Vous êtes sur la page 1sur 11

D.

Roeleveld1
Department of Mechanical &
Industrial Engineering,
Ryerson University,
350 Victoria Street,
Toronto, ON M5B 2K3, Canada
e-mail: droeleve@ryerson.ca

D. Naylor
Department of Mechanical &
Industrial Engineering,
Ryerson University,
350 Victoria Street,
Toronto, ON M5B 2K3, Canada

W. H. Leong
Department of Mechanical &
Industrial Engineering,
Ryerson University,
350 Victoria Street,
Toronto, ON M5B 2K3, Canada

Free Convection in
Asymmetrically Heated
Vertical Channels
With Opposing Buoyancy
Forces
Laser interferometry and flow visualization were used to study free convective heat transfer inside a vertical channel. Most studies in the literature have investigated buoyancy
forces in a single direction. The study presented here investigated opposing buoyancy
forces, where one wall is warmer than the ambient and the other wall is cooler than the
ambient. An experimental model of an isothermally, asymmetrically heated vertical channel was constructed. Interferometry provided temperature field visualization and flow visualization was used to obtain the streamlines. Experiments were carried out over a range
of aspect ratios between 8.8 and 26.4, using temperature ratios of 0, 0.5, and 0.75.
These conditions provide a modified Rayleigh number range of approximately 5 to 1100.
In addition, the measured local and average Nusselt number data were compared to numerical solutions obtained using ANSYS FLUENT. Air was the fluid of interest. So the Prandtl
number was fixed at 0.71. Numerical solutions were obtained for modified Rayleigh numbers ranging from the laminar fully developed flow regime to the turbulent isolated
boundary layer regime. A semi-empirical correlation of the average Nusselt number was
developed based on the experimental data. [DOI: 10.1115/1.4026218]
Keywords: interferometry, flow visualization, numerical modeling, natural convection,
vertical channel

Introduction

Free convection in a heated vertical channel is a classical heat


transfer problem that has been studied extensively in the literature.
There are many applications of this problem, such as electronics
cooling, simulation of flow in nuclear reactors, and fenestration
systems (i.e., windows with blinds). Many of these studies have
investigated buoyancy driven flow in a single direction inside an
asymmetrically heated vertical channel. The current study investigates opposing buoyancy driven flow, where the fluid tends to
flow in opposite directions inside the vertical channel.
A schematic diagram of the vertical channel geometry is shown
in Fig. 1. An open-ended vertical channel is created by two isothermal channel walls of height L separated by a channel spacing
b. The aspect ratio of the channel is defined as A L/b. The cold
wall has a temperature TC and the hot wall has a temperature TH.
The cold wall temperature is set below the ambient temperature
T1 and the hot wall is set above the ambient temperature. Aung
[1] defined a temperature ratio as
RT

TC  T1
TH  T1

(1)

Most existing studies of free convection inside a heated vertical


channel have investigated temperature ratios in the range of
0  RT  1 (i.e., TC, TH > T1). This study will investigate opposing buoyancy forces, where the temperature ratio is in the range of
1 < RT  0 (i.e., TH > T1 and TC < T1).

1
Corresponding author.
Contributed by the Heat Transfer Division of ASME for publication in the
JOURNAL OF HEAT TRANSFER. Manuscript received September 11, 2013; final
manuscript received November 28, 2013; published online March 7, 2014. Assoc.
Editor: Zhixiong Guo.

