You are on page 1of 15

International Journal of Heat and Mass Transfer 50 (2007) 27722786

www.elsevier.com/locate/ijhmt

Experimental investigation on natural convection in a


convergent channel with uniformly heated plates
Nicola Bianco a, Oronzio Manca b,*, Sergio Nardini b
a
Dipartimento di Energetica, Termouidodinamica applicata e Condizionamenti ambientali, Universita degli Studi di Napoli Federico II,
Piazzale Tecchio 80, 80125, Napoli, Italy
b
Dipartimento di Ingegneria Aerospaziale e Meccanica, Seconda Universita degli Studi di Napoli, Via Roma 29, 81031, Aversa (CE), Italy

Received 21 July 2006


Available online 16 January 2007

Abstract

An experimental investigation on natural convection in air in vertical convergent channels with wall uniform heat ux is presented.
Wall temperatures show that by increasing the spacing the eect of the convergence angle decreases. For the lowest spacing, a signicant
decrease of maximum wall temperature occurs passing from the parallel vertical plate congurations to convergence angles h P 2 .
Dimensionless wall maximum temperature and Nusselt numbers are evaluated and correlated to the Rayleigh number or the channel
Rayleigh number. The correlations are in the ranges 2:85 6 Ra0bmin 6 1:22  105 and 0 6 h 6 10 .
Flow visualization is carried out both numerically and experimentally. Numerical and experimental data are compared and a good
agreement has been found.
2006 Elsevier Ltd. All rights reserved.

1. Introduction those channel congurations that can improve its thermal


performance [15]. Among them a very simple conguration
The thermal control of many systems, such as electron- is the convergent vertical channel with two principal paral-
ics, often requires an accurate prediction of the perfor- lel walls [313,16]. The rst numerical and experimental
mance of natural convection in channels [1,2]. As far as study of water natural convection in a convergent vertical
cooling of electronic equipment is concerned, an interesting channel was carried out by Sparrow et al. [3]. The principal
problem is the heat transfer in a convergent channel with converging walls were maintained at the same uniform
two uniformly heated at plates [313]. The determination temperature and the angles from the vertical varied
of the thermal performance of these congurations is between 0 and 15. The authors found that the Nusselt
rather dicult because of the large number of thermal numbers for the convergent channels could be brought into
and geometric variables. Experimental and numerical very close agreement with those for the parallel-walled
investigations can therefore give useful information on channel by employing correlation variables based on the
the eects of the heat ux, the inclination angle and the maximum interwall spacing as the characteristic dimen-
spacing. sion. The study was extended to the diverging channels
Natural convection in channels has been widely investi- by Sparrow and Ruiz [4]. They found that the correlation
gated both numerically [2] and experimentally [14] for dif- for vertical channels with parallel walls could be adapted
ferent boundary conditions at the principal walls of the to converging and diverging channels, by using the maxi-
channel. The present research eorts are devoted to dene mum interwall spacing. Kihm et al. [16] experimentally
investigated air natural convection in a uniform wall tem-
*
Corresponding author. Tel.: +39 0815010217; fax: +39 0815010204. perature converging channel using a specklegram tech-
E-mail address: oronzio.manca@unina2.it (O. Manca). nique. Local and average heat transfer coecients were

0017-9310/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2006.11.017
N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786 2773

Nomenclature

a thermal diusivity, m2 s1 m kinematic viscosity, m2 s1


b channel spacing (bmin or bav or bmax), m q density, kg m3
g acceleration of gravity, m s2
Gr channel Grashof number, Eq. (1) Subscripts
k thermal conductivity, W m1 K1 av average
L channel length, m b channel spacing (bmin or bav or bmax)
Nu Nusselt number, Eq. (3) bav average channel spacing
p pressure, Pa bmax maximum channel spacing
Pr Prandtl number bmin minimum channel spacing
q heat ux, W/m2 c convective
r regression coecient f uid
R dependent variable, Eq. (7) k conductive
Ra Rayleigh number, Eq. (1) min minimum
Ra0 Ra (b/L), channel Rayleigh number max maximum
T temperature, K o ambient air
T+ dimensionless temperature, Eq. (5) r radiative
u, v velocity components, m s1 s solid
W channel width, m w wall
x, y coordinates along the plate, m X Ohmic dissipation
Xi independent variable, Eq. (7)

Greek symbols
b volumetric coecient of expansion, K1
h inclination angle,

evaluated for ve inclination angles (060) and eight chan- accomplished in [7]. The inuence of converging angle on
nel exit openings. Haung et al. [11] accomplished an exper- local and average Nusselt numbers was studied. A blended
imental study of mixed convection in a vertical convergent correlation for average Nusselt number in isothermal chan-
channel. The channel was heated asymmetrically at uni- nels was proposed in the Rayleigh number range 1106 and
form heat ux. The heated wall was vertical and the oppo- convergence angle range 030. Results for asymmetric
site wall was insulated and convergent with an angle of 3. heating conditions were also given and, for these congura-
Flow visualization was employed in order to visualize the tions, a region of downow and recirculation appeared at
ow structure. Local and average Nusselt numbers were the channel outlet zone for high Rayleigh number. This
evaluated and correlated in terms of relevant dimensionless eect tended to diminish when converging angle increased.
parameters. A numerical investigation on natural convection in air in
Natural convection in air in uniform wall temperature vertical diverging and converging channels was carried
converging vertical channels was studied numerically by out in [8]. A nonuniform wall heating was considered and
Said [5]. The best correlation for the Nusselt number at it was modelled as an isothermal zone with its length lower
low Rayleigh numbers (<102) was achieved by the use of or equal to the wall length. The adiabatic zone was placed
the maximum channel spacing as the characteristic dimen- either at the bottom end of the channel or at the top end.
sion. Shalash et al. [6] investigated numerically and exper- The Rayleigh number was based on the wall length and
imentally, the same conguration studied in [5]. The the numerical simulations were obtained in the Rayleigh
computational domain was extended upstream and down- number range between 105 and 106. The results showed
stream of the channel. A Mach-Zender interferometer was that the optimal angle between the two walls was approxi-
used to perform the experiments. The authors found that, mately zero when the Rayleigh number was large. Natural
at low Rayleigh numbers, increasing the angle signicantly convection in air in a convergent channel with the two prin-
increases the Nusselt number, whereas, at high Rayleigh cipal at plates at uniform heat ux, with nite thickness
numbers, increasing the angle, the Nusselt number and thermal conductivity, was numerically studied in [9].
decreases. When the inclination angle was increased, the An extended computational domain was adopted. Temper-
local Nusselt number decreased near the channel inlet ature proles were strongly aected by the convergence
and increased near the channel outlet, particularly at low angle at low Rayleigh numbers, whereas the opposite occu-
Rayleigh numbers. A two-dimensional numerical simula- red at high Rayleigh numbers. At the lower minimum gap,
tion of natural convection in a converging channel was streamlines and isotherms highlighted a low pressure zone
2774 N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786

in the channel, due to a choked ow in its upper region.

