You are on page 1of 8

TURBULENCE MODELLING AND ITS IMPACT ON CFD PREDICTIONS FOR COOLING OF ELECTRONIC

COMPONENTS

Kulvir K Dhinsa, Chris J Bailey, Koulis A Pericleous
Centre for Numerical Modelling and Process Analysis
University of Greenwich,
Old Royal Naval College
Greenwich, London, SE10 9LS, UK
Phone: +44(0)20-8331 8141
Fax: +44(0)20-8331 8665
Email: k.k.dhinsa@gre.ac.uk


ABSTRACT

This paper will discuss Computational Fluid Dynamics (CFD)
results from an investigation into the accuracy of several
turbulence models to predict air cooling for electronic
packages and systems. Also new transitional turbulence
models will be proposed with emphasis on hybrid techniques
that use the k model at an appropriate distance away from
the wall and suitable models, with wall functions, near wall
regions. A major proportion of heat emitted from electronic
packages can be extracted by air cooling. This flow of air
throughout an electronic system and the heat extracted is
highly dependent on the nature of turbulence present in the
flow. The use of CFD for such investigations is fast becoming
a powerful and almost essential tool for the design,
development and optimization of engineering applications.
However turbulence models remain a key issue when tackling
such flow phenomena. The reliability of CFD analysis
depends heavily on the turbulence model employed together
with the wall functions implemented. In order to resolve the
abrupt fluctuations experienced by the turbulent energy and
other parameters located at near wall regions and shear layers
a particularly fine computational mesh is necessary which
inevitably increases the computer storage and run-time
requirements. The PHYSICA Finite Volume code was used
for this investigation. With the exception of the k and
k models which are available as standard within
PHYSICA, all other turbulence models mentioned were
implemented via the source code by the authors. The LVEL,
LVEL CAP, Wolfshtein, k , k , SST and kl k /
models are described and compared with experimental data.

KEY WORDS: Thermal Management, Heat Transfer, Low
Reynolds Number (Transitional) Flows, Flow Separation and
Reattachment.

NOMENCLATURE

a
1
SST model constant, 0.31
C
1
k- model constant 1.44
C
2
k- model constant 1.92
C

k- model constant 0.09


CD
k
cross diffusion term
D maximum local length scale, m
E LVEL integrating constant, 9.0
G turbulent generation rate
i internal energy
k turbulent kinetic energy, m
2
s
-2

k
~
thermal conductivity, Wm
-1
K
-1

L distance to the nearest wall, m
l

turbulent dissipation length


l

turbulent mixing length


p pressure, Pa
Re Reynolds number
S source term
S SST model strain rate
t time, s
u velocity vector, ms
-1

u, v, w velocity components, ms
-1

u
+
wall velocity
y
*
Wolfshtein model Reynolds number
y
+
wall distance

Greek symbols
revised LVEL model constant, 0.01
k- model constant, 0.075
* k- model closure coefficients
* SST model constant, 0.09
turbulent dissipation rate, m
2
s
-3

von Karman constant, 0.41
diffusion coefficient
dynamic viscosity, Nsm
-2

t-max
maximum turbulent dynamic viscosity, Nsm
-2

kinematic viscosity, m
2
s
-1

+
effective viscosity
fluid density, kgm
-3

k
k- model constant, 1.0

k
k- model constant, 2.0

k- model constant, 1.3

k- model constant, 2.0


dissipation function
general variable
specific dissipation rate, s
-1


Subscripts
l laminar
t turbulent
+ dimensionless
487
2004 Inter Society Conference
on Thermal Phenomena 0-7803-8357-5/04/$20.00 2004 IEEE


INTRODUCTION

The electronics industry is now developing at a rapid rate and
with it the problems associated with the cooling of electronic
packages is becoming increasing complex.

The electronics packaging design community is now
commonly using CFD to thermally characterize the
performance of electronic packages and systems. These
simulation tools solve both fluid flow and temperature
throughout the system. For fluid flow the classical Navier
Stokes equations are solved, with suitable turbulence models,
to accurately capture the flow around a package.

