Thermal Science and Engineering Progress 17 (2020) 100484

Contents lists available at ScienceDirect

Thermal Science and Engineering Progress

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

Optimizing rectangular fins for natural convection cooling using CFD T

a,⁎ b a
R.C. Adhikari , D.H. Wood , M. Pahlevani
a
Department of Electrical and Computer Engineering, University of Calgary, Calgary, Alberta, Canada
b
Department of Mechanical and Manufacturing Engineering, University of Calgary, Calgary, Alberta, Canada

ARTICLE INFO ABSTRACT

Keywords: Efficient heat transfer from finned-surfaces under natural convection is a critical consideration in the design of
Natural convection cooling heat sinks for microelectronics and thermal devices. Previous experimental studies on natural convection from
Rectangular fins rectangular fins on horizontal surfaces have shown that maximum heat transfer per unit base area occurs within
Heat transfer and flow patterns a narrow range of fin spacing in tall fins, and over a wider range of fin spacing in shallow fins. However, the heat
Optimum fins
transfer and flow characteristics of optimum fin configurations and the combined effects of fin geometrical
Dynamic-Q algorithm
parameters have not been sufficiently scrutinized in the literature. This work is primarily focused on char-
acterizing the optimum heat transfer and flow patterns using three-dimensional, steady-state, laminar, conjugate
heat transfer simulations, which serve as the fundamental basis for determining optimum fins. The main fin
design parameters are the fin spacing, height, and length. Multi-parametric computational fluid dynamics (CFD)
optimization was conducted to determine the combined effects of fin spacing, height, and length on heat
transfer. To find a global optimum fin design, we performed a multi-parametric optimization using the dynamic-
Q algorithm, rather than through a single parameter variation as is commonly done in previous studies. It was
found that the fin spacing, height, and length strongly affect the flow through the fin channel ends, and hence the
temperature gradient and heat transfer rate from the fin surfaces. The results demonstrate that the dynamic-Q
optimization could efficiently find the global optimum CFD solutions for the design objectives: the maximum
heat transfer per unit base area and the maximum heat transfer per unit base area and fin weight. The results are
validated through the experimental results available for typical fin designs, as well as an experimental mea-
surement of an example fin design in this work.

1. Introduction only lead to lower heat transfer than the surfaces without fins but also
increase the footprint area and weight of the finned-surfaces
Efficient cooling is one of the key design challenges for thermal and [3,4,7–11].
electronics devices due to increasing miniaturization. Natural convec- Whereas fin designs may vary widely depending upon the specific
tion-cooling through finned-surfaces is the most desirable technique as device design and applications, the most common configuration is the
it does not require any moving parts such as fans or pumps, which are rectangular array of straight fins on a horizontal surface as depicted in
the points of failure and cost. Natural convection occurs due to buoy- Fig. 1. Natural convection from such fins has been studied extensively,
ancy forces as a result of differential heating of air around the fin sur- and various correlations have been suggested for estimating the heat
faces and heat transfer from the finned-surface across the thermal transfer rate [3,4,12,13]. However, these correlations do not often
boundary layer [1,2]. A circulatory thermal flow field develops around converge to consistent results, particularly when determining the op-
the finned-surfaces as a result of downward flow of denser air and timum design of fins [4,7,14]. For instance, Jones and Smith [4]
upward flow of warm air. The upward flow results in a vertical con- showed that the empirical correlations recommended by Harahap and
verging draft of air, commonly referred as a “chimney” or “plume”, on MacManus [12] over-predicted by more than 94% their measured heat
which the natural convection heat transfer relies [1,3–6]. As a result, transfer for similar fin designs. Jones and Smith [4] argued that the use
the flow field exhibits a very complex three-dimensional fluid motion of n in the correlation, and consideration of fin length (L) as the char-
with heat transfer. Therefore, deeper understanding of the flow and acteristic dimension, instead of fin spacing (s), which was shown to be
heat transfer characteristics around the finned-surfaces is crucial to important in their study, could be the main reason for this significant
optimize convection cooling. Incorrectly designed finned-surfaces not deviation. This reveals that the correlations may not capture the

⁎
Corresponding author.
E-mail address: [email protected] (R.C. Adhikari).

https://doi.org/10.1016/j.tsep.2020.100484

2451-9049/ © 2020 Elsevier Ltd. All rights reserved.

R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

Nomenclature s Fin spacing, mm

sopt Optimal fin spacing, mm
Ab Base surface area of fin array (= ntL + (n 1) sL ), m2 T Fin surface temperature, °C or K
At Total surface area of fin array (= T Ambient temperature, °C or K
2nLH + (n 1) sL + ntL ), m2 Tf Film temperature (= (T + Ts )/2 ), °C or K
h Convection coefficient (= Q/ At T ), W/m2K t Fin thickness, mm
H Fin height, mm u Air velocity in x-direction, m/s
K Thermal conductivity of fin material (= 167 W/mK) v Air velocity in y-direction, m/s
k Thermal conductivity of air, W/mK W Width of fin array, m
L Fin length, mm Wt Weight of fin array, kg
lc Characteristic length, mm w Air velocity in z-direction, m/s
M Molecular weight of air, gm/mol Thermal diffusivity of air, k / cp , m2/s
n Number of fins Coefficient of thermal expansion of air (= 1/ Tf ), 1/K
p Pressure, Pa T Temperature difference between fin surface and ambient,
Q Total heat transfer, W/m2K (=Ts T ) °C or K
Qb Total heat transfer, W/m2K Kinematic viscosity of air, m2/s
Qopt Optimal heat transfer, W/m2K Density of air or fin material, kg/m3
Ra Rayleigh number (= g Tlc3/( )) 0 Operating density of air, kg/m3
R Ideal gas constant, J/mol K

Fig. 1. Schematic illustration of natural convection from a rectangular fin array on a heated horizontal surface. The upward flow of heated air and downward flow of
denser air results in the convection of heat as a chimney or plume.

