You are on page 1of 8

Thermally Optimum Spacing of

A. Bar-Cohen ertical3 Natural Gonwection


Department of Mechanical Engineering,
Ben-Gurion University of the Negev,
Beer-Sheva, Israel
Cooled, Parallel Plates
Fellow ASME While component dissipation patterns and system operating modes vary widely,
many electronic packaging configurations can be modeled by symmetrically or
W. iVl. Rohsenow asymmetrically isothermal or isoflux plates. The idealized configurations are
amenable to analytic optimization based on maximizing total heat transfer per unit
Department of Mechanical Engineering, volume or unit primary area. To achieve this anlaytic optimization, however, it is
Massachusetts Institute of Technology, necessary to develop composite relations for the variation of the heat transfer
Cambridge, Mass. 02139 coefficient along the plate surfaces. The mathematical development and verification
Fellow ASME
of such composite relations as well as the formulation and solution of the op-
timizing equations for the various boundary conditions of interest constitute the
core of this presentation.

Introduction
Vertical two-dimensional channels formed by parallel confirmed by the laborious numerical calculations of Bodia
plates or fins are a frequently encountered configuration in and Osterle [5], and extended to asymmetric heating by Aung
natural convection cooling in air of electronic equipment, [7] and Miyatake et al. [9, 10]. In a subsequent section of this
ranging from transformers to main-frame computers and discussion, the limiting relations for fully developed laminar
from transistors to power supplies [1, 2, 3]. Packaging flow, in a symmetric isothermal or isoflux channel, as well as
constraints and electronic considerations, as well as device or in a channel with an insulated wall, will be rederived by use of
system operating modes, lead to a wide variety of complex a straightforward integral formulation.
heat dissipation profiles along the channel walls. In many The analytic relations for the isolated plate (or inlet region)
cases of interest, however, a symmetric isothermal or isoflux limit and the fully developed (or exit region) limit can be
boundary representation, or use of an isothermal/isoflux expected to bound the Nu values over the complete range of
boundary together with an insulated boundary condition flow development. Intermediate values of Nu can be obtained
along the adjoining plate, can yield acceptable accuracy in the from detailed experimental and/or numerical studies or by
prediction of the thermal performance of such con- use of the correlating expression suggested by Churchill and
figurations. Usagi [12] for smoothly varying transfer processes. This
Elenbaas [4] was the first to document a detailed study of correlation technique relies on the analytic expressions at the
the thermal characteristics of one such configuration, and his two boundaries and a limited number of data points to derive
experimental results for isothermal plates in air were later a highly accurate composite correlation and its use will be
confirmed numerically [5] and shown to apply as well to the demonstrated in later sections.
constant heat flux conditions [6]. More recently, Aung and
coworkers [7, 8] and Miyatake and coworkers [9, 10] extended Fully Developed Limit
the available results to include both asymmetric wall tem-
perature and heat flux boundary conditions, including the Momentum Considerations. In laminar, fully developed,
single insulated wall. two-dimensional flow between parallel platesas shown in
From these and complementary studies emerges a unified Fig. 1the pressure drop is given by [11]
picture of thermal transport in such a vertical channel. In the dP
inlet region and in relatively short channels, individual = -12 i*w/pb3 (1)
dX loss
momentum and thermal boundary layers are in evidence
along each surface and heat transfer rates approach those For free-convection flow, this flow resistance is balanced by
associated with laminar flow along isolated plates in infinite the buoyant potential expressible as [11]
media. Alternately, for long channels, the boundary layers
merge near the entrance and fully developed flow prevails = (~Pf-Po)g=-f>Pg(Tf-T0) (2)
along much of the channel. dx buoy

In this fully developed regime, the local heat transfer


coefficient is constant (neglecting the temperature dependence Equating equations (1) and (2), the flow rate per unit width,
of fluid properties) and equal to the well-documented forced" w, in the channel, is found equal
convection values [11]. However, since the local fluid tem- w = p2g0bHT/-To)/12fi (3)
perature is not explicitly known, it is customary to reexpress
the fully developed heat transfer coefficient in terms of the Nusselt NumberSymmetric, Isothermal Plates. An
ambient or inlet temperature. The Nu for isothermal plates energy balance on the differential volume, shown in Fig. 1,
appropriate to this definition can be derived from the "in- equating heat transferred from two isothermal walls with that
compressible natural convection" form of the Navier-Stokes absorbed in the flow, yields
equations. This was done semianalytically by Elenbaas [4], wcpdT=2h(Tw-Tf)dx (4)
From continuity considerations the flow rate, w, is constant,
Contributed by the Heat Transfer Division for publication in the JOURNAL OF
and in fully developed flow with temperature-independent
HEAT TRANSFER. Manuscript received by the Heat Transfer Division November properties, the local heat transfer coefficent, h, as well as cp,
1,1982. is constant. Consequently, wcp/2h can be considered

