Applied Thermal Engineering 20 (2000) 141±154

www.elsevier.com/locate/apthermeng

Turbulent natural convection cooling of electronic

components mounted on a vertical channel
R. Bessaih*, M. Kadja
Institut de GeÂnie-MeÂcanique, Universite de Constantine, Route d'Ain El. Bey, Constantine, 25000, Algeria
Received 14 May 1998; accepted 20 December 1998

Abstract

The paper presents a numerical simulation of conjugate, turbulent natural-convection air cooling of
three heated ceramic components, which are identical and mounted on a vertical adiabatic channel. A
two-dimensional, conjugate heat transfer model and the standard k±E turbulence model were used to
obtain the dynamic and thermal ®elds. The ®nite-volume method has been used to solve the model
equations throughout the entire physical domain (solid and ¯uid). After validation of the method with
available measurements for a single source, it was applied to investigate the e€ects on cooling of spacing
between the heated electronic components and of the removal of heat input in one of the components.
The former modi®cation led to better cooling while the latter can be partially advantageous only when
the non-powered components are mounted between the powered ones: this reduces the temperature of
the powered components situated downstream from the non-powered component. # 1999 Elsevier
Science Ltd. All rights reserved.

Keywords: Air cooling; Natural convection; Electronic components; Turbulence

1. Introduction

In many electronic cooling situations, arrays of heat-dissipating components are mounted on

vertical (or inclined) parallel plate channels that are open to the ambient at opposite ends. The
simplest method of cooling these arrays is by circulating air vertically via natural convection.
This method of cooling of electronic equipment continues to play an important role in their
thermal management, because it provides the advantage of low noise and high system

* Corresponding author. Tel.: +213-4-92-31-00; fax: +213-4-94-10-66/+213-4-92-56-08.

1359-4311/00/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S 1 3 5 9 - 4 3 1 1 ( 9 9 ) 0 0 0 1 0 - 1
142 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154

Nomenclature

g gravitational acceleration
K non-dimensional turbulent kinetic energy
k non-dimensional thermal conductivity
L the length scale
P nondimensional pressure
Pr Prandtl number
Prt turbulent Prandtl number
q heat density (heat input per unit volume)
Ra Rayleigh number
U nondimensional horizontal component of the velocity
V nondimensional vertical component of the velocity (parallel to g )
X nondimensional horizontal coordinate
Y nondimensional vertical coordinate (parallel to g )

Greek letters
a molecular thermal di€usivity
a t nondimensional turbulent thermal di€usivity
b thermal expansion coecient
t nondimensional time
E nondimensional dissipation
n molecular kinematics viscosity
n nondimensional kinematics viscosity
n t nondimensional eddy viscosity
r ¯uid density
y nondimensional temperature

reliability. Also it does not require maintenance, and is not accompanied by electromagnetic
interference [1].
A large number of studies on natural convection cooling of electronic components have been
done in recent years. Unfortunately most of the numerical ones do not present any validation
of experimental data due to its scarcity. For example, Afrid and Zebib [2] conducted a
numerical study on single and multiple uniformly heated devices. They used a two-dimensional,
conjugate laminar ¯ow model and the analysis of their results led to qualitative suggestions for
improving the overall cooling of a multicomponent system. A similar con®guration was
considered by Said and Muhanna [3], but the components were here taken as protruding from
one of the two vertical walls forming a channel. Sathe and Joshi [4] studied numerically heat
dissipation by natural convection from a heat generating protrusion, mounted on a substrate
inside a square enclosure.
More recently, Huang and Aggarwal [5] investigated numerically the e€ects of wall
conduction on cooling of a centred heat source in a two-dimensional rectangular enclosure.
 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154 143

Heindel et al. [6] performed two- and three-dimensional calculations on laminar ¯ow induced
by a 3  3 array of discrete heat sources ¯ush-mounted to one vertical wall of a rectangular
cavity whose opposite wall was isothermally cooled. The two- and three-dimensional
predictions were found to be approximately identical for heater aspect ratios Ahtr e3. In a
companion paper [7] three-dimensional predictions were compared with experimental data and
heat transfer correlations developed. The work of Desai et al. [8] concerned cooling in
rectangular enclosures with multiple protruding heaters mounted on one side wall, with the top
wall being cooled, and the opposing vertical wall and the bottom wall being insulated. Their
predictions compared well with previous experimental and numerical work.
Among the few experimental studies conducted in this ®eld it is worth mentioning that of
Kang and Jaluria [9], who performed measurements of heat transfer from a heat source
module of ®nite thickness, mounted on a vertical or horizontal surface. The results obtained
indicate that the natural convection ¯ow and the associated heat transfer characteristics vary
strongly with the rate of energy input and the source thickness.
The main objective of this study is to determine the e€ects of spacing between the heated
electronic components and the non-powering of one of the components, in order to give
qualitative suggestions that may improve the thermal design of printed board assemblies.

