Sie sind auf Seite 1von 6

8th THERMINIC Workshop

1-4 October 2002, Madrid

A Compact Model for Heat Sinks with Experimental Verification

Cristina Autuori Tomazeti and Carlos A. C. Altemani


Department of Energy- Faculty of Mechanical Engineering – State University of Campinas
Campinas – SP - Brazil
Altemani@fem.unicamp.br

rates always decreased with the interfin spacing. The


Abstract thermal resistance however attained a minimum and then
increased again, thus indicating an optimum spacing.
A compact model was developed for the thermal design This behavior was predicted by their analysis, but their
of a parallel plates heat sink mounted in a duct with model presented a large overprediction of the flow rate
bypass regions. It was based on an electrical circuit and a large underprediction of the thermal resistance.
analogy and a selection of correlations for pressure losses In the presence of bypass regions around the heat sink,
and convective heat transfer. The results were obtained as indicated in Figure 1, just a fraction of the duct flow
by an iterative solution of a set of nonlinear algebraic rate will pass through the interfin channels. This
equations. An experimental investigation was performed fraction must be determined to evaluate the heat sink
to compare results and to improve the model predictions. thermal performance. The experimental results of Wirtz
Comparisons were also made with data from the et al. [4] showed that heat sink overall heat transfer
literature. decreased with the amount of flow bypass. They
suggested a thermal design procedure based on heat
1. Introduction transfer correlations for developing flow between parallel
Air is the most convenient fluid for electronics cooling, plates. Butterbaugh and Kang [5] showed that the
but its thermal conductivity and specific heat are thermal resistance of a heat sink with distinct bypass
relatively small. To overcome this difficulty, many times regions correlates directly with the pressure drop across
the air cooling of electronic components and systems is the heat sink, for any amount of bypass. They presented
associated to extended surface devices [1]. Heat sinks a network flow model using correlations for the laminar
probably constitute the most commonly used, in many flow of parallel plates channels. Their predictions were
sizes and shapes, made of distinct materials, and within 10 per cent of most of their experimental results.
operating under either forced or natural airflow This finding was very important from the point of view
conditions. Over the years, the trend of a large increase of heat sink design. It indicated the need to complement
in the associated heat fluxes has demanded a proper and the thermal resistance data with information concerning
efficient air cooling thermal design. Thus, the thermal the pressure drop across the heat sink. This procedure
behavior of heat sinks must predicted with great care, in could help forced convection heat sink manufacturers to
order to achieve the limits of air-cooling. set a standard for heat sinks specification. Biber and
Compact models constitute a very convenient tool to Belady [6] presented pressure drop data for five distinct
select a heat sink, mainly due to their ability to give very
fast answers. Several heat sink configurations can be
tested in a short time and the best results can be selected tf s
to be further investigated by other methods, such as CFD
and experiments. The development of reliable models h
has been the subject of many investigations. For no
clearance around the heat sink, the interfin velocities hf
may be readily obtained from the duct flow. Goldberg [2]
tested three copper plates heat sinks limited to laminar
airflow. According to his analysis, the design with the
smallest interfin width presented the largest pressure wm
drop and the smallest thermal resistance. Knight et al.
[3] tested three aluminum fin heat sinks under the same w
pressure drop. They showed that the volumetric flow Figure 1 - Duct Cross Section with Heat Sink
Tomazeti, C.A. and Altemani, C.A.C.
A Compact Model for Heat Sinks with Experimental Verification

