Вы находитесь на странице: 1из 8

Increasing the thermal performance of convective systems through

boundary layer shaving


Alexandre K. da Silva
a,
, Louis Gosselin
b,1
a
Department of Mechanical Engineering, University of Texas at Austin, 1 University Station, C2200, Austin, TX 78712, USA
b
Dpartement De Gnie Mcanique, Universit Laval, Quebec City, QC G1V 0A9, Canada
a r t i c l e i n f o
Article history:
Received 29 January 2013
Received in revised form 25 June 2013
Accepted 27 June 2013
Keywords:
Forced convection
Boundary layer
Thermal efciency
a b s t r a c t
This is a numerical study where we propose and quantify a technique for increasing the efciency of
convective systems. The concept is based on trimming or shaving and locally discarding the
warmest part of a developing thermal boundary layer such that uid at moderate temperatures is
always absorbing thermal energy from the heated surfaces, which are not allowed to overheat. For
the present study, a parallel plate channel that is subjected to an internal forced convective cooling
is the conguration considered. Therefore, geometrically speaking, the proposed trimming technique
simply consists in optimizing a blunt reduction of the spacing between the upper and lower plates
composing the channel along its axial location, such that the new conguration has a step-like geom-
etry, allowing the heated coolant to be discarded at the transition between the two sections. The
numerical results, which are directly compared with the performance of an optimally spaced ow
between two parallel plates, show that the thermal performance can be increased by over 20% with
two levels of trimming along the ow axis for forced convection i.e., three sections with different
plate-to-plate spacings. The numerical predictions are also compared with scaling-based results
showing a good agreement. Additionally, numerical simulations also considered buoyancy driven
ows (i.e., natural convection); however, no signicant performance gains were observed when the
boundary layer trimming was used in that case.
2013 Elsevier Ltd. All rights reserved.
1. Introduction
Because convection heat transfer is one of the most traditional
methods for transferring thermal energy, this process has been
widely considered in several studies, e.g., [13]. Generally speak-
ing, most studies aim to parametrically characterize several
aspects of systems and congurations subjected to convective
thermal losses/gains (e.g., [4,5]), while others aim to improve
convective thermal performance, e.g., [6,7].
With respect to performance improvement methodologies,
several techniques were proposed. For instance, numerous studies
detail optimized geometric aspects for convective systems in
forced, natural and mixed convection. Among the parameters con-
sidered, focus has been given to the determination of optimal
hydraulic diameters of channels [8,9], ideal shapes of enclosures
and loops with internal convective transfer [10], etc. It is funda-
mental to mention that, in many of these analyses, the optimiza-
tion shows itself extremely relevant as changes in performance
can be fairly substantial when the optimal geometric values are
disregarded e.g., [1].
Attempts to optimize boundary conditions of convective sys-
tems have also been suggested. Studies like the ones shown in Refs.
[1114] detail the optimization of convective systems that are
heated by a series of individual heaters. In general, the results sug-
gest that a larger density of heaters should be present in locations
with smaller thermal resistances (i.e., thin boundary layers.) While
this optimization is unquestionably geometry-based, if the ap-
proach is taken to the limit where each heater is simply an inni-
tesimal or a punctual source, out of numerous sources, these
results translate into surfaces with a variation of heat ux in the
ow direction [15]. In fact, this trend has been further validated
by numerically considering the Graetz problem from a purely
numerical standpoint, with the goal of determining the optimal
longitudinal distribution of heat ux on the walls [16]. The perfor-
mance gains show that the heat ux tends to decrease towards the
outlet, which is equivalent, in a way, to the reduction of the density
of punctual heaters commented above. Another important class of
design improvements proposed in literature consists in lling the
unheated space at the inlet of a parallel plate channel with plates
of different lengthscales [17,18].
0017-9310/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2013.06.068