116/Vol. 106, FEBRUARY 1984 Transactions of the ASME


Copyright 1984 by ASME
Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms
CpphP&S l-e'
q= 12 n
(Tw-T0){\
TL )}
(Tw-T0)(l-e-^)] (7)

Inserting equation (7) into (6) with the surface area, A, equal
to 2LS, the desired Nusselt number is found as
1 rcpp2gpb\Tlv-T0)
Nu
24 ]
fikL
TL

[0-^)<'--K>]
\-e~
(8)

X The combination of parameters appearing in the first


h bracketted term of equation (8) is recognizable as the channel
Rayleigh number, i.e., Ra b/L = Ra'. Consequently, at the
fully developed limit, where L o and the second term in
equation (8) approaches unity, Nu0 is seen to approach
Ra'/24. This result agrees exactly with the previously cited
analytical and numerical results [4, 5,7].
Fig. 1 Schematic of flow in a vertical channel
Nusselt NumberAsymmetric, Isotherm Plates. For
constant along the channel, and equation (4) can be simply fully developed flow in a channel formed by a single
integrated to yield the local fluid or bulk temperature isothermal plate and an insulated plate, only the isothermal
surface is involved in heat transfer, and Nu0 must be based on
T'/='-
f (T-T0)e (5)
that surface alone. Modifying equations (4, 5) and (6) in this
where T has replaced 2h/wcp. vein, the derivation of Nu0 is found to exactly parallel the
To accommodate the desire to obtain a Nusselt number development of equation (8) and yield Nu0 = Ra'/12 at the
based on the temperature difference between the wall and the limit of L oo in agreement with [10]. Comparison with the
ambient fluid, Nu0 can be defined as results obtained by Aung [7] reveals this value to be ap-
proximately 7 percent lower then the NuD associated with the
r q/A
Q/A 1i b (6) second surface being at ambient temperature (rather than
adiabatic).
The transfer rate, q, can be determined from the flow rate and
the temperature rise in the channel by the use of equations (3)
and (5), with the latter evaluated at x = L to find the exit Nusselt NumberSymmetric, Isoflux Plates. In a
temperature. The average fluid temperature in the channel channel formed by two constant heat flux plates, the fluid
can be found by integrating equation (5) from x = 0 to x = L temperature increases lineary along the channel and equation
and dividing by the length of the channel, L. Following these (5) can be replaced by
operations Tf=T0+2q"x/wcp (9)

QT= total heat flow rate, W z = generalized parameter in


A = area, m 2 ; coefficient Ra = Rayleigh number, = equation (19)
b = plate spacing, m p2g@cpL3 AT/nk, dimen- Z = modified generalized
B = coefficient sionless parameter in equation (19),
C
P = specific heat at constant Ra* = modified Rayleigh = Bzp~"/A
pressure, J/kgC number, ==
C = coefficient p2gPcpL4q"/nk2, dimen- Greek Letters
d = fin thickness, m sionless P = volumetric coefficient of
g = gravitational acceleration, Ra' = channel Rayleigh number, thermal expansion, K~'
m/s 2 = p2gPcpb4AT/fiL, r = thermall parameter,
ft = heat transfer coefficient, dimensionless 2h/wcp,m~
W/m2oC Ra" = modified Channel p = density, kg/m
3

k = thermal conductivity, <.Rayleigh number, = /* = dynamic viscosity, kg/m-s


2 s 2
W/mC P gl3cpb q"/fiLk A = difference
L = channel or plate length, m S = plate width, m
m = number of fins T = temperature, C Subscripts
n = exponent w = mass flow rate per unit buoyant = natural convection driving
Nu 0 = channel Nusselt number, width, kg/m-s force
= q"b/k(Tw, -T0) W = width of prime area, m / = fluid
dimensionless X = length coordinate, m loss = pressure drop
P = exponent y = generalized parameter in 0 = entrance or ambient value
2
P = pressure, N/m equation (19) opt = optimum
q = exponent in equation (17); Y = modified generalized w = wall
heat flow rate, W parameter in equation (19), X = local value
2
q" = heat flux, W/m = ylAz" L = at the x^L location

Journal of Heat Transfer FEBRUARY 1984, Vol. 106/117

Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms


4-
^ * * ^
H, elation, EQ 28

+>--'
!3p Hh Data of Wirtz a 3 t u t z n K i n [ l 7 ]