Correlation
Marcondes et al. [12] carried out a numerical investigation

None

None

None
on natural convection in parallel, convergent and divergent

Yes

Yes

Yes

Yes

Yes

Yes

Yes
Yes
channels using a fully elliptic procedure considering only
the simple channel as computational domain. Both the

4  103 7  104

1  105 1  107

1  105 1  108

2:851:22  105
6:44:8  104

4:42:9  108

4:41:5  105

4:61:8  108
channel walls were at uniform temperature and results were

11  108

12  104
given for Prandtl number ranging from 0.7 to 88. A corre-

11106
lation for average Nusselt number, Rayleigh number, in

Ra0
terms of maximum channel width and channel aspect ratio,
and Prandtl number was proposed. They found that for

1.65, 3.82,
11.4, 22.9
convergent channels a recirculation region in the outlet

6, 8.5, 12
2.514

0.550

2.060
zone was observed. Design charts for the evaluation of

1258

1058

1058
11.4
L/b
thermal and geometrical parameters of vertical convergent


channels were presented by Bianco et al. [13]. Correlations
for dimensionless maximum wall temperature and average

Reference
Nusselt number, in terms of Rayleigh numbers and geo-

bmin =2
length
bmax

bmax

bmax

bmax

bmax
metrical parameters, for natural convection in air, in a ver-

bmin

bmin

bmin

bmin
Lw
tical convergent channel, with symmetrically heated walls,

0, 15, 30, 45,


were proposed. The results were obtained elaborating the

0, 2, 5, 10,

0, 2, 5, 10

0, 2, 5, 10

0, 2, 5, 10

0, 2, 5, 10
numerical data carried out in [9,10]. A simple procedure

0, 2, 4, 8

0, 5, 15
to evaluate the optimal geometrical congurations in terms
of channel spacing and convergence angle was presented.

060

010
05
15

60
In the authors opinion, there is a lack of experimental
d

data on natural convection in air (Pr = 0.71) in vertical


convergent channels with uniform heat ux at the principal
Radiative

walls as shown in Table 1. Therefore, an experimental


None

None

None

None

None
None
None

None

None
eect

Yes

Yes
investigation has been carried out. In this paper, results
in terms of wall temperature proles are presented as a
function of the walls inclination angle, the distance
Conduction on

between the walls and the heat ux. The experimental


results are compared with the ones carried out with the
None

None

None

None

None
None

None
walls

numerical model given in [9,13]. Therefore, the present


Yes

Yes

Yes
Yes
paper allows to verify experimentally the results and con-
clusions reported in [9,13]. The comparison is accom-
Comparison among dierent studies of natural convection in convergent channels

plished in terms of stream function elds obtained


Boundary conditions on

numerically and visualization obtained experimentally.


Nusselt numbers and dimensionless maximum wall temper-
atures are evaluated and correlated to the channel Rayleigh
channel walls

numbers and dimensionless geometrical parameters.


UWHF

UWHF

UWHF
UWHF

2. Experimental apparatus
UWT

UWT

UWT

UWT

UWT
UWT

UWT

The investigated test section is shown in Fig. 1, where x


Type of investigation

and y are the coordinates along the length and the width of
Experimental and

the plate, respectively. The channel is made of two princi-


Numerical and

Numerical and

Numerical and

Numerical and

Numerical and

pal plates, symmetrically heated at uniform heat ux and


experimental

experimental

experimental

experimental

experimental

two unheated plexiglass side walls (Fig. 1a). It is open to


Numerical
Numerical
Numerical

Numerical

Numerical

numerical

the ambient along the top and the bottom edges.


The cross-section of the isoux principal walls of the
channel is sketched in Fig. 1b. Each wall was made with
Bianco et al. (2006) [10]

Bianco et al. (2006) [13]


Sparrow et al. (1988) [3]

Marcondes et al. (2006)


Shalash et al. (1997) [6]

a 3.2 mm thick, 406 mm long and 530 mm wide phenolic


Kihm et al. (1993) [16]

Kaiser et al. (2004) [7]


Bejan et al. (2004) [8]
Bianco and Nardini

berboard plate, with a typical thermal conductivity of


0.17 W m1 K1. Its surface facing the channel was coated
Said (1996) [5]

Present work

with a 16 lm thick copper layer, which was the heater. It


(2005) [9]
Reference

was obtained by cutting the copper in a 2.5 mm wide


Table 1

[12]

and 38 m long serpentine track and its expected electric


resistance was about 8.0 X. A thick copper bar, bolted to
N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786 2775

bmin

Polystyrene Phenolic fiberboard

3.2 mm
plates
Copper

Flow direction
L

Polystyrene
W

y
b max

Fig. 1. (a) Sketch of the test section; (b) particular of the heated walls.