combined effects of fin geometrical parameters on the flow and heat h is small. Thus, the experimental test cases of Jones and Smith [4] are
transfer. It is not sufficiently studied what heat transfer and flow field considered in the present study. Other experimental studies include
characteristics constitute optimum fin designs. For instance, the ex- [15,16]. Shobhan et al. [15] studied the variation of heat flux and
perimental results of Jones and Smith [4] showed that maximum heat temperature in the central fin of a three fin array, with a focus on
transfer per unit base area occurs within a narrow range of fin spacing, qualitative descriptions of the flow in transient heating and cooling
particularly in tall fins, but the flow and heat transfer mechanisms were regimes. They obtained a similar trend in heat flux over a narrow range
not investigated in detail. of fin spacing as obtained by Jones and Smith [4] for the particular fins
In the literature, both experimental and numerical studies have at- studied. Yüncu and Anbar [16] measured the effect on h of varying H
tempted to characterize the heat transfer of rectangular fin arrays on from 6 mm to 26 mm and s from 6.2 mm to 83 mm at constant
horizontal surfaces in natural convection. Early experimental studies L = 100 mm and t = 3 mm. They presented a correlation in terms of
include [3,4,12]. These studies were conducted on similar fin designs, non-dimensional parameters for the particular fins studied, however, no
and have derived correlations in terms of average heat-transfer coeffi- explanation for the conditions of optimum fin performance are pro-
cient h, and other non-dimensional parameters. Jones and Smith [4] vided. Although not the subject of this work, the thermal performance
studied the effects of fin height H and s on h at constant fin length L. of moving, and more complex, fin shapes, such as exponential and
They found that H has little influence on h, whereas s has significant trapezoidal, can be better than stationary, rectangular, and straight fins.
effects. They correlated the data of s and H for optimum fins as Turkyilmazoglu [17,18] studied the efficiency of moving trapezoidal
Hs 360 mm2 for fins with L = 254 mm. They compared their results and exponential shaped fins as compared to rectangular fins. He de-
with Harahap and McManus [12], and found inconsistencies in the veloped closed-form solutions for the temperature field and the fin ef-
results for similar fin designs. Here, the performance of two fin arrays ficiency for both shapes, and showed that longitudinal trapezoidal fins
which are similar except for the number of fins n, can be compared. with thicker tips have better heat transfer than rectangular fins. Simi-
Jones and Smith [4] used n = 7 for all fin configurations, and Harahap larly, he showed that straight rectangular fins could have lower or
and McManus [12] conducted experiments by varying n from 17 to 33. higher performance depending upon the exponential fin configurations.
At n = 17, L = 254 mm, H = 38 mm, and s = 6.35 mm, the results of However, it is not clear whether the comparison was based on the
Harahap and McManus [12] for h agreed well within 10% to that of optimum rectangular fins or not, and whether this conclusion holds true
Jones and Smith [4] for n = 7 , L = 254 mm, H = 36.57 mm, and for stationary fins. Nevertheless, it indicates the significance of design
s = 6.35 mm for all temperature differences tested (see Fig. 4 of Ref. optimization of rectangular fins, as well as the above fin shapes.
[4]). Despite this increase in width W by more than twice, the effect on Previous numerical studies using steady, laminar, heat transfer

2
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

simulations include [14,19–27]. Tari et al. [14] studied heat transfer transfer simulations of the fins studied experimentally in [3,4], which
from inclined plate fins, including the horizontal orientation. They are the most systematic experimental studies that we could find. An-
studied the flow structure and developed correlations. Dogan et al. [19] other novelty is that, rather than seeking a single-parametric optimi-
computed the general flow patterns of different shapes of fin arrays on a zation as is commonly done in the literature, we undertook a multi-
horizontal surface, and also suggested correlations. Baskaya et al. [20] parametric computational fluid dynamics (CFD) optimization using the
studied the general flow and heat transfer features using CFD by dynamic-Q algorithm, which efficiently searches for a global optimum
varying the fin geometrical parameters. They concluded that optimum in a computationally expensive problem. A simple experimental test
performance could not be obtained by simply focusing on variation of was also conducted to verify the optimization results. To summarize the
one or two parameters, e.g. variation of single parameter at a time as in review and statement of our objectives: the purpose of this study is to
most experimental measurements. However, they did not investigate characterize the heat transfer and flow features of optimum fins and
the specific characteristics of optimum combination of geometrical perform multi-parametric CFD optimization to determine an optimum
parameters. Mobidiy et al. [21] simulated the quasi three-dimensional combination of geometrical parameters.
flow over short rectangular fins on a horizontal surface, and char- The remainder of the paper is organized as follows. Section 2 de-
acterized the single and multiple chimney flows, that are shown in scribes the computational model used in the study. Section 3 outlines
Fig. 1. Karvinen and Karvinen [22] examined the optimization of plate the model validation. Section 4 contains the main results, with a par-
fins for maximizing heat flux based on an approximate analytical so- ticular focus on heat transfer characteristics and flow patterns in op-
lutions. Similarly, Huang et al. [23] studied the dynamic chimney flow timum fins. Extending these results, a multi-parametric CFD optimiza-
behavior of rectangular fins in natural convections for the fins with tion using a dynamic-Q algorithm is conducted on example design
L = 56–500 mm with H = 6.4 and 38 mm and a fixed s = 6.4 mm. problems. Results are verified through the experimental results avail-
Their important finding is that the effect of the downward flow of cool able in the literature, and experimental testing on an example fin design
air on heat transfer is not significantly different for taller compared to conducted in this study. Finally, Section 5 summarizes the main find-
shallow fins due to thicker boundary layers in the fin channels. Wong ings.
and Huang [24] conducted a parametric study on the dynamic natural
convection from long horizontal fins with dimensions L = 128, 254 and 2. Computational model
380 mm using a 3-D unsteady numerical analysis. They characterized
the oscillating chimney flow with reference to variation in L and H We analyze the conjugate heat transfer and flow using three-di-
variation and its impact on heat transfer in general. Shen et al. [25] mensional, steady-state, laminar flow simulations. The simulations
studied the heat transfer from rectangular fins at different orientations were conducted on the commercial Computational Fluid Dynamics
of the fin array. The literature survey revealed that the predictive (CFD) code ANSYS Icepak-Fluent [28]. The governing equations are the
capability of the correlations suggested in the literature for determining conservation of mass (continuity), momentum, and energy for the flow,
heat transfer characteristics is limited in general, and the results are and the heat conduction equation for the solid, which are written in
often conflicting. Also the numerical studies are presented in general- Cartesian co-ordinates as follows. As shown in Fig. 4, x is in the di-
ized form, rather than as specific characteristics of optimum fin con- rection of the fins, y is vertical, and z is in the transverse direction
figurations. Thus, the characteristics of heat transfer and the underlying normal to the fins.
flow patterns are very important in guiding the optimization of fins. The continuity equation for the steady flow is
The first novel feature of this work is the investigation of the effects
u v w
of fin height, spacing, and length on the heat transfer and flow, with + + =0
particular attention to determining the characteristics of optimum fins. x y z (1)
We conducted three-dimensional, steady-state, laminar, conjugate heat where is the density, and u , v and w are the velocities in the x,y , and z

Fig. 2. Variation of air properties with temperature used in the simulations. Experimental data taken from Cebeci and Bradshaw [6].