Journal of Heat Transfer

There have been numerous studies on free convection in isothermally heated vertical channels. Elenbaas [2] was one of the
first to study heat flow in a symmetrically (i.e., TH TC, RT 1),
isothermally heated vertical channel. Using two square plates separated by various channel spacings, he was able to obtain experimental data for a wide range of modified Rayleigh numbers.
Some analytical work determined that the Nusselt number
approaches two asymptotes at the upper and lower modified Rayleigh numbers. An overall channel average Nusselt number correlation was developed using the analytical work and experimental
data.
Aung et al. [3] studied the conditions of a uniform heat flux and
a uniform wall temperature in an asymmetrically heated vertical
channel. Numerical solutions were obtained over a wide range of
modified Rayleigh numbers and some experimental work was performed to verify the results. They showed that, for uniform wall
temperatures, a nearly universal curve can be used to relate the
Nusselt numbers and the modified Rayleigh numbers for a wide
range of temperature ratios. This is the case if the Nusselt number
and Rayleigh number are defined by using the appropriate characteristic temperature difference. This characteristic temperature
difference is defined as
DT

TH TC
 T1
2

(2)

The modified Rayleigh number based on this temperature difference is


 
gbDT q2 b3 b
Rab=L
(3)
Pr
L
l2
where g is gravity, b is the fluid thermal expansion coefficient, q
is the fluid density, l is the fluid dynamic viscosity, and Pr is the

C 2014 by ASME
Copyright V

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

JUNE 2014, Vol. 136 / 062502-1

asymptotes at low and high modified Rayleigh numbers to develop correlations. They developed correlations for both isothermal and isoflux conditions in symmetrically and asymmetrically
heated vertical channels. The new correlations were developed to
fit various experimental and numerical data. A review of all the
correlations was performed by Raithby and Hollands [5]. It was
determined that the best correlation for the overall channel average Nusselt number for isothermally heated channel walls is
2
0
!1:9 
1=1:9

1:9
@0:618Rab=L1=4
Nu 4 Nufd

Fig. 1 Schematic of the problem geometry

Prandtl number. Note that the absolute value of the characteristic


temperature difference is taken because in the context of this paper, the modified Rayleigh number is always positive. The overall
channel average Nusselt number is defined as
Nu

qH qC b
2kDT

(4)

where qH and qC are the average convective heat fluxes at the hot
and cold walls in the vertical channel and k is the fluid thermal
conductivity. The hot wall heat flux is always positive and the
cold wall heat flux is usually negative in the negative temperature
ratio cases. All air properties are evaluated at the film temperature,
unless otherwise noted. The film temperature is defined as
Tf

TH TC =2 T1
2

(5)

Fully developed flow in an asymmetrically heated vertical


channel was studied analytically by Aung [1]. It was found that at
low modified Rayleigh numbers (Ra(b/L) ! 0), the asymptote of
the Nusselt number varied depending on the temperature ratio.
The average Nusselt number at the fully developed limit is
Nufd

4R2T 7RT 4
901 RT 2

Rab=L

(6)

Various channel heating configurations were studied by BarCohen and Rohsenow [4] using analytical expressions for the

(7)