the electric supply wire, was soldered to the ends of each ments; the measured dierences in the air ambient temper-
heater. The dissipated heat ux was evaluated with an ature in the proximity of the inlet and the exit sections of
accuracy of 2%, by measuring the voltage drop and the the apparatus were less than 0.8 K.
current through the electric resistance. A maximum varia- The principal walls of the channel were 406 mm long
tion of 10% in the electrical resistivity of the copper and 450 mm wide. The values of the minimum channel
was evaluated in the worst conditions, when the maximum gap were bmin = 7.0 mm, 10.0 mm, 20.0 mm, 32.3 mm and
dierence in the wall temperatures was 30 K. Therefore, a 40.0 mm. The convergence angle, h, varied in the 010
uniform wall heat ux was assumed, with a 5% maximum range; consequently, the maximum channel spacing,
deviation from its average value. Heat losses from the back bmax, varied in the 7.0148.0 mm range when bmin was
of the heated walls were reduced by sticking a 150 mm 7.0 mm and in the 40.0181.0 mm range when bmin was
polystyrene block on the rear surface of each wall. The 40.0 mm.
total normal emissivity of the walls was 0.1 and was Wall temperature measurements were carried out by
obtained by sticking a self-adhesive 25 lm thick aluminum 0.50 mm OD iron-constantan (type J) thermocouples,
foil on the surfaces facing the channel. The emissivity was embedded in the berboard in the very proximity of the
evaluated by means of radiometric direct measurements. back side of the copper layer and bonded with a 3 M epoxy
The electric insulation between the copper surfaces and glue. They were run horizontally, parallel to the surfaces,
the aluminum foil was assured by uniformly spraying an thereby lying along isotherms in order to minimize conduc-
electrically insulating varnish onto them before coating. tion heat losses in the leads. Sixteen thermocouples were
Side walls were two trapezoidal plexiglass plates, 4.0 mm placed in the centerline of each plate at the following loca-
thick, placed between the principal walls at their lateral tions: x 2:5;12:7;38:1;73:7;88:9;104:1;124:5;139:7; 154:9;
edges. They were machined within an accuracy of 175:3;190:5;205:7;241:3;281:9;342:9;403:9 mm.
0.03 mm. The distance between the principal walls was At 300 mm from the leading edge of one of the
measured with an accuracy of 0.25 mm using a dial-gauge walls, eight additional thermocouples were horizontally
caliper which can resolve 0.025 mm. The inclination located outward from the centerline at y 75:0 mm,
angle was set and veried by measuring the maximum 100.0 mm, 125.0 mm and 150.0 mm, in order to pro-
and the minimum spacings. The walls were fastened vide indications of the horizontal variation of the wall tem-
together by a steel frame, which was designed in such a perature. The ambient air temperature was measured by
way as not to obstruct the uid ow in the proximity of the same type of thermocouples located in the proximity
the channel inlet. A hinge at the top of the principal walls of the leading edge of the channel. Fifteen thermocouples
allowed their rotation in order to obtain a convergent were axed to the rear surface of the plates and embedded
channel. The channel was aligned vertically, with horizon- in the polystyrene block, in order to evaluate the conduc-
tal leading edges, by means of a plumb line and a level. The tive heat losses. Thermocouple voltages were recorded to
entire apparatus was located in an enclosed room, accu- 1 lV. Each thermocouple was calibrated in a 0.01 K ther-
rately sealed in order to eliminate extraneous air currents mostatic bath using a reference standard thermometer (Pt
and air drafts were further reduced by vertical screens, 100). The calibration of the temperature measuring system
2.5 m high. A large fraction of the lower part of the screens showed an estimated precision of the thermocouple-read-
was made up of a 0.20 m high mesh. The range of ambient out system of 0.2 K. A National Instruments SCXI mod-
temperature varied from 21 to 22 C during the experi- ule data acquisition system and a personal computer were
2776 N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786

Camera In the following step reference will be made to three dif-


ferent Rayleigh numbers, based on the characteristic length
bmin, bmax and bav = (bmax + bmin)/2, with the aim of nd-
ing the correlation which more accurately merges the con-
Laser sheet vergent channel results with those for the parallel-walled
Plenum channel.
Incense sticks
Nusselt numbers will be based on the dierence between
the wall and the inlet uid temperatures, rather than on
Compressed Air that between the wall and the bulk uid temperatures, since
Heat exchanger
the latter cannot be easily measured in practical applica-
Cylindrical lens
tions. The channel Nusselt number is dened as
qc b
Laser Source
Nu 3
T W  T o k
where Tw is the average wall temperature
Z
1 L
Tw T w x dx 4
L 0
Fig. 2. Flow visualization equipment.
The dimensionless temperature is dened as
used for the data collection and reduction. The data acqui- T  To
sition was performed through the LabViewTM software. T qc b
5
Tests showed the wall temperature to be symmetric k

between the two heated plates at the same x location, the Also for Nusselt numbers, reference will be made to
dierences being within 0.2 C. Wall temperature was bmin, bmax and bav.
assumed to be independent of y coordinate within The properties of the air are evaluated at the reference
100 mm, since its maximum deviation from the centerline temperature (Tw + To)/2.
temperature was found to be no greater than 1.2 C when Local convective heat ux, qc(x), is not uniform because
the latter was 80.0 C. of radiation and conduction. Experimental data are
Smoke for visualization was generated by burning reduced rst by introducing, in the equations presented
incense sticks in a steel tube, connected to a compressor above, the local heat ux
(Fig. 2). The smoke was injected through a glass heat
qc x qX x  qk x  qr x 6
exchanger to reduce the temperature of the smoke. The
smoke passed into a plenum and its temperature was con- In Eq. (6), the overall heat rate divided by the area of the
trolled using a thermocouple. This value was close to that wall surface is the local heat ux due to Ohmic dissipation,
of the incoming ambient air into the channel. Then it was qX(x), which was assumed to be uniform along the heated
driven in the test section through a small slot situated along plates; qk(x) denotes the local conduction heat losses from
the leading edge of the channel. The visualization was the plates and qr(x) is the local radiative heat ux from
made possible by means of a laser sheet, generated by a the plates. For each run, the terms qk(x) were calculated
HeNe laser source. The laser sheet was produced by plac- by a nite dierence numerical procedure, assuming a
ing a mirror near the end of the test section at an angle of two-dimensional distribution of the temperature in the
45 with respect to the direction of the main ow, after polystyrene. The predicted temperatures in signicant con-
which a cylindrical lens was placed to enlarge the beam gurations of the system had been previously compared to
as needed. A ne regulation was allowed by means of a those measured by the thermocouples embedded in the
micrometer screw system, in order to obtain photos at polystyrene insulation and the agreement was very good,
the y = 0 plane. the maximum deviation being 4%. The qr(x) terms were
calculated for each temperature distribution in the walls,
3. Data reduction ambient temperature and channel spacing, dividing each
plate into 10 equal length strips along the channel, accord-
The channel Rayleigh number and the modied channel ing to the procedure described by Webb and Hill [17]. For
Rayleigh number are dened as: all the investigated congurations the conductive heat
losses were about 9% of the Ohmic wall heat ux, whereas
gbqc b4 b the radiative heat losses ranged between about 3% and 5%.
Ra Gr Pr 2
Pr and Ra0 Ra 1
mk L The uncertainty in the calculated quantities was deter-
where qc is the spatially-averaged convective heat ux mined according to the standard single sample analysis rec-
Z ommended in [18]. The uncertainty of a dependent variable
1 L R as a function of the uncertainties in the independent vari-
qc q x dx 2
L 0 c ables Xi is given by the relation
N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786 2777