3
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

directions respectively. The three momentum equations are steady state Laplacian for the temperature with constant thermal dif-
fusivity.
u u u p u u u
u +v +w = + µ + µ + µ Eqs. (1)–(5) were solved using the steady-state laminar flow model.
x y z x x x y y z z In natural convection problems in fins, Rayleigh numbers Ra < 108
(2) indicate a laminar flow, with transition to turbulence occurring over the
range of 10 8 < Ra < 1010 [28]. As Ra 9.46 × 107 in the fins examined
v v v
u +v +w in this work (considering the fin length as a characteristic length), these
x y z
flows might be turbulent or transitional. All simulations, however, as-
p v v v sumed laminar flow which can be justified on the grounds of the good
= + µ + µ + µ +( 0) g
y x x y y z z (3) agreement to the measurements that is demonstrated in the next sec-
tion. Radiative heat transfer was considered using the surface-to-surface
u
w
+v
w
+w
w
=
p
+ µ
w
+ µ
w
+ µ
w radiation model [14,19,28]. For brevity, we have omitted the model
x y z z x x y y z z equations, which can be found in references [19,28]. To account for the
(4) variation of air properties, such as density, viscosity, thermal con-
ductivity, and diffusivity due to changes in T and p, the ideal gas law
where p is the pressure and is the viscosity. These equations are
was used with experimental data for the property variations. µ was
solved for the three velocity components and p. The single energy
computed using Sutherland’s law [6] as
equation for the temperature, T, is

T T T T T T 8 T 3/2
µ = 1.45 × 10
u +v +w = + + +S T + 110 (6)
x y z x x y y z z (5)

where is the thermal diffusivity, and S is the source term for radiation. where T is expressed in degrees Kelvin and µ in kg/ms. Similarly,
The temperature distribution within the fins was computed by the density is calculated from the ideal gas law as

Fig. 3. A three-dimensional hexahedral computational mesh used in the simulations (a). The blue region is the fin array. Zoomed views of a typical mesh resolution
on the fin walls (b) and a cross-section of the mesh resolution around the fins (c).

4
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

p between the computed and experimental results, which is a fairly good

= R
M
T (7) agreement between the two in view of the uncertainties incurred in
both the experimental measurements and the numerical calculations.
where R is the universal gas constant and p is the (atmospheric) pres- Jones and Smith’s experimental results have about ±10% uncertainty.
sure. The variation of air properties used in the simulations are pre- Importantly, the computed Q/ Ab has its maximum at the same s as the
sented in Fig. 2. measurements, despite the difference in magnitude. Similarly, the
The computational domain was discretized with fully structured comparison of the experimental results of Starner and Harahap [3] and
hexahedral grids with high resolution in the near wall region to capture the CFD results for h is shown in Fig. 6. A maximum relative error of
the effects of thermal boundary layer more accurately. A typical com- about 8.3% was found between the computed and experimental results.
putational mesh used in the simulation is shown in Fig. 3. The second A more detailed investigation of the heat transfer characteristics and
order upwind numerical scheme was adopted, as this offers a more the flow features including a multi-parametric optimization is pre-
accurate finite difference approximation of the spatial derivative and is sented in the following sections. We have considered the experimental
more robust [28]. Convergence criteria of 10 4 for the velocities and test cases of Jones and Smith [4] in the analysis as their experimental
10 7 for the energy were used. “Open type” boundary conditions were study extended the results of Starner and MacManus [3].
used at all sides of the computational domain, Fig. 4, with the buoyancy
enabled, and temperature on the boundary was that used in the ex-
periments. This figure shows that the fin was placed at the centre of an 4. Results and discussion
open computational domain whereas in all experiments, the fin array
sat on a solid surface at some height. There was no heat transfer from 4.1. Heat transfer behavior and flow patterns
the back of the array. No experimental information, however, was given
on any boundary layer development over the surface, so we chose to Simulations were conducted at experimentally tested H = 23.9,
model the array as shown in the figure for simplicity. A grid con- 36.6, and 49.3 mm, and s = 4.05, 6.35, 12.7, 25.4, 36.88, and
vergence study was performed to confirm the adequacy of the grid re- 49.28 mm. Fig. 5 shows that the taller fins exhibit the maximum Q/ Ab
solution by gradually refining the grid size until an acceptable con- within a narrow range of fin spacing, sopt = 6–8 mm. Another re-
vergence was achieved for the convection coefficient h (= Q/ Ab T ), markable heat transfer behavior is that shallow fins, e.g. H = 36.58 and
which was systematically assessed by computing h for an experimental 23.88 mm, have a relatively wider range of optimum spacing, sopt , or
case of Jones and Smith [4]. A total of 463,334 elements were required flat performance curve as depicted in Fig. 7. This reveals that fin op-
to produce grid independent results for h, which corresponds to less timization for taller fins is relatively more complex than shallow fins,
than 1.2% numerical uncertainty between the last three results. The and thus is an important design consideration. As a general guideline,
results of grid independence study are summarized in Table 1. All the Jones and Smith [4] correlated the experimental data for optimum fins
computations were performed with at least this number of grid ele- as Hsopt 360 mm2. It is evident that this correlation may be realistic
ments. We confirmed the predictive capability of the computational only for tall and wider fins.
model for a range fin design problems by comparing the computed At H = 49.28 mm and s < 6 mm, it can be seen that Q/ Ab has de-
results for the convection coefficient, h, with the experimental results creased sharply, which is due to the flow restriction in the narrow fin
for different fin geometries available in references [3,4]. channels. The reason could be that the flow velocity is too low to de-
velop a thermal boundary layer in the channel walls. Similarly, for
3. Model validation s > sopt , Q/ Ab has decreased but h has increased sharply to a maximum.
Here, the relative increase in Q is greater than the increase in At for
The computational model was validated against the measured heat constant H, which has increased h. In contrast, as s increases, the re-
transfer from rectangular fins on horizontal surfaces by Starner and lative increase in Ab is greater than the increase in Q, leading to overall
MacManus [3] and Jones and Smith [4], which are possibly the most reduction in Q/ Ab . The results in Fig. 5 illustrate that it may be possible
systematic experimental studies in the literature. The details of geo- to increase Q by increasing H up to some value, but determining sopt
metrical parameters and experimental results are available for different becomes more complex as maximum heat transfer in tall fins occurs
fin configurations, which allowed us to make comparisons between the over a narrow range of s. It can be readily inferred that for each s, there
CFD and experimental results to assess the predictive capability of the is an optimum height Hopt , at which Q/ Ab is maximum.
CFD model and investigate the heat transfer and flow characteristics.
The details of the fin design parameters and temperature variations in
the experimental measurements used in the CFD simulations are sum-
marized in Tables 2 and 3.
All the fin configurations tested by Jones and Smith [4] had
L = 254 mm, t = 3.17 mm, and n = 7. Only H and s were varied, and
thus the width of the fin array, W, was different for each s. In the
measurements of Jones and Smith [4], average surface temperature and
ambient were considered to calculate the temperature difference. It is
also experimentally verified by Jones and Smith [4] that the heat sink
was nearly isothermal with only a few degrees difference. The tem-
perature difference in the CFD calculations is based on the difference
between the maximum temperature on the fin surface and the ambient
temperature. To quantify the thermal performance of a particular fin
configuration, the heat transfer per unit base area, Q/ Ab , and convec-
tion coefficient, h, were used as by Jones and Smith [4]. Here,
h = Q /At T , where Q is the total heat transfer, At is the total fin surface
area and the base, and T is the temperature difference between the fin
surface and the surrounding air. A comparison of the experimental re-
sults of Jones and Smith [4] and the CFD results for Q/ Ab and h is shown
in Fig. 5. A maximum relative error of about 11.34% was found Fig. 4. Computational domain with the boundary conditions.

