of the ts are nonlinear (power t), the squared
correlation coefcients R
2
, measuring goodness of t, are calcu-
lated using the following formula:
R
2
= 1
MSE(n 1)
R(y y)
(4)
where MSE is the mean squared error, n is the sample size, and y is
the average of y.
For all of the three tting approaches, we report both R
2
values
and 95% condence intervals for estimated parameters (Appendix
A).
2.7. Inclined heat sink
We directly use the approach (the model, the mesh and the
solution scheme) that is validated for the vertical case by varying
the direction of the gravitational acceleration (g) in order to create
the effect of inclination without changing any of the validated
model parameters. This approach makes the room to rotate with
the heat sink (see Fig. 2). Since the heat sink is very small com-
pared to the large air volume of the computational domain (the
Table 4
The present estimates (Sim) and the calculated ones in Ref. [14] of convective heat transfer rate (Q
c
) for L=250 mm.
Fin height H (mm) Fin spacing S (mm) Number of ns Convective heat transfer rate, Q
c
(W)
Q
in
= 25 W Q
in
= 50 W Q
in
= 75 W Q
in
= 100 W Q
in
= 125 W
Sim [14] Sim [14] Sim [14] Sim [14] Sim [14]
25 85.5 3 16.41 10.01 33.77 20.89 51.09 30.72 68.16 40.61 84.91 50.75
25 32.4 6 17.54 10.57 36.09 21.45 54.68 31.67 73.13 42.39 90.9 52.96
25 14.7 11 18.33 10.13 37.96 21 57.7 31.68 77.42 43.15 97.45 54.81
25 8.8 16 17.85 9.9 37.4 20.93 57.18 31.82 76.99 43.04 94.52 53.78
25 5.85 21 16.87 9.45 35.61 19.85 54.63 30.36 73.67 40.93 92.63 51.25
15 85.5 3 15.79 10.31 32.53 20.62 49.16 30.56 65.51 39.54 81.49 49.08
15 32.4 6 16.66 10.01 34.34 20.31 52.02 30.06 69.52 39.89 86.74 50.13
15 14.7 11 17.41 9.96 36.07 20.26 54.81 31.12 73.45 41.45 91.91 51.49
15 8.8 16 17 9.47 35.57 19.85 54.31 30.5 72.99 40.65 91.53 50.07
15 5.85 21 16.18 9.16 34.07 18.99 52.12 28.54 70.07 38.88 87.86 48.14
5 85.5 3 14.82 10.33 30.54 20.49 46.1 30.3 61.37 39.59 76.17 48.74
5 32.4 6 15.11 10.02 31.23 20.39 47.2 30.13 62.84 39.41 78.09 48.7
5 14.7 11 15.47 10.14 32.05 20.2 48.56 30.36 64.77 39.51 81.14 49.2
5 8.8 16 15.37 9.7 32.02 19.8 48.39 29.49 64.97 39.02 81 48.37
5 5.85 21 15 9.3 31.35 18.82 47.65 28.08 63.66 36.85 79.28 45.39
578 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
room) and placed at the center of one of the identical walls, and
since all of the walls of the cubical room are at the same uniform
temperature, changing the direction of g does not affect the air cir-
culation within the room away from the heat sink (in the far eld)
while affecting the air circulation only in the vicinity of the heat
sink (in the near eld), thus creating the desired effect. It is con-
rmed with the simulation results that the far eld air ow does
not change with the inclination and air always returns to the heat
sink at 20 C.
For the n spacing of 11.75 mm by equally spacing 13 ns and
xing the length to 250 mm, steady state solutions are obtained for
the inclination angles of 4, 10, 20, 30, 45, 60, 75,
+80, 85, 90 from the vertical. Three different heater input
power values (Q
in
) of 25, 75 and 125 W are investigated.
3. Results and discussion
Simulations are run until convergence of the temperature val-
ues at six thermocouple locations (marked in Fig. 3). During the
post processing, the average of these temperatures is taken as
the wall temperature (T
w
). Considering that the room (the compu-
tational domain) is very large compared to the heat sink and the
walls of the room are maintained at 20 C, ambient air temperature
(T
a
) is taken as 20 C. The heat sink base-to-ambient temperature
difference is dened as T = T
w
- T
a
. For evaluating the properties
of air, the lm temperature (T
f
) is dened as the average of T
w
and T
a
.
The convective heat transfer rate from the heat sink (Q
c
) is
determined by subtracting the radiative transfer rate of the heat
sink (Q
r
) from its total heat transfer rate, both being obtained from
the simulation results.
In Sections 3.1-3.3, we report our observations related to effects
of heat sink geometric parameters on the ow and temperature
elds. These observations serve towards further validating our
model and verifying the results. In Sections 3.4-3.6, we develop
an analytical close form for a correlation, suggest a set of correla-
tions and compare our correlations with the literature. Finally, in
Sections 3.7-3.13, we report our results for the inclined heat sink
case.
3.1. Heat sink length dependence
On all surfaces of the hot vertical plate-n heat sink, boundary
layers develop starting from the bottom, as in the case of natural
convection from a vertical at plate. Boundary layer thicknesses in-
crease throughout the length of the heat sink. Therefore, the aver-
age heat transfer coefcient depends on the heat sink length.
Shorter lengths give rise to higher heat transfer coefcients due
to lower boundary layer thicknesses. However, increasing the
length also increases the heat transfer area hence the convective
heat transfer rate. In Fig. 4, in order to observe the heat sink length
dependence, the heat convection rates per unit base area versus
the n spacing are plotted for two different lengths of heat sinks
of 15 mm n height. Here, the heater input power is taken as 75 W.
The length of 250 mm corresponds to higher convective heat
transfer rates per unit base area. The same effect is observed for
the remaining four heater input powers.
3.2. Fin height dependence
As the n height increases, interaction between each n surface
boundary layer and the base surface boundary layer changes.
Moreover, inlet of fresh air from the ambient to the heat sink chan-
nels along the length changes with the n height. To show this ef-
fect, a shorter heat sink of 100 mm length with 14.7 mm n
spacing is simulated with heater input power of 25 W. The velocity
vectors for H = 5, 15 and 25 mm are presented in Fig. 5. As the n
height increases, it is observed that distinguishably more air enters
from the open side of the heat sink throughout its length. For
H = 25 mm, air entrance continues all the way to the top of the heat
sink, whereas for H = 5 mm, entrance of air is limited to the very
bottom part of the heat sink.
The differences in ow structure directly affect the convective
heat transfer from the heat sinks. When the n height is increased
while keeping all the other parameters constant, the convective
heat transfer rate increases due to the increased extended surface
area. The cross-sectional temperature contours (on the xz plane)
for all of the three n heights are shown in Fig. 6. Here, L = 250
mm, S = 14.7 mm and Q
in
= 50 W are kept constant. All three sub-
gures have the same temperature scale. The heat sink with
H = 25 mm transferred more heat, as a result, has a signicantly
lower (about 30 C) temperature. Fig. 6 also conrms that keeping
the n thickness at 3 mm and the base thickness at 5 mm are good
choices for obtaining a uniform temperature distribution in the
heat sink. For all three values of H, the heat sink temperature is
uniform, both through the height of the ns and the width of the
heat sink base.
In Fig. 7, heat convection rates per unit base area versus the n
spacing for three different n heights are plotted for L = 250 mm
and Q
in
= 75 W. The following observations are made: Firstly,
5 mm ns are not very effective. Secondly, going from 15 mm to
25 mm n height does not give as much improvement as going
from 5 mm to 15 mm. Finally, for 5 mm n height, the effect of
n spacing on convective heat transfer rate is very weak.
3.3. Optimum n spacing
As depicted in Figs. 4 and 7, with increasing n spacing, heat
convection rate as a function of n spacing at rst increases up
to a maximum value, and then decreases monotonically, when
the heater input power is xed at 75 W. Similar behavior is also ob-
served when the heater power is xed at any of the four remaining
values. Moreover, for all ve input powers, the optimum n spac-
ing is around 12 mm.
In experimental studies, investigators are limited with their
experimental set up. In contrast, in CFD simulations, one can easily
vary geometric parameters. To determine the optimum n spacing,
six different n spacing values around 12 mm are tried and the
optimum value is determined by following the procedure that
was explained in Section 2.3.
The pair of plots in Fig. 8, respectively depict the wall tempera-
ture and the convective heat transfer rate as a function of n spac-
ing; in both cases, L = 250 mm and Q
in
= 25 W. In order to nd the
optimum n spacing values, a polynomial curve is tted to each
1080
1100
1120
1140
1160
1180
1200
1220
1240
0 20 40 60 80 100
Q
c
/
(
W
L
)