2. Geometry and model equations

The three similar ceramic components under study are mounted on a vertical adiabatic wall
as illustrated in Fig. 1. The components are 0.10 L high and 0.05 L thick. The spacing between
them is 0.10 L. The spacing between the two adiabatic vertical walls forming the channel is
equal to L, where L is the length scale chosen in all our computations.
For an incompressible ¯uid, with constant thermo-physical properties, except for variation of
density with temperature in the buoyancy force term (i.e. the Boussinesq approximation is
valid), the governing equations for turbulent ¯ow can be written in non-dimensional form as
follow:
Continuity equation
@U @V
‡ ˆ0 …1†
@X @Y
Momentum equation in X direction
    
@U @…UU † @…VU † @P @   @U @   @U
‡ ‡ ˆÿ ‡ Pr …n ‡ nt † ‡ …n ‡ nt † …2†
@t @X @Y @X @X @X @Y @Y
Momentum equation in Y direction
@V @…UV † @…VV †
‡ ‡
@t @X @Y
    
@P @   @V @   @V
ˆÿ ‡ Pr …n ‡ nt † ‡ …n ‡ nt † ‡ Ra
Pr
y …3†
@Y @X @X @Y @Y
144 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154

Fig. 1. The geometry studied: three electronic components mounted on a vertical wall.

Energy equation
    
@y @…Uy† @…Vy† @   @y @   @y
‡ ‡ ˆ …k ‡ at † ‡ …k ‡ at † ‡S …4†
@t @X @Y @X @X @Y @Y

Turbulent kinetic energy equation

      
@K @…UK † @…VK † @  nt @K @  nt @K
‡ ‡ ˆ Pr n ‡ ‡ n ‡
@t @X @Y @X sk @X @Y sk @Y
" 2  2  2 # …5†
 @U @V @U @V Pr2  @y
‡ Prnt ‡ ‡2 ‡2 ÿ E ÿ Ra n
@Y @X @X @Y Prt t @Y

Rate of dissipation of turbulent kinetic energy equation

      
@E @…UE† @…VE† @  nt @E @  nt @E
‡ ‡ ˆ Pr n ‡ ‡ n ‡
@t @X @Y @X sE @X @Y sE @Y
" 2  2  2 # …6†
 E @U @V @U @V E2 Pr2  E @y
‡ C1 Prnt ‡ ‡2 ‡2 ÿ C2 ÿ Ce Ra n
K @Y @X @X @Y K Prt t K @Y
 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154 145

The scales used for the non-dimensionalization of length, time, velocity, pressure, temperature,
kinetic energy and dissipation are: L, L 2/a, a/L, r (a/L )2, qL 2/kS, (a/L )2, a 3/L 4, respectively.
Ra and Pr are the Rayleigh and Prandtl numbers, de®ned respectively as:
gb‰qL2 =kS ŠL3
Ra ˆ
an

n
Pr ˆ
a
where q is the heat density, kS the thermal conductivity of heated components, g the magnitude
of gravitational acceleration, b the thermal expansion coecient of the ¯uid, r its density, n the
kinematic viscosity and a the thermal di€usivity. The non-dimensional viscosity n  is n/nair and
is equal to 1 in the ¯uid region and 1 within the solid regions corresponding to components.
The non-dimensional turbulent eddy viscosity n t is (Cm/Pr )
K 2/E in the ¯uid domain and zero
within the solid regions. The non-dimensional thermal conductivity is k =k/kair, the non-
dimensional turbulent thermal di€usivity a t is (Pr/Prt)
n t and zero in the solid regions. The
energy source S equals 0 for air (no heat generation), and equals k =1000.00 in the heated
components.
The constants of the standard k±E model are those given by Jones and Launder [10] and are:
C1=1.44; C2=1.92; Cm=0.09; sE=1.00; sk=1.30; and Prt=1.00. The constant Ce in the
buoyant term of the dissipation equation was chosen as 0.70.
The boundary conditions used for the components of velocity and for temperature are:
at X = 0.00
@y
UˆVˆ ˆ0 …adiabatic wall†
@X
at X = 1.00
@y
UˆVˆ ˆ0 …adiabatic wall†
@X
at Y = 0.00
@V @y
Uˆ ˆ ˆ0 …channel inlet†
@Y @Y
at Y = 3.00
@U @V @y
ˆ ˆ ˆ0 …channel exit†
@Y @Y @Y
The equations of the turbulent kinetic energy and its dissipation, Eqs. (5) and (6), were solved
only in the ¯uid region. The turbulent kinetic energy is set to zero at solid walls, and its
normal gradients are prescribed as zero at the other boundaries. The equation of dissipation is
not solved at nodes which are adjacent to the wall; in this region, the production and
dissipation of turbulent kinetic energy are equal, which leads to an approximation of the
146 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154