where Nufd is determined from Eq. (6). There have been many
other studies on asymmetrically, isothermally heated vertical
channels with positive temperature ratios [610]. The existing
correlations from the literature cannot accurately predict the heat
transfer of the negative temperature ratio cases.
There are three studies in the literature that investigate opposing buoyancy forces in free convection of an antisymmetrically
(i.e., RT 1) heated vertical channel. The antisymmetrical case
is where the temperature difference between the hot wall and the
ambient is the same as the temperature difference between the ambient and the cold wall (i.e., TH T1 T1 TC). Habib et al.
[11] studied turbulent flow in an antisymmetrically (RT 1)
heated vertical channel with a channel aspect ratio of 3.125. The
Rayleigh number was 2.0  106, with the hot wall 10  C above
ambient and the cold wall 10  C below ambient. (The Rayleigh
number was defined based on the channel height and the temperature difference between the hot wall and the ambient temperature.)
Velocity profiles of the flow were determined using a laser
Doppler anemometer. The results showed a large vortex flow,
with the air flowing up the hot wall and down the cold wall similar
to flow inside a tall vertical cavity.
Ayinde et al. [12] also investigated turbulent flow in an antisymmetrically heated vertical channel. Two temperature differences of TH TC 15  C and 30  C were used between the two
channel walls. Two aspect ratios of 6.25 and 12.5 and two
Rayleigh numbers of 1.0  108 and 2.0  108 were studied. (The
Rayleigh number was defined based on the channel height and the
temperature difference between the hot wall and the ambient.) A
particle image velocimeter was used to determine velocity profiles
and a correlation for dimensionless flow rate inside the channel
was developed. The results indicated that the flows entering at the
top and bottom of the channel were mixed with the recirculated
flow inside the channel before exiting the other side of the channel. The results also showed that the flow pattern inside the channel was similar to a sealed tall enclosure. In these two
experimental studies, velocity field measurements were made for
the antisymmetrical case, in relatively low aspect ratio channels
(A < 13). In contrast, the present work investigates the convective
heat transfer rates in higher aspect ratio channels (typically
A > 13) at lower Rayleigh numbers and over a wider range of negative temperature ratios.
The antisymmetrical case (RT 1) is a special case that has
been recently investigated by Roeleveld et al. [13]. Similar to the
current study, flow visualization and laser interferometry were
used to study free convection in an open-ended vertical channel.
The heat transfer rates were also determined using laser interferometry and a numerical model was developed to solve over a
wide range of Rayleigh numbers. The Rayleigh number was
defined based on the channel spacing and the temperature difference between the hot and cold walls. The special nature of the
RT 1 case can be illustrated by noting that the modified Rayleigh and Nusselt numbers will always be zero using the definitions in the current paper, Eqs. (3) and (4). The flow- and
temperature-field were found to have similarities to that of a tall
sealed enclosure. It was determined that the average convective
heat transfer of the antisymmetrical case can be approximated

062502-2 / Vol. 136, JUNE 2014

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

Transactions of the ASME

approach, the entire range of possible cases for free convection


inside a heated vertical channel can be reduced to 1  RT  1
and Eq. (8) can be used if the temperature ratio falls outside this
range.
Another concern in this paper is that when studying a case that
is outside the 1  RT  1 range, the characteristic temperature
difference can become negative. Using the same example from
above, the characteristic temperature difference is DT 15  C.
The traditional definition of modified Rayleigh number based on
the surface to ambient temperature difference gives a negative
value. In order to circumvent this problem, the absolute value of
the characteristic temperature difference is, therefore, used in Eq.
(3). It should be noted that the characteristic temperature difference can be negative in the Nusselt number definition in Eq. (4),
so absolute values are not required in this equation.
2 Experimental Apparatus and Methodology
Fig. 2 Schematic of two equivalent vertical channel cases with
different temperature ratios; one with negative buoyancy forces
and one with positive buoyancy forces relative to the gravity
vector

using an existing correlation from the literature for a tall sealed


enclosure.
In the current study, an experimental model was constructed to
be used for laser interferometry and flow visualization. Flow visualization was used to determine the streamlines and laser interferometry was used to determine the isotherms. Experimental data
were obtained for three different temperature ratios (RT 0, 0.5
and 0.75) at four different aspect ratios (A 26.4, 17.6, 13.2,
and 8.8). These conditions provide a modified Rayleigh number
range of 4.7 < Ra(b/L) < 1084. The temperature field data were
then analyzed to determine the local and average Nusselt numbers. General purpose computational fluid dynamics (CFD) software ANSYS FLUENT [14] was used to develop a numerical model of
free convection inside an asymmetrically, isothermally heated
vertical channel. Both laminar and turbulent numerical predictions
were validated against the experimental data. Additional numerical solutions were obtained for RT 0.5 and 0.75 over a wide
range of modified Rayleigh numbers (0.1  Ra(b/L)  104).
In the literature, most of the studies on free convection inside a
heated vertical channel with buoyancy forces in a single direction
have examined cases where these buoyancy forces are in the opposite direction of the gravity vector (i.e., TH, TC > T1). These
studies can also be applied to cases where the buoyancy forces are
in the same direction of the gravity vector (i.e., TH, TC < T1), but
there is an inherent problem when using these correlations for
negative temperatures relative to the ambient. The problem lies in
calculating the temperature ratio and the modified Rayleigh number. For example, if the cold wall is 20  C, the hot wall is
10  C, and the ambient is 0  C, then the temperature ratio is
RT 2. This temperature ratio is outside the range of 0  RT  1
of most existing studies in the literature. This can be overcome by
understanding that every case of free convection in an open-ended
vertical channel has an equivalent case with the buoyancy forces
in the opposite direction. Figure 2 shows a schematic of two
equivalent cases, one with negative buoyancy forces (RT 2) and
the other with positive buoyancy forces (RT 0.5) relative to the
gravity vector. This figure shows that a case with RT 2 is equivalent to a case with RT 0.5 assuming the air properties are constant. So if the temperature ratio is outside the range of existing
studies (0  RT  1), it can be modified by
RT RT 1