Flow characterization within electronic products is represented
by the relationship between the turbulent phenomena and the
flow structures within the wake of an electronic package. The
association between the flow structure and the generation of
heat encourages the flow to separate and recirculate locally
between each board-mounted package. The consequence of
these features has the potential to cause the electronic product
to malfunction due to the temperature exceeding some
operational or reliability limit.

Extensive documentation has been published which
characterizes the flow structures present when studying fluid
flow around a surface mounted cubical obstacle, whether
referring to a single or matrix array of obstacles. The
interested reader is referred to the works of Meinders and
Hanjali [1], who concentrate on low Reynolds number flow.
Also the work of Martinuzzi and Tropea [2] and Hussein and
Martinuzzi [3] who consider higher Reynolds number flow
phenomena.

In duct flows, if the Reynolds number (Re) is below 1000 the
flow is said to be laminar; hence the viscous forces dominate.
Above Re = 4000 the inertia forces dominate over the viscous
forces and flow perturbations grow, leading to turbulence.
The range between the two limits is termed the transitional
region. This is the region of most relevance to the electronics
packaging community.

The uncertainty of mathematically modelling turbulence is
reflected in the large variety of models available. These range
from Prandtls zero-equation mixing length model [4] to more
complex higher order models. At present the vast majority of
CFD calculations for this application area use Launder and
Spaldings k model [5], which is optimized for high
Reynolds number flows. For the purpose of this investigation
CFD results will also be shown for the low Reynolds number
implementation of the Wilcox k model [6, 7], Menters
SST model [8], which combines k and k , and a
newly formulated two-layer hybrid kl k / turbulence model
for the prediction of flow around a heated package.

It has now become a priority for electronic cooling CFD
predictions to recognize the areas of low Reynolds number
flow within a system and to exploit transitional turbulence
models which should ideally hold the attributes of being both
accurate and computationally fast with regards to solution
time [9].

Low Reynolds number turbulence modelling tends to be
associated with the key disadvantage of requiring a fine
computational mesh in near-wall regions, for example, this
disadvantage holds true for the Wilcox k turbulence
model. This fine mesh constraint stems from the formulation
of such models abandoning the use of wall functions which
therefore require the solution of the viscosity-affected sub-
layer close to the wall. Such techniques increase the computer
storage and run-time requirements significantly greater than
those methods which employ the wall function approach. In a
typical congested environment of an electronics package, such
attention to wall regions becomes impractical.

Taking the above model disadvantage into consideration
Eveloy et al. [10] recently published work comparing CFD
predictions with experimental data for a particular set of
components on a printed circuit board. These comparisons
showed that the Shear-Stress Transport (SST) turbulence
model [8], which is based on a hybrid approach formulation
that employs the Wilcox k model at near wall regions
and the k model at some distance away from the wall,
results in better predictions for leading edge heat transfer and
component thermal interaction. It was also demonstrated in
[10] that the comparisons between CFD predictions and
experimental data showed that the Spalart-Allmaras (SA) one-
equation model [11] could be considered as a computationally
cheaper alternative to the SST model.

The aim of this research is to formulate a low Reynolds
number turbulence model that can be used in a CFD code for
accurate prediction of flow and temperature in electronic
systems. To be applicable the model should:
be relevant for congested domains containing a mix
of low and high aspect ratio geometries as found in
electronic systems (Figure 1)
compute within a reasonable time, so that it can be
used by electronic thermal design engineers.







Figure 1. Schematic of low/high aspect ratio geometries

The formulated model will be compared with other models to
see how it satisfies the above aim. This includes the k-, k-,
SST and LVEL models although at present we are not
investigating the SA model. This model requires the use of a
trip source in the eddy viscosity transport equation (see [11])
and hence the definition of a trip point in each boundary layer
region. This may be practical for flat-plate type geometries but
488
2004 Inter Society Conference
on Thermal Phenomena
lacks general applicability and should be used with caution for
electronic systems where flow separation and re-attachment is
important.