" 2  2  2 #1=2 b Ly
oR oR oR
dR dX 1 dX 2  dX n R Q
oX 1 oX 2 oX n

M N
7 I P
L O
The uncertainty in the values of the air thermophysical
properties were assumed to be negligible. The maximum Solid
percentage uncertainties in the values of the independent L
variables were: 0.5% for Tw, 0.93% for To, 1.1% for
(TwTo), 1.2% for bmin, 2.0% for qX, 4.0% for qk, 3.0% n
for qr and 5.0% for qc. On the basis of Eqs. (1), (3), (5)
the maximum uncertainty in Ra0 bmin was 6.7% whereas A B C D E F
the maximum uncertainty in Nubmin turned out to be 3.4%.
x

Lx
4. Mathematical model and numerical procedure
o y
H
G
The numerical model presented in [9] allows the solution
of the physical domain shown in Fig. 3a. It consists of two Fig. 3. Sketch of the conguration (a) and computational domain (b).
non-parallel plates which form a convergent channel. Both
plates are conductive and heated at uniform heat ux qX.
The ow in the channel is assumed to be two-dimensional, Table 2
Boundary conditions for the domain
laminar, incompressible, with negligible viscous dissipa-
tion. The working uid is air (Pr = 0.71) and all thermo- Wall u v T
physical properties of the uid are assumed to be Fluid domain
ou ov
constant, except for the dependence of density on the tem- AH and FG oy 0 oy 0 T To
ou ov
perature (Boussinesq approximation) which gives rise to HG ox 0 ox 0 T To
oT
the buoyancy forces. The thermophysical properties of AB, EF, IL, OP u=0 v=0 ox 0
the uid are evaluated at the ambient temperature, To, BC, DE, LM, NO u=0 v=0 k f oT oT
ox k s ox
which is assumed to be 300 K in all cases. DN u=0 v=0 k f oT oT
on k s on qX
The governing equations, for the uid region in steady CM u=0 v=0 k f oT oT
on k s on  qX
state regime, are RQ ou
ox 0 ov
ox 0 oT
ox 0 if u > 0; T T o if u < 0
ou ov oT
ou ov IR oy 0 oy 0 oy 0 if v < 0; T T o if v > 0
0 8 QP ou
0 ov
0 oT
0 if v > 0; T T o if v < 0
ox oy oy oy oy
 2  Solid domain
ou ou 1 op o u o2 u
u v  m  gbT  T o 9 DN k f oT oT
on k s on qX and T s T f
ox oy qf ox ox2 oy 2
 2  CM k f oT oT
on k s on  qX and T s T f
ov ov 1 op o v o2 v BL, OE oT
0
u v  m 10 on
ox oy qf oy ox2 oy 2 BC, DE, LM, NO k f oT oT
ox k s ox and T s T f
 2 2 
oT oT oT oT
u v af 11
ox oy ox2 oy 2
domains are reported in Table 2. The pressure defect equals
A two-dimensional conduction model in the walls is em- zero on the inlet and outlet boundaries.
ployed; radiative heat transfer is neglected. The heat trans- A nite volume method was employed to solve Eqs.
fer equation in the steady state regime with constant (8)(12) and the commercial code Fluent [20] was used to
thermophysical properties is carry out numerical results. Several preliminary tests were
o2 T s o2 T s carried out with dierent methods of solution. Since results
2 0 12 were nearly equal for the same conditions, the quicker
ox2 oy
method was employed. The segregated solution method
Since the two plates are placed in an innite medium, the was chosen to solve the governing equations, which were
problem has been solved by taking a computational do- linearized implicitly with respect to the equations depen-
main of nite extent, as depicted in Fig. 3b, and by follow- dent variable. The rst-order upwind scheme was chosen
ing the approach given in [19]. The upstream and for the unsteady energy and momentum equations. The
downstream reservoirs have been employed to simulate semi implicit method for pressure-linked equations (SIM-
the free-stream condition of the ow away from the region PLE) scheme was chosen to couple pressure and velocity.
of the thermal disturbance induced by the heated plates. Computation starts with zero values for the velocity com-
The imposed boundary conditions for the uid and solid ponents and with pressure and temperature values equal
2778 N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786

to the ambient values. The convergence criteria were 105 18


for the residuals of the velocity components and 106 for =0
16
the residuals of the energy equation. =2
=5
Preliminary results were obtained to evaluate the eect of 14
=10
the grid. The conguration examined was the case with 12

Tw-To [K]
bmin = 32.3 mm, h 10 at qX 120 W=m2 . The average
10
Nusselt numbers were evaluated for dierent grids, with
the following number of nodes in the channel: 50  100, 8
100  200, 150  300, 80  400, 100  400 and 200  400.
6 q =30 W/m 2
Dierences between Nusselt numbers for the 100 400 and b min =7.0 mm
200  400 were nearly 4% and the former was chosen in 4
order to obtain the results. 0 100 200 300 400
As to the dimensions of the reservoirs, Lx L and x [mm]
Ly 10bmin were chosen, since they do not aect the veloc-
30
ity and temperature in the channel, as it was reported in =0
[19]. A more detailed description of the numerical model 25 =2
and procedure is given in [9,10]. =5
=10
20

Tw-To [K]
5. Results
15
Experiments were carried out in air Pr 0:71 for
bmin 7:040:0 mm, h 0 , 2, 5 and 10 and qX 30, 10 q =60 W/m 2
60, 120 and 220 W/m2, corresponding to a Ra0 bmin from b min =7.0 mm
2.85 to 1.22  105. 5