(
W
/
m
2
)
S (mm)
L=250 mm
L=340 mm
Fig. 4. Length dependence. Comparison of convective heat transfer rates per unit
base area for two heat sink lengths for the heater input power of 75 W and for the
n height of 15 mm.
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 579
data set. By differentiating the polynomials, the specic n spacing
values at which either the convection heat transfer is at maximum
or the wall temperature at minimum can be found. The optimum
n spacing values maximizing the convective heat transfer rate
and minimizing the wall temperature are presented respectively
in Tables 5 and 6. The differences between the two tables are
due to changing view factors with n spacing and related changes
in radiation heat losses. Since minimizing the wall temperature in-
cludes the changes in radiative transfer, it may be a better ap-
proach. For comparison, the optimum n spacing values obtained
in [14] are tabulated in Table 7 (note that the rst column data
is DT, instead of Q
in
). Table 7 results were obtained for the heat
sinks having S values of 5.85, 8.8, 14.7, 32.4 and 85.5 mm while
the S values for Table 6 results are 8.8, 9.6, 10.6, 11.8, 13.1 and
14.7 mm. Therefore, when the agreement is not very good, consid-
ering that the optimum is around 12 mm, our Table 6 results
should be trusted.
3.4. Analytical derivation for a correlation
Let us consider the buoyancy driven ow of air in the channel
between two adjacent ns. The coordinate system is the same as
Fig. 5. Fin height dependence. Speed vectors for 3 different n heights, H = 5 mm (top left), 15 mm (top right) and 25 mm (bottom).
580 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
the one used in the numerical model (Fig. 2). Flow velocity in
x-direction (along the width of the heat sink, perpendicular to n
side surfaces) is u = 0. The gravitational acceleration g is in y-direc-
tion (along the length); ow velocity in y-direction is v. z-direction
is along the n height; ow velocity in z-direction is w. The govern-
ing equations can be written as the following:
Continuity equation
@v
@y

@w
@z
= 0 (5)
x-momentum equation
@p
@x
= 0 (6)
y-momentum equation
qv
@v
@y
qw
@v
@z
=
dp
dy
l
@
2
v
@x
2

@
2
v
@y
2

@
2
v
@z
2
!
qgbDT (7)
z-momentum equation
qv
@w
@y
qw
@w
@z
=
dp
dz
l
@
2
w
@x
2

@
2
w
@y
2

@
2
w
@z
2
!
(8)
Energy equation
qc
p
v
@T
@y
qc
p
w
@T
@z
= v
dp
dy
w
dp
dz

k
@
2
w
@x
2

@
2
w
@y
2

@
2
w
@z
2
!
(9)
In order to obtain a form for dimensionless correlation, we need to
non-dimensionalize Eqs. (5)(9). The non-dimensional coordinates
are (X, Y, Z)
Fig. 6. Temperature distributions for three different n heights with the same temperature scale, H = 5 mm (top), 15 mm (middle) and 25 mm (bottom) for L = 250 mm,
S = 14.7 mm and Q
in
= 50 W.
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 581
X =
x
S
Y =
y
L
Z =
z
H
(10)
The non-dimensional velocity components are given as
V =
v
v
/
W =
w
w
/
(11)
The characteristic velocity in the y-direction is proportional to the
body force qg, the volumetric expansion coefcient b, the tempera-
ture excess T T
a
of air between two adjacent ns and the ambient
air, the channel cross sectional area in y-direction SH, and inversely
proportional to dynamic viscosity l. To non-dimensionalize y-
direction velocity, let us dene
v
/
=
qgb(T
w
T
a
)SH
l
(12)
To satisfy the continuity equation in non-dimensional form
v
/
@V
L@Y
=
w
/
@W
H@Z
i:e:
v
/
L
=
w
/
H
thus w
/
=
H
L

v
/
(13)
Non-dimensional temperature is dened as
/ =
T T
a
T
w
T
a
(14)
Non-dimensional pressure is dened as
P =
p
q(v
/
)
2
(15)
After putting these denitions in Eqs. (5)(9), we obtain the non-
dimensional form of the governing equations as the following:
Continuity equation
@V
@Y