(8)

to convert into an equivalent temperature ratio inside the range of


existing studies. This modified temperature ratio can be used with
the existing definitions and correlations. Using this same

2.1 Apparatus. A vertical channel was constructed to be used


with a Mach-Zehnder Interferometer (MZI) and flow visualization. Two aluminum plates were mounted vertically, with an adjustable channel spacing in order to study various aspect ratios.
The aspect ratio was adjusted in order to study various modified
Rayleigh numbers. The aluminum plates had a width of
W 355 mm and a length of L 264 mm. The plates were
roughly 38 mm thick, with beveled edges filled with polystyrene
at the top and bottom of the channel. Two constant temperature
water baths controlled the temperatures of the two aluminum
plates, by running water from the baths through grooves machined
into the back of each plate. One wall was cooled and the other
wall was heated creating the opposing buoyancy forces between
the plates. The cold wall was typically cooled 7.5  C below the
ambient room temperature except in the RT 0 case where
TC T1. This temperature was set such that it would not fall
below the dew point of the ambient air and create condensation on
the surface of the cold wall. The hot wall was typically heated 10
to 15  C above the ambient temperature depending on the temperature ratio being studied. The temperature differences between the
two channel walls were between 15 and 22.5  C. These temperature differences were used to obtain sufficient interference fringes
in the output of the interferometer. The thermocouples and thermopiles were Type T constructed out of copper and constantan
wire with special limits of error. The thermopiles were made with
12 junctions, 6 in each surface of interest. One thermopile monitored the temperature difference between the cold and hot walls of
the vertical channel and another monitored the temperature difference between the hot wall and the ambient. Six thermocouples
were used as a reference temperature for the two thermopiles in
the cold wall. The thermocouples and thermopiles were uniformly
spaced and embedded inside the channel walls to within approximately 1.6 mm (1/16 in.) of the surface. The thermocouples were
calibrated with thermometers traceable to national standards and
the thermopiles were checked against the NIST standard tables.
The plates were measured to be isothermal to within 0.2  C. In
order to prevent air entrainment, optical windows or acrylic panels
were mounted to the sides of the experimental model. The experimental model was placed in a smoke room during the experiments where no drafts or outside ventilation could disturb the flow
pattern inside the vertical channel.
2.2 Flow Visualization. Flow visualization was conducted to
observe the streamlines between the two vertical channel walls. A
plane laser sheet produced by a cylindrical lens introduced from
the top of the vertical channel was used to illuminate the smoke in
a cross section of the model. The smoke was sulfuric acid aerosol,
generated using a Drager tube, which was blown through hoses to
the experimental model from outside the smoke room. The smoke
was introduced slowly into the top and bottom of the vertical
channel and allowed to settle for a few seconds before an image
was captured. The flow was observed to be steady for the two

Journal of Heat Transfer

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

JUNE 2014, Vol. 136 / 062502-3

cases that were investigated. So a still image was taken with a


DSLR camera to show the streamlines inside the channel.