0 1 00 200 300 400


5.1. Wall temperatures x [mm]

50
Wall temperature proles as a functions of the axial
=0
coordinate are reported in Figs. 46 for the investigated =2
values of spacing, of the convergence angle h and heat ux 40
=5
qX . =10
For bmin 7:0 mm and qX 30 W=m2 , Fig. 4a, at the 30
Tw-To [K]

exit section the percentage decrease of wall temperature


value with respect to the maximum wall temperature is 20
about 16% for h 0 whereas it is between 17% and 20%
for h P 2 . When the value of the heat ux increases, 10 q =120 W/m 2
Fig. 4b and c, the percentage decrease becomes higher. b min =7.0 mm
For convergence angles h P 2 , wall temperature values 0
are higher than the ones for h 0 in the lower part of 0 100 200 300 400
x [mm]
the heated wall because in this zone the velocity is lower,
even in boundary layer, due to a larger transversal section. Fig. 4. Wall temperature rise above the ambient temperature vs. the
In a convergent channel the boundary layers adjacent the vertical coordinate, for bmin 7:0 mm and h = 0, 2, 5, 10: (a) qX
30 W=m2 ; (b) qX 60 W=m2 ; (c) qX 120 W=m2 .
walls interact at a lower x value than the one for h 0 .
Furthermore, for each convergence angle h when the heat
ux value increases, the value of the coordinate x, where 30 W=m2 and h 0 and 18% for qX 120 W=m2 and
the interaction between the boundary layers is attained, is h 0 . Furthermore, percentage dierences between
higher. In fact, when the value of the heat ux, hence the maximum temperature values for h 0 and h P 2 are
Rayleigh number, increases, the boundary layer becomes lower than the ones for bmin 7:0 mm. In fact, the percent-
thinner. For h P 2 temperatures are lower than the ones age dierence is 11% for qX 30 W=m2 , 15% for qX
for h 0 at the higher part of the channel walls because 60 W=m2 , 9% for qX 120 W=m2 , 4% for qX
uid velocity adjacent the walls increases. Discrepancies 220 W=m2 . For bmin 10:0 mm the boundary layer is thin-
between maximum wall temperature in the range 210 ner than the one for bmin 7:0 mm because Rayleigh num-
are lower than the ones between 0 and 2. ber is higher. Therefore, the point, where the temperature
When the value of bmin = 10.0 mm, Fig. 5, maximum proles for h 0 and h P 2 intersect, is closer to the
temperature values are lower than the ones for bmin = channel exit section and lies in the channel upper half.
7.0 mm for each value of heat ux and convergence angle. For bmin 20:0 mm, Fig. 6, dierences between maxi-
In particular, the decrement is equal to 8% for qX mum temperature values for investigated convergence
N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786 2779

14 22
=0 20 =0
=2 =2
12 18
=5 =5
=10 16 =10
10

Tw-To [K]
Tw-To [K]

14
12
8
10
q =30 W/m 2
6 8 q =60 W/m 2
b min =10.0 mm
6 b min =10.0 mm
4 4

0 100 200 300 400 0 100 200 300 400


x [mm] x [mm]

40 60
=0 =0
35 =2 50 =2
=5 =5
30 =10
=10
40
25
Tw-To [K]

Tw-To [K]
20 30

15 q =120 W/m 2 q =220 W/m 2


20
b min =10.0 mm b min =10.0 mm
10
10
5
0 100 200 300 400 0 100 200 300 400
x [mm] x [mm]

Fig. 5. Wall temperature rise above the ambient temperature vs. the vertical coordinate, for bmin 10:0 mm and h = 0, 2, 5, 10: (a) qX 30 W=m2 ; (b)
qX 60 W=m2 ; (c) qX 120 W=m2 ; (d) qX 220 W=m2 .

angles are lower than the ones for bmin 10:0 mm, maxi- not advantageous because they imply an increase of system
mum percentage dierence being equal to 3% between volume without a signicant decrease of maximum wall
h 0 and 2 for qX 60 W=m2 . This suggests that for temperature, according to [8,21,22]. For low bmin values,
bmin values greater than 20.0 mm, a convergent channel is optimal convergence angle is at h 2 whereas when bmin
not recommended in order to contain wall maximum tem- increases, optimal angle is at about h 0 . However, max-
perature. A signicant advantage in the use of a convergent imum wall temperature values attained for h 2 are close
channel is for low bmin values and convergence angle to the lower ones. In either case, the best conguration
almost 2. This is in accordance with numerical results regarding the volume is h 2 .
reported in [810] and theoretical analysis given in [13].
Fig. 7 shows maximum wall temperature, referred to the
5.2. Correlations for dimensionless maximum wall
ambient temperature, as a function of bmin for the investi-
temperature and average Nusselt number
gated convergence angles and for several heat ux values.
This gure allows for a detection of optimal thermal con-
Maximum wall temperatures reported in the previous
gurations with respect to maximum wall temperature for
gure are made dimensionless by Eq. (5) where the refer-
converging channels. For each heat ux value, the changes
ence length is bmin, bmax or bav. Correlations between
of maximum wall temperature with convergence angle
dimensionless maximum wall temperature and channel
decrease when bmin increases and this discrepancy is higher
Rayleigh number can be obtained by means of least square
than 2 K for bmin values up to about 10 mm. In each case,
method. The correlations have the following expression
the worst conguration for bmin 6 10 mm is the one with
parallel heated walls. For bmin higher than 10 mm maxi- b
T 0
b aRab 13
mum wall temperature are attained at h < 10 . When bmin
increases, the values of the maximum wall temperature The values of the coecient a and b are reported in
diminish, but the percentage drop passing from Table 3 as functions of the reference lengths bmin, bmax
bmin = 20.0 mm to bmin 40:0 mm is much lower than and bav. Experimental data and the corresponding correla-
percentage drop passing from bmin 7:0 mm to bmin tions are reported in
20:0 mm. This suggests that high bmin values are not  Fig.
 8. A better correlation was
2=5
bmin
favourable for the thermal control. In order to reduce obtained with Rabmax bmax
as the independent variable,
maximum wall temperature values, high bmin values are as suggested in [13]. The corresponding expression is
2780 N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786

20 Channel Nusselt number and channel Rayleigh number


=0 were also correlated by the following monomial equation
18
=2
=5 Nub nRam
b 16
16
=10
whose coecients, for bmin, bav and bmax, are reported in
Tw-To [K]