closely ducted heat sinks. The results were obtained from


experiments, CFD modeling and a method based on
correlations. The method based on correlations was
considered an excellent starting point for heat sink
design because it could be very fast to analyse heat sinks
by a parametric procedure. They considered, however,
that there was a need to assess the validity of the flow
correlations used for the heat sinks. Reis and Altemani
[7] included turbulent flow and heat transfer correlations
in a network flow model and compared their prediction
with experimental data from the literature. The
comparisons with the experimental data of [5] showed
larger deviations from the data for larger bypass regions
and smaller approach air speeds. Figure 2 – Flow network model.
In order to improve the heat sink pressure drop and
thermal resistance predictions, the compact model of [7] increases were associated to the form drag of the flow
was reviewed with the support of experimental data. around the heat sink, considered like a permeable bluff
Since there is no standard procedure to test a heat sink, it body. Ahead of the sink, there is a pressure increase
was necessary to select one from the literature. The (∆Phs) due to stagnation of the approach flow and behind
pressure drop was measured in [6] employing the same the heat sink there is a flow separation and a subsequent
duct with and without the heat sink and subtracting these pressure increase (∆Phb) to the downstream airflow.
two values. The present experiments were performed Instead of treating these terms as pressure increases
using the same procedure. In addition to comparisons along the interfin flow path, they were subtracted from
with the present experimental data, the modified model the entire flow network and were treated as pressure
was also compared with previous experimental data and drops in both the tip and lateral flow channels. The
with correlations from the literature. advantage is that the calculated pressure drop is directly
In the presence of bypass regions, the compact model that across the heat sink. They were expressed by
generates a system of non-linear algebraic equations.
∆Phs = 0.8 ρ (U 2 − U hs
2
)/2 (1)
They state that the pressure drop across the heat sink is
the same for any parallel flow branch. These equations
were assembled for a known total airflow rate in the duct ∆Phb = Cρ (U 2 − U hb
2
)/2 (2)
housing the heat sink. The solution was the pressure The constant C in equation (2) was 0.1 for either tip or
drop across the heat sink and the flow rates in the lateral bypass only and 0.2 for combined bypass [5]. The
interfin channels and the bypass regions. Due to the mean airflow velocity Uhs in the frontal area just ahead of
non-linearity of the algebraic equations, the new model the heat sink differed from Uhb just behind the heat sink
made use of an iterative solution procedure based on the due to flow leakage to the tip clearance. These velocities
SIMPLE algorithm used for CFD [8]. were obtained by continuity at the upstream and
2. Modeling and Experiments downstream ends of the heat sink, expressed by
Qhs = Qhi and Qhb = Qhe (3)
Initially the flow network model used to obtain the heat
sink pressure drop and the airflow rate in the interfin Two conditions were imposed on the flow network to
channels will be described. This result was used to generate the desired solution. First, the mass
obtain the heat sink thermal resistance, as described next. conservation of nodes 1 to 3:
Finally, the experimental apparatus and the test Qn,in − Qn, out = 0 (4)
procedure will be presented.
Second, the same pressure drop across the heat sink by
2.1. Network flow model any parallel branch:
The flow around a heat sink mounted in a duct with tip ∆Ph = ∆Pt = ∆Pl = P14 (5)
(t) and lateral (l) bypass regions was modeled according
The head losses in both the interfin and the bypass
to the flow network indicated in Figure 2. The duct total
channels were associated to those in rectangular ducts
volumetric airflow rate (Q) is subdivided into three
with aspect ratio α ≤ 1, through the hydraulic diameter
parallel branches across the heat sink. The model adopts
analogy. The friction pressure losses were evaluated with
discrete head losses in each branch, due to contraction,
the apparent Fanning friction factor fa, so that the
friction and expansion. Similarly to [5], pressure
Tomazeti, C.A. and Altemani, C.A.C.
A Compact Model for Heat Sinks with Experimental Verification