from the thermopiles and thermocouple readings. The overall


channel average Nusselt number was calculated from
Nu

2.3 Laser Interferometry. A 200 mm diameter beam MachZehnder Interferometer was used to obtain temperature field
measurements. A 15 mW helium-neon laser was used in a standard MZI as described in Goldstein [15]. The interferometer has
two settings: infinite fringe mode and finite fringe mode. The infinite fringe setting was used for temperature field visualization,
because a line of constant fringe shift is an isotherm in this mode.
The finite fringe mode was used for the heat transfer measurements. For steady cases, the interferogram was photographed with
a high resolution (39 megapixels) still image camera and for the
unsteady cases, a high speed digital movie camera was used.
The local Nusselt numbers on each channel wall were determined using the gradient measurement technique as described by
Poulad et al. [16]. The temperature gradient at the cold surface
was calculated using


Rk0 TS2 @e 
@T 

(9)
@x x0 WPG @xx0
where R is the gas constant, k0 is the wavelength of the laser light
in a vacuum, Ts is the absolute surface temperature (in this case,
Ts TC), P is the absolute pressure and G is the Gladstone-Dale
constant. (It should be noted that R 287 J/kg K for air,
k0 632.8 nm for a helium-neon laser, and G 0.226  103 m3/
kg for a helium-neon laser in air.) The fringe gradient normal to
the measurement surface, @e=@xjx0 , was extracted from the finite
fringe interferograms using a custom MATLAB [17] image processing code developed by Poulad et al. [16]. A scan of pixel intensity was taken along a horizontal line at the y-location of interest.
A nonlinear regression technique similar to Slepicka and Cha [18]
was used to unwrap the phase from the pixel intensity data. The
pixel intensity near the surface was expressed as
Ix Iavg F sinDx  x1 /

(10)

where Iavg is the average pixel intensity, F is the amplitude, D is


the amplitude of spatial fringe intensity, / is the phase shift, and
x1 is the location of the first pixel in the scan. Equation (10) is fit
to the extracted pixel intensity data by adjusting the four constants
iteratively until the sum squared error between the data and I(x)
was minimized. The fringe shift gradient at the surface was calculated from

@e 
D
(11)

@xx0 2p
For the hot surface at x b, the temperature and fringe gradients
were similarly calculated using Eqs. (9) and (11), respectively.
Using the temperature gradient, the local heat fluxes on the cold
and hot walls of the channel were calculated using


@T 
@T 
and qy;H ks 
(12)
qy;C ks 
@x x0
@x xb
where ks is the thermal conductivity of the air at the surface temperature. The hot wall heat flux is always positive and the cold
wall heat flux is typically negative in this study. The local Nusselt
number was defined as
Nuy;C

qy;C b
qy;H b
and Nuy;H
DTk
DTk

(13)

where k is the fluid thermal conductivity calculated at the film


temperature (Eq. (5)). The temperature difference was calculated

1
2L

 L

Nuy;C dy


Nuy;H dy

(14)

where the integrals were evaluated using the trapezoidal rule.


When studying the unsteady cases, the temperature field was
three-dimensional and time dependent. The output of the MZI is
two-dimensional. So the temperature field was beam-averaged in
the z-direction. A sequence of interferograms was recorded at 30
fps for 20 s using a high-speed digital movie camera. The timeaveraged heat flux became nearly stationary after 20 s. Each interferogram was analyzed using the same custom MATLAB image
processing code developed by Poulad et al. [16] in order to obtain
the instantaneous fringe gradient. The instantaneous heat fluxes of
both the cold and hot walls were determined and then integrated
in order to obtain the time-averaged heat fluxes. The timeaveraged heat fluxes were then substituted into Eqs. (13) and (14)
in order to determine the local and average Nusselt numbers.
As with any experiment, there is uncertainty in the results. A
detailed error analysis was conducted based on the Kline and
McClintock [19] method. Consider an experimental result Y calculated from n independent variables a1, a2,,an. Each variable
has a total uncertainty of da1, da2,,dan. If the uncertainty in
each variable were given the same odds, then the uncertainty of
the result dY at these odds is
v
u n 
2
uX @Y
(15)
dai
dY t
@ai
i1
Each variable has two components of uncertainty: bias error and
random error. The two types of uncertainty were combined using
a root sum square method with a 95% confidence level. Table 1
shows the typical values of the variables and their total uncertainties. The uncertainty in the air property data was estimated from
the property data scatter plots presented by Touloukian et al. [20],
Touloukian and Makita [21], and Touloukian et al. [22]. The
uncertainty in the fringe gradient is the largest source of error in
these experiments. This is due to the regression algorithm used in
the custom MATLAB image processing code and the optical
imperfections in the MZI. The estimated uncertainty in the modified Rayleigh number is 65%. The error in the local Nusselt numbers is estimated at 612%. The integration process averages out
some of the random error in the local Nusselt number data. For
this reason, the overall channel average Nusselt number is more
accurate than the individual local Nusselt number data and the
estimated uncertainty is 69%. Further information on the uncertainty analysis is given in Roeleveld [23].