14
Table 3. Experimental data as well as the regression curves,
12 based on bmin, bav and bmax, are reported in Fig. 11ac,
q =60 W/m 2
respectively. The best correlation of experimental data were
10
b min =20.0 mm obtained by adopting bmax as a characteristic length. The
8 regression coecients at any spacing values are larger than
those in Table 4. However, it is worth remarking that the
6
0 100 200 300 400 coecients in Eq. (15) are very similar to those for the sin-
x [mm] gle plate. One can conclude that the thermal performance
of a convergent channel with a maximum spacing bmax is,
=0 for practical purposes, equal to that of a parallel-walled
50 channel with a spacing b bmax .
=2
=5
=10
40
5.3. Flow visualization and numerical stream function and
Tw-To [K]

temperature elds

30 In the following, a comparison between experimental


q =220 W/m 2
b min =20.0 mm and numerical data are accomplished in terms of experi-
mental ow visualization and numerical stream function
20
elds. The comparison is given in Figs. 1214 for conver-
gence angles equal to 2, 5, and 10, for bmin 7:0, 20.0
0 100 200 300 400
and 32.3 mm and for qX 30 and 120 W/m2.
x [mm]
Pictures of the ow patterns in the channel and stream
Fig. 6. Wall temperature rise above the ambient temperature vs. the function elds for a convergence angle equal to 2 and a
vertical coordinate, for bmin 20:0 mm and h = 0, 2, 5, 10: (a) qX ohmic heat ux equal to 120 W/m2 are reported in Fig. 12
60 W=m2 ; (b) qX 220 W=m2 .
for bmin 7:0 mm (a, b) and 20.0 mm (c, d). Both numerical
and experimental results show a laminar boundary layer
near the channel walls. At the channel inlet, the boundary
"  2=5 #0:27
bmin layers do not interact both for bmin 7:0 mm and
T
max;bmax 8:50 Rabmax 14 20.0 mm. It is observed, but not shown here, that for the
bmax
parallel walled channel, the interaction between the two
with r2 0:995 and valid for 196 < Rabmax < 4:89  108 boundary layers starts in the zone near the channel inlet.
and 4:73  102 < bbmax
min
< 1 ranges. This correlation allows For a convergence angle equal to 5, pictures of the ow
to evaluate the dimensionless maximum wall temperature patterns and stream function elds for qX = 30 and 120 W/
when the emissivity, e, of the channel walls is 0, whereas m2 are reported in Fig. 13 for bmin 7:0 mm (Fig. 13a, b, e,
the correlation proposed in [13] is valid for e > 0. f) and 20.0 mm (Fig. 13c, d, g, h). For bmin 7:0 mm, both
Fig. 9 shows the good agreement between the correla- numerical and experimental results show a recirculating
tion of Eq. (14) and the experimental data. zone in the central part of the channel. This is due to the
As proposed in [3], the channel Nusselt number and the two symmetric boundary layers that merge in the exit
channel Rayleigh number, Ra0 Rab=L, were referred to region of the channel and the convergence angle as well.
the three characteristic lengths bmin, bav and bmax and cor- The velocity component along y is not zero and tends
related in the following form toward the centerline of the channel. These two eects
choke the channel ow and generate the recirculating
Nub nRa0m
b 15 region. By comparing Fig. 13a and b, related to qX
The coecients in Eq. (5) were obtained by means of the 30 W=m2 , with Fig. 13e and f, related to qX 120 W=m2 ,
least square method and are reported in Table 4, together it is observed that the recirculating zone moves up when
with the regression coecients. It is worth noticing that the ohmic heat ux increases. This eect can be explained
the worst correlation r2 0:983 was obtained by basing with the decrease of the boundary layer thickness. The
the Nusselt and Rayleigh numbers on the average spacing. recirculating zone is not present for bmin 20:0 mm, even
Channel Nusselt and modied Rayleigh numbers, based on if a separation of the streamlines in the central part of
bmin, bav and bmax, as well as the regression curves are the channel is detected for qX 120 W=m2 , Fig. 13h. These
reported in Fig. 10ac. The correlations are in the follow- gures show a good agreement between numerical and
ing range: 2:85 6 Ra0bmin 6 1:22  105 and 0 6 h 6 10 . experimental results, also in accordance with [9,10,12].
N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786 2781

16 26

15 =0 =0
=2 24 =2
Tw,ma x - To [K] 14 =5 =5

Tw,ma x - To [K]
=10 =10
13 22

12 q =30 W/m 2 20 q =60 W/m 2


11
18
10

9 16
0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
b min [m] b min [m]

42
54
40 =0
=2 52
38 =5
Tw,ma x - To [K]

50

Tw,ma x - To [K]
=10
36
48
34 q =120 W/m 2
46
32 =0
44 q =220 W/m 2 =2
30 =5
42 =10
28 40
0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
b min [m] b min [m]

Fig. 7. Maximum wall temperature rise above the ambient temperature as a function of bmin, for h = 0, 2, 5, 10: (a) qX 30 W=m2 ; (b) qX 60 W=m2 ;
(c) qX 120 W=m2 ; (d) qX 220 W=m2 .

Table 3
Coecients in Eqs. (13) and (16)
0 b
T
b aRab ; Eq. (13) Nub n Ram
b ; Eq. (16)
2
a b r n m r2
b bmin 2.80 0.22 0.988 0.15 0.26 0.991
b bav 2.49 0.20 0.978 0.20 0.24 0.989
b bmax 2.39 0.20 0.985 0.20 0.24 0.994

In Fig. 14, pictures of the ow patterns and stream func- bmin 32:3 mm are reported to show that, for this mini-
tion elds for a convergence angle equal to 10, for qX = 30 mum channel spacing value, the channel is not chocked
and 120 W/m2 are reported. In this case, for qX = 30 W/m2 and no recirculating zones are present. Here the boundary
results for bmin 7:0 mm and 20.0 mm are reported layers do not interact, as shown in Fig. 14h. Also in this
(Fig. 14ad), whereas for qX = 120 W/m2, results are case the gures show a very good agreement between
related to bmin 7:0 mm and 32.3 mm (Fig. 14eh). These experimental and numerical results.
gures show that for h 10 and for bmin 7:0 mm, the In Fig. 15, numerically obtained temperature elds for
recirculating zones are larger and situated at a larger x than h 5 , qX 30 W=m2 and 120 W/m2 are reported for
for h 5 . In fact, in this case, due to the larger conver- bmin 7:0 mm (Fig. 15a and c) and 20.0 mm (Fig. 15b
gence angle, the boundary layers meet at a larger x coordi- and d). The thermal boundary layer thickness decreases
nate and the vortexes move up when the heat ux increases. with the ohmic heat ux; in fact the isotherm related to
Fig. 14c and d show that for this convergence angle the ambient temperature penetrates more inside the chan-
h 10 , a recirculating zone is also present for bmin nel for qX 120 W=m2 than for qX 30 W=m2 . This eect
20:0 mm. In this case, this zone is closer to the outlet of is more pronounced for bmin 20:0 mm than for
the channel than the corresponding case with bmin bmin 7:0 mm. For bmin 7:0 mm, a maximum relative
7:0 mm. It was observed, but not shown here, that the temperature on the centerline is observed in the central part
channel is also chocked for bmin 20:0 mm and qX of the channel. This is clearly due to the recirculating zone
120 W=m2 . For this heat ux value, ow patterns for that moves hot air downward from the upper zones. This
2782 N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786