Dhinsa et al. [12, 13] has demonstrated results for the
Meinders single cube test configuration, using LVEL, LVEL-
CAP, Wolfshtein, k-, and the k- models. In this paper
details have also been presented on the development of the
hybrid turbulence based on k- and k-l which at present looks
encouraging. This paper extends this work by comparing the
results for the Shear-Stress Transport (SST) model [14]
against experimental data from Meinders [15].


MATHEMATICAL MODELS

This section will outline the formulation of all the turbulence
models used in this investigation subject to the general
conservation equation of the quantity . All simulations in
this investigation have been constrained to steady state
conditions therefore allowing the transient term to be ignored
from equation (1), and all subsequent equations.


(1)

With regards to equation (1), it has been necessary to hide
uncommon terms for all the solved variables into the source
term, a summary of which is given in Table 1.

Table 1. Governing equations of flow for a compressible
Newtonian fluid

Equation


S

Continuity 1 0 0
x-
Momentum

u


S
x
p
Mx
+



y-
Momentum

v


S
y
p
My
+



z-
Momentum

w


S
z
p
Mz
+



Energy

i k
~

S
u p
i
+ + div

The source term contains any extra phenomena taking place in
the system, such as the application of wall functions,
gravitational and pressure effects.

LVEL Turbulence Model
This is a simple algebraic turbulence model which does not
require the solution of any partial differential equations. The
model depends on the calculation of the distance to the nearest
wall, the local velocity and the laminar viscosity to determine
the effective viscosity.
The first step of the LVEL model is the solution of the
following Poisson equation, which therefore allows the
calculation of the maximum local length scale and the local
distance to the nearest wall to be completed with relative ease.

(2)


(3)

(4)

With the local speed, V, and L a Reynolds number is
calculated, then using the Reynolds number relationship stated
below together with a version of the universal law of the wall
the calculation of the dimensionless effective viscosity is
obtain by analytical differentiation of the wall function
relationship.
(5)


(6)


(7)


This model results in a turbulent viscosity which varies from
element to element. Further mathematical derivation of the
model can be found in the work of Agonafer et al. [16].

Revised LVEL Model
A single value for the turbulent viscosity based on a user
specified velocity and length scale is calculated, but allows the
value to vary according to the log-law of the wall for any
elements close to solid surfaces.

Model derivation is identical to the LVEL model discussed
above, but in this instant an upper bound constrains the
turbulent viscosity.
(8)

Wolfshtein Turbulence Model
The one-equation Wolfshtein turbulence model [17] solves a
transport equation for the kinetic energy and then uses
empirical functions to describe the turbulent mixing and
dissipation lengths.

A number of modifications have been suggested for the
empirical functions used. One such model modification was
formulated in 1975 by Norris and Reynolds [18], but will not
be discussed any further in this investigation.


(9)





(10)
( )
( ) ( ) ( )
S
u
t



+ = +

grad div div


wall. at the 0 where 1 = =
2
2
+ = D
= D L
y u
+ +
= Re
( ) 435 . 0 ln
1
min , , = |
.
|

\
|
=
+ + +

y E y u
u
y
d
d
v
+
+
+
=
V L
t
=
max
( )

+ = +

|
|
.
|

\
|
|
|
.
|

\
|
+
G
v
k
v
k u
t
k
t
k
t
l
grad div div
l
k C

2 3 4 3
=
489
2004 Inter Society Conference
on Thermal Phenomena


(11)


(12)


Standard k Turbulence Model
Launder and Spaldings two-equation k model is
unarguably the most widely used and validated model
employed for turbulent fluid dynamics to date. The extensive
use of the model has highlighted both the capabilities and
shortcomings of the model.

The model has achieved notable success when dealing with
thin shear layers and recirculating flows without the need for
case-by-case modification of the model constants. Also
success of the model is noted for confined flows where the
normal Reynolds stresses are relatively unimportant compared
to the Reynolds shear stresses which are of utmost importance.