contraction losses were already included [9]. Thus, the TABLE 1 – Coefficients for equation (10)
head losses were expressed by Expansion Ke
Laminar Turbulent
L V2 V2
∆Pc + ∆P f + ∆Pe = 4 f a ρ + Keρ (6) A0 1.000 1.000
Dh 2 2 A1 -2.400 -2.083
Due to airflow leakage into the tip bypass region, the A2 1.000 1.005
friction pressure losses along the tip clearance and the B0 1.000 1.000
interfin channels were evaluated using the average B1 -2.800 -2.125
velocities at the entrance and exits. The leakage pressure B2 1.000 0.976
loss between nodes 2 and 3 in Figure 2 were evaluated as
friction losses for a length equal to half the fin height. section, as indicated in Figure 1. Since the heat sink was
The critical Reynolds number in the interfin and in the centered in the duct, the lateral channels were treated as
bypass channels was evaluated by a correlation derived a single channel. Both bypass channels were associated
[7] from data presented in [10] for flow in rectangular to rectangular ducts with the same cross section and with
ducts with abrupt entrance: aspect ratio (α ≤ 1) obtained as follows. For the lateral
Re c = 3035 − 4498α + 10719α 2 − 11285α 3 + 4233α 4 (7) bypass channel the aspect ratio α was either equal to the
ratio [(w - wm)/(2hf)] or its reciprocal and for the tip
For laminar flow the apparent friction factor was
bypass channel, it was either [(h - hf)/wm] or its
approximated [7] by
reciprocal.
3.435 16φ −1 + 1.25(4 X ) −1 − 3.435 X −0.5 Equation (4) applied to nodes 1-3 together with eq. (5)
f a . Re = + (8)
X 0.5 1 + 0.00021X − 2 to the 3 parallel branches and equation (5) comprised a
system of 8 equations associated to the flow network
2 11 L/ D
where φ = + α (2 − α ) and X = indicated in Figure 2. In the presence of bypass regions,
3 24 Re they were solved by an iterative procedure based on the
For turbulent flows, the apparent friction factor was
SIMPLE algorithm from the knowledge of the total flow
obtained from a correlation presented in [9]: rate Q in the duct. Basically, a pressure drop (∆P14)
across the heat sink was guessed and the corresponding
f a = A(φ . Re) B (9) flow rates Qi in each parallel flow branch indicated in
Figure 2 were obtained. If the sum of the parallel Qi was
1.01612
where A = 0.09290 + not equal to the known Q, a correction was made in the
( L / D) previous pressure drop and the process was repeated until
0.31930 convergence was attained.
and B = −0.26800 −
( L / D)
2.2. Convective heat transfer
The expansion coefficient Ke was obtained [7] from the
values presented in Kays (1950) for parallel plates (with The convective heat transfer rate from the heat sink
aspect ratio α = 0) and for ducts of square (α = 1) cross was evaluated with the previously obtained flow
section. It was expressed by distribution in the presence of bypass regions. The
incoming airflow (Ti) and the heat sink base (Tw)
K e = (1 − α ).( A0 + A1σ + A2σ 2 ) + α .( B 0 + B1σ + B 2σ 2 ) (10) temperatures were assumed uniform. The convective
The coefficients Ai and Bi used in this equation are heat transfer rate was expressed by
presented in Table 1. The contraction area ratio σ was
q s = h * Atη0 (Tb − Ti ) (12)
obtained either from the geometry or from mass
conservation: The convective heat transfer coefficient h* was expressed
s U U in terms of the incoming airflow temperature, and the
σf = , σl = , σt = (11) heat sink overall thermal efficiency η0 was
s+t U l U t
Af
respectively for the interfin (f), the lateral (l) and the tip ηo = 1 − (1 − η f ) (13)
(t) airflow paths. At
The hydraulic diameter concept was employed for all The finned (Af) and total (At) areas were obtained from
the flow channels across the heat sink. The interfin the heat sink geometry, while the fin efficiency (ηf) was
channels were those of rectangular cross section with the the standard relation for an adiabatic fin tip with a
fin height (hf) and spacing (s). The lateral and the tip correction for the fin length.
bypass channels were assumed those of trapezoidal cross For laminar flows in the interfin channels, the average
Tomazeti, C.A. and Altemani, C.A.C.
A Compact Model for Heat Sinks with Experimental Verification