10 10

+ -0.22 =0
T = 2.80 Ra'
2
ma x,bmin bmin
=2
r =0.988 =5
1 =10

ma x,bma x
ma x,bmin

+
+
T

T
=0 0.1
=2
=5 T+ma x,bmax = 8.50 [R a bmax (bmin /b ma x)2/5 ]-0.2 7
=10 r2 =0.996
0.1 0.01
10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8 10 9
Ra' bmin 2/5
Ra bmax (b min /bma x)

10
Fig. 9. Dimensionless maximum wall temperature rise above the ambient
+
temperature vs. Rabmax bmin =bmax 0:4 .
-0.2 0
T max ,bav = 2.49 Ra' bav
r2 =0.978
1
Table 4
ma x,bav

Coecients in Eq. (15)


+
T

b bmin b bav b bmax


0.1 =0
=2 n 0.48 0.57 0.59
=5 m 0.21 0.19 0.19
=10 r2 0.987 0.983 0.989

0.01
10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8
Ra 'ba v two considered h values. Isotherms in Fig. 16b h
10 ; qX 30 W=m2 ; bmin 20:0 mm show the presence
10 of the recirculating zone detected by the ow patterns pre-
+ -0.2 0
sented in Fig. 14c and d. For bmin 32:3 mm, Fig. 16d, the
T max ,bmax = 2.39 Ra' bmax
2
r =0.985
ambient isotherm penetrates inside the channel, very close
1
to the channel exit and the thermal boundary layers meet
only at almost the exit of the channel. In this case, iso-
max ,bmax

therms show that no recirculating zones are present.


+
T

0.1 =0
=2 6. Conclusions
=5
=10
An experimental investigation on natural convection in
0.01 air Pr 0:71 in a convergent channel with symmetric
10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8 10 9
heating was accomplished, in order to analyze the eects
Ra' bmax
of the channel spacing, convergence angle and heat ux.
Fig. 8. Dimensionless maximum wall temperature rise above the ambient The channel walls were heated at uniform heat ux. Results
temperature vs. the modied Rayleigh number: (a) b bmin ; (b) b bav ; in terms of wall temperature proles as a function of the
(c) b bmax . walls inclination angle, the channel spacing and the heat
ux were given. Nusselt numbers and dimensionless maxi-
mum temperatures were evaluated and correlated to the
eect is not detectable for bmin 20:0 mm; in this case the modied Rayleigh number based on minimum value of
vortexes are not present as shown in Fig. 13. spacing between the plate, bmin, in the 2.851.22  105
Numerically obtained temperature elds for h 10 , range and 0 6 h 6 10 .
qX 30 W=m2 and 120 W/m2 are reported in Fig. 16. For the lowest spacing value, the parallel plate congu-
Here, for qX 30 W=m2 , results for bmin 7:0 mm and ration, h 0 , presented the highest maximum wall tem-
20.0 mm are reported (Fig. 16a and b), whereas for perature and for a small increase of the convergence
qX 120 W=m2 , results are related to bmin 7:0 mm and angle, h 2 , a remarkable decrease was noticed. For
32.3 mm (Fig. 16c and d). By comparing Fig. 16ac, with h > 2 , maximum wall temperature presented a slight
Fig. 15ac, it is observed that the distance from the inlet decrease. These results conrmed the numerical analysis
at which the thermal eld can be considered fully devel- given in [810]. Moreover, for larger channel spacings,
oped does not depend on the convergence angle, for the bmin P 20 mm, wall temperatures increased at increasing
N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786 2783

100 10

=0 =0
=2 =2
=5 =5
10 =10 =10
Nu bm in

Nu bmin
1

1 Nubmin=0.15 Rabmin0.26
Nubmin=0.48 Ra'bmin0.21

r2=0.991
r2=0.987

0.1 0.1
10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 2 10 3 10 4 10 5 10 6 10 7
Ra bmin
Ra 'bmin
100
100 =0
=0 =2
=2 =5
=5 10 =10
10 =10

Nu bav
Nu bav

1 0.24
Nubav=0.20 Rabav
1
0.19
Nubav=0.57 Ra'bav r2=0.989

r2=0.983 0.1 2
10 10 3 10 4 10 5 10 6 10 7 10 8 10 9
0.1
10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8 Ra bav
Ra 'bav
100
100 =0
=0 =2
=2 =5
=5 10 =10
10 =10
Nu bmax
Nu bmax

1
Nubmax=0.20 Rabmax0.24
1
0.19
Nubmax=0.59 Ra'bmax
r2=0.994

r2=0.989 0.1
10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8 10 9 10 10
0.1
10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8 10 9 Ra bmax
Ra 'bmax
Fig. 11. Channel Nusselt number vs. the Rayleigh number: (a) b bmin ;
Fig. 10. Channel Nusselt number vs. the modied Rayleigh number: (a) (b) b bav ; (c) b bmax .
b bmin ; (b) b bav ; (c) b bmax .

emissivity of the channel walls, e = 0 and, together with the


converging angles at any spacings, whereas they were
correlation proposed in [13], completes the e range to eval-
almost independent of the spacing at larger values of the uate maximum wall temperatures. The best correlation for
heat ux. However, it is worth remarking that the depen- the Nusselt number was obtained in terms of channel Ray-
dence of wall temperature on the inclination angle was
leigh number with bmax as reference length. The correla-
smaller in the exit region of the channel.
tions were evaluated in the ranges: 195 6 Rabmax 6 4:89
Dimensionless maximum wall temperatures and average
108 and 0 6 h 6 10 .
Nusselt numbers were correlated with channel Rayleigh
A comparison between experimental ow visualization
numbers in monomial equations. It was found that, for
and stream function elds obtained numerically was
the dimensionless maximum wall temperatures, the best
accomplished and a good agreement was observed. Numer-
monomial correlation in terms of Ra was obtained with ical and experimental results showed a recirculating zone in
bmin as reference length and a very good correlation was the central part of the channel for convergence angles
 2=5
carried out in terms of Rabmax bbmax
min
. It is suitable for greater than 5 and for several minimum channel spacing
2784 N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786

Fig. 12. Flow pattern pictures and stream function elds for qX 120 W=m2 , h 2 , and for several bmin values.