The model is favoured for industrial applications due to its
relatively low computational expense and generally better
numerical stability than more complex turbulence models such
as the Differential Stress Equation Model (DSM) introduced
by Launder et al. [19].

The predominant drawback of the standard k turbulence
model, for this application area, is that the model was
designed for high Reynolds number flows therefore resulting
poorly in terms of model accuracy when considering fluid
flow over populated Printed Circuit Boards (PCB) which is
usually classified as being low Reynolds number flow due to
the small velocities and length scales encountered [20].

The formulation for Launder and Spaldings turbulence model
consists of two transport equations, one equation to describe
the kinetic energy of turbulence and a second related to the
rate of turbulent dissipation.


(13)







(14)



With the eddy viscosity defined below.

(15)


Wilcox k Turbulence Model
This turbulence model was first introduced by Kolmogorov in
1942, a number of improvements have been suggested for the
original model including those of Spalding [21] and Saiy [22].
The most popular suggestions for model improvements came
from Wilcox [23] in 1988 who formulated a low Reynolds
number alternative to the standard k turbulence model.

Wilcox suggested using an equation to represent the frequency
of the vorticity fluctuations rather than using an equation
describing the turbulent dissipation rate. The formulation of
the turbulent kinematic viscosity and the closure coefficients
can be found in [12, 23].



(16)





(17)




It has been suggested by Menter [24, 25] that the core
deficiency of the k turbulence model stems from the
sensitivity of the model to the free stream values of . A
possible solution to this deficiency is to use a combination of
the k model equations implemented near wall regions
and the k turbulence model to be employed in the bulk
flow region. This lead Menter to formulate the Shear-Stress
Transport (SST) turbulence model [14, 26].

Shear-Stress Transport Turbulence Model
The formulation of this two-equation hybrid type turbulence
model is stated below.



(18)





(19)
The blending function F
1
is defined as:



(20)



l k C vt
2 1 4 1
=
( ) ( )
* *
016 . 0 263 . 0
exp , exp 1 1
y y
L L
l l

= =

( )

+ = +

|
|
.
|

\
|
|
|
.
|

\
|
+
G
v
k
v
k u
t
k
t
k
t
l
grad div div
( )
k
C
k
G
v
C
v
u
t
t
t
l
2
2
1
grad div div

+ = +

|
|
.
|

\
|
|
|
.
|

\
|
+



2
k
C
t
=
( )
k G
v
k
v
k u
t
k
t
k
t
l

*
grad div div

+ = +

|
|
.
|

\
|
|
|
.
|

\
|
+
( )

2
grad div div

+ = +

|
|
.
|

\
|
|
|
.
|

\
|
+
G
k
v
u
t
t
l
( )
( ) ( ) | | k k
P
k u
t
k
t k l k
grad div
~
div
*

+ + = +

( )
( )
( ) | | ( )


grad grad
1
1 2 grad div
div
2 1
2 2
+ + +
= +

k
F
S
u
t
t l

|
|
.
|

\
|
=
4
2
2
2 *
1
4
,
500
, max min tanh
L
CD
k
L
v
L
k
F
k
l


490
2004 Inter Society Conference
on Thermal Phenomena


(21)

The turbulent eddy viscosity is calculated as follows:


(22)
The second blending function F
2
is defined by:




(23)
To prevent the build-up of turbulence in the stagnation regions
a production limiter is used




(24)
The model constants are calculated using the F
1
blending
function

(25)
Model constants stated below.





The disadvantage of this model is highlighted when
considering recirculating flow, separation and reattachment.
Menter et al. [14].

Two-Layer Hybrid kl k Turbulence Model
This newly formulated hybrid turbulence model exploits the
advantages of the standard k model by using the original
model equations at a sufficient distance away from the
enclosure walls and any electronic package and then drops
down to a single equation turbulence model in the vicinity of
solid devices.

The model divides the test geometry into two regions; the
allocation of the division is determined by using a critical
Reynolds number. Region A solves the standard k model
equations; Region B on the other hand switches to the l k
model and solves an appropriate set of equations which
represent the turbulent dissipation rate. The l k model also
employs the use of wall functions and damping functions to
make this model economical in terms of run-time. To bridge
the gap between the two turbulence models a matching
technique has been used.