10 10' 4 6 8 |0 3
Ra"
Fig. 5 NuoL variation for symmetric isof lux platesdata of [17]

determine the empirical variation of the heat transfer rate for (QT/2LSWAT0k) = (b + d) -lb-l(516/P2bs
the symmetric, isoflux, air-cooled channel. The results were + 2.873/P * 2 )- 0 - 5
05
(31)
reported in terms of Nu based on the temperature difference
at xL and are compared in Fig. 5 with the composite Nu where
relation, equation (28), based on the same definition. PmCpipfgPATo/pkL
Examination of Fig. 5 reveals the predicted values to lie within
the experimental error band (Nu 5 percent, Ra 16 Differentiating equation (31) with respect to b, setting the
percent) of the data for all but the lowest values of Ra ". derivative to zero and cancelling common terms leads to

Nu 0 , L = |(48/Ra") + 2.51/(Ra") 0 - 4 )- 0 - 5 (28) - (b + d) " ' -b-1 + y (576/P 2 * 8 +2.873/P- 5 6 2 )-'


Asymmetric, Isoflux Plates. When a vertical channel is
formed by a single isoflux plate and an insulated plate, the (8 576/.P2 b9 + 2 2.873/P 0 5 b3) = 0 (32)
desired composite relation for Nu, based on the midheight Following additional algebraic operations, equation (32) is
temperature difference, can be found by appropriately found to reduce to
combining equations (17) and (27) (with n = 2) to yield
(26+ 3c?-0.005 P 15 6 7 ) 0 pi = 0 (33)
Nu 0 , i / 2 = {6/Ra" + 1.88/(Ra") 0 - 4 ) " 1 / 2 (29) Solution of equation (33) should now yield the value of b
which maximizes QT, i.e., the bopl value.
In general, bm is seen to be a function of both the plate/air
Optimum Plate Spacing
parameter, P, and the plate thickness, d, but for negligibly
The composite relations derived in the previous section can thick plates
be used to predict the value of the heat transfer coefficient for 6 o p t =2.714//> 1 / 4 (34)
each of the four thermal configurations examined. No less
important, however, is their potential use in optimizing the This result exceeds the Elenbaas optimum spacing by only 4
spacing between vertical, heat-dissipating plates when two- percent and yields optimum values of channel Rayleigh and
dimensional flow can be assumed to prevail. Nusselt numbers of 54.3 and 1.31, respectively.
In electronic cooling applications, it is often of interest to
maximize the rate of heat transfer from individual plates or
Symmetric, Isothermal Plates. The total heat transfer rate
component carrying, printed circuit boards. This can be
from an array of vertical plates, QT, is given by
achieved by spacing the plates in such a manner that the
QT=(2LSAT0)(m)(Nxiok/b) (30) isolated plate Nu prevails along the surface. To achieve this
where m, the number of plates, equals W/(b + d), b equals aim precisely requires an infinite plate spacing, but setting Nu
the spacing between adjacent plates, and d is the thickness of (via equation (23)) equal to 0.99 of the isolated plate value
each plate. yields Ra' = 463 and bmm equal to 4.64/P 174 . This result is in
general agreement with [18] where the identically defined
Examination of Fig. 2 shows that the rate of heat transfer,
maximum plate spacing was determined to occur at Ra'
from each plate decreases as plate spacing is reduced. Since
approximately greater than 600. It is perhaps of interest to
the total number of plates or total plate surface area increases
note that at Ra' = 600, the composite Nu is found to reach
with reduced spacing, QT may be maximized by finding the
0.993 of the isolated plate value.
plate spacing at which the product of total plate surface area
and local heat transfer coefficient is maximum. Based on his As might have been anticipated, the bmax spacing can be
experimental results, Elenbaas determined that this optimum shown to correspond to approximately twice the boundary
spacing for negligibly thick plates could be obtained by setting layer thickness along each surface at the channel exit, i.e, x =
Ra' 0pt = 46 yielding a Nu 0 of 1.2 [4]. L. By comparison 6 opt corresponds to nearly 1.2 boundary
Using equation (23) to determine Nu 0 and dividing both layer thicknesses a t x = L.
sides of equation (30) by the product of total fin area, tem-
perature difference, thermal conductivity, and width of the Asymmetric, Isothermal Plates. In analyzing the
base area, yields asymmetric, isothermal configuration, equation (31) can

120/Vol. 106, FEBRUARY 1984 Transactions of the ASME

Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms


10
8
6
"*"
1 J
Nu
2 /
A

A lis
'o
.'S \
,:

.FULLY-DEVELOPE ) / s .*^ \
lillT /,-y ^ , I
LIMIT
*/ """^ INTEGRAL SOLUTION
/y
2 ~f
in1
2 5 10 2 5 10 2 5 |03 2 5 I04