@W
@Z
= 0 (16)
x-momentum equation
@P
@X
= 0 (17)
y-momentum equation
V
@V
@Y
W
@V
@Z
=
dP
dY

m
Hw
/
H
S

2
@
2
V
@X
2

@
2
V
@Y
2

H
L

2
@
2
V
@Z
2
" #

m
v
/
S
(18)
z-momentum equation
V
@W
@Y
W
@W
@Z
=
L
H

2
dP
dZ

m
Hw
/
H
S

2
@
2
W
@X
2

@
2
W
@Y
2

H
L

2
@
2
W
@Z
2
" #
(19)
Energy equation
m
Sw
/
V
@/
@Y
W
@/
@Z

=
gbL
c
p
V
dP
dY
W
dP
dZ

mk
qc
p
SHw
/2
H
S

2
@
2
/
@X
2

@
2
/
@Y
2

H
L

2
@
2
/
@Z
2
" #
(20)
We obtain dimensionless groups by rearranging the groups in Eqs.
(18)(20) as
1000
1050
1100
1150
1200
1250
1300
1350
0 20 40 60 80 100
Q
c
/
(
W
L
)


(
W
/
m
2
)
S (mm)
H= 25 mm
H=15 mm
H=5 mm
Fig. 7. Comparison of convective heat transfer rates per unit base area for three
heat sink heights for the heater input power of 75 W and heat sink length of
250 mm. Optimum n spacing is around 12 mm.
40
42
44
46
48
50
52
54
56
58
60
8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0
T
w
(
o
C
)
S(mm)
H=5mm
H=15mm
H=25mm
15.0
15.5
16.0
16.5
17.0
17.5
18.0
18.5
19.0
8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0
Q
c
(
W
)
S(mm)
H=5mm
H=15mm
H=25mm
Fig. 8. Determining optimum n spacing: minimizing wall temperature (top) and
maximizing convective heat transfer rate (bottom). Examples are for L = 250 mm
and Q
in
= 25 W.
Table 5
Optimum n spacing values for maximizing convection heat transfer rate.
Q
in
(W) Optimum n spacing, S
opt
(mm)
L = 250 mm L = 340 mm
H = 25 mm H = 15 mm H = 5 mm H = 25 mm H = 15 mm H = 5 mm
25 12.6 12.5 12.2 12.7 12.6 12.5
75 12.5 12.3 11.8 12.4 12.3 12.1
125 12.1 11.9 11.7 12 12.1 11.9
582 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
m
Hw
/
=
m
Sw
/

S
H

(21)
m
Sw
/
=
m
S
gb(TwTa)SH
m
h i
H
L

=
1
gb(TwTa)S
3
m
2
h i
H
S

H
L

(22)
mk
qc
p
SHw
/2
=
1
Sw
/
m

2
H
S

lcp
k

(23)
The dimensionless groups are the following:
P
1
=
H
L
P
2
=
S
H
P
3
=
[gb(TwTa)S
3
[
m
2
= Gr; P
4
=
(lcp)
k
= Pr and P
5
=
gb
/
L
cp
.
The rst term on the right-hand-side of Eq. (20) is not important
for small temperature differences (no signicant pressure gradient
in either direction) therefore P
5
is negligible. By combining P
1
, P
2
and P
3
we can dene a modied Grashof number:
Gr
/
= Gr
H
L

m
1
S
H

m
2
(24)
For a similar modication, the powers are previously suggested in
[10] as m
1
= 1/2 and m
2
= 1 , thus,
Gr
/
=
[gb(T
w
T
a
)S
3
[
m
2
( )
S
H