Table 1 Measured
uncertainties
Parameter
Channel height
Channel width
Channel spacing
Specific heat
Dynamic viscosity
Thermal conductivity
Pressure
Surface temperature
Temperature difference
Fringe gradient

062502-4 / Vol. 136, JUNE 2014

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

variables

Symbol

and

their

estimated

Measured value

L
264.2 mm
W
355.1 mm
b
10 mm 30 mm
cp
l
k, ks
P
747.6 mm Hg
288.0 K 310.0 K
TS
1.25 K 7.5 K
DT
@e
@xx0;xb 200 m1 1500 m1

total

Total uncertainty
60.3 mm
63.2 mm
60.1 mm
60.8%
60.5%
61%
60.2 mm Hg
60.35 K
60.10 K
610.2%

Transactions of the ASME

determined to be sufficient for the current simulations. Roeleveld


[23] has further details on the grid sensitivity study.
Aung [1] solved fully developed flow solutions analytically for
laminar free convection inside an asymmetrically heated vertical
channel. This analytical solution presents equations for both the
dimensionless velocity and temperature profiles of the fully developed flow. Figure 4 shows that the dimensionless velocity profiles
of the analytical solution and the numerical model at Ra(b/
L) 0.5 and y/L 0.5 compare favorably. Using this analytical
solution and setting dV/dx 0 at x 0, it was determined that
reverse flow starts at approximately RT 0.5 in the fully developed flow regime. At such low modified Rayleigh numbers, the
heat transfers by pure conduction, so the temperature profile is a
linear function of distance between the two channel walls. The analytical work was developed for asymmetrically heated vertical
channels with positive temperature ratios (0  RT  1), but this
graph shows that this analytical solution also applies for negative
temperature ratios (1 < RT < 0). Further numerical validation
against the current experimental data will be presented in Sec. 4.

4
Fig. 3 Boundary conditions and computational domain

Numerical Solution

A numerical model was developed to determine the convective


heat transfer. The flow was assumed to be two-dimensional and
incompressible. At first a steady laminar solution was sought. But
convergence could not be obtained over the entire modified Rayleigh number range of interest. For the higher modified Rayleigh
numbers, a standard k-e turbulence model was used with enhanced
wall functions. Walsh and Leong [24] have shown that the
enhanced wall functions give better results for free convection on
a vertical wall. The standard governing equations used are given
in Versteeg and Malalasekera [25]. A control-volume formulation
with a second-order upwind scheme for evaluation of the convective terms, the PRESTO option [26], and the SIMPLEC algorithm
[27] were used in the numerical model. The fluid properties were
assumed to be constant and viscous dissipation was neglected in
the energy equation. The Boussinesq approximation was used to
account for variation in fluid density due to temperature variation.
For this study, the Prandtl number was fixed at 0.71, since the
fluid of interest was air.
The domain of the numerical model and the boundary conditions are shown in Fig. 3. No slip and impermeability conditions
were applied to the walls. An adiabatic boundary condition was
used on the horizontal surfaces near the inlet and outlet of the
channel. This