GrPr b/L
Fig. 3 Nu variation for parallel platesone isothermal, one insulated

10

r^2
"sOLATEO PLATE LIMIT
^x 5 ~ '
^ 3 _
q ^
X
i - co^
' " .4P0SITE
y-^i -- '
FULLY DEVELOPED L I M I T ^ ^ *

* X- SOBEL, LANDIS +MUELLEP, - DAT*


0-EN8EL+ MUELLER-. CALCULATION

I01 1 M i l l II
lO 2 3 5 7 10 2 3 5 7 lO1 23 5 7 O2 2 3 5 7 KD3 2 3 5
Ra"
Fig. 4 Nu 0 U 2 variation for symmetric isoflux platesdata of [6]

and Usagi [12], the correlating exponent, n, is found to equal Symmetric, Isoflux Plates. Natural convection heat
approximately 2, yielding a composite relation for two transfer from an isolated, uniform heat flux, vertical plate is
isothermal surfaces as generally correctable in the form
Nu0 = (576/(Ra') 2 +2.873/Via 7 )- 1 / 2 (23) Nux = C4(Ra*)' (25)
The close proximity of the Elenbaas data points to the While theoretically C4 for air has been shown to equal 0.519
composite relation, and the asymptotic equations at both [15], the empirical large-spacing asymptote for channel heat
limits, indicated in Fig. 2, serves to validate this approach. transfer is generally higher [6, 8, 9], yielding
Nu o =0.73(Ra") 1/5 (26)
Asymmetric, Isothermal Plates. For vertical channels
formed by an isothermal plate and an insulated plate, the for Nu0 based on the midheight temperature difference or
asymptotic limits were previously shown to be Nu0 = Ra'/12 Nu0 = 0.63(Ra")1/5 when the maximum channel wall to inlet
for Ra' - 0 and Nu0 = 0.59 Ra1/4 for Ra' - oo. Inserting air temperature difference is used.
these limiting expressions into equation (20) and assuming Much of the available Nu data for channels formed by
that despite channel asymmetry the symmetric correlating isoflux plates is presented in terms of the temperature dif-
exponent n = 2 applies to this configuration as well, the ference between the wall, at the channel midheight, and the
composite relation for asymmetric isothermal plates is found inlet air, e.g., [6, 8]. Superposing the two relevant asymp-
to be totes, equations (15) and (26), the composite Nu relation
appropriate to this definition is found as
Nu0 = [144/(Ra')2 +2.873/ (24)
-0.5
Nu0,L/2 = ((12/Ra")+ 1.88/(Ra") (27)
Comparison of equation (24) with the limited data of
Nakamura et al. [14] reported in [10] and the numerical Comparison in Fig. 4 of equation (27) with typical data of
solution of Miyatake and Fujii [10], as in Fig. 3, shows Sobel et al. [6] and the results of the Engel and Mueller
equation (24) to offer near-excellent agreement with the data numerical calculation presented in [8] reveals the composite
and to improve somewhat on the predictive accuracy of the isoflux relation to have a high predictive accuracy and no
numerical solution in the region where Nu displays the ef- further adjustment of the correlating exponent appears to be
fects of both fully developed and developing flow. Figure 3 necessary. The larger than anticipated Nu values at the low
and equation (24) also reveal the Nu0 from the thermally Ra" data points of Sobel et al. [6] may be explained by
active surface in an asymmetric channel to be higher than unaccounted-for radiation and conduction losses at the
from a comparable surface in a symmetric configuration, for channel exit, as noted by the authors.
a fixed channel width or Rayleigh number, at low values of In a recent study [17], both direct temperature
Ra'. measurements and analysis of interferograms were used to

Journal of Heat Transfer FEBRUARY 1984, Vol. 106/119

Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms


4-
^ * * ^
H, elation, EQ 28

+>--'
!3p Hh Data of Wirtz a 3 t u t z n K i n [ l 7 ]

10 10' 4 6 8 |0 3
Ra"
Fig. 5 NuoL variation for symmetric isof lux platesdata of [17]

determine the empirical variation of the heat transfer rate for (QT/2LSWAT0k) = (b + d) -lb-l(516/P2bs
the symmetric, isoflux, air-cooled channel. The results were + 2.873/P * 2 )- 0 - 5
05
(31)
reported in terms of Nu based on the temperature difference
at xL and are compared in Fig. 5 with the composite Nu where
relation, equation (28), based on the same definition. PmCpipfgPATo/pkL
Examination of Fig. 5 reveals the predicted values to lie within
the experimental error band (Nu 5 percent, Ra 16 Differentiating equation (31) with respect to b, setting the
percent) of the data for all but the lowest values of Ra ". derivative to zero and cancelling common terms leads to