H
L

1=2
=
[gb(T
w
T
a
)S
4
[
m
2
(HL)
1
2
h i (25)
The average Nusselt number based on S should be a function of
Gr
/
and Pr; it takes the simple form:
Nu
S
=
hS
k
= C(Gr
/
Pr)
n
(26)
Leung and Probert [10] obtained a similar relation in which they
had an exponential n effectiveness term that we do not need
due to the high thermal conductivity of aluminum. They used
Table 6
Optimum n spacing values for minimizing average temperature.
Q
in
(W) Optimum n spacing, S
opt
(mm)
L = 250 mm L = 340 mm
H = 25 mm H = 15 mm H = 5 mm H = 25 mm H = 15 mm H = 5 mm
25 11.6 11.5 11.1 11.7 11.6 11.4
75 11.3 10.8 10.8 11.4 11.5 11
125 11.4 10.6 10.5 11.3 11.4 11
Table 7
Optimum n spacing values from Ref. [14].
DT (K) Optimum n spacing, S
opt
(mm)
L = 250 mm L = 340 mm
H = 25 mm H = 15 mm H = 5 mm H = 25 mm H = 15 mm H = 5 mm
50 11 10.9 11.9 11.8
75 10.9 10.8 10.7 11.8 11.7 11.6
100 10.8 10.7 10.6 11.7 11.6 11.5
125 10.7 10.6 10.5 11.6 11.4 11.4
Fig. 9. Analysis of entire data for the power function ts in the form of Eq. (26).
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 583
n =
1
2
for Gr
/
Pr < 250
1
3
for 250 < Gr
/
Pr < 10
6
(
(27)
3.5. Average Nusselt number correlation
Using the form, Eq. (26), we process our entire data in the two
respective ranges in Eq. (27). Based on power curve ts (Fig. 9), we
obtained C constant coefcients of 0.0929 and 0.2413 which are
the mid points of the respective 95% condence intervals of
0.09290.0016 and 0.24130.0016, in Gr
/
Pr < 250 and 250 < Gr
/
Pr
< 10
6
ranges, respectively. The squared correlation coefcients
(R
2
) of the ts are 0.9607 and 0.9932 for the rst and second
ranges, respectively, that is, the suggested form ts very well to
our data. Thus, we suggest the following correlations:
Nu
S
= 0:0929(Gr
/
Pr)
1=2
for Gr
/
Pr < 250 (28a)
Nu
S
= 0:2413(Gr
/
Pr)
1=3
for 250 < Gr
/
Pr < 10
6
(28b)
In addition to the two ranges in Eq. (28b), we have a third range
of data points where Gr
/
Pr > 10
6
(marked with triangles in Fig. 9).
The data in this nal region corresponds to a n spacing of
85.5 mm. Because the n spacing is very large, the ns act like
Fig. 10. Analysis of data in 250 < Gr
/
Pr < 10
4
range. The results for H = 5 mm (marked with h) depends on L; all of the points above the dashed line are for L = 250 mm, while
the ones on the other side are for L = 340 mm. No L dependence is observed for H = 15 and 25 mm.
Fig. 11. Comparison of Nu
S
parity plots using present data for present Eq. (29), Elenbaas [5] correlation, and Leung and Probert [10] correlation.
584 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
individual plates. As a result, their height and length become
important. Observe that the data appears to be both n height
and length dependent. Note, however, that congurations with
small numbers of ns, hence, large n spacing values may not be
practical choices; thus, we do not suggest any correlation for that
region.
In the same gure (Fig. 9), the entire data that is clustered in
10
4
10
6
range of Gr
/
Pr is obtained for the n spacing of 32.4 mm.
Again, due to this large n spacing, the ns continue to act like
individual plates; thus, the height and length dependence of Nu
S
does still exist. The practical choices of n spacing are expected
to fall in Gr
/
Pr < 10
4
range.
In Fig. 10, the data restricted to 250 < Gr
/
Pr < 10
4
range is further
analyzed. It is observed that the ns of H = 5 mm show heat sink
length dependence; all of the points (marked with h) above the
dashed line are for L = 250 mm, while the ones on the other side
are for L = 340 mm. This very small n height is not practically very
useful. (Nevertheless, a correlation where C = 0.2301 and n = 1/3
can be obtained). In contrast, H = 15 and 25 mm data in the same
Gr
/
Pr range (marked with h) does not show dependence to either
the n height or the length. Additionally, this n height range is
very practical for engineering applications. Therefore, it may be
worth considering the range 250 < Gr
/
Pr < 10
4
, H P15 mm
separately. Fitting power curve to the data in this range,
Fig. 12. Comparison of Nu
S
parity plots for present Eq. (29), Elenbaas [5] correlation, and Leung and Probert [10] correlation. Data from Leung et al. [79] is used.
Fig. 13. Comparison of Nu
S
parity plots for present Eq. (27), Elenbaas [5] correlation, and Leung and Probert [10] correlation. Data from Starner and McManus [2] is used.
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 585
C = 0.252 is obtained as the mid point of the 95% condence inter-
val 0.25200.0026. The squared correlation coefcient R
2
is 0.9705.
Notice that C value here is higher than the C value in Eq. (28b).
Interestingly, for the data obtained for the heat sinks with
S = S
opt
= 11.75 mm and H = 15 and 25 mm, the constant coefcient
C is determined to be 0.2543, with the 95% condence interval of
0.25430.0056; R
2
is 0.9295. That is, also around S
opt
, C value is
higher than the C value in Eq. (28b).
As a result, for the smaller range of Gr
/
Pr, we recommend the
following alternative:
Nu
S
= 0:252(Gr
/
Pr)
1=3
for 250 < Gr
/
Pr < 10
4
; H _ 15 mm (29)
3.6. Comparison with experimental literature
A good way to judge the accuracy of a suggested correlation is
to compare Nusselt numbers calculated directly from the model
parameters and the data with those calculated from the correla-
tion. In Figs 1113, we provide several parity plots comparing
our correlation with the previous vertical correlations due to Elen-
baas [5] and Leung and Probert [10]. In each gure, there are 3 par-
ity plots, corresponding to the two previously suggested ones and
ours. The x-axis values depict direct results whereas the y-axis val-
ues depict the results from the respective correlations. For each of
the three parity plots to be compared, we also draw their best t-
ting line to give an idea of the overall trend for each correlation. An
ideal correlation is expected to lie along the diagonal (y = x), called
parity line. To rule out the effect of the choice of data, we repeat the
procedure using two more data sets, in Figs. 12 and 13. The data set
used in Fig. 12 is due to Leung et al. [79] and the one in Fig. 13 is
due to Starner and McManus [2].
For the rst data set (Fig. 11) our correlation is almost on the
parity line, the diagonal. Elenbaas [5] correlation is also quite good,
with its trend line deviating slightly from the parity line. The trend
line for the Leung and Probert [10] correlation, on the other hand,
signicantly deviates from the parity line. Note that the considered
data fall within the range of all three correlations 250 < Gr
/
Pr < 10
4
.
When the procedure is repeated for Leung et al. data [79], trend