Nusselt number was obtained from a correlation upstream wall. The heat sink pressure drop was obtained
developed by Sparrow [11] for the entrance region of a from the difference of the gage pressures in the chamber
flat plate rectangular duct. It considered a duct with downstream the rectangular duct with and without the
uniform temperature and velocity profiles at the entrance heat sink. The airflow rate was obtained from the
and isothermal walls: pressure difference between the two chambers in
1/ 2 conjunction with a calibrated metering nozzle.
* h * Dh 0,664 Gz  Pr 
Nu = = 1 + 7,3( )1 / 2  (14) The tests to determine the thermal resistance were
k 1 / 6  Gz  made with an aluminum heat sink with an electrical
Pr
Re Pr heater attached below the heat sink base. A guard heater
where Gz = and Pr ≈ 1 was employed below the main heater to minimize the
L / Dh
thermal losses.
For turbulent flow, the adopted correlation for the Three heat sinks were built for the experimental tests -
average Nu was that presented by Bhatti and Shah [12] two made of plexiglass and one of Aluminum. The three
for the abrupt entrance of circular ducts and Pr = 0.7: heat sinks were used for the pressure drop tests and that
 made of aluminum was employed also for the thermal
hDh 2.4254 
Nu = = Nu∞ 1 +  (15) resistance tests. All heat sinks had a base width wm = 53
k  ( L / D ) 0.676 
 h  mm and a length L = 50 mm. The other dimensions
The value of Nu∞ was obtained from Gnielinski equation: (mm) of the short plexiglass (SP), the tall plexiglass
(TP), and the Aluminum (A) heat sinks are shown in
( f a / 2)(φ Re − 1000) Pr Table 2.
Nu∞ = (16)
1 + 12.7( f a / 2) 0.5 (Pr 0.67 − 1) The rectangular duct housing the heat sinks during the
tests had a variable cross section. The duct width W was
where the fully developed friction factor was obtained adjusted to either 109 mm or 53 mm and the height h
from eq. (9) assuming an infinite (L/D). was either 25 mm or 40 mm. Thus, four cross sections
Since h in eq. (15) was based on the fluid mean bulk were obtained to test the three heat sinks as follows. The
temperature, the value of h* was then obtained from short plexiglass (SP) and the Aluminum (A) heat sinks
m& c p h At were tested with no bypass, with top and lateral bypass
h* = [1 − exp( − )] (17) only and with combined bypass. The tall plexiglass (TP)
At m& c p
heat sink was tested for configurations with no bypass
This is a standard conversion to the heat transfer and with the lateral bypass only.
coefficient using the fluid entrance temperature as The experimental tests were performed under steady
reference. state conditions. A differential pressure transducer
measured the heat sink pressure drop with a variable
2.3. Experimental Tests capacitance-sensing element, calibrated for a span of 0 -
The experimental apparatus used for the experimental 5 cm H2O. The temperatures of the heat sink base and
tests was built as indicated in Figure 3. It consisted of a heaters and of the incoming air were obtained by
wood chamber with a (730x610) mm cross section and a calibrated type T thermocouples. The heaters (main and
length of 1100 mm. An internal wall housed a metering guard) under the heat sink base dissipated electric power
nozzle for the airflow forced in suction mode by a fan from two independent dc power supplies. A calibrated
located downstream. The heat sink was located in a digital multimeter measured all the thermocouples and
rectangular duct with variable cross section in the power supply outputs.
The uncertainties of the experimental results were
obtained by the method of sequential perturbations
described by Moffat [13]. The estimated uncertainties for
the results were (+/-) 4 % for the approach air velocity to
the heat sink, 5 % for the convective thermal resistance
inflow
and 3 Pa for the heat sink pressure drop.

TABLE 2- Heat sinks dimensions (Figure 1)


PS PH A
outflow S 1.9 2.4 2.9
Rectangular duct nf 13 11 13
with heat sink nozzle tf 2.0 2.2 1.0
Figure 3 – Experimental apparatus hf 25.0 40.0 25.0
Tomazeti, C.A. and Altemani, C.A.C.
A Compact Model for Heat Sinks with Experimental Verification