Nu 0 , L = |(48/Ra") + 2.51/(Ra") 0 - 4 )- 0 - 5 (28) - (b + d) " ' -b-1 + y (576/P 2 * 8 +2.873/P- 5 6 2 )-'


Asymmetric, Isoflux Plates. When a vertical channel is
formed by a single isoflux plate and an insulated plate, the (8 576/.P2 b9 + 2 2.873/P 0 5 b3) = 0 (32)
desired composite relation for Nu, based on the midheight Following additional algebraic operations, equation (32) is
temperature difference, can be found by appropriately found to reduce to
combining equations (17) and (27) (with n = 2) to yield
(26+ 3c?-0.005 P 15 6 7 ) 0 pi = 0 (33)
Nu 0 , i / 2 = {6/Ra" + 1.88/(Ra") 0 - 4 ) " 1 / 2 (29) Solution of equation (33) should now yield the value of b
which maximizes QT, i.e., the bopl value.
In general, bm is seen to be a function of both the plate/air
Optimum Plate Spacing
parameter, P, and the plate thickness, d, but for negligibly
The composite relations derived in the previous section can thick plates
be used to predict the value of the heat transfer coefficient for 6 o p t =2.714//> 1 / 4 (34)
each of the four thermal configurations examined. No less
important, however, is their potential use in optimizing the This result exceeds the Elenbaas optimum spacing by only 4
spacing between vertical, heat-dissipating plates when two- percent and yields optimum values of channel Rayleigh and
dimensional flow can be assumed to prevail. Nusselt numbers of 54.3 and 1.31, respectively.
In electronic cooling applications, it is often of interest to
maximize the rate of heat transfer from individual plates or
Symmetric, Isothermal Plates. The total heat transfer rate
component carrying, printed circuit boards. This can be
from an array of vertical plates, QT, is given by
achieved by spacing the plates in such a manner that the
QT=(2LSAT0)(m)(Nxiok/b) (30) isolated plate Nu prevails along the surface. To achieve this
where m, the number of plates, equals W/(b + d), b equals aim precisely requires an infinite plate spacing, but setting Nu
the spacing between adjacent plates, and d is the thickness of (via equation (23)) equal to 0.99 of the isolated plate value
each plate. yields Ra' = 463 and bmm equal to 4.64/P 174 . This result is in
general agreement with [18] where the identically defined
Examination of Fig. 2 shows that the rate of heat transfer,
maximum plate spacing was determined to occur at Ra'
from each plate decreases as plate spacing is reduced. Since
approximately greater than 600. It is perhaps of interest to
the total number of plates or total plate surface area increases
note that at Ra' = 600, the composite Nu is found to reach
with reduced spacing, QT may be maximized by finding the
0.993 of the isolated plate value.
plate spacing at which the product of total plate surface area
and local heat transfer coefficient is maximum. Based on his As might have been anticipated, the bmax spacing can be
experimental results, Elenbaas determined that this optimum shown to correspond to approximately twice the boundary
spacing for negligibly thick plates could be obtained by setting layer thickness along each surface at the channel exit, i.e, x =
Ra' 0pt = 46 yielding a Nu 0 of 1.2 [4]. L. By comparison 6 opt corresponds to nearly 1.2 boundary
Using equation (23) to determine Nu 0 and dividing both layer thicknesses a t x = L.
sides of equation (30) by the product of total fin area, tem-
perature difference, thermal conductivity, and width of the Asymmetric, Isothermal Plates. In analyzing the
base area, yields asymmetric, isothermal configuration, equation (31) can

120/Vol. 106, FEBRUARY 1984 Transactions of the ASME

Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms


Table 1 Summary of heat transfer relations for vertical natural convection arrays
Condition Heat transfer rate Optimum spacing" Optimum Nu0"
Isothermal plates 576 2.873
Symmetric Nu = (Nu0)opt = 1.31
(RaT VRa"7 J