lines for the parity plots of all three correlations deviate from the
parity line almost the same amount (Fig. 12). Whereas Leung and
Probert correlation overestimates, both our correlation and Elen-
baas one underestimate.
When the procedure is repeated for Starner and McManus [2]
data, which falls in Gr
/
Pr < 250, the trend line of the parity plot
for our correlation (Eq. (28a)) is almost on the parity line
(Fig. 13). Whereas Elenbaas correlation underestimates, Leung
and Probert one overestimates.
3.7. Upward inclination ow structure
When a hot plate-n heat sink is inclined from the vertical in
upward facing orientation, since the wall temperature is higher
Fig. 14. Temperature contours for upward inclined heat sink at 45, 75 and 90 (from top to bottom) for Q
in
= 75 W and H = 25 mm (side views of the heat sink).
586 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
than the ambient air temperature, cooler air entering from the bot-
tom side or (at inclinations close to horizontal) from both the top
and bottom sides of the heat sink heats up. Consequently, a plume
of air rises from the heat sink at a location that depends on the
inclination angle. Note that there is also cool air entrance along
the heat sink from the open side of the channel formed by the con-
secutive ns. Since the driving force behind the phenomenon is the
buoyancy force, regardless of the inclination angle, a plume rises in
the opposite direction of the gravitational acceleration, as can be
seen in Fig. 14 depicting uid temperature contours at the inclina-
tion angles of 45, 75 and 90 for the heater power of 75 W
and n height of 25 mm. The heat sink n is shown as a wireframe
in order to make uid velocity in the channel between two adja-
cent ns visible. The heat sink base is also visible due to its higher
temperature. When the subgures are compared, it is observed
that the base temperature increases with the increasing angle,
Fig. 15. Streamlines for (a) the vertical (0) and upward inclined heat sinks at (b) 45, (c) 60, (d) 75, (e) 85 and (f) 90 for Q
in
= 75 W and H = 25 mm (side views of
the heat sink and insulation). Flow separation locations are marked with red rectangles. (For interpretation of the references to colour in this gure legend, the reader is
referred to the web version of this article.)
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 587
Fig. 16. Streamlines for (a) the vertical and downward inclined heat sinks at (b) 45, (c) 60, (d) 75, (e) 80, (f) 85 and (g) 90 for Q
in
= 75 W and H = 25 mm (side views of the
heat sink and insulation). Flow separation locations are marked with red rectangles. (For interpretation of the references to colour in this gure legend, the reader is referred
to the web version of this article.)
588 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
indicating lower heat transfer rates at higher angles. At 45, the
plume is at the top end of the heat sink (Fig. 14a); at 75, it rises
from a location on the heat sink; and at upward horizontal (90),
it is at the center of the heat sink.
At each inclination angle, the density of the uid changes with
the uid temperature. As a result, different buoyancy driven ow
structures are formed. Such ow structures at the inclination an-
gles of 0, 45, 60, 75, 85 and 90 for Q
in
= 75 W,
H = 25 mmare shown in Fig. 15. All of the subgures have the same
scale, which is shown in Fig. 15a.
As inclination increases, ow velocities at the exit (top end) of
the heat sink reduce. As apparent from the gures, e.g., from
Fig. 15a and b, the color changes from red to orange. In Fig. 15c
and d, ow separation locations are marked with red rectangles.
It is observed that for the upward case the separation location
starts to move from the tip towards the center after 60 of
inclination. At 90, the ow is symmetric around the center of
the heat sink, placing the separation location to the center.
3.8. Downward inclination ow structure
When a hot plate-n heat sink is inclined from the vertical in
downward direction, since the wall temperature is higher than
the ambient air temperature, cooler air entering into the channel
between two consecutive ns heats up along the heat sink and
then escapes from the top side or (at inclinations close to horizon-
tal) from both the top and bottom sides of the heat sink. Such ow
structures at the inclination angles of 0, +45, +60, +75, +80,
+85 and +90 for Q
in
= 75 W, H = 25 mm are shown in Fig. 16.
In the downward horizontal orientation (+90) (see Fig. 16g), air
separates at the mid point of the heat sink length. As the inclina-
tion angle decreases, the separation point moves towards the lead-
ing edge of the heat sink (see the location for +85 in Fig. 16f
marked with the rectangle). After reaching the leading edge, at
Fig. 17. Variations of surface average temperature (top), convective heat transfer
rate (middle) and radiative heat transfer rate (bottom) with respect to inclination
angle, for all of the three n heights and Q
in
= 125 W in upward inclinations.
Fig. 18. Variations of surface average temperature (top), convective heat transfer
rate (middle) and radiative heat transfer rate (bottom) with inclination angle for all
of the three n heights and Q
in
= 125 W in downward inclinations.
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 589
around +80 (Fig. 16e), the separation point stays there for smaller
inclination angles, hence, no separation along the heat sink is ob-
served for h < 80 (Fig. 16ad). These observations agree with the
observations of Mittelman et al. [1].
The separation location and whether it is located on the heat
sink or at the tip are important for the performance of the heat
sink. We will revisit them in Section 3.12.
3.9. Inclination angle dependence of T
w
, Q
c
and Q
r
Variations of surface average temperature, convective heat
transfer rate and radiative heat transfer rate with respect to incli-
nation angle are shown for all of the three n heights and
Q
in
= 125 W in upward and downward inclinations in Figs 17 and
18, respectively.
From Fig. 17, we make the following observations for upward
inclinations of the heat sink:
v Since the radiative heat transfer rate (Q
r
) depends on the fourth
power of the average heat sink temperature (T
w
), both Q
r
and T
w
have similar dependence to the inclination angle.
v Starting from vertical up to 30, the changes in T
w
, Q
c
and Q
r
are very small; after 30, Q
c
decreases until reaching a
minimum.
v T
w
, Q
c
and Q
r
at 85 are very close to the values at 90.
From Fig. 18, we make the following observations for down-
ward inclinations of the heat sink:
v Q
c
stays almost the same in 0-30 range, though at angles very
close to vertical, it may be slightly larger than it is in the vertical