Corresponding author. Tel.: +1 (512) 232 0866; fax: +1 (512) 471 1045.
E-mail address: akds@mail.utexas.edu (A.K. da Silva).
1
Tel.: +1 (418) 656 7829; fax: +1 (418) 656 7415.
International Journal of Heat and Mass Transfer 67 (2013) 272279
Contents lists available at ScienceDirect
International Journal of Heat and Mass Transfer
j our nal homepage: www. el sevi er . com/ l ocat e/ i j hmt
In this study, we intend to contribute to the performance
enhancement portfolio of convective systems by describing a
method that eliminates overheated volumes of coolant from these
systems, allowing for better temperature management. The con-
cept, which can be seen as the shaving and local discarding of
the warmest fraction of the boundary layer developing onto heat
dissipating surfaces, is applied to straight channels composed of
two parallel plates, which is subjected to forced and natural con-
vection. The shaving can be considered equivalent to a channel
with a porous wall (e.g., [19,20]), where the heated uid close to
the wall is being continuously expelled along the channel.
2. The concept for forced convection cooling
It is well understood that for any thermal system that is subject
to a given amount of heat dissipation, its thermal performance is
inversely proportional to its maximum temperature [1,6]. For a
conventional symmetrically heated channel with internal forced
convection, such as the one shown in Fig. 1A, for example, the max-
imal temperature will occur at x = L. Therefore, aiming to reduce
the systems maximum temperature while using a given amount
of coolant, the present study proposes geometric changes to the
channel. The idea is to shave the warmest part of the thermal
boundary layer (i.e., the volume of uid that is the closest to the
heated walls) by imposing a reduction to the channel spacing as
shown in Fig. 1B note that because the conguration shown in
Fig. 1B has only one blunt transition, this design is identied as
S = 1, where S in the number of trimming levels (for the congura-
tion shown in Fig. 1A, we have S = 0). According to the method, the
heated part of the coolant is discarded at the blunt transition
between the two sections and, ideally, centrally located coolant
with a reduced temperature is pushed over the remaining heated
section of the channel see Fig. 1B. In the limiting case where
small fractions of coolant are continuously withdrawn from the
channels wall (see Fig. 1C), we would, most likely, reach a
converging channel with a permeable wall.
While it is acceptable that the maximal temperature be reduced
by eliminating the heated coolant, one can also realize that, at each
transition, coolant located at the center of the channel, which has a
higher velocity than the uid near to the walls, is once again sub-
ject to higher levels of shear stress since the hydrodynamic bound-
ary layer is, in a way, re-developing at each transition.
Consequently, for a system subject to a given pressure drop, there
might be a limit where the coolant is being slow down to the point
where the system temperature increases. Therefore, the optimiza-
tion of this problem can be seen as a search for a continuously
developing thermal boundary layer without inicting overwhelm-
ing gains in uid friction.
Recall, however, that the comparison between congurations
with different S values becomes relevant if, at least two constraints
are enforced. The rst one is the pressure drop used to push the
coolant through the channel. For the present study, we used the
Nomenclature
Be dimensionless pressure drop
c
p
specic heat (J/kg K)
D plate-to-plate spacing (m)
g gravity (m/s
2
)
k thermal conductivity (W/m K)
L plate length (m)
P pressure (Pa)
DP pressure drop (Pa)
q
00
heat ux (W/m
2
)
Ra Rayleigh number
S number of trimming sections
t
i
channel trimming height (m)
T temperature (K)
u velocity, x-direction (m/s)
U mean velocity, x-direction (m/s)
v velocity, y-direction (m/s)
V
0
coolant volume per unit of length (m
2
)
Greek symbols
a thermal diffusivity (m
2
/s)
b expansion coefcient (1/K)
d boundary layer thickness (m)
l dynamic viscosity (kg/ms)
q density (kg/m
3
)
Subscript
$ non dimensional variable
0, 1, 2 channel sub-section index
opt optimal
max maximum
Fig. 1. Parallel plate channel in forced convection: (A) traditional channel (i.e.,
S = 0), (B) channel with shaving level (i.e., S = 1), (C) channel with large number of
shaving steps (i.e., S = N).
A.K. da Silva, L. Gosselin/ International Journal of Heat and Mass Transfer 67 (2013) 272279 273
well-known dimensionless pressure drop referred to as the Bejan
number, or simply Be, which is dened as Be = (DPL
2
l
1
a
1
) [1].
The second constraint is the volume of coolant owing inside the
channel at a given instant, V
0
. Note that V
0
is in fact, a volume of
coolant per unit of length since all congurations considered in this
study are 2-D. The volume constraint can be written as:

n
i0
L
i
D
i
V
0
1
where L
i
and D
i
are the local length and the plate-to-plate spacing
for each of the sections, as indicated in Fig. 1B. Furthermore, note
that the constraint given by Eq. (1) can also be written as a function
of D
i
and t
i
the latter is indicated in Fig. 1.
Finally, to clearly quantify the gains of the proposed technique,
we choose to use, as a reference, a previously optimized channel
composed of two parallel plates of length L and spacing D
opt
, where
D
opt
represents the optimal spacing for such conguration as given
in Refs. [1,9]. Since D
opt
is a function of the dimensionless pressure
drop imposed, i.e., Be, and three values of Be were considered in
this study, we were able to restrict the volume of uid needed
for each value of Be, regardless of the number of sections or shav-
ing levels imposed to the channel. Therefore, any performance gain
(i.e., reduction of the maximal temperature within the channel)
will be relative to the already optimized straight channel with no
boundary layer trimming (i.e., S = 0).
According to the numerical optimization of a simple channel
composed of parallel plates, the volume of uid per unit of length
perpendicular to Fig. 1A available for each pressure drop number is
presented in Table 1.
Note that the relation between V
0
and Be shown in Table 1 is not
new see Refs. [1,9]. In fact, the relation shown in Table 1 between
V
0
and Be will be used later in the text (see Section 4) to validate
the numerical code used in this study against a previously pub-
lished results. Finally, using L and the standard length scale (see
Eq. (2) in the next section), V
0
can be written as a dimensionless
variable