144 2.873 ^ --5


(Nuo)opt = 1.04
-[ ( R a T + VRl']
Asymmetric Nu,

Isoflux plates* T 12 1.88 "1


Symmetric (Nu oX/2 ) opt =0.62.
Nu 0 , i / 2 =[ + ( ^^J

Asymmetric + (Nu o , L/2 ) opt =0.49


o.L/2 '
L Ra^ (Ra^T3 J
"For negligibly thick plates
* Based on the plate temperature at x=L/2

again be used to calculate the total heat transfer from a given


base area and to determine the optimum spacing between
plates when m, the number of thermally active plates, is now
set equal to W/2(b + d). Proceeding as before, the governing
relation for the optimum spacing is found to be 10
L5 7
(2b + 3d-0.02P b )opt=0 (35)
?2 a
For negligibly thick plates, bopi is then given by
bm=2A54/P1'* (36)
At this optimum spacing, Ra' o p t = 21.5 and Nu opl = 1.04.
To maximize the heat transfer rate from each individual, h z
u
thermally active plate, it is again desirable to set the plate
spacing such that fully developed flow does not develop in the
channel and that, as a consequence, the isolated plate Nu limit
is attained along the entire surface. Calculating via equation
m C
10
8
i
/// LEGEND-
(24), Ra' m a x at the 0.99 limit is found to equal approximately 6 III Smooth plates
184and6 m a x = 3.68/P 1 M .
1

1
iC

0
Vertical grooves
Horizontal grooves
Small 2-D grooves
Symmetric, Isoflux Plates. When the boundary con-
ditions along the surfaces of the parallel plates are identically ) 1 large 2-D grooves
or approximately equal to uniform heat flux, total heat
transfer from the array can be maximized simply by allowing
the number of plates to increase without limit. In most electric
z 4 6 8 4 6 8
cooling applications, however, the plate, printed circuit ' 1.0
board, or component surface must be maintained below a b
critical temperature and, as a consequence, plate spacing and (cm)
Nu 0 values cannot be allowed to deteriorate to very small Fig. 6 Influence of grooves on the heat transfer coefficient from
values. isothermal, parallel plates [20]
Recalling the Nu 0 definition of equation (15) and rewriting
equation (29), the relationship between the midheight tem-
perature difference and the other parameters is found to be difference. Thus, when equation (29) is used to evaluate Nu 0
expressible as in the equation (30) formulation of total array heat transfer,
q"b\ 12 1.88 the optimizing equation for the symmetric, isoflux con-
A77./2 = + (37) figuration takes the form
Ra" Ra"
Thus, when both the surface heat flux and the allowable d / QT
temperature difference are specified, equation (37) can be db\2LSWATL/2k
used to solve for the requisite interplate spacing.
Alternately, when only the heat flux is specified, it is of
interest to determine the plate spacing yielding the lowest -!(<*+*-'(*-'--'-)"> 8>
possible surface temperature. This condition corresponds to a where
spacing which is sufficiently large to avoid boundary layer
interference and, by the method previously described, is R=cpp2g0q"/liLk1
found to occur at Ra" equal approximately to 17000 and &max Following differentiation, the governing relation for bopl is
= 1.02R--2. found as
In distinction to the bm3X value and the plate spacing ob- /12 3.76ft4 36
tained via equation (37), the optimum b value for an array of 1
(R
0
R + i*)
(39)
=
isoflux plates an be defined to yield the maximum volumetric
R /opt
(or prime area) heat dissipation rate per unit temperature For negligibly thick plates, i.e., d = 0, the optimum isoflux