case. This can be explained with the thinning of the boundary
layer in small downward inclinations.
v T
w
, Q
c
and Q
r
changes monotonically until the inclination gets
very close to +90 (downward horizontal).
v Q
c
from downward horizontal heat sink is always signicantly
smaller than Q
c
from both vertical and upward horizontal heat
sinks.
3.10. Fin height dependence
Variation of convection heat transfer rate with n height in up-
ward inclination is shown for Q
in
= 125 W in the middle sub-gure
of Fig. 17.
From Fig. 17, we make the following n height related observa-
tions for upward inclinations of the heat sink:
v With increasing H, the difference in Q
c
between vertical and
upward horizontal heat sinks gets smaller.
v As the n height increases, angle dependence of T
w
, Q
c
and Q
r
decrease. This can be attributed to vertical plate like behavior
of ns with large height protruding from the horizontal base.
v As the n height increases, the gain from further increasing the
n height decreases.
v After 60, the angle at which minimum of Q
c
is observed
depends on the n height (H).
v For H = 5 mm, the minimum Q
c
is attained at 90 (upward hor-
izontal). The angles at which a minimum is attained get smaller
with increasing H.
Variation of the convection heat transfer rate with respect to
the n height in downward inclination for Q
in
= 125 W is shown
in the middle sub-gure of Fig. 18. The rst three observations gi-
ven above (based on Fig. 17) related to n height dependence for
upward inclinations are also valid for downward inclinations.
3.11. Validity range of modied vertical case correlations
When a bare at plate is inclined from the vertical, it has been
observed that the correlations for vertical case are still valid up to
45 by only replacing Ra with Ra cosh [26]. Behind this approach,
there is an assumption that the ow structure stays the same,
and the only change is in the body force, where the effective body
force becomes qgcosh. With the same rationale, the vertical case
correlation for plate-n heat sinks with the same modication is
expected to be valid at small inclination angles, as long as the ow
structure stays the same.
Modifying the correlations suggested in Section 3.5 for the ver-
tical case by multiplying Gr
/
with cosh, we obtain
Nu
S
= 0:0929(Gr
/
Pr cos h)
1=2
for Gr
/
Pr cos h < 250 (30)
Nu
S
= 0:2413(Gr
/
Pr cos h)
1=3
for 250 < Gr
/
Pr cos h < 10
6
(31)
where Nu
S
= (hS)=k is the Nusselt number based on n spacing S, Pr
is the Prandtl number, Gr
/
= [gbDTS
4
[=[m
2
(LH)
0:5
[ is the modied
Grashof number, and h is the inclination angle measured from the
vertical.
Recall that in Eq. (29), an alternative correlation valid for H = 15
and 25 mm in a narrower Gr
/
Pr range covering all practical S values
has been suggested. Modifying it yields
Fig. 19. Comparison between convection heat transfer rates obtained by simulation
and vertical correlation, Eq. (32), in upward (top) and downward (bottom)
inclinations for H = 15 and 25 mm.
590 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
Nu
S
= 0:252(Gr
/
Pr cos h)
1
3
for 250 < Gr
/
Pr cos h < 10
4
(32)
In Fig. 19, we compare the estimation of convection heat transfer
rate obtained using the Nu
s
correlation given by Eq. (32) to the rate
given by simulation. Simulation data are marked with , the rates
based on Eq. (32) with h. Both upward (top gure) and downward
(bottom gure) inclinations are examined. Three different trends in
each of the subgures correspond to three different heater power
values, i.e. 125, 75 and 25 W. Only H = 15 and 25 mm cases are cov-
ered, as Eq. (32) is valid for these heights.
It is observed (top subgure) that the two rates, i.e. simulated
and estimated from correlation, agree very well for the upward
inclination angles ranging 060; for those larger than 60, using
Eq. (32) does not seem suitable. On the other hand, for the down-
ward case (bottom subgure), the rates agree in a wider inclination
angle range of 080. That is, Eq. (32) is observed to be valid from
60 to +80.
Compared to the narrower range of similar correlations for bare
at plates, the surprisingly wide range of 60 to +80 may be
attributed to the ow channels in between the consecutive ns,
which are keeping the structure of the ow two dimensional for
a very wide range of inclinations.
To better judge the validity of Eq. (32) in the above mentioned
angle ranges, the best tting power curves for the inclined data in
the respective ranges are shown in Fig. 20. The estimated constant
coefcients, which are respectively 0.2585 and 0.2538 for the
upward and downward inclinations, are very close to 0.252, the
coefcient of Eq. (32). This supports the validity of Eq. (32) in the
respective ranges. The constant coefcients are the mid points of
the respective 95% condence intervals of 0.25850.0031 and
0.25380.0043. The squared correlation coefcients (R
2
) of the ts
are 0.9389 and 0.9484 for the upward and downward inclinations,
respectively. Note that in Section 3.5 for the vertical case with
S = S
opt
= 11.75 mm, we determined the constant coefcient as
0.25430.0056. Since both 0.2585 and 0.2538 fall within the 95%
condence interval range, we conclude that S
opt
does not change
with inclination in 60 < h < +80 interval.
We further verify our suggested correlations using experimen-
tal results by Mittelmann et al. [1] for downward case and Starner
and McManus [2] for upward inclination of 45; these constitute all
of the available inclined case data in the literature. The Mittelman
et al. [1] data is in a very good agreement with our correlations,
both Eqs. (31) and (32). See Fig. 21 where the correlation curve is
plotted together with the data from both Mittelman et al. and
our simulations. The mean absolute error of Mittelman et al. data
from Eq. (32) is only 5.8%. Such a good match with this experimen-
tal data indicates the validity of Eq. (32) in downward 60-80
range. An equivalent verication for Starner and McManus [2] data
is depicted in Fig. 22. This experimental data falls in (Gr
/
Pr cos
h < 250) range. Therefore, the comparison is made with Eq. (30).
Although the number of available data points is small, the data
cover most of the range. The data agree very well with the correla-
tion, yielding a mean absolute error of 8.8%. Note that both Mittel-
man et al. [1] and Starner and McManus [2] data are obtained for
very different geometric parameters and conditions than both each
Fig. 20. Upward (top) and downward (bottom) curve ts to simulation results for
angles in 060 and 080 ranges, respectively, and for H = 15 and 25 mm.
Fig. 21. Simulation results and Mittelman et al. [1] results in downward 6080