5
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

Table 1
Effect of mesh resolution on heat transfer coefficient, h [H = 49.28 mm,
s = 50.8 mm, and T = 151 °C ]
Grid elements Convection coefficient, h [W/m2K]

120,000 6.08
351,034 7.20
383,234 7.87
425,210 7.36
447,427 7.53
463,334 7.45

Table 2
Fin design parameters (Starner and MacManus, [3])
Length, L Height, H Spacing, s Width, W No of Thickness, t
[mm] [mm] [mm] [mm] fins, n [mm]

254 38.1 6.35 127 17 1.01

254 25.8 7.95 127 14 1.01 Fig. 6. Comparison of the experimental results of Starner and McManus [3] and
254 12.4 6.35 127 17 1.01 the CFD results for the convection coefficient (h = Q/ At T ).
254 6.35 6.35 127 17 1.01

Table 3
Fin design parameters (Jones and Smith [4])
Length, L Height, H Spacing, s [mm] No of Temperature
[mm] [mm] fins, n difference, °C

254 23.88 4, 6.35, 8.38, 12.7, 7 151

25.4, 50.8
254 36.58 4, 6.35, 12.7 7 151
254 49.28 4, 6.35, 12.7, 19, 7 151
31.75, 50.8

In natural convection, the temperature distribution is coupled to

velocity field as the air density is related to the temperature difference
between the surface and the surrounding air in the buoyancy term. To
analyze the influence of s on heat transfer and flow features in tall fins,
the temperature contours and velocity vectors are plotted in Figs. 8–10
for H = 49.28 mm at s = 4.05, 6.35, and 12.7 mm. In general, it can be
observed that a three-dimensional converging vertical flow is devel-
Fig. 7. Comparison of the experimental results of Jones and Smith [4] and the
oped above the centre of the fin array as a result of horizontal inflow of
CFD for the effect of height H on Q / Ab [L = 254 mm and T = 111 °C].
air from each channel ends. This type of flow is called “chimney or
plume” and was also experimentally visualized by Jones and Smith [4]

Fig. 5. Comparison of the experimental results of Jones and Smith [4] and the CFD results for the heat transfer per unit base area, Q / Ab , and the convection
coefficient, h = Q/ At T , [H = 49.28 mm, L = 254 mm and T = 151 °C].

6
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

Fig. 8. Temperature contours at the mid-planes, taken at transverse plane (left) and longitudinal plane (right) of each case, illustrating the temperature resulting into
a single plume [H = 49.28 mm, n = 7, and T = 151 °C]. At s = 4.05 mm, the temperature gradient is almost zero in the inter-fin spacing, whereas at
sopt = 6.35 mm, there is considerable temperature gradient.

Fig. 9. Velocity vectors plotted at the mid-plane in the fin channel in longitudinal direction, illustrating the air flow from the channel ends toward the array centre
and development of a chimney flow [H = 49.28 mm, n = 7, and T = 151 °C].

Fig. 10. Temperature contours at the mid-

plane in the fin channel in transverse di-
rection (a) and velocity vectors at a mid-
plane of fin channel along the longitudinal
direction (b), illustrating the temperature
gradient and air flow from the channel ends
toward the array centre and development of
a chimney flow [H = 49.27 mm,
s = 12.7 mm, n = 7, and T = 151 °C].

and Harahap and MacManus [12], but no experimental measurements axisymmetric turbulent plume separating sharply from the ambient air
of the flow (e.g. velocity) were made. As the heated air moves upward [5]. In such flow structures, part of the heat transfer takes place due to
from around the centre of the fin array, the velocity of air in the fin “entrainment” or mixing of the surrounding air into the plume [5].
channels would increase the heat transfer. It can be observed that the To quantify the influence of s on Q/ Ab at H = 49.28 mm, the
temperature contours have a smoother boundary separating nearly temperature and air velocity in the fin channels, at mid-height and top
uniform buoyant air from the surroundings, which resembles an of the fins, are plotted in Figs. 11 and 12 respectively. Fig. 11 shows

7
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

also improves h, but not necessarily Q/ Ab . It is noted here that Ab will

increase as s is increased for the same number of fins. Jones and Smith
[4] conducted experiments at varying H and s but at constant length
L = 254 mm. They stated that L could have significant influence on
Q/ Ab , and suggested further investigation. In the next section, the in-
fluence of L is studied as an optimization problem.