V
0


D
opt
.
3. Scaling formulation for laminar forced convection
The present scale analysis is based on the S = 1 case, which can
easily be extended to a more complex geometry, i.e., S > 1. For the
scale analysis, let us consider that the uid average velocity in the
rst section of the system (0 < x < L
0
) is U
0
, and in the second por-
tion, U
1
. As different sets of design variables are considered, the
heat transfer and uid ow features of the channel can change sig-
nicantly. For example, when t
0
is small, a boundary layer will
grow on the rst wall, and only a fraction of it will be shaved. War-
mer uid will then continue its way to the second wall, which is
undesirable since the idea here is to bring fresh uid to the second
portion of the system. Clearly, the rst step must be thick enough
to remove the warmer uid. On the other hand, t
0
should not be
too large since that would reduce the wall-to-wall spacing in the
second portion of the channel and could lead to a developed ow
in that portion which would be detrimental. The ideal situation
is thus to have a shaving such that the opening t
0
is equal to the
boundary layer thickness at x = L
0
. Therefore, considering the vari-
ables bellow,

L
i
;

t
i
;

D
i

L
i
; t
i
; D
i
L
;

T
T T
0
q
00
L=k
;

U
U
i
DPL=l
2
the optimal conguration for the L
0
plate section can be written as
[1]:
d
0
t
0
$ 1 4:92

L
1=2
0

t
0

U
1=2
0
Pr
Be
_ _
1=2
: 3
Such a design would lead to the growth of a second boundary layer
on the second wall. Similarly, at the outlet of the system (at
x = L
0
+ L
1
), the ideal situation is to have the boundary layers touch
at the very outlet of the channel, i.e.,:
d
1
t
1
$ 1 4:92

L
1=2
1

t
1

U
1=2
1
Pr
Be
_ _
1=2
4
Comparing both equations, it follows that in the optimal design:

L
0

t
2
0

U
0
$

L
1

t
2
1

U
1
5
Now, let us try to estimate the hot spot temperature. Assuming
again that boundary layers grow on each wall in the optimal design,
it can be shown that the temperature at L
0
and L
1
are respectively
[1]:

T
max
x

L
0
$
Pr

L
0

U
0
Be
_ _
1=2
6
and

T
max
x

L
1
$
Pr

L
1

U
1
Be
_ _
1=2
: 7
In an optimized design, any local overheating is prohibited, and the
hot spot temperature at each of the outlets will have the same order
of magnitude, i.e.,:

T
max
x

L
0
$

T
max
x

L
1
8
If that were not the case, it would be possible to modify the design
to enhance its performance. For example, assume that a certain
design provides a hot spot at x

L
0
. This means that the second
portion of the system works well with a large spacing, whereas
the rst portion is likely too thin. Therefore, more of the space
allocated to the second portion could be given to the rst portion
in order to maximize overall performance. In the end, a good
balance in the system should lead to similar hot spot temperatures
at each outlet.
Now, comparing Eqs. (6), (7), (8), and (5), it follows that at the
optimum, we have:

L
0

U
0
$

L
1

U
1
9

t
0
$

t
1
10
Eq. (10) in itself is useful for design. Eq. (9), though, needs to be re-
worked since in the present problem,

U
0
and

U
1
are unknowns:
they depend on the imposed pressure drop and on the geometry
of the system.
The pressure drop due to a section of duct with distinct bound-
ary layers is DP 0:664 lU
3=2
L
1=2
_ _
= Dm
1=2
_ _
[1]. Considering the
total pressure drop by summing the drops for both segments, it
follows that:
Table 1
Relation between pressure drop and volume
per unit of length constraints.
Be
V
0
=L
2
=

Dopt
10
5
0.169
10
6
0.096
10
7
0.054
274 A.K. da Silva, L. Gosselin/ International Journal of Heat and Mass Transfer 67 (2013) 272279
1 0:664
Be
Pr
_ _
1=2

U
3=2
0

L
1=2
0

D
0

U
3=2
1

L
1=2
1

D
1
_ _
11
From geometrical considerations, it is noted that

D
0

V
0
2

L
1

t
1

L
0
2

t
0

t
1
$ 4

t
1
and

D
1
2

t
1
. Combining these relations
with Eq. (11) yields:

t
1
0:332
Be
Pr
_ _
1=2

U
3=2
0

L
1=2
0
2


U
3=2
1

L
1=2
1
12
Now with Eq. (9), it follows that:

U
0
$
2

t
1
0:664
Be
Pr
_ _
1=2

L
1=2
0
2

L
1=2
1

L
3=2
0
_ _
_

_
_

_
2=3
13
Using the fact that

L
0

L
1
1, and remembering that we want to
maximize the ratio

U
0
=

L
0
see Eq. (6), one can observe that maxi-
mizing

U
0
=

L
0
can be achieved by modifying geometrical parame-
ters, since:

U
0

L
0
/

L
2
0
2
1

L
0
_ _
2
_ _
2=3
14
Taking the derivative with respect to

L
0
and equating to zero yields:

L
0;opt

2
3
and

L
1;opt

1
3
15
Interestingly, these optimal lengths do not depend on the Bejan
number. Finally, the optimal

t
0
and

t
1
values are derived from the
volume constraint. Remembering that

t
0;opt

t
1;opt
:

V
0

2
3
4

t
1
3
2

t
10
3

t 16
or simply

t
0;opt
$

t
1;opt
$ 0:3

V
0
17
Finally, the optimal geometry can be inserted in the hot spot tem-
perature expression, Eq. (5), in order to estimate the minimized
maximum temperature:

T
max;min
$ Be
1=4
18
In developing Eq. (18), it was assumed that

V
0
scaled as Be
1/4
which was the dependence originally found in the literature, e.g.,
[1], and also shown in Table 1.
4. Numerical formulation for forced convection cooling
The mathematical model for the system in forced convection
described previously is based on the 2-D, constant property formu-
lation of the conservation equations of mass, momentum and en-
ergy. In fact, the implementation of the system of differential
equations used here for forced convection follows the dimension-
less scales used in Ref. [7], which, in a dimensional form, yields:
@u=@x @v=@y 0 19
q u@u=@x v@u=@y @P=@x l @
2
u=@x
2
_ _
@
2
u=@y
2
_ _ _
20a
q u@v=@x v@v=@y @P=@y l @
2
v=@x
2
_ _
@
2
v=@y
2
_ _ _
20b
qc
P
u@T=@x v@T=@y k @
2
T=@x
2
_ _
@
2
T=@y
2
_ _ _
21
for the conservation of mass, momentum in the x direction,
momentum in the y direction, and energy, respectively.
A total of six boundary conditions were needed to bound the
numerical domain, three hydrodynamic and three thermal. The
thermal boundary conditions are: uniform heat ux at all walls,

T = 0 at the inlet and zero normal heat ow at the outlet, while


the hydrodynamic boundary conditions are: zero velocity at all
walls, applied pressure at the inlet and zero normal stress at the
exit planes.
The conservation equations were solved with Comsol Multi-
physics [21]. A standard validation procedure was applied to the
model created, which required a mesh study and a direct compar-
ison with third party results. While the mesh validation was easily
obtained and required smaller element sizes the higher the ow
strength (i.e., Be), the direct comparison was done with well estab-
lished correlations for the optimal spacing and maximal conduc-
tance (C) for a straight channel with forced convection as the one
shown in Fig. 1A, e.g., [6,9], where C $ 1=D

T
max

D
opt
. For instance,
for forced convection Ref. [1] suggests that

D
opt
3:2Be
1=4
and
C 0:4Be
1=2
. The maximum difference between the values calcu-
lated numerically and the ones proposed by the correlations for
10
5
6 Be 6 10
7
are shown below in Table 2. The comparisons indi-
cate a very good agreement for the range of Be considered, hence
validating the numerical methodology.
5. Results for laminar forced convection
We start the optimization process by analyzing and describing
the effect of some key variables on the performance of a channel
with a single shaving level (i.e., S = 1). In Fig. 2, we vary

L
0
while
enforcing the equality

t
0
=

t
1
and respecting the constraint given
by Eq. (1) for all cases simulated. This leaves only one design var-
iable, namely

L
0
. As can be seen in Fig. 2,

T
max
is drastically affected
by

L
0
with fairly large slopes lying on both sides of

L
0;opt
indicating
the relevance of the optimization. Furthermore, the occurrence of
an optimal value of

L
0
can be understood by realizing that, in either
limit (i.e.,

L
0
!1 and

L
1
!1) the channel with a single step-like
geometry approaches a traditional channel (i.e., S = 0) where high-
er temperatures are expected. In fact, it can be observed that, in
both limits,

T
max
approaches the maximum temperature obtained
when a step-less channel (i.e., S = 0, Fig. 1A) has its plate-to-plate
spacing optimized, which, according to the present numerical cal-
culations is, 0.0948.
In Fig. 3, we present the results for a channel with a single shav-
ing level (i.e., S = 1) for three

L
0
-constant curves. For a given iso-

L
0
curve, there is thus only one design variable left, which was chosen
as

t
1
. The results, which relate the variation of

T
max
with respect to

t
1
, show that there is an optimal value of

t
1
that minimizes

T
max
for
each

L
0
-constant curve. One should note, however, that for the
three

L
0
-constant curves shown,

T
max;min
represents a local mini-
mal temperature. Additionally, the shape of the curves presented
in Fig. 2 shows that