dissipation, E=C 0.75

m K 3/2/k dl, where dl is the distance from the wall to the ®rst node, and k is
the von Karman constant and is equal to 0.41.

3. Numerical method

A ®nite-volume method was used to discretize the two-dimensional partial di€erential

equations of the mathematical model. Scalar quantities (P,y, k and E ) are stored in the centre
of these volumes, whereas the horizontal and vertical components of the velocity (U and V )
are stored on the faces. Power law pro®les are used for the spatial variation of the dependent
variables to ensure realistic results for a wide range of the grid Peclet numbers. The viscosity
and thermal conductivity at the solid±¯uid interface were determined by harmonic averaging.
The SIMPLER algorithm described by Patankar [11] was used to determine the pressure from
the continuity equation. The discretized equations were solved iteratively in each direction
along the axes using the line-by-line tri-diagonal matrix algorithm (TDMA). The conjugate
conduction in the solid regions was handled numerically by solving the same full set of
momentum and energy equations throughout the entire region, but with a large value of
viscosity, i.e. n =1020 speci®ed in each node of the grid covering the components.
The steady-state ¯ow and thermal ®elds are considered to be attained when the maximum of
the variation in the velocity and scalar quantities between two consecutive time steps (Dt ) is
less than 10ÿ6. Runs took approximately 2 h CPU time on a Pentium 133 MHz PC.
Convergence of the numerical solution at consecutive time steps was reached when the mass,
momentum and energy residuals were small enough. The grid used had 32  90 nodes and was
chosen after performing grid independency tests.

4. Results and discussion

4.1. Validation of the method with experimental data

The numerical method developed in this study was ®rst validated using the experimental
data from Cheesewright et al. [12] corresponding to a protrusion-free geometry. They studied
the turbulent natural convection ¯ow in a closed cavity, di€erentially heated from the sides
with adiabatic horizontal walls. Fig. 2 shows the comparison between the air vertical velocity
measured by a Laser-Doppler Anemometer system, for a Rayleigh number equal to 5  1010. It
is clear that the predicted velocity pro®le contains the three distinct regions of the ¯ow: the
upward and downward buoyancy generated near wall regions and the core region which is
stagnant. These predictions are in excellent agreement with measurements.
The method was further validated by the experimental data from Kang and Jaluria [9] on
the cooling of electronic components by natural convection. The authors gave measurements of
the dimensionless local temperature distribution from a protruding thermal source located on a
vertical surface. The predicted variation of the temperature at the plate surface is compared
with measurements in Fig. 3. The di€erence between measurements and predictions can be seen
to be very small and concern only the region downstream from the source.
 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154 147

Fig. 2. Vertical velocity pro®le in the middle height of the cavity: comparison between computations and
experimental data [12].