4.2. Multi-parametric optimization using dynamic-Q

Fin design optimization for natural convection is a constrained,

multi-parametric non-linear problem, which is often computationally
expensive. The optimization objective is to maximize the total heat
transfer per unit base area, Q/ Ab , for minimum weight, i.e. minimum
H , L , s , and n. The fin design parameters are H , L , s, n , and Q, and T
is a constraint. Due to a large number of design parameters, it is often
difficult or impossible to obtain a global optimal solution using single-
parametric studies as is done in the literature. Further, CFD-based op-
timization is computationally expensive for problems involving a large
Fig. 11. Comparison of temperature and velocity measured at the top and mid- number of design parameters. Here, we conducted multi-parametric
height sections of the fins at s = 4.05 mm, illustrating the heat transfer rate. conjugate heat-transfer CFD simulations, in which all possible combi-
The zero velocity indicates the fin positions.
nations of geometrical parameters are evaluated to determine an op-
timal combination.
The objective function for heat transfer per unit base area, Q/ Ab , is
defined as
Q
maximize f (Q , s, n , L) = [((n 1) s + nt ) L]
subject to minimum H , L , s and n for H > 0, L > 0, s > 0, n > 0,
(8)
where Q, s, L and n are either design constraints or variables. The multi-
parametric simulations were first tested with the experimental case of
Jones and Smith [4], and then simulations were conducted with all the
geometrical parameters s, H , and L as variables for a given T . Simi-
larly, in a design problem where weight is critical, the objective func-
tion is defined for heat transfer per unit base area per unit fin weight as
Q
maximize f (Q , s, n , L , H ) = [((n 1) s + nt + nH ) Ab Lt ]
subject to minimum H , L , s, and n for H > 0, L > 0, s > 0, n > 0.
(9)

Fig. 12. Comparison of temperature and velocity measured at the top and mid- Eqs. (8) and (9) were solved using the dynamic-Q optimization al-
height sections of the fins at sopt = 6.35 mm, illustrating heat transfer rate. The gorithm, which applies a dynamic-trajectory-optimization algorithm to
zero velocity indicates the fin position. successive quadratic approximations of the optimization problem

that the heat transfer to the air in the fin channel is almost zero for
s = 4.05 mm. This clearly indicates the restriction of flow in channels,
and hence the cooling effect. Here it is evident that the velocity is too
low to form the thermal boundary layers, in which the heat transfer
occurs. Thus, the fin surface area for heat transfer is equivalently re-
duced to the flat plate surface area or the base area, Ab , the thickness of
the flat plate being equal to H. This is the main reason for the significant
drop in heat transfer. Furthermore, the integral of the vertical velocity
over the channel cross-section (the results are not shown) shows the
volume flow rate of cool air into the channel is small for narrow fin
channels, e.g. s = 4.05 mm. As expected, when s was increased to
sopt = 6.35 mm, both the temperature gradient and air velocity in the
fin channels increased as shown in Fig. 12. The reason is the increase in
base surface area, Ab , and the air inflow in the channels from the ends.
The velocity of the horizontal inflow from the channel ends, i.e. cool air
replenishing the warm air from the walls of the fins, caused a significant
temperature gradient across the channel section. At wider spacing,
s = 12.7 mm, the flow is well distributed in the channel with a larger Fig. 13. Comparison of temperature and velocity measured at the top and mid-
temperature gradient as shown in Fig. 13, but as Ab has also increased, height sections of the fins at s = 12.7 mm and H = 49.28 mm, illustrating the
the total heat transfer per unit base area, Q/ Ab , has decreased. This heat transfer rate. The maximum temperature and zero velocity indicate the fin
implies that increasing s provides more air flow into the channel and positions.