T
max
is strongly affected by

t
1
when this vari-
able is far from its ideal value. Keeping in mind that

t
0

V
0
=2

t
1
=

L
0
, and that the inequality

t
0
> 0 must always be
respected, it can be realized that for small values of

t
0
(i.e.,

t
0
?0), minimal shaving is imposed on the boundary layer forcing
the channel to resemble a step-less structure (i.e., S = 0 as shown in
Table 2
Maximum differences between predicted and calculated values for forced convection.
C
Dopt
Max. difference
(%)
Avg. difference
(%)
Max. difference
(%)
Avg. difference
(%)
4.45 3.09 6.48 5.75
A.K. da Silva, L. Gosselin/ International Journal of Heat and Mass Transfer 67 (2013) 272279 275
Fig. 1A), which increases

T
max
. Differently, in the limit were

t
0
)

t
1
; the downstream channel is over constrained vertically,
which also promotes a surge of

T
max
. Now focusing on the effect
of

L
0
, it is clear that by determining the local

T
max;min
for a series
of

L
0
curves, one can double minimize the maximum temperature
(i.e., minimize with respect to

L
0
and

t
1
simultaneously) and nd
the global minimum for the given value of Be, i.e., 10
5
in the case
of Fig. 3 the same process must be repeated for Be = 10
6
and
10
7
. For Be = 10
5
,

L
0;opt
is 0.71, which is also graphically shown in
Fig. 3.
Fig. 4 summarizes the optimal geometry for the main dimen-
sions of a channel with a single shaving level (S = 1), which are

L
0
,

t
0
and

t
1
note that

L
1
1

L
0
and because of that it is not
shown in this gure. The results plotted in Fig. 4 show that

L
0;opt
is nearly constant between 10
5
< Be < 10
7
, where it varies from
0.71 to 0.68. As for

t
0;opt
and

t
1;opt
, the results show that these
are nearly identical. Combining the results for

L
0;opt
,

t
0;opt
and

t
1;opt
it can be realized that, geometrically, a channel with one step
level has its plate-to-plate spacing cut in half at the transition,
while having, roughly, a 70/30 length ratio between the upstream
and the downstream sections for values of Be simulated. Further-
more, these numerical results are supported by the scaling deriva-
tion shown in Section 3. Note that, according to Eq. (15),

L
0;opt
is
independent of Be and scales as 2/3 this results is shown by
the constant dashed curve in Fig. 4. Also, as indicated by Eq. (17),

t
0;opt
and

t
1;opt
scale as 0.3

V
0
; which, given the volume data listed
in Table 1, returns optimal

t values of 0.0507, 0.0288 and 0.0162,
for Be equal to 10
5
, 10
6
and 10
7
, respectively. The agreement be-
tween the numerical and scaling results is such that the scaling
curve, which is also plotted in Fig. 4 as a dashed line, can barely
be seen.
In Fig. 5, we show the optimal length scales of the two-step
channel conguration (i.e., S = 2). One important detail of the opti-
mization process developed here is the fact that the ratio between
the channel thicknesses at each level was assumed to be equal (i.e.,

t
0

t
1

t
2

t). This assumption, which was based on the results


obtained for the S = 1 conguration (Fig. 4 more specically), facil-
itated the optimization process, which now has only two degrees
of freedom (the lengths of two of the sections), rather than the
original four. The data reported on Fig. 5 shows a modest effect
of Be on

L
0;opt
and

L
1;opt
(the same is valid for

L
2;opt
, since

L
2;opt
1

L
0;opt

L
1;opt
). As for

t
opt
, a fairly noticeable reduction
can be seen as Be increases recall that, as shown in Table 1,

V
0
opt
also decreases with Be. Moreover, differently from the
Fig. 2. Effect of the axial location of the step transition (i.e.,

L0) on the maximal
temperature.
Fig. 3. Effect of

t1 on the maximum temperature for three

L0-constant curves.
Fig. 4. Summary of the optimal geometric dimension of a S = 1 structures (Fig. 1B).
Fig. 5. Optimized dimensions for a step channel with two shaving sections (i.e.,
S = 2).
276 A.K. da Silva, L. Gosselin/ International Journal of Heat and Mass Transfer 67 (2013) 272279
summary presented in Fig. 4 for the S = 1 conguration, where all
variables were considered during the optimization characterizing
a global minimal, in Fig. 5 a simplication was imposed on

t
0
,

t
1
, and

t
2
(i.e., these are all equal to

t), which characterizes the

T
max
as a local minimal. Finally, it is important to mention that
the curves shown in Figs. 4 and 5 were plotted for three values of
the Be number these are given by the open symbols shown in
these gures. While the number of x-axis points can be considered
small, it is important to notice that all curves show a nearly con-
stant slope throughout the range of Be considered. Therefore, it is
safe to assume that the behaviors of the variables reported are ex-
pected to follow the same trend in between the calculated values.
Fig. 6 quanties the benets of the shaving technique by di-
rectly comparing the reduction of

T
max
due to the creation of
step-like channels once again, it is fundamental to recall that