The main advantage of such a hybrid model is that it is
computationally cheap in terms of the mesh requirement
which therefore makes this a fast and economical model to
run. Also the k model advantages will filter through to
the kl k / model; these advantages include outstanding
performance for many industrially significant flows such as
confined flow phenomena.

It should be noted that it has not yet been determined if the
disadvantages of the standard k model also filter through
into the hybrid model. Initial model testing suggests that the
disadvantages of the k model will not impose themselves
onto the hybrid model. Further model derivation can be found
in the work of Dhinsa et al. [12].


MODEL CONFIGURATION

Fluid flow and heat transfer CFD predictions around a single
cube mounted on the base of a low Reynolds number
enclosure are compared against Meinders experimental data.
The numerical analysis has been undertaken using the finite
volume CFD code PHYSICA [27], a multi-physics code
developed at the University of Greenwich.

The test channel has dimensions (1215 x 61 x 600)mm and
uses an inlet velocity of 4.47m/s resulting in a low Reynolds
number of approximately 4440 based on the cube height. The
cube of size H = 15mm is mounted x/H = 50 downstream of
the inlet boundary on the centreline allowing the use of a
symmetry boundary condition in the z-direction.

The test channel base plate is constructed from phenol-
formaldehyde which is 10mm in thickness and has a thermal
conductivity of 0.33W/mK. The structure of the cube can be
decomposed into two materials; the core of the cube
measuring 12mm is constructed from copper which is kept at a
constant temperature of 75C. The second material is an
epoxy resin which encapsulates the copper and has a uniform
thickness of 1.5mm.

The base of the cube does not have an epoxy layer attached
therefore resulting in the overall dimensional structure of the
cube measuring (15 x 13.5 x 15)mm, a schematic of which is
shown in Figure 2.















Figure 2. Schematic representation of the test geometry
Copper Core
75C
Epoxy Layer
1.5mm
x
z
y
Flow
Symmetry boundary condition
Inflow
Outflow
856 . 0 , 5 . 0 0 . 1 , 85 . 0
, 0828 . 0 , 40 3 , 44 . 0 , 9 5
2 1 2 1
2 1 2 1
= = = =
= = = =


k k
|
.
|

\
|

=
10
, grad grad
1
2 max
10
2

k CD
k
( )
F
S
a
k
a
vt
2 1
1
, max
=
(
(

|
|
.
|

\
|
=
2
2 *
2
500
,
2
max tanh
L
v
L
k
F
l
( ) k
P P
x
U
x
U
x
U
P k k
i
j
j
i
j
i
t k
*
. 10 , min
~
=

=
|
|
.
|

\
|
( )
2 1 1 1
1
F F
+ =
491
2004 Inter Society Conference
on Thermal Phenomena


RESULTS AND DISCUSSION
The CFD predictions for this investigation have been
conducted on a computational mesh density of approximately
350,000 elements. For any k turbulence model
simulations the computational mesh density has been
increased to approximately 575,000 elements. This increased
mesh density is necessary at locations of critical importance,
such as around the cube, as the k model does not utilize
wall functions and therefore needs a relatively fine
computational mesh at solid interfaces where viscous sub-
layers exist. It should also be mentioned that a grading
technique has been used to obtain a computationally fine mesh
at any solid/fluid interface.

The flow over surface mounted bluff bodies is often associated
with the separation of shear layers at the side and top faces of
the body in question. These shear layers may reattach either
on the channel wall or between neighbouring obstacles when
considering matrix arrays.

It has been identified in previous work [1, 2, 3] conducted in
this area that when simulations of this type of phenomena are
being undertaken four main flow features are observed.

With reference given to Figure 3, the most pronounced flow
feature that emerges is the structure of the horseshoe vortex
which originates upstream of the windward face and extends
around both lateral sides of the cube whilst weakening in the
streamwise direction. The flow in proximity to the horseshoe
vortex is characterized as being in an unsteady turbulent state
while the core flow in the corridor above the cube remains
almost undisturbed.