8
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

[28,29]. This algorithm was chosen as it efficiently solves optimization to 49.28 mm. These results clearly indicate that shorter and wider fins
problems in CFD with minimum storage requirements, where function would transfer more heat per unit base area per unit weight, Q/ Ab Wt ,
evaluations are computationally expensive as is the case of the present than longer, taller and wider fins. These results are consistent with the
problem [28,29]. The description of the dynamic-Q optimization al- experimental results of Jones and Smith [4], and empirical relations for
gorithm is presented in the Appendix. In this work, three problems were taller fins, Hs 360 mm2. Similarly, taller and longer fins transfer more
analyzed. The first is aimed at finding computationally the optimum total heat, Q, than shallow and shorter fins. This indicates that H is
geometry determined experimentally by Jones and Smith [4]. The important in maximizing Q/ Ab Wt . It may thus be concluded that L , H ,
second is not, strictly, an optimization problem. It is to test the accuracy and s have significant effects on Q, Q /Ab and Q/ Ab Wt of the fin array,
of the CFD results for varying T in a fixed geometry by comparison to the optimization of which requires a multi-parametric CFD optimiza-
new experiments. The first two problems are, therefore, validation ex- tion as presented in this work. It is worth mentioning here the com-
ercises. The third problem is the most complex as the optimization in- putational time for typical optimization problems. Fig. 17 shows that
cludes minimizing the weight of the fins. there are 16 CFD test evaluations during the optimization, and Fig. 18
Problem I was addressed using two separate simulations. First, a shows that there are 9 CFD test evaluations to finally reach the op-
multi-parametric optimization was applied to verify the optimal con- timum values. The total computational time for each test evaluation
dition of Fig. 5 at H = 49.28 mm and T = 151 °C by varying s and depends on the initial and boundary conditions, number of geometrical
keeping the other parameters constant. A second optimization was parameters, and the objective function. For the computational model
conducted by varying L and s to quantify the effect of L on Q/ Ab at and mesh resolution used in this study, the total recorded computa-
H = 49.28 mm and T = 151 °C. The influence of L on Q/ Ab is shown tional time for the optimization problem reported in Fig. 17 is
in Fig. 14. The results for L = 260 mm followed the experimental re- 49.07 min and 66.37 min for the optimization problem reported in
sults of Jones and Smith [4] at L = 254 mm, indicating that the dy- Fig. 18 on an eight-core parallel computer.
namic-Q algorithm accurately reproduces the optima found in the
measurements. The results show that the increase in L from 200 to 5. Conclusions
260 mm significantly decreased Q/ Ab . The physical basis for this result
is that in longer fin arrays, velocity of the flow entering from the We have presented an analysis of heat transfer characteristics and
channel ends would not be sufficiently developed to reach the centre of flow patterns of optimum rectangular fin configurations in natural
the array or the extra frictional resistance prevents sufficient air getting convection for fin design optimization. Three-dimensional, steady-state
into form a plume, and thus decreasing the cooling effects or the heat laminar heat transfer simulations were conducted on experimentally
transfer rate. tested rectangular fins reported in the literature to analyze the optimum
Problem II in Table 4 determines Q as a function of T for a given fins. Multi-parametric CFD optimization was conducted using dynamic-
fin design, i.e. constant H , L , and s. An experimental setup was devel- Q optimization algorithm to determine the optimum geometrical
oped as shown in Fig. 15. A regulator was attached at the base of the parameters, followed by measurement of an example fin design to
fins as a heat source, and the base was insulated. There was no heat verify the CFD predictions of the total heat transfer dependence on the
transfer from the back of the array. The increase in T was measured by surface temperature. The key findings can be summarized as follows.
increasing Q. The experiment was conducted in a large room with
stagnant air and the room temperature was constant at 24.2 °C. A FLIR • The results of heat transfer simulations showed a good agreement
E75 thermal imaging camera was used to measure the fin surface with the experimental results reported in the references [3,4]. Dif-
temperature. As per manufacturer’s data-sheet, the maximum experi- ferent fin geometries with varying fin spacing s and height H were
mental measurement error of the thermal camera is ±2%. For each test tested at constant L to assess the predictive capability of the com-
case, the temperature was recorded after the heat sink attained a steady putational model for the heat transfer. From the CFD results, air
surface temperature. This occurred at around 20–30 min. To ensure velocity vectors, and temperature contours in the fin channels were
reproducibility of the measurement, temperature was measured at analyzed to characterize the heat transfer behavior in optimum
different sections of the fin surface by changing the location of the conditions.
camera pointer. The maximum temperature difference between the • In high fins, optimal heat transfer occurs within a narrow range of
different sections at same fin height was 0.1 °C. In the CFD optimiza- fin spacing, whereas in shallow fins, optimal heat transfer occurs
tion, Q was computed at experimentally measured temperature differ- over a wider range of fin spacing. A generic three-dimensional flow
ence values, T . The comparison of the experimental and the optimi-
zation results are summarized in Fig. 16, which shows a good
agreement between the two. The results demonstrate the good accuracy
of the dynamic-Q algorithm over a wider range of temperature differ-
ence.
In most design applications, Q/ Ab per unit weight of fin material is
an important consideration. A third optimization problem was solved
for the maximum heat transfer per unit base area for minimum weight,
Q/ Ab Wt . Here, Wt = (tb Ab + ntLH ) is the mass of the fin array.
Simulations were conducted to maximize Q/ (tb Ab2 + Ab ntLH ) for
Tmax = 166 °C by varying s, H and L. Fig. 18 shows the results of op-
timization for minimum weight, in which it is evident that the optimum
points for maximum Q/ Ab and Q/ (tb Ab2 + Ab ntLH ) are different. Op-
timum value for Q/ Ab Wt was obtained for the minimum H = 36 mm
and L = 200 mm, and s = 9.12 mm. At L = 254 mm, sopt = 6–8 mm for
maximum Q/ Ab , whereas for maximum Q/ (tb Ab2 + Ab ntLH ) , the op-
timum s and L were found as 9.12 mm and 200 mm respectively. The
total heat transfer rate at the optimum points, Qopt = 184.7 W for
L = 254 mm and sopt = 6–8 mm, whereas at sopt = 9.12 mm and
L = 200 mm, Qopt was computed as 111 W. It is noted here that L was Fig. 14. Optimization results for the effect of length L on Q / Ab at varying s for
varied from 200 to 260 mm in the optimization problem and H from 36 the optimization problem I in Table 4 [H = 49.28 mm and T = 151 °C].

9
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

Table 4 pattern as a vertically directed heat convection plume or chimney

Design parameters for the multi-parametric optimization problems was observed above the centre of the fin array. The flow patterns
Parameter Problem I Problem II Problem III and temperature contours indicated that the most important factors
that affect the heat transfer rate are the size of air flow inlet area of
L, mm 200–260 75 200–300 the fin channel, i.e. fin spacing and height. In narrow non-optimal
n 7 6 7
fins, the lack of a thermal boundary layer in the fin channels due to
s, mm 4–20 6 4–20
H, mm 49.28 17 36–49.28
low air velocity was found to be the main cause of poor heat
t, mm 3.17 1.25 3.17 transfer.
T , °C
Function to be maximized
151 variable 151 • At optimal fin spacing, both the temperature gradient and air ve-
[(n
Q
1) s + nt ] L [(n
Q
1) s + nt ] L [(n
Q
1) s + nt + nH ] Ab tL
locity in the fin channels increased. The increased velocity of the
horizontal inflow from the channel ends replenishing the warm air
from the channels caused a significant temperature gradient. At
wider spacing, the flow is well distributed in the channel with more
temperature gradient, but as the base area Ab has also increased, the
total heat transfer per unit base area, Q/ Ab , is decreased. This im-
plies that increasing s provides more air flow into the channel and
also improves h, but not necessarily Q/ Ab . It is noted that Ab will
increase as s is increased. Therefore, optimal Q/ Ab depends both on
the temperature gradient and the total channel or fin surface area.
• Multi-parametric conjugate heat-transfer CFD simulations using
dynamic-Q optimization algorithm were conducted to determine
optimum fin parameters. Validation of the optimization results was
by comparison to previous experimental results. It was found that
the computed optimum geometry was close to the experimentally
determined one. In addition, an experimental test was conducted on
a fixed fin geometry to verify the computed heat transfer rate over a
wide range of temperature differences. The optimization results
showed that optimum geometrical parameters for the maximum
heat transfer per unit base area and for the maximum heat transfer
per unit base area and weight are different. The dynamic-Q opti-
mization algorithm was found to be and efficient and accurate op-
timization method for simulating heat transfer in rectangular fins
from natural convection.
• This paper emphasizes the determination of optimum fin config-
urations using numerical multi-parametric optimization, rather than
suggesting a generalized empirical correlation. We believe it is not
possible to determine a specific optimum fin-array for all combi-
nations of fin geometries and temperature differences through em-
Fig. 15. Experimental setup for the fin array in problem II, Table 4 pirical correlations. In other words, the effects of parameter varia-
[H = 17 mm, L = 75 mm, and s = 6 mm]. tion are best determined computationally. The techniques
developed in the paper should yield optimal fins in related situa-
tions, such as in the presence of combined forced and natural con-
vection, or cases where the fins are inclined. It may be possible to
develop a semi-empirical correlation for optimum fin geometry, but
extensive simulations on a wide range of fin configurations and
Rayleigh numbers are needed.