4.2. Predictions for the case of three equally spaced, equally powered components

From the foregoing formulation, the problem considered in the present study is governed by
the following parameters: the Rayleigh number, Ra, the Prandtl number, Pr, the thermal
conductivity ratio, k , and the geometry parameters. Numerical simulations have been
performed for the values Pr = 0.71 (corresponding to air), and for two Rayleigh numbers,
Ra = 7.3  105 and 3.65  106 (corresponding to the power input of 1 and 5 W, respectively).
The non-dimensional thermal conductivity of ceramic was taken as k S=1000.0. The results
obtained are shown in Figs. 4±7.
In order to reveal important details of the ¯ow structure, the velocity vectors are presented
in Fig. 4(a). The ¯ow ®eld starts with a parabolic-like velocity pro®le, the ¯ow velocities along
the wall are seen to increase in the vertical direction, specially in the region, corresponding to
1.50 < Y < 3.0. In this region, the hot ¯uid is accelerated resulting in an enhanced heat
transfer by convection.
The temperature in each component from the wall towards its edge is almost constant (see
Fig. 4(b)). Also, the temperature variations are concentrated around the heated components. In
the horizontal direction, the temperature drops to ambient temperature not far from the
components, thus forming thin thermal boundary layers.
The turbulent kinetic energy ®eld and its dissipation are illustrated in Fig. 4(c) and (d). In
the solid regions, K is zero (Fig. 4(c)). The jumps of K around the heated components
lead to a complex variation, and reaches its maximum values where the turbulence is enhanced.
However, in the horizontal direction it decreases at the middle of the channel, then increases
close to the plate before dropping to zero at the wall. Regarding the dissipation rate, it is clear
148 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154

Fig. 3. Temperature distribution along the vertical wall: comparison between computations and experimental data
[9].

from Fig. 4(d), that higher values of E are concentrated around the heated components.
Here also, the variation is complex because of the dissipation jumps near the heated
regions.
The eddy viscosity distribution in all the domain (¯uid and solid) is presented in Fig. 4(e).
This ®gure displays an eddy viscosity which is zero at solid regions and maintains an almost
constant value in the vertical direction, especially in the region extending from the inlet to
approximately Y = 1. In the horizontal direction, n t increases progressively between the left
and right surfaces to reach a maximum of 21 times the molecular kinematics viscosity.
The adimensional temperature (y=(TÿTambient)/qL 2/kS) distribution along the wall (X = 0)
predicted for the two Rayleigh numbers, Ra = 7.3  105 and 3.65  106 (corresponding to
power inputs equal to 1 and 5 W) is presented in Fig. 5. The problem geometric parameters
considered here are: component height, 0.10 L; component thickness, 0.05 L; spacing between
components, 0.10 L.
As mentioned in Section 4.2, the temperature of the wall is the maximum temperature of the
component because the temperature distribution in the component is such that it decreases
from the wall to the edge of the component (the edge being in contact with air and therefore is
the coolest part of the component). This is the reason why one is interested in the wall
temperature rather than the `component temperature'. The wall temperature must therefore be
less than the `critical temperature', necessary for thermal control. Also, the results presented in
Fig. 5 are for the dimensionless temperature y=(TÿTambient)/qL 2/kS, which is adimensionlized
using the heat density q. This is why the results for q = 5 W/m3 appear to be lower than those
for q = 1 W/m3. A simple calculation, using Fig. 5, will show that the opposite is true for the
 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154 149

Fig. 4. (a) Velocity vectors; (b) isotherms; (c) turbulent kinetic energy; (d) dissipation of kinetic energy; (e) eddy
viscosity.
150 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154

Fig. 4 (continued).

dimensional temperature, for example:

ymax for continuous line  0:55ˆ)Twall ÿ Tambient ˆ 0:55qL2 =kS ˆ 0:55  5 L2 =kS

ˆ 2:75 L2 =kS

ymax for broken line  1:02ˆ)Twall ÿ Tambient ˆ 1:02 qL2 =kS ˆ 1:02  1:0 L2 =kS

ˆ 1:02 L2 =kS

It is clear from Fig. 5 that for both power inputs considered here, the temperature rise between
the third and second electronic components is smaller than that between the ®rst and second
components. This is due to the progressive increase of the vertical velocity, which enhances the
convective heat transfer coecient, resulting in more heat removal. This result is con®rmed by
the experimental work of Ortega and Mo€at [13], who also reported that the temperature rise
of an array of heated devices mounted on a vertical wall becomes constant after the fourth
device.

4.3. E€ect of non-powered components

Fig. 6(a) presents the variation of the temperature at the wall when all components are
equally powered except the middle one; and Fig. 6(b) gives the temperature variation when all
components are equally powered except the top one. It is clear from Fig. 6(b) that a non-
powered component placed above powered components leads to higher temperatures of the
®rst and second components, i.e. to worse cooling of the components. This is due to a lower
 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154 151

Fig. 5. Pro®les of temperature (y=(TÿTambient)/qL 2/kS) along the wall for the two heat inputs.

buoyancy force created by only these two sources and therefore lower ¯uid vertical velocity,
which in turn decreases the heat transfer coecient. When the middle component is not
powered (Fig. 6(a)), the temperature of the bottom component diminishes for the same reason
as before, while the top component temperature is reduced because there is heat transfer to the
middle component and also because of an increased temperature gradient between the top
component and air. In this case, only the components situated downstream from the non-
powered component are better cooled.