The separation of the top shear layer results in a bound
recirculation vortex located at the leading top edge of the cube
and is observed only for the first cube when considering an
array or a matrix of cubes. The footprint left by the vortex is
identified as being two counter rotating circles [28].





Figure 3. Schematic representation of the flow around a
surface mounted cube (Hussein & Martinuzzi [3])

The separated side shear layers are the result of high vorticity
recirculations close to the leading edge. Their origins are
located at the channel floor with the vortex tubes covering a
significantly large surface area of both the side faces. Also the
vortex tubes are observed to be confined by the main flow and
the presents of the cube.

Finally an arc-shaped vortex confined to the depth and height
of the cube dominates the wake flow. This vortex is caused by
a strong up wash close to the leeward face. A similar counter
rotating footprint, as that found on the top surface of the cube,
is also observed.

The post processing tool Tecplot [29] was used to identify the
four recirculation regions with the aid of flow streamlines.
The CFD simulated results are presented in Figure 4 for the
two-layer hybrid kl k / turbulence model.

Referring to Figure 4 the streamline plots A and B represent
the xy-plane with plot B focusing on the bound recirculation
vortex. This visualization plane is taken at the location of the
symmetry boundary condition. Plots C and D correspond to
the xz-plane with plot D focusing on the side recirculation
vortex tube. This plane is located at y/H = 0.5.

It can be concluded at this point that all the flow features
observed in Meinders experimental work have been
reproduced in this investigation.


A) Horseshoe, bound and wake vortices.

B) Bound vortex.

C) Side and wake vortices.

D) Side vortex.

Figure 4. Streamline plots identifying the flow
recirculation regions

CFD x-direction velocity results have been extracted for the
nearest element centre of each control volume extending along
the normal direction. The results, shown in Figure 5, for all
turbulence models considered in this investigation have been
validated against Meinders experimental data for the
streamwise velocity profile in the normal direction at a
Side vortex
recirculation tube
Arc-shaped
vortex
Horseshoe
vortex
Bound
recirculation
492
2004 Inter Society Conference
on Thermal Phenomena
location of x/H = 6.7 downstream of the inlet at the centreline
of the channel.



Figure 5. x-velocity profile at a location of x/H = 5.7
downstream of the cubes leeward face.

There are several conclusions which can be drawn from the
results presented in Figure 5. It can clearly be identified that
for all turbulence models considered the greatest discrepancies
are located at the wake region of the cube and at the top test
channel plate.

Secondly both the LVEL type models together with
Wolfshteins model seem to produce the most accurate results,
especially in the two problematic regions identified above.
The SST model is still recovering after the wake recirculation
region; this poor flow recovery rate has also been identified by
Menter et al. [30].

Finally it has become evident that the standard k
turbulence model significantly over predicts the profile in the
bulk flow, which suggests that this model experiences thicker
boundary layers compared to the other models investigated.
Also the model is noted to under predict in the wake region of
the cube most severely in comparison to all other turbulence
models.

Surface temperature profiles have also been taken along the
vertical path around the cube. It should be noted that the
temperature results for the Wolfshtein one-equation turbulence
model have not yet been completed and will therefore not be
mentioned any further.



Figure 6. Surface temperature measurements along the vertical
path ABCD

CFD temperature results are located at the first element centre
in the epoxy resin material, not on the surface itself which is
where the experimental results are taken.

As analysis is undertaken of the temperature profile shown in
Figure 6 it is worth noting that if inaccuracies exist in the flow
domain these will also filter through to the temperature
domain, suggesting that an initial temperature result forecast
can already be made for the standard k model. It has
been shown that the flow results for the k model poorly
predict the actual phenomena taking place within the system,
this leads onto the proposal that the k model is likely to
demonstrate poor temperature predictions as with flow

.