CRediT authorship contribution statement

R.C. Adhikaria: Conceptualization, Methodology, Validation,

Investigation, Writing - original draft, Writing - review & editing. D.H.
Wood: Conceptualization, Methodology, Supervision, Writing - review
& editing. M. Pahlevani: Conceptualization, Resources, Project ad-
ministration, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial

interests or personal relationships that could have appeared to influ-
ence the work reported in this paper.

Fig. 16. Comparison of the CFD and the present experimental results for the Acknowledgement
heat transfer and temperature rise for the fin design problem II in Table 4
[H = 17 mm, L = 75 mm, and s = 6 mm].
The authors would like to acknowledge the CMC Microsystems
Canada for providing ANSYS license to perform thermal flow simula-
tions.

10
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

Fig. 17. Optimization of heat transfer per unit base area, Q / Ab , for Tmax = 166 °C and varying s for the optimization problem I in Table 4. Optimum Q / Ab = 9,986 W/
m2 was obtained at L = 254 mm, s = 7.18 mm, and H = 49.28 mm.

Fig. 18. Optimization of heat transfer per unit base area per unit weight, Q/ Ab Wt , for Tmax = 166 °C at varying s, H and L. Optimum Q/ Ab Wt = 12,880 W/m2kg was
obtained at s = 9.12 mm, L = 200 mm, and H = 36 mm.

Appendix A. Appendix

A non-linear constrained optimization problem can be written as

minimize f (x ); (x1, x2 , ….,xn )
subject to gj (x ) 0, j = 1, 2, …, p,
hk (x ) = 0, k = 1, 2, …, q (10)
where f (x ), gj (x ) , and hk (x ) are scalar functions of x. The dynamic-Q algorithm generates sub-problems P [i], i = 0, 1, 2, 3 …, at successive
approximations x i of the solution x , by constructing spherically quadratic extrapolations f (x ), gj (x ) , and hk (x ) of functions f (x ), gj (x ) , and hk (x )
respectively [28], which are written as
1
f (x )=f (x i ) + Tf (x i )(x x i ) + 2 (x x i )T A (x x i );
1
gj (x )=gj (x i ) + T g (x i )(x
j x i ) + 2 (x x i )T Bj (x x i); j = 1, 2, …, p ;
1
hk (x )=hk (x i ) + Th
k (x
i )(x xi) + 2
(x x i )T Ck (x x i ); k = 1, 2, …, q (11)
where A , Bj , and Ck are the Hessian matrices expressed as:

11
R.C. Adhikari, et al. Thermal Science and Engineering Progress 17 (2020) 100484

A = diag (a, a, …, a) = aI; Bj = bj I; Ck = Ck I (12)

Eq. (12) shows that the sub-problems, P [i], are spherically quadratic [28]. In solving the problem, a linear approximation is made by setting the
curvatures a, bj , and ck to zero for the first sub-problem, i.e. i = 0. In the next step, a, bj , and ck are chosen in such a way that the approximating
functions, i.e. Eqs. (11) and (12), interpolate their corresponding functions at both x i and x i 1, which can be written for i = 1, 2, 3, …as:

2[f (x i 1) f (x i ) f (x i )(x i 1 x i )] 2[gj (x i 1) gj (x i) gj (x i )(x i 1 x i )] 2[hk (x i 1) hk (x i ) hk (x i )(x i 1 x i )]

a= ; bj = ; ck =
||x i 1 x i ||2 ||x i 1 x i||2 ||x i 1 x i||2 (13)
In Eq. (13), if the gradient vectors f, gj , and hk can not be computed analytically, they are approximated from the data by means of first-order
forward finite difference schemes [28]. It is imperative to note that as the second derivatives of the objective function and constraints are ap-
proximated using function and gradient data, the (n2 order) calculations and storage, which would be required for the second derivatives, are not
needed. The computational and storage requirements are thus reduced to order (n) . These savings become significant when the number of design
parameters is large such as in this paper. In summary, following optimization steps are followed in the dynamic-Q algorithm [28]:

1. Define a starting point x 0 and a step . Initialize i = 0 .

2. Compute f (x i ), gj (x i), hk (x i ), f (x i ), gj (x i ) , and hk (xi ) . If optimization criteria are satisfied, then stop iteration.
3. Construct a local approximation for the sub-problems, P [i], to the optimization problem at x i .
4. Solve the approximated sub-problem P [i] using the constrained leap-frog method as described below with x 0 = x i and then compute x*i .
5. Iterate with i = i + 1, x i = x *(i 1) and return to step 2 for more iterations until the optimization criteria are satisfied.