V
0
opt
f Be was held constant, limiting

T
max
as the only gure of
merit of interest in this simulation process. The results of Fig. 6
show that, on average, the minimized hot spot temperature drops
roughly by 20% and 8.8% when a standard channel (S = 0) is con-
verted into a one step-conguration (S = 1), and when an S = 1
design becomes an S = 2, respectively, for all Be values simulated.
Not only is the increase in thermal performance substantial, but
also, the diminishing return in terms of the temperature reduction
as S increases (S can be seen as a measure of the complexity of the
geometry) has been indentied in other studies that deal with the
geometric optimization of thermal systems e.g., [6]. Also interest-
ing is to note that the

T
max
values shown in Fig. 6 were multiplied
by Be
1/4
, which is the scale obtained in Eq. (18) for the shaving
method considering the S = 1 geometry. As expected, assuming
that the relation between

T
max
and Be as shown in Eq. (18) is accu-
rate, all numerical data for S = 1 should collapse. Therefore, the fact
that all curves are basically superimposed in Fig. 6 conrms the
correctness of Eq. (18).
Fig. 7 offers the numerical temperature distribution interpreta-
tion of the effect caused by creating a step transition in half of the
channel with forced convection. With the channels simulated
being symmetric, only the upper half of the domain is shown in
Fig. 7. Additionally, note that the maps shown in Fig. 7 respect
the geometric scales determined previously (i.e., these are opti-
mized congurations), and are plotted while using a single temper-
ature color range. In other words, the color temperature scale
shown for all three frames of Fig. 7 are that of an S = 0 congura-
tion, where

T
low
0 (dark blue) and

T
high


T
max
0:0948 (red).
Therefore, given the results shown in Fig. 6, which indicate that
the maximal temperature can be reduced as S increases, it is ex-
pected that the S = 1 and 2 frames of Fig. 7 do not show intense
red regions, which are present in the S = 0 frame. Indeed, the
appearance of two hot spots located near the exit of the channel
for the S = 0 conguration represents the dimensionless tempera-
ture value of 0.0948. As the channel complexity S increases, the
hot spots vanish indicating an increase in the efciency of the
system.
6. Scaling formulation for turbulent forced convection
A analysis similar to the one shown Section 3 can be performed,
this time assuming a turbulent regime. Although no turbulent ow
simulations were performed for the present paper, it is instructive
to present a brief scale analysis to explain how turbulence might
affect the results. Assuming a turbulent ow over a at plate in-
stead of a laminar one [1], Eq. (3) becomes:
d
0
t
0
$ 1
0:382

L
4=5
0

t
0

U
1=5
0
Pr
Be
_ _
1=5
22
Similarly, d
1
/t
1
assumes the same form as Eq. (22), and is also of the
order of 1. Applying d
0
/t
0
$ d
1
/t
1
one nds:

L
4=5
0

t
0

U
1=5
0
$

L
4=5
1

t
1

U
1=5
1
23
The hot spot temperature occurring at the end of the rst plate can
be estimated by:

T
max
x

L
0
$

L
4=5
0
Pr

U
0
Be
_ _
1=5
24
A similar expression would be found for the hot spot at the end of
the second plate. As for the laminar regime, one can assume that
both hot spot temperatures have the same order of magnitude,
i.e.,

T
max
x

L
0
$

T
max
x

L
1
. Combining previous results, it is
found that the optimal conguration in turbulent regime is given
by:

L
4
1

U
1
$

L
4
0

U
0
25

t
0
$

t
1
26
It is interesting to nd that the design rule of Eq. (26) is exactly
the same as that found previously under laminar regime. In order
to nd the optimal lengths, it is necessary to apply the same ideas
as in the laminar regime. The relations

D
0
$ 4

t
1
and

D
1
$ 2

t
1
still
apply since they are based on geometrical considerations and since
Eq. (26) is valid in the turbulent regime. Expressing the total
pressure drop through the channels with the help of previously
developed relations, one nds an expression for

U
0
that can be used
in the expression of the hot spot temperature. Then, this hot spot
temperature is minimized with respect to

L
0
. The optimal
dimensions that are obtained through this procedure are:
Fig. 6. Effect of the number of shaving sections on the maximal temperature of a
channel in forced convection.
Fig. 7. Temperature map of the three types of channel: (A) regular channel, and (B)
and (C) channels with shaved boundary layers for Be = 10
5
.
A.K. da Silva, L. Gosselin/ International Journal of Heat and Mass Transfer 67 (2013) 272279 277

L
0;opt

1
1
2
_ _
1=4
1
% 0:5432

L
1;opt
% 0:4568 27
Finally, the optimal spacing is obtained from the volume constraint,
and is:

t
0;opt
$

t
1;opt
$ 0:3240

V
0
28
Compared to the optimal design in the laminar regime, it is found
that the optimal spacing is basically the same. As for the optimal
lengths of the plates,

L
0
and

L
1
tend to be almost equal for turbulent
ow, although they were more dissimilar when the ow was lami-
nar. Finally, the minimized hot spot temperature is proportional to

T
max;min
/ Be
1=5
.
The last question to answer is when the turbulent and laminar
regimes respectively apply. The scale of the Reynolds number
based on D
0
in the optimal laminar design is:
U
0
D
0
m
$ 0:8680