Fig. 13. Flow pattern pictures and stream function elds for h 5 and for several qX and bmin values: (ad) qX 30 W=m2 ; (eh) qX 120 W=m2 .
N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786 2785

Fig. 14. Flow pattern pictures and stream function elds for h 10 and for several qX and bmin values: (ad) qX 30 W=m2 ; (eh) qX 120 W=m2 .

Fig. 15. Temperature elds for h = 5 and for several qX and bmin values: Fig. 16. Temperature elds for h = 10 and for several qX and bmin values:
(a, b) qX 30 W=m2 ; (c, d) qX 120 W=m2 . (a, b) qX 30 W=m2 ; (c, d) qX 120 W=m2 .
2786 N. Bianco et al. / International Journal of Heat and Mass Transfer 50 (2007) 27722786

values according with [9,10,12]. These values increased with temperature by numerical investigation, Int. J. Heat Fluid Flow 25
the convergence angle. The vortexes moved up when the (2004) 671682.
[8] A. Bejan, A.K. da Silva, S. Lorente, Maximal heat transfer density in
heat ux or the convergence angle increased. The distance vertical morphing channels with natural convection, Numer. Heat
from the inlet at which the temperature elds could be con- Transfer Part A 45 (2004) 135152.
sidered fully developed increased with the ohmic heat ux, [9] N. Bianco, S. Nardini, Numerical Analysis of natural convection in
but it did not depend on the convergence angle for the two air in a vertical convergent channel with uniformly heated conductive
considered values. The recirculating zone induced a relative walls, Int. Commun. Heat Mass Transfer 32 (2005) 758769.
[10] N. Bianco, L. Langellotto, O. Manca, V. Naso, Numerical analysis of
maximum of the temperature on the channel centerline. radiative eects on natural convection in vertical convergent and
symmetrically heated channels, Numer. Heat Transfer Part A 49
Acknowledgements (2006) 369391.
[11] T.M. Huang, C. Gau, W. Aung, Mixed convection ow and heat
This research is supported by Seconda Universita degli transfer in a heated vertical convergent channel, Int. J. Heat Mass
Transfer 38 (1995) 24452456.
Studi di Napoli and by Ministero dellIstruzione dellUni- [12] F. Marcondes, V. de Souza Melo, J.M. Gurgel, Numerical analysis of
versita e della Ricerca with the 2005 PRIN Grant research natural convection in parallel, convergent, and divergent open-ended
program. channels, Int. J. Numer. Meth. Heat Fluid Flow 16 (3) (2006) 304
323.
[13] N. Bianco, L. Langellotto, O. Manca, S. Nardini, Thermal design and
References optimization of vertical convergent channels in natural convection,
Appl. Therm. Eng. 26 (2006) 170177.
[1] G.P. Peterson, A. Ortega, Thermal control of electronic equipment [14] N. Bianco, B. Morrone, S. Nardini, V. Naso, Air Natural Convection
cooling, in: J.P. Hartnet, T.F. Irvine (Eds.), Advances in Heat between Inclined Parallel Plates with Uniform Heat Flux at the Walls,
Transfer, Academic Press, New York, 1990, pp. 181314. Int. J. Heat Technol. 18 (2000) 2346.
[2] O. Manca, B. Morrone, S. Nardini, V. Naso, Natural convection in [15] G.A. Ledezma, A. Bejan, Optimal geometric arrangement of
open channels, in: B. Sunden, G. Comini (Eds.), Computational staggered vertical plates in natural convection, ASME J. Heat
Analysis of Convection Heat Transfer, WIT Press, Southampton, Transfer 119 (1997) 700708.
2000, pp. 235278. [16] K.D. Kihm, J.H. Kim, L. Fletcher, Investigation of natural convec-
[3] E.M. Sparrow, R. Ruiz, L.F.A. Azevedo, Experimental and numer- tion heat transfer in vertical converging channel ows using a
ical investigation of natural convection in convergent vertical specklegram technique, ASME J. Heat Transfer 115 (1993) 140148.
channels, Int. J. Heat Mass Transfer 31 (1988) 907915. [17] B.W. Webb, D.P. Hill, High Rayleigh number laminar natural
[4] E.M. Sparrow, R. Ruiz, Experiments on natural convection in convection in an asymmetrical heated vertical channel, ASME J. Heat
divergent vertical channels and correlation of divergent, convergent, Transfer 111 (1989) 649656.
and parallel-channel Nusselt numbers, Int. J. Heat Mass Transfer 31 [18] R.J. Moat, Describing the uncertainties in experimental results, Exp.
(1988) 21972205. Therm. Fluid Sci. 1 (1988) 317.
[5] S.A. Said, Investigation of natural convection in convergent vertical [19] Andreozzi, O. Manca, Thermal and uid dynamic behaviour of
channels, Int. J. Energy Res. 20 (1996) 559567. symmetrically heated vertical channels with auxiliary plate, Int. J.
[6] J.S. Shalash, J.D. Tarasuk, D. Naylor, Experimental and numerical Heat Fluid Flow 22 (2001) 424432.
studies of natural convection heat transfer in vertical converging [20] Fluent Incorporated, Fluent 6.1, User Manual, 2003.
channel ows, in: M. Giot, F. Mayinger, G.P. Celata (Eds.) [21] A. Bejan, Shape and Structure from Engineering to Nature,
Proceedings of Fourth Experimental Heat Transfer, Fluid Mechanics Cambridge University Press, New York (NY), 2000.
and Thermodynamics, Brussels, 1997, pp. 21672174. [22] A.K. da Silva, A. Bejan, Constructal multi-scale structure for
[7] A.S. Kaiser, B. Zamora, A. Viedma, Correlations for Nusselt number maximal heat transfer density in natural convection, Int. J. Heat
in natural convection in vertical converging channels at uniform Fluid Flow 26 (2005) 3444.