Journal of Heat Transfer FEBRUARY 1984, Vol. 106/121

Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms


plate spacing is then found to equal this configuration be preferred over a symmetric distribution
bopl = 1.412 R--2 (40) of the heat dissipation on the array of parallel plates.
Examination of the results for both maximum and optimum
The value of Ra" opt is thus 6.9 and Nu oL/2 at the optimum plate spacing reveals the error inherent in such an approach.
spacing is found to equal 0.62. For isothermal plates bmia was found to equal 4.64 p 0 - 2 5
Asymmetric, Isoflux Plates. By analogy to the symmetric, in the symmetric configuration and 3.68 p-a-25 in the
isoflux configuration, the requisite plate spacing for specified asymmetric configuration. Similarly, 6max equals 7.02 R~0-2
values of q" and ATL/2 on the thermally active surface can be for symmetric, isoflux plates and 5.58 R "-2 when the channel
obtained by appropriate solution of equation (29). is formed by an isoflux plate and an adiabatic plate. Since the
Similarly, equation (29) can be used to determine the lowest plate spacing required for maximum heat transfer from each
Ra" at which the prevailing Nu0 is indistinguishable from the surface in the asymmetric configuration is thus substantially
isolated plate limit. This condition is found to occur at Ra" greater than 50 percent of the symmetric value, the total
equal approximately to 5400 and to yield the plate spacing dissipation of an asymmetric array subject to the same
required to obtain the lowest surface temperature, ftmax, equal constraints must fall below the heat dissipation capability of a
to5.58i?--2. symmetric array.
Finally, when the relation governing the total heat It can be shown that, for a given array base area or volume,
dissipation of an array of alternating isoflux and insulated an optimum array of negligibly thick isothermal plates
plates is differentiated relative to the plate spacing and the alternating with insulated plates cannot dissipate more than
derivative set equal to zero, the optimum value of b for this 63 percent of the heat dissipated by an optimum array of
configuration can be found by solving isothermal plates. This finding is reenforced by the results
obtained by Aung [16], which indicate that thermal asym-
/ 6 , 3.76 64 18 \ metry reduces total heat dissipation to approximately 65
i^b Ki~+-^d) =0 (41)
\R 0A
R R /opt v
' percent of the comparable symmetric configuration when
every second plate is at the ambient temperaure.
For negligibly thick plates
Use of the derived optimum spacing and optimum Nu0
Z7opt = 1.169 i ? 0 2 (42) values for symmetric and asymmetric isoflux channels yields a
The optimum modified channel Rayleigh Number is thus 2.2, nearly identical reduction in total heat dissipation for the
yielding a Nu o t / 2 of 0.49. asymmetric configuration as encountered in isothermal
plates.
Discussion
Three-Dimensional Flow and Geometric Effects. In the
The preceding has established an analytical, albeit ap- present development of design equations for the spacing
proximate, structure for determining the channel width, or between isothermal and isoflux plates no attempt has been
spacing between surfaces forming a two-dimensional channel, made to address the influence of three-dimensional flow, i.e.,
appropriate to various thermal constraints for symmetric and side in-flow or lateral edge effects, on the anticipated Nu0
asymmetric, isothermal, and isoflux boundary conditions. values nor on the recommended optimum spacings. Clearly
With the developed relations summarized in Table 1 and such effects can be anticipated to become progressively
subject to the stated assumptions, it is thus possible to select
greater as the ratio of interplate spacings to channel height is
the interplate spacing which will maximize heat transfer from reduced. In [19] the lateral edge effects for 7.6 cm square
the individual, thermally active surfaces or, alternately, plates were found to be of no consequence for Ra' values
choose the spacing which yields the maximum heat dissipation greater than 10 but to produce deviations of up to 30 percent
from the entire array. In the absence of a large body of or more in the equivalent NuD when Ra' was below 4. Fur-
verified experimental results, the agreement found between thermore, for larger square plates (15.2 x 15.2 cm) the two-
both the composite and optimum spacing relations for dimensional theory has been found to apply for all Ra' values
symmetric, isothermal plates, and the classic Elenbaas [4] greater than 2 [20]. Consequently, while the asymptotic
data, serves to verify the credibility and engineering accuracyapproach of the Elenbaas data [4] to the analytical, two-
of the approach described herein. Several noteworthy features dimensional, fully developed flow limitas shown in Fig.
of the composite and optimizing relations are discussed 2may be fortuitous there is little likelihood of three-
below. dimensional flow effects in the Ra' region corresponding to
the optimum and maximum interplate spacings derived.
Asymmetric Versus Symmetric Fully Developed
Limit. Comparison of the derived relations for the fully While smooth plates may serve as a convenient idealization
developed Nu0 reveals the asymmetric value to exceed the for component-carrying, Printed Circuit Boards (PCB's), in
symmetric value by a factor of two for isothermal surfaces reality such PCB's are better represented by plates with both
and a factor of V2 for the isoflux condition. At first glance horizontal and vertical grooves. This configuration was
this experimentally verified result [9, 10] appears coun- studied in [20] where heat transfer coefficients from two
terintuitive since the thicker thermal boundary layer in the dimensional, grooved, parallel plates were found to exceed
asymmetrically heated channel (equal to the interplate the smooth plate values at small interplate spacings and to
spacing) could be expected to yield lower heat transfer* equal the smooth plate values for spacings appropriate to the
coefficients than encountered with the thinner boundary isolated plate limit. As shown in Fig. 6, the enhancement of
layers of the symmetrically heated configuration. While this the heat transfer rate for small spacings appears to be
assertion is correct for Nusselt numbers based on the local dependent on the groove geometry.
wall-to-fluid temperature difference, it must be recalled that
the Nu0 is defined in such a way as to include the temperature Summation
rise in the convecting air. As a result, Nu0 can be expected to
reflect the "helpful" influence of reduced heat addition in the The complexity of heat dissipation in vertical parallel plate
asymmetric case and to yield the observed higher values. arrays encountered in electronic cooling applications
frequently dissuade thermal analysts and designers from
Asymmetric Versus Symmetric Optimum Arrays. The attempting an even first-order analysis of anticipated tem-
higher Nu0 to be expected in asymmetric configurations has perature profiles and little theoretical effort is devoted to
led some thermal designers to suggest that whenever possible thermal optimization of the relevant packaging con-