V
05=3
Be
2=3
Pr
2=3
29
In the numerical simulations, it is found that this number is always
smaller than the laminarturbulent transition at $2000.
7. Results for laminar natural convection
The design optimization procedure outlined above in forced
convection was also applied to the channel with self-driven ows.
For the natural convection simulations, the same set of differential
equations as shown in Section 4 was used, including the constant
property approximation. In order to allow for the development of
natural convection, the well-known Boussinesq approximation
was used [1]. Under this approach the buoyancy term is added to
the right-hand side of the momentum equation representing the
x direction see Fig. 1A. Also, given this frame of reference, the
gravitational force, which is not shown in Fig. 1, must also be
aligned with the x-direction note the differently from the forced
convection case where we had a specied pressure boundary con-
dition at the inlet of the channel, for the natural convection case
the inlet pressure was zero. The non-dimensional set of equations
employed can be seen in Ref. [15].
In principle, the same concept discussed in Section 2 is also
applicable to buoyancy driven systems, assuming that gravity is
aligned with the ow direction. Therefore, it is worth to develop
a natural convection version of Table 1, which would use the Ray-
leigh number, Ra = (gbq
00
L
4
k
1
a
1
m
1
), as the primary dimension-
less variable, see Ref. [1]. The numerically determined V
0
for the
natural convection is given by Table 3.
The results shown in Table 3 were also used to validate the
numerical implementation against the theoretical results reported
in Ref. [8], which suggests that

D
opt
2:12Ra
1=5
and
C 0:3434Ra
2=5
. The comparison between the correlation and
our results, which is shown in Table 4, shows a very good agree-
ment between both results.
While results for forced convection suggest that the boundary
layer shaving is a promising method, the results for natural con-
vection indicate little to no positive effect for a step-like channel
structure. In fact, most of the congurations tested exhibited a
performance decrease. This behavior can be visualized in Fig. 8
for a vertical channel with a single shaving level (i.e., S = 1).
According to the upper frame of this gure, the performance,
which is indicated by

T
max
since the coolant volume per unit length
remained unchanged and equal to 0.216 for Ra = 10
5
, shows that
the maximum temperature only approaches the traditional chan-
nel (S = 0) maximum temperature as

t
1
!

V
0
=2. Furthermore, ana-
lyzing these results from a

V
0
constraint perspective, we can
observe that large

t
1
forces

t
0
to be small, ultimately representing
a step-less straight channel.
Results were also obtained for stronger buoyancy effects with
Rayleigh numbers up to 10
8
, and for three

L
0
curves. Generally
speaking, the overall trends are similar to the ones shown in
Fig. 8; however, the results indicate slightly lower values for

T
max
(around 3%) as

t
1
!

V
0
=2. Even though there are indications that

T
max
can be reduced, gains are not close to the substantial temper-
ature drops seen in the forced convection case.
Also interesting is to further discuss the distinct behavior pre-
sented for the natural and forced convection cases. While the con-
cept, at rst, appears to be independent of the ow driving force,
results suggest that this is not the case. One possible reason for
the behavior presented for natural convection is that, while the
shaving method aims to eliminate the overheated boundary layer,
it also discards, in the natural convection case, the layer of the ow
with the highest density difference (i.e., buoyancy). For instance, as
the volume of uid is shaved from one step to another, the central
uid volume (away from the walls) upstream of the shaving loca-
tion, which is cold compared to the uid nearby the vertical
walls of the channel, needs to be pulled up by the channel down-
stream where shaving occurs.
Table 3
Maximum coolant volume per unit of length as
a function of Ra.
Ra
V
0
=L
2
=