The convection from the heat sink the airflow in each was a large decrease in the pressure drops and the airflow
thermal test was evaluated subtracting other losses from in the interfin channels was laminar in all these tests.
the electric power dissipation in the main heater under The largest deviations of the model predictions were
the heat sink base. The other losses included conduction those for heat sink A, with a maximum of 0.6 Pa.
due to contact of the base with the duct wall and the The results of the tests with tip bypass only, made with
thermal insulation under the heat sink, radiation of the the heat sinks A and SP, are shown in Figure 6. The
heat sink surfaces and conduction through the connecting deviations from the model predictions were larger for
and thermocouple wires. heat sink A, but they were (2.3 Pa maximum) within the
experimental uncertainty (3 Pa).
3. Results The combined bypass configuration was also tested
The measured values of the heat sink pressure drop and with the heat sinks A and SP. The results shown in
thermal resistance will be compared with the results Figure 7 indicate the very small pressure drops due to the
predicted by the described compact model. Additional large bypass. The experimental data agreed quite well
comparisons of the model predictions will also be made with the model predictions – the largest deviation was
with data obtained from the literature. 0.7 Pa for heat sink A. The empirical correlation of
In the absence of bypass, the pressure drop increase Jonsson and Moshfegh [14] was also included in Figure 7
with the approach air velocity to the three heat sinks is for heat sink SP, indicating an overprediction of the
presented in Figure 4. The change of slope observed in measured data.
the curves of heat sinks A and SP indicate a flow The thermal resistance was defined as
transition in the interfin channels. The largest deviation T − Ti 1
(5 Pa) from the model predictions occurred for heat sink R= w = (18)
*
qconv h Atη 0
SP at the transition. Note that in this configuration the
heat sinks have an interfin space with the duct walls at The measured values and the model predictions for heat
both sides. sink A are shown in Figure 8. The configuration with tip
The results for the configuration with lateral bypass bypass only presented deviations bigger than the others,
only are presented in Figure 5. Due to flow bypass there but within the (5%) experimental uncertainty, except the
0 1 2 3 4 5

140 14 20 SP 20
Short plexiglass HS (SP)
120 Tall plexiglass HS (TP) 12 A
Pressure Drop [Pa]

15 Model 15
Pressure Drop [Pa]

100 Aluminum HS (A) 10

80 Model Prediction 80
10 10

60 60

40 40 5 5

20 20
0 0

0
0 1 2 3 4 5
0 1 2 3 4 5 6 7 8
U [m/s]
U [m/s]
Figure 4 – Pressure drop: confined configuration Figure 6 – Pressure drop: tip bypass only
0,0 0,5 1,0 1,5 2,0 2,5
0,0 0,5 1,0 1,5 2,0 2,5 3,0 3,5 4,0 6 6
10 10

SP SP
5 5
8 TP
8 A
Pressure Drop [Pa]

4 [14]-SP 4
Pressure Drop [Pa]

A
6 6
Model
Model 3 3

4 4
2 2

2 2
1 1

0 0
0 0

0,0 0,5 1,0 1,5 2,0 2,5 3,0 3,5 4,0 0,0 0,5 1,0 1,5 2,0 2,5
U [m/s] U [m/s]

Figure 5 – Pressure drop: lateral bypass only Figure 7 – Pressure drop: combined bypass
Tomazeti, C.A. and Altemani, C.A.C.
A Compact Model for Heat Sinks with Experimental Verification

1 2 3 4 5 6 7 8
2,5 2,
4. Conclusions
No bypass
2,0 Lateral Only 2,
The compact model used in this work was described
Tip only with all the correlations and the method of solution. The
model predictions for the heat sink pressure drop and
R [K/W]

1,5 Combined 1,
Model thermal resistance compared most favorably with data
obtained from experiments. Comparisons were also
1,0 1,
made with data and correlations from the literature.