inclination angle range plotted together with Eq. (32). Agreement is very good.
Fig. 22. Starner and McManus [2] results at 45 plotted together with our
correlation in Gr
/
Pr cos h < 250 range, Eq. (30).
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 591
other
/
s and ours. Thus, their agreement with our correlation set is
an indication of the generality of our suggested correlations.
3.12. Importance of ow separation location
For inclined at plates, Fujii and Imura [26] have shown the
validity of vertical case correlations with the cosh modication
up to 45. Later, Bejan and Kraus [27] recommended the cosine
modication up to 60. For plate-n correlations, however, there
is no work examining validity of the cosine modication.
In the case of upward inclinations of our plate-n heat sink, we
observe that the modied vertical correlations are valid up to 60.
For the downward inclinations, the validity range is even larger.
This difference between upward and downward inclinations de-
serves a discussion.
Let us consider one of the channels bounded by two adjacent
parallel plate ns and the base of the heat sink. The side of the
channel facing the room is open. It was observed that in the verti-
cal case, there is air inlet to the channel from the open side
throughout the length (Fig. 15a). This cooler air inlet enhances
the convective heat transfer rate. As shown in Fig. 15b, at 45
inclination, there is still an inlet of cool air from the open side of
the channel, but the hot air exit from the upper tip of the channel
is not as strong as it is in the vertical case. As the inclination angle
increases in upward inclinations, we observe that hot air starts to
exit from the open surface of the channel instead of the upper
tip. It is observed that at 60, ow separation is about to occur
(Fig. 15c). At 75 inclination (Fig. 15d), the exit of hot air is mainly
from the open side. This indicates a big change in ow structure,
explaining why the cosine modication is valid up to 60 inclina-
tion. As the inclination further increases, hot air exit (separation)
location moves downward from the upper tip, nally reaching to
the center of the channel at 90 (Fig. 15f). The ow structure after
60 is more similar to the ow structure in the upward horizontal
than the one in the vertical.
In downward inclinations, however, the phenomenon is differ-
ent (Fig. 16). The ow is bounded by the heat sink at the top. When
the heat sink is inclined from the vertical, air still enters the chan-
nels between the ns from the bottom and the open side of the
heat sink (Fig. 16b). As the inclination increases, the entrance from
the bottom gets weaker (Fig. 16c). A backward circulation forms on
the insulation block around +75 inclination (marked by the rect-
angle in Fig. 16d). This is the start of a ow separation, but it is
not yet occurring on the heat sink. At +80 inclination, the separa-
tion occurs at the bottomtip of the heat sink (Fig. 16e). On the right
side of the red triangle marking the ow separation, the ow is to
the right and against the buoyancy force component that is parallel
to the heat sink. For the separation to occur, the ow should over-
come this force. This explains the delayed start of the separation
(compared to the upward inclination case). As the inclination fur-
ther increases, the ow separation location moves upward from
the bottom tip, nally reaching to the center of the channel at
+90 (Fig. 16g). The ow structure after +80 is more similar to
the ow structure in the downward horizontal than the one in
the vertical.
3.13. Inclination angles signicantly deviating from the vertical
We have demonstrated the applicability of the vertical correla-
tion for inclination angles up to 80 for the downward case and up
to 60 for the upward case.
For inclination angles signicantly deviating from the vertical,
the modied vertical correlations, Eqs. (30)(32), do not seem to
be valid. This naturally raises the question whether the ranges
6090 and 8090, respectively for the upward and downward
cases, can be explained by horizontal models. Because our efforts
are concentrated on generating a detailed validated model for
the vertical case, we do not have enough data at 90 and +90
to suggest horizontal case correlations. Our limited data, neverthe-
less, is in agreement with the Mittelman et al. [1] data. The data
sets align very well, Mittelman et al. data covering a higher Gr
/
Pr
range.
4. Conclusion
In the rst part of the paper, we studied the natural convection
from rectangular cross section vertical plate ns on a vertical base,
after modeling a recent experimental set-up and conditions in [14].
We validated our numerical model by comparing the simulation
results with the experimental ones. This validation forms a basis
for using the model in inclined heat sink studies.
By visualizing the simulation data, we observed airow struc-
tures within the channels formed by the ns as well as in the vicin-
ity of the heat sink. The relation between entrance of cooler
ambient air to the channels (which helps to increase convective
heat transfer) and the n height is observed from the simulation
data. By visualizing temperature distributions, effectiveness of
the selected base and n thicknesses is demonstrated and the ef-
fect of n height on the performance of the heat sink is observed.
Especially at values of n spacing signicantly higher than the
optimum one, the n height affects the heat sink performance.
The n spacing is determined to be an important parameter in heat
sink performance. All of these observations are consistent with the
experimental literature.
Using our simulation results, we suggest the correlation in Eq.
(28b), via which it is possible to obtain the average Nusselt number
based on the n spacing from a modied form of Grashof number
including all the geometric parameters of the heat sink. For a tigh-
ter range, 250 < Gr
/
Pr < 10
4
, we suggest Eq. (29). The parity plots
depict the accuracy of the suggested correlations.
In the second part of the paper, we investigated the steady-state
natural convection from hot heat sinks with parallel plate ns pro-
truding from an inclined base in both upward and downward fac-
ing orientations. The examined inclination angle range includes ten
angles in each orientation. As a result, the angle dependence of the
phenomenon is thoroughly investigated.
It is observed that, within small inclinations from the vertical in
both directions, the inclination does not reduce the convection
heat transfer rate. The heat transfer rate stays almost the same.
It even slightly increases in very small downward inclinations,
due to thinner boundary layer.
For upward facing inclinations, we observed that the ow sep-