Dopt
10
5
0.216
10
6
0.135
10
7
0.085
10
8
0.054
Table 4
Maximum differences between predicted and calculated values for natural
convection.
C
Dopt
Max. difference
(%)
Avg. difference
(%)
Max. difference
(%)
Avg. difference
(%)
3.35 1.43 1.85 1.21
Fig. 8. Effect of the boundary layer shaving on self-driven ows.
278 A.K. da Silva, L. Gosselin/ International Journal of Heat and Mass Transfer 67 (2013) 272279
8. Conclusions
In this study we simulated two-dimensional channels in natural
and forced convection created by the combination of two parallel
plates where, at a given location along the ow direction, the chan-
nel goes through an abrupt change in plate-to-plate spacing, allow-
ing for the overheated uid, near the heat dissipating walls, to be
eliminated. To analyze the benet or drawbacks of this methodol-
ogy, in all numerical calculations, the volume of coolant available
was held xed and equal to the volume of coolant needed such that
traditional channel would have the highest global conductance, as
reported previously [1,8]. This simplication allowed us to simply
consider the maximum temperature of the numerical domain as
our gure of merit.
The numerical results for forced convection showed a signi-
cant reduction of the maximum temperature with the inclusion
of shaving steps. Also, the more shaving steps were added, the
higher the maximum temperature drop, which agrees, from an
increasing complexity perspective, with the existing literature
e.g., [6]. The results also show that for a channel with a single shav-
ing level, the geometric variables (i.e., L
i
and t
i
) strongly affect the
gure of merit. Furthermore, the numerically optimized geometric
variables matched fairly well the results derived via scale analysis.
As for the natural convection case, results indicated that the shav-
ing methodology would not be as helpful as the channels perfor-
mance is barely increased by the inclusion of a shaving level. In
fact, for most cases, the shaving step increases the maximum
temperature.
While the present numerical study shows promising results for
forced convection channels these are still to be validated experi-
mentally. Also, the numerical domain and boundary conditions
can be further modied so as to mimic a permeable wall (see
Fig. 1C) where the boundary layer is being continuously drained
and not allowed to overheat.
Acknowledgements
Gosselins work was supported by the Natural Sciences and
Engineering Research Council of Canada (NSERC).
References
[1] A. Bejan, Convection Heat Transfer, third ed., John Wiley & Sons, Inc., NewYork,
2004.
[2] A. Faghri, Y. Zhang, J.R. Howell, Advanced heat and mass transfer: Global
Digital Press, 2010.
[3] F.P. Incropera, Convection heat-transfer in electronic equipment cooling, J.
Heat Transfer-Trans. ASME 110 (4B) (1988) 10971111.
[4] G. De Vahl Davis, Natural convection of air in a square cavity: a bench mark
numerical solution, Int. J. Numer. Methods Fluids 3 (1983) 249264.
[5] W.H. Leong, K.G.T. Hollands, A.P. Brunger, Experimental Nusselt numbers for a
cubical-cavity benchmark problem in natural convection, Int. J. Heat Mass
Transfer 42 (11) (1999) 19791989.
[6] A. Bejan, Shape and Structure: From Engineering to Nature, Cambridge
University Press, Cambridge, 2000.
[7] A.K. da Silva, L. Gosselin, Evolutionary placement of discrete heaters in forced
convection, Numer. Heat Transfer Part A-Appl. 54 (1) (2008) 2033.
[8] A. Bar-Cohen, W.M. Rohsenow, Thermally optimum spacing of vertical,
natural-convection cooled, parallel plates, J. Heat Transfer-Trans. ASME 106
(1) (1984) 116123.
[9] A. Bejan, E. Sciubba, The optimal spacing of parallel plates cooled by forced-
convection, Int. J. Heat Mass Transfer 35 (12) (1992) 32593264.
[10] A.K. da Silva, L. Gosselin, Volumetric maximization of coolant usage in closed
self-driven circuits, Int. J. Heat Mass Transfer 53 (1920) (2010) 39693976.
[11] C.Y. Wang, Optimum placement of heat-sources in forced-convection, J. Heat
Transfer-Trans. ASME 114 (2) (1992) 508510.
[12] F.P. Incropera, J.S. Kerby, D.F. Moffatt, S. Ramadhyani, Convection heat-transfer
from discrete heat-sources in a rectangular channel, Int. J. Heat Mass Transfer
29 (7) (1986) 10511058.
[13] M. Keyhani, V. Prasad, R. Cox, An experimental-study of natural-convection in
a vertical cavity with discrete heat-sources, J. Heat Transfer-Trans. ASME 110
(3) (1988) 616624.
[14] H.J. Shaw, W.L. Chen, Laminar forced convection in a channel with arrays of
thermal sources, Warme- Und Stoffubertragung 26 (4) (1991) 195201.
[15] A.K. da Silva, S. Lorente, A. Bejan, Optimal distribution of discrete heat sources
on a wall with natural convection, Int. J. Heat Mass Transfer 47 (2) (2004) 203
214.
[16] A.K. da Silva, L. Gosselin, The numerical optimization of a Neumann-type
boundary for the Graetz problem, Numer. Heat Transfer Part A-Appl. 59 (8)
(2011) 577593.
[17] T. Bello-Ochende, A. Bejan, Maximal heat transfer density: plates with
multiple lengths in forced convection, Int. J. Therm. Sci. 43 (12) (2004)
11811186.
[18] A. Bejan, Y. Fautrelle, Constructal multi-scale structure for maximal heat
transfer density, Acta Mech. 163 (12) (2003) 3949.
[19] W. Breugem, B.J. Boersma, R.E. Uittenbogaard, The laminar boundary layer
over a permeable wall, Transp. Porous Media 59 (3) (2005) 267300.
[20] A. Ishak, J.H. Merkin, R. Nazar, I. Pop, Mixed convection boundary layer ow
over a permeable vertical surface with prescribed wall heat ux, Zeitschrift Fur
Angewandte Mathematik Und Physik 59 (1) (2008) 100123.
[21] Comsol multiphysics

v. 3.4, Burlington, MA, 2007.


A.K. da Silva, L. Gosselin/ International Journal of Heat and Mass Transfer 67 (2013) 272279 279

Вам также может понравиться