0,5 0,
References
1 2 3 4 5 6 7 8
U [m/s]
[1] Kraus, A.D. and Bar-Cohen, A.: “Thermal Analysis and
Figure 8 – Thermal resistance: HS A Control of Electronic Equipment”, Hemisphere, New
-5 0 5 10 15 20 25 30 35 40 45 50
2,4 2, York (1985).
[2] Goldberg, N.: “Narrow Channel Forced Heat Sink”, IEEE
2,1 Combined 2,
Trans. CHMT, vol. 7, n.1, pp.154-159 (1984).
Lateral only [3] Knight, R.W., Goodling, J.S., and Gross, B.E.: ”Optimal
1,8 1,
Tip only Thermal Design of Air Cooled Forced Convection Finned
R [K/W]

1,5 No Bypass 1,
Heat Sinks – Experimental Verification”, IEEE Trans.
1,2 1,
CHMT, vol.15, n.5, pp.754-760 (1992)
[4] Wirtz, R.A., Chen, W., and Zhou, R.: "Effect of Flow
0,9 0, Bypass on the Performance of Longitudinal Fin Heat
Sinks", ASME Journal of Electronic Packaging, vol.116,
0,6 0,
pp.206-211 (1994).
0,3 0, [5] Butterbaugh, M.A., and Kang, S.S.: "Effect of Airflow
-5 0 5 10 15 20 25 30 35 40 45 50 Bypass on the Performance of Heat Sinks in Electronic
Pressure Drop [Pa] Packaging", Proceedings, ASME InterPACK'93, vol.2,
Figure 9 – R x Pressure Drop correlation pp.843-848 (1995).
6 9 12 [6] Biber, C.R., and Belady, C.L.: "Pressure Drop Prediction
3,0 3,0 for Heat Sinks: What is the Best Method?", Proceedings,
Q Exp (Knight) ASME InterPACK'97, vol.2, pp.1829-1835, (1997).
2,5 2,5
Q Model (Knight) [7] Reis, E. and Altemani, C.A.C.: “Design of Heat Sinks
2,0 Q Present Model 2,0
and Planar Spreaders with Airflow Bypass”, Proceedings,
100 Q [kg/s]
R [K/W]

1,5 1,5 ASME InterPACK’99, vol.1, pp.477-484 (1999).


1,0 1,0 [8] Patankar, S.V.: “Numerical Heat Transfer and Fluid
Flow”, Taylor&Fracis, (1980).
0,5 0,5 [9] Phillips, R.J.: “Microchannel Heat Sinks”, in Advances in
0,0 R Exp (Knight) 0,0 Thermal Modeling Electronic Components and Systems,
R Model (Knight) vol. 2, Ch. 3, Eds. Bar-Cohen, A., and Kraus, A.D., IEEE,
-0,5 -0,5
R Present Model (1990).
-1,0 -1,0 [10] Shah, R.K., and Bhatti, M.S.: "Laminar Convective Heat
6 9 12
Number of interfin channels Transfer in Ducts", in Handbook of Single Phase
Convective Heat Transfer, John Wiley, New York,
Figure 10 – Comparison with the data of [3]. (1987).
[11] Sparrow, E.M.: “Analysis of Laminar Forced Convection
Heat Transfer in Entrance Region of Flat Rectangular
point with the lowest approach velocity, with a deviation Ducts”, NACA TN 3331, (1955).
of 8,5 %. These results are redrawn in terms of the [12] Bhatti, M.S. and Shah, R.K.: "Turbulent and Transition
pressure drop in Figure 9. They indicate, as was done in Flow Convective Heat Transfer in Ducts", in Handbook
[5], the same thermal resistance – pressure drop of Single Phase Convective Heat Transfer, John Wiley,
correlation for any amount of bypass. New York, (1987).
[13] Moffat, R.J.: “Using Uncertainty Analysis in the Planning
The experimental data and the simulations presented in
of an Experiment”, J. Fluids Engineering, vol.107,
[3] were compared with the present model predictions, as pp.173-178, (1985).
indicated in Figure 10. With the present model, the [14] Jonsson, H., and Moshfegh, B.: “Modeling of the
airflow rate and thus the thermal resistance are in better Thermal and Hydraulic Performance of Plate Fin, Strip
agreement with data than the original predictions. Fin, and Pin Fin Heat Sinks – Influence of Flow Bypass”,
Proceedings, IEEE ITherm 2000, vol.I, pp.185-192.

Das könnte Ihnen auch gefallen