The sub-problems, P [i], for the constrained optimization are solved using the dynamic-trajectory “leap-frog” method with the penalty function
formulations [28]. In its unconstrained form, the leap-frog optimizer determines the minimum of a function, f (x ) , by formulating the motion of a
particle of unit mass in an n -dimensional conservative force field [28]. The potential energy of a particle at a point x (t ) at time t is defined by the
function f (x ) . The trajectory of the particle requires the solution of the following equations of the motion [28]:
X¨ (t ) = f (X (t )) (14)
with initial conditions
X (0) = X0 , X (0) = 0 (15)
The solution of the above equations over the time interval [0, t] is expressed as
1 1
||X (t )||2 ||V (0)||2 = f (X0 ) f (X (t )); f (t ) + T (t ) = f (0) + T (0) = K
2 2 (16)
where T (t ) is intantaneous kinetic energy of the particle, and K is a constant determined by the initial conditions. It is noted that energy is conserved
in the above equation. It is also evident that f = T , and as T increases, f decreases. This forms the basis of the dynamic-trajectory method, which
means that the ”leap-frog” optimization algorithm computes an approximation to the trajectory followed by the particle in the force field [28]. The
above leap-frog optimization algorithm is modified to handle constrained problems, such as the fin design problems of this paper, by means of the
penalty-function, the details of which can be found in [28].

References [16] H. Yüncü, G. Anbar, An experimental investigation on performance of rectangular

fins on a horizontal base in free convection heat transfer, Heat Mass Transfer 33
(5–6) (1998) 507–514.
[1] W.M. Kays, Convective Heat and Mass Transfer, McGraw-Hill Education, Tata, [17] M. Turkyilmazoglu, Efficiency of the longitudinal fins of trapezoidal profile in
2012. motion, J. Heat Transfer 139 (9) (2017) 094501.
[2] J.H. Lienhard, A Heat Transfer Textbook, Courier Corporation, 2013. [18] M. Turkyilmazoglu, Heat transfer from moving exponential fins exposed to heat
[3] K. Starner, H. McManus, An experimental investigation of free-convection heat generation, Int. J. Heat Mass Transfer 116 (2018) 346–351.
transfer from rectangular-fin arrays, J. Heat Transfer 85 (3) (1963) 273–277. [19] M. Dogan, M. Sivrioglu, O. Yílmaz, Numerical analysis of natural convection and
[4] C.D. Jones, L.F. Smith, Optimum arrangement of rectangular fins on horizontal radiation heat transfer from various shaped thin fin-arrays placed on a horizontal
surfaces for free-convection heat transfer, J. Heat Transfer 92 (1) (1970) 6–10. plate-a conjugate analysis, Energy Convers. Manage. 77 (2014) 78–88.
[5] J. Turner, Buoyant convection from isolated sources, Buoyancy Effects Fluids [20] S. Baskaya, M. Sivrioglu, M. Ozek, Parametric study of natural convection heat
(1973) 165–206. transfer from horizontal rectangular fin arrays, Int. J. Therm. Sci. 39 (8) (2000)
[6] T. Cebeci, P. Bradshaw, Physical and Computational Aspects of Convective Heat 797–805.
Transfer, Springer Science & Business Media, 2012. [21] M. Mobedi, H. Yüncü, A three dimensional numerical study on natural convection
[7] A. Bar-Cohen, Fin thickness for an optimized natural convection array of rectan- heat transfer from short horizontal rectangular fin array, Heat Mass Transfer 39 (4)
gular fins, ASME J. Heat Transfer 101 (1979) 564–566. (2003) 267–275.
[8] A. Bar-Cohen, W. Rohsenow, Thermally optimum spacing of vertical, natural con- [22] R. Karvinen, T. Karvinen, Optimum geometry of plate fins, J. Heat Transfer 134 (8)
vection cooled, parallel plates, J. Heat Transfer 106 (1) (1984) 116–123. (2012) 081801.
[9] N.S. Effendi, K.J. Kim, Orientation effects on natural convective performance of [23] G.-J. Huang, S.-C. Wong, Dynamic characteristics of natural convection from hor-
hybrid fin heat sinks, Appl. Therm. Eng. 123 (2017) 527–536. izontal rectangular fin arrays, Appl. Therm. Eng. 42 (2012) 81–89.
[10] K. Ong, C. Tan, K. Lai, K. Tan, Heat spreading and heat transfer coefficient with fin [24] S.-C. Wong, G.-J. Huang, Parametric study on the dynamic behavior of natural
heat sink, Appl. Therm. Eng. 112 (2017) 1638–1647. convection from horizontal rectangular fin arrays, Int. J. Heat Mass Transfer 60
[11] B. Freegah, A.A. Hussain, A.H. Falih, H. Towsyfyan, Cfd analysis of heat transfer (2013) 334–342.
enhancement in plate-fin heat sinks with fillet profile: investigation of new designs, [25] Q. Shen, D. Sun, Y. Xu, T. Jin, X. Zhao, Orientation effects on natural convection
Therm. Sci. Eng. Progress (2019) 100458. heat dissipation of rectangular fin heat sinks mounted on leds, Int. J. Heat Mass
[12] F. Harahap, H. McManus, Natural convection heat transfer from horizontal rec- Transfer 75 (2014) 462–469.
tangular fin arrays, J. Heat Transfer 89 (1) (1967) 32–38. [26] S. Mandal, A. Deb, D. Sen, Mixed convective heat transfer with surface radiation in
[13] F. Harahap, H. Lesmana, I.A.S. Dirgayasa, Measurements of heat dissipation from a rectangular channel with heat sources in presence of heat spreader, Therm. Sci.
miniaturized vertical rectangular fin arrays under dominant natural convection Eng. Progress 14 (2019) 100423.
conditions, Heat Mass Transfer 42 (11) (2006) 1025–1036. [27] M. Bayat, M.R. Faridzadeh, D. Toghraie, Investigation of finned heat sink perfor-
[14] I. Tari, M. Mehrtash, Natural convection heat transfer from inclined plate-fin heat mance with nano enhanced phase change material nepcm, Therm. Sci. Eng.
sinks, Int. J. Heat Mass Transfer 56 (1–2) (2013) 574–593. Progress 5 (2018) 50–59.
[15] C. Sobhan, S. Venkateshan, K. Seetharamu, Experimental analysis of unsteady free [28] ANSYS, Ansys academic research icepak, release 18, ANSYS.
convection heat transfer from horizontal fin arrays, Warme-und Stoffubertragung [29] J. Snyman, A. Hay, The dynamic-q optimization method: An alternative to sqp,
24 (3) (1989) 155–160. Comput. Math. Appl. 44 (12) (2002) 1589–1598.

12