Considering the rear face (AB) of the cube in Figure 6 it can
be concluded that the two-layer hybrid kl k / turbulence
model produces the best comparison with the experimental
data, suggesting that the wake flow vortex that is resident in
this region is accurately predicted. For the top face (BC) all
turbulence models investigated fail to capture the sharp peak
in the data which coincides with the bound recirculation
region. Nevertheless both LVEL type models, k and the
SST turbulence model do seem to show a slight increase in
temperature close to the top leading edge of the cube.

Finally at the front face (CD) of the cube both LVEL type
models and the two-layer hybrid kl k / model predict the
closest agreement at the mid face of the cube. It is evident
that the standard k model on the other hand performs
poorly, noticeably at the top and front faces.

It should be mentioned that the experimental data was
collected with a technique referred to as Infrared
Thermography (IR) which may have attributed to possible
inaccuracies at the corners of the cube, due to the restriction of
such a data retrieval system.


CONCLUSIONS

The major contributor to electronic malfunction is the
presence of the flow recirculations discussed in this
investigation. As the recirculation vortices develop they
encapsulate heat from a package and restrict further heat being
extracted and removed from the system.

From the results presented in this paper it can be concluded
that the standard high Reynolds number k turbulence
model struggles to accurately predict real world phenomena
even for simplistic cases. Other models investigated give
better results, but no clear winner at present emerges.

Although velocity data around the cube itself where not available at
the time of writing

493
2004 Inter Society Conference
on Thermal Phenomena
This is work in progress, and further studies will concentrate
on developing the two-layer hybrid kl k / turbulence model
which shows at this stage, considerable promise.


ACKNOWLEDGMENT

This research, undertaken at the University of Greenwich, is
financially supported by the Engineering and Physical
Sciences Research Council (EPSRC) through the PRIME
Faraday organization as an Industrial Case Award. The
authors would also like to acknowledge helpful discussions
with Flomerics Ltd. relating to this application field.


REFERENCES

[1]. E. R. Meinders and K. Hanjali, Vortex Structure and
Heat Transfer in Turbulent Flow over a Wall-Mounted Matrix
of Cubes, International Journal of Heat and Fluid Flow, vol.
20, pp. 255-267, 1999.
[2]. R. J. Martinuzzi and C. Tropea, The Flow around
Surface-Mounted, Prismatic Obstacles Placed in a Fully
Developed Channel Flow, Journal of Fluid Engineering, vol.
115, pp. 85-92, 1993.
[3]. H. J. Hussein and R. J. Martinuzzi, Energy Balance for
Turbulent Flow around a Surface Mounted Cube Placed in a
Channel, Physics of Fluids, vol. 8, pp. 764-780, 1996.
[4]. J. O. Hinze, Turbulence, 2
nd
Edition, New York, NY,
1975.
[5]. B. E. Launder and D. B. Spalding, The Numerical
Computation of Turbulent Flows, Computer Methods in
Applied Mechanics and Engineering, vol. 3, pp. 269-289,
1974.
[6]. D. C. Wilcox, The Remarkable Ability of Turbulence
Model Equations to Describe Transition, DCW Industries
Inc., La Canada, CA, 1993.
[7]. D. C. Wilcox, Turbulence Modelling for CFD, DCW
Industries Inc., La Canada, CA, 1993.
[8]. F. R. Menter, Two-Equation Eddy Viscosity Turbulence
Models for Engineering Applications, AIAA Journal, vol. 32,
no. 8, pp. 1598-1605, 1994.
[9]. P. Rodgers, Prediction of Microelectronics Thermal
Behavior in Electronic Equipment: Status, Challenges and
Future Requirements, Proc 4
th
International Conference on
Thermal and Mechanical Simulation and Experiments in
Micro-Electronics and Micro-Systems, EuroSimE 2003.
[10]. V. Eveloy, P. Rodgers and M. S. J. Hashmi, An
Experimental Assessment of Computational Fluid Dynamics
Predictive Accuracy for Electronic Component Operational
Temperature, Proc ASME Heat Transfer Conference, Las
Vegas, Nevada, USA, 2003, Paper Number HT2003-47282.
[11]. P. Spalart and S. Allmaras, A One-Equation Model for
Aerodynamic Flows, Technical Report AIAA-92-0439, 1992.
[12]. K. Dhinsa, C. Bailey and K. Pericleous, Investigation
into the Performance of Turbulence Models for Fluid Flow
and Heat Transfer Phenomena in Electronic Applications,
20
th
IEEE SEMI-THERM Symposium, pp. 278-285, 2004.
[13]. K. Dhinsa, C. Bailey and K. Pericleous, Low Reynolds
Number Turbulence Models for Accurate Thermal
Simulations of Electronic Components, Proc of the
International Conference on Thermal and Mechanical
Simulation and Experiments in Micro-Electronics and Micro-
Systems, EuroSimE 2004. (in press)
[14]. F. R. Menter, Ten Years of Industrial Experience with
the SST Turbulence Model, Proc of the 4
th
International
Symposium on Turbulence, Heat and Mass Transfer, pp. 625-
632, 2003.
[15]. E. R. Meinders, Experimental Study of Heat Transfer in
Turbulent Flows over Wall-Mounted Cubes, PhD Thesis,
Faculty of Applied Science, Delft University of Technology,
Delft, Netherlands, 1998.
[16]. D. Agonafer, Gan-Li and D. B. Spalding, The LVEL
Turbulence Model for Conjugate Heat Transfer at Low
Reynolds Numbers, Application of CAE/CAD Electronic
Systems, EEP ASME vol. 18, 1996.
[17]. M. Wolfshtein, The Velocity and Temperature
Distribution in One-Dimensional Flow with Turbulence
Augmentation and Pressure Gradient, International Journal of
Heat and Mass Transfer, vol. 12, no. 3, pp. 301-318, 1969.
[18]. L. H. Norris and W. C. Reynolds, Turbulent Channel
Flow with a Moving Wavy Boundary, Report No. FM-10,
Stanford University, Mechanical Engineering Department,
USA, 1975.
[19]. B. E. Launder, G. J. Reece and W. Rodi, Progress in the
Development of a Reynolds Stress Turbulence Model,
Journal of Fluid Mechanics, vol. 68, no. 3, pp. 537-566, 1975.
[20]. V. Eveloy, P. Rodgers and M. S. J. Hashmi, Numerical
Prediction of Electronic Component Heat Transfer: An
Industry Perspective, 19
th
IEEE SEMI-THERM Symposium,
2003.
[21]. D. B. Spalding, Mathematical Models of Turbulent
Transport Processes, Imperial College, Mechanical
Engineering Department, HTS/79/2, 1979.
[22]. M. Saiy, Turbulent Mixing of Gas Streams, PhD
Thesis, Imperial College, University of London, 1974.
[23]. D. C. Wilcox, Reassessment of the Scale Determining
Equation for Advanced Turbulence Models, AIAA Journal,
vol. 26, no. 11, pp. 1299, 1988.
[24]. F. R. Menter, Influence of Free Stream Values on
k Turbulence Model Predictions, AIAA Journal, vol. 30,
no. 6, pp. 1657-1659, 1992.
[25]. F. R. Menter, Improved Two-Equation k
Turbulence Model for Aerodynamic Flows, NASA TM-
103975, 1992.
[26]. F. R. Menter, Zonal Two-Equation k Turbulence
Models for Aerodynamic Flows, AIAA Technical Paper, 93-
2906, 1993.
[27]. PHYSICA Version 2.12, Multi-Physics Software Ltd.,
(http://www.multi-physics.com/)
[28]. E. R. Meinders, T. H. van der Meer and K. Hanjali,
Local Convective Heat Transfer From an Array of Wall-
Mounted Cubes, International Journal of Heat and Mass
Transfer, vol. 41, no. 2, pp. 335-346, 1998.
[29]. Tecplot Version 9.0, Amtec Engineering Inc.,
(http://www.amtec.com/)
494
2004 Inter Society Conference
on Thermal Phenomena

You might also like