4.4. E€ects of spacing between components

The temperature distribution along the wall for two di€erent spacings between the
components is presented in Fig. 7. It is clear from this ®gure that, when the spacing between
the components is doubled, the temperature of each component is reduced, meaning better
cooling of the components. This must be due to an increase of the recirculation speed between
the two components leading to better side cooling, which enhances the convection heat
transfer. However, this result can only be taken into account in the thermal design of
electronic equipments when miniaturization is not a constraint.

5. Conclusions

Two-dimensional turbulent natural-convection cooling of three heated components mounted

on a vertical wall was investigated numerically. The ®nite-volume method was used to solve the
modelling equations. For all parameters considered, steady-state ¯ow and thermal ®elds were
152 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154

Fig. 6. E€ect of non-powered components on the temperature at the wall, for Ra = 3.65  106.
 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154 153

Fig. 7. E€ect of spacing between the components, for Ra = 3.65  106.

reached. The temperature ®eld in each component was found to be almost uniform. Increasing
the spacing between the components led to a decrease of their temperature, i.e. better cooling.
Worse cooling was obtained when all powered components were placed upstream of a non-
powered component. Finally for the case where non-powered components were mounted
between powered components, only the components situated downstream from the non-
powered component were better cooled.

References

[1] L.T. Yeh, Review of heat transfer technologies in electronic equipment, Journal of Electronic Packaging 117
(1995) 333±339.
[2] M. Afrid, A. Zebib, Natural convection air cooling of heated components mounted on a vertical wall,
Numerical Heat Transfer, Part A 15 (1989) 243±259.
[3] S.A. Said, A. Muhanna, Investigation of natural convection in a vertical parallel-walled channel with a single
square obstruction, in: A.F. Emery (Ed.), Simulation and Numerical Methods in Heat Transfer, vol. HTD 157,
ASME, 1990, pp. 73±80.
[4] S.B. Sathe, Y. Joshi, Natural convection liquid cooling of substrate-mounted protrusion in a square enclosure:
e€ects of thermophysical properties, geometric dimensions and boundary conditions, in: R.A. Wirtz, G.L.
Lehmann (Eds.), Thermal Modelling and Design of Electronic Systems and Device, vol. HTD 153, ASME,
1990, pp. 73±80.
[5] Y. Huang, K. Aggarwal, E€ect of wall conduction on natural convection in an enclosure with a centered heat
source, Journal of Electronic Packaging 117 (1995) 301±306.
[6] T.J. Heindel, S. Ramadhyani, F.P. Incropera, Laminar natural convection in a discretely heated cavity: I-
Assessment of three-dimensional e€ects, Journal of Heat Transfer 117 (1995) 902±909.
154 R. Bessaih, M. Kadja / Applied Thermal Engineering 20 (2000) 141±154

[7] T.J. Heindel, S. Ramadhyani, F.P. Incropera, Laminar natural convection in a discretely heated cavity: II-
Comparisons of experimental and theoretical results, Journal of Heat Transfer 117 (1995) 910±916.
[8] C.P. Desai, K. Vafai, M. Keyhani, On the natural convection in a cavity with a cooled top wall and multiple
protruding heaters, Journal of Electronic Packaging 117 (1995) 34±45.
[9] B.H. Kang, Y. Jaluria, Natural convection heat transfer characteristics of a protruding thermal source located
on horizontal and vertical surfaces, International Journal Heat Mass Transfer 33 (1990) 1347±1357.
[10] W.P. Jones, B.E. Launder, The calculation of low-Reynolds number phenomena with a two-equation model of
turbulence, International Journal Heat Mass Transfer 16 (1981) 1119±1130.
[11] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, McGraw-Hill, New York, 1981.
[12] R. Cheesewright, K.J. King, S. Ziai, Experimental data for the validation of computer codes for the prediction
of two-dimensional buoyancy cavity ¯ows, in: Winter Annual Meeting, HTD 60, ASME, 1986, pp. 75±81.
[13] A. Ortega, R.J. Mo€at, Heat transfer from an array of simulated electronic components: experimental results
for free convection with and without a shrouding wall, in: Heat Transfer in Electronic Equipment, vol. HTD
48, ASME, 1985, pp. 5±15.