122/Vol. 106, FEBRUARY 1984 Transactions of the ASME

Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms


figurations. The foregoing has aimed at establishing an 9 Miyatake, O., Fujii, T., Fujii, M., and Tanaka, H., "Natural Convective
analytical structure for such analyses while presenting and Heat Transfer Between Vertical Parallel PlatesOne Plate with a Uniform
Heat Flux and the other Thermally Insulated," Heat TransferJapanese
verifying useful relations for heat distribution patterns Research, Vol. 4,1973, pp. 25-33.
identical to or approaching isothermal or isoflux boundary 10 Miyatake, O., and Fujii, T., "Free Convective Heat Transfer Between
conditions. Vertical PlatesOne Plate Isothermally Heated and the Other Thermally
Insulated," Heal TransferJapanese Research, Vol. 3,1972, pp. 30-38.
11 Rohsenow, W. M., and Choi, H., Heat, Mass and Momentum Transfer,
References Prentice Hall, New Jersey, 1961.
12 Churchill, S. W., and Usagi, R., " A General Expression for the
1 Bar-Cohen, A., "Fin Thickness for an Optimized Natural Convection Correlation of Rates of Transfer and Other Phenomena," AIChE Journal,
Array of Rectangular Fins," ASME JOURNAL OF HEAT TRANSFER, Vol. 101, Vol. 18, No. 6, 1972, pp. 1121-1128.
1979, pp.564-566. 13 McAdams, W. H., Heat Transmission, McGraw Hill, New York, 1954.
2 Kraus, A. D., Cooling Electronic Equipment, McGraw Hill, New York, 14 Nakamura, H., et al., Paper No. 126, Proceedings of the 42nd National
1962. Meeting of the Japan Society of Mechanical Engineering, Vol. 5, 1964.
3 Aung, W., Kessler, T. J., and Beitin, K. L., "Free-Convection Cooling of 15 Sparrow, E. M., and Gregg, J. L., "Laminar Free Flow Convection from
Electronic Systems," IEEE Transaction on Parts, Hybrids and Packaging, Vol. a Vertical Plate with Uniform Surface Heat Flux," ASME Transaction C,
PHP-9, No. 2, 1973, pp. 75-86. 1956, pp. 435-440.
4 Elenbaas, W., "Heat Dissipation of Parallel Plates by Free Convection," 16 Aung, W., "Heat Transfer in Electronic Systems with Emphasis on
Physica, Vol. 9, No. 1, Holland, 1942. Asymmetric Heating," Bell System Technical Journal, Vol. 52, 1973, pp.
5 Bodoia, J. R., and Osterle, J. F., "The Development of Free Convection 907-925.
between Heated Vertical Plates," ASME JOURNAL OF HEAT TRANSFER, Vol. 84, 17 Wirtz, R. A., and Stutzman, R. J., "Experiments on Free Convection
1964, pp.40-44. Between Vertical Plates with Symmetric Heating," ASME JOURNAL OF HEAT
6 Sobel, N., Landis, F., and Mueller, W. K., "Natural Convection Heat TRANSFER, Vol. 104,1982, pp. 501-507.
Transfer in Short Vertical Channels Including the Effect of Stagger," 18 Levy, E. K., "Optimum Plate Spacings for Laminar Natural Convection
Proceedings - Third International Heat Transfer Conference, Vol. 2, 1966, pp. Heat Transfer from Parallel Vertical Isothermal Flat Plates," ASME JOURNAL
121-125. OF HEAT TRANSFER, Vol. 93, 1971, pp. 463-465.
7 Aung, W., "Fully Developed Laminar Free Convection between Vertical 19 Sparrow, E. M., and Bahrami, P . A., "Experiments on Natural Con-
Plates Heated Asymmetrically," International Journal of Heat and Mass vection from Vertical Parallel Plates with Either Open or Closed Edges,"
Transfer, Vol. 15, 1972, pp. 1577-1580. ASME JOURNAL OF HEAT TRANSFER, Vol. 102, 1980, pp. 221-227.
8 Aung, W., Fletcher, L. S., and Sernas, V., "Developing Laminar Free 20 Horton, S. F . , "Natural Convection from Parallel Plates With Grooved
Convection Between Vertical Flat Plates with Asymmetric Heating," In- Surfaces," MSc thesis, Department of Mechanical Engineering, Massachusetts
ternational Journal of Heat and Mass Transfer, Vol. 15, 1972, pp. 2293-2308. Institute of Technology, Aug. 1981.

Journal of Heat Transfer FEBRUARY 1984, Vol. 106/123

Downloaded From: http://heattransfer.asmedigitalcollection.asme.org/ on 04/07/2014 Terms of Use: http://asme.org/terms