aration location plays an important role. Up to 60 from the verti-
cal, the separation location stays at the top edge leading to a single
ow direction in the channels between the ns. The effective body
force is qgcosh. The set of vertical case correlations, Eqs. (28b) and
(29), remains valid up to 60 by multiplying the body force term
with cosh. For downward facing inclinations, the ow separation
does not occur until +80, thus, the effective body force remains
qgcosh from vertical up to 80. That is, the modied vertical case
correlation remains valid in a very wide inclination angle range.
Based on these observations, we suggest Eqs. (30)(32) in
60 6 h 6 +80.
Finally, we indirectly observed that the optimum n spacing
does not signicantly change with inclinations in
60 6 h 6 +80 interval, indicating that optimally designed verti-
cal heat sinks can be inclined within this range without being af-
fected by the n spacing.
The very good agreement of our correlations with the literature
data further validates the numerical model and supports our
claims. Since the investigated ranges of parameters are suitable
592 I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593
for electronic device cooling, the suggested correlations have a
practical use in electronics cooling applications.
Acknowledgment
The work was completed and the manuscript written during a
sabbatical stay of the rst author at the Institute for Nanoelectron-
ics, Technische Universitt Mnchen. The rst author thanks Paolo
Lugli for his hospitality during that stay. He also thanks DAAD-
TUM fellowship for a partial nancial support. Both of the authors
thank Amas Ullamann for providing their experimental data from
[1] and the anonymous reviewers for their constructive comments.
Appendix A
In Table A1, we present the obtained constant coefcients C,
95% condence intervals and R
2
values for the least squares, Tukey
and Huber estimates. The selected estimates are shown in bold
face.
References
[1] G. Mittelman, A. Dayan, A. Dado-Turjeman, A. Ullmann, Laminar free
convection underneath a downward facing inclined hot n array, Int. J. Heat
Mass Transfer 50 (2007) 25822589.
[2] K.E. Starner, H.N. McManus, An experimental investigation of free convection
heat transfer from rectangular n arrays, J. Heat Transfer 85 (1963) 273278.
[3] I. Tari, Passive cooling assembly for at panel displays with integrated high
power components, IEEE Trans. Consumer Electron. 55 (2009) 17071713.
[4] I. Tari, F.S. Yalcin, CFD analyses of a notebook computer thermal management
system and a proposed passive cooling alternative, IEEE Trans. Components
Packag. Technol. 33 (2010) 443452.
[5] W. Elenbaas, Heat dissipation of parallel plates by free convection, Physica 9
(1942) 665671.
[6] J.R. Welling, C.V. Wooldridge, Free convection heat transfer coefcient from
rectangular n arrays, J .Heat Transfer 87 (1965) 439444.
[7] C.W. Leung, S.D. Probert, Thermal effectiveness of short protrusion rectangular,
heat exchanger ns, Appl. Energy 34 (1989) 18.
[8] C.W. Leung, S.D. Probert, M.J. Shilston, Heat exchanger: optimal separation for
vertical rectangular ns protruding from a vertical rectangular base, Appl.
Energy 19 (1985) 7785.
[9] C.W. Leung, S.D. Probert, Natural-convective heat exchanger with vertical
rectangular ns and base: design criteria, in: IMechE Proceedings, vol. 201,
1987, pp. 365372.
[10] C.W. Leung, S.D. Probert, Heat-exchanger performance. Effect of orientation,
Applied Energy 33 (1989) 235252.
[11] A. Gven, H. Ync, An experimental investigation on performance of
rectangular ns on a vertical base in free convection heat transfer, Heat
Mass Transfer 37 (2001) 409416.
[12] F. Harahap, H. Lesmana, Measurements of heat dissipation from miniaturized
vertical rectangular n arrays under dominant natural convection conditions,
Heat Mass Transfer 42 (2006) 10251036.
[13] B. Yazicioglu, H. Ync, Optimum n spacing of rectangular ns on a vertical
base in free convection heat transfer, Heat Mass Transfer 44 (2007) 1121.
[14] B. Yazicioglu, Performance of rectangular ns on a vertical base in free
convection heat transfer, MS thesis, Middle East Technical University, Ankara,
2005.
[15] A. Dayan, R. Kushnir, G. Mittelman, A. Ullmann, Laminar free convection
underneath a downward facing hot n array, Int. J. Heat Mass Transfer 47
(2004) 28492860.
[16] Y.M. Ko, C.W. Leung, S.D. Probert, Steady-state free-convective cooling of heat
exchangers with vertical rectangular ns: effect of n material, Appl. Energy
34 (1989) 181191.
[17] H. Ync, G. Anbar, An experimental investigation on performance of
rectangular ns on a horizontal base in free convection heat transfer, Heat
Mass Transfer 33 (1998) 507514.
[18] H. Ync, M. Mobedi, A three dimensional numerical study on natural
convection heat transfer from short horizontal rectangular n array, Heat Mass
Transfer 39 (2003) 267275.
[19] S. Baskaya, M. Sivrioglu, M. Ozek, Parametric study of natural convection heat
transfer from horizontal rectangular n arrays, Int. J. Thermal Sci. 39 (2000)
797805.
[20] H.G. Yalcin, S. Baskaya, M. Sivrioglu, Numerical analysis of natural convection
heat transfer from rectangular shrouded n arrays on a horizontal surface, Int.
Commun. Heat Mass Transfer 35 (2008) 299311.
[21] L. Dialameh, M. Yaghoubi, O. Abouali, Natural convection from an array of
horizontal rectangular thick ns with short length, Appl. Thermal Eng. 28
(2008) 23712379.
[22] ANSYS Fluent 12.0 Users manual, ANSYS, Inc., 2008.
[23] W.H. McAdams, Heat Transmission, McGraw-Hill, New York, 1954.
[24] S.W. Churchill, H.H.S. Chu, Correlating equations for laminar and turbulent free
convection from a vertical plate, Int. J. Heat Mass Transfer 18 (1975) 1323
1329.
[25] W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterling, Numerical Recipes:
The Art of Scientic Computing, Cambridge University Press, New York, 2007.
Section 15.7. Robust Estimation.
[26] T. Fujii, H. Imura, Natural-convection heat transfer from a plate with arbitrary
inclination, Int. J. Heat Mass Transfer 15 (1972) 755767.
[27] A. Bejan, A.D. Kraus, Heat Transfer Handbook, John Wiley and Sons Inc.,
Hoboken, New Jersey, 2003.
Table A1
Coefcients, condence intervals and R
2
values of power curve ts. Selected ts are marked using bold face.
Fig. Eq. Least squares Tukey Huber
C 95%CI R
2
C 95%CI R
2
C 95%CI R
2
Fig. 9 (28a) 0.0929 0.0016 0.9607 0.0922 0.0015 0.9674 0.0926 0.0015 0.9661
(28b) 0.216 0.0046 0.9456 0.2413 0.0016 0.9932 0.229 0.0025 0.9839
Fig. 10 0.2301 0.0049 0.9293 0.2308 0.0053 0.9171 0.2314 0.0054 0.9142
(29) 0.252 0.0026 0.9705 0.2518 0.0027 0.9664 0.2518 0.0027 0.9675
I. Tari, M. Mehrtash/ International Journal of Heat and Mass Transfer 56 (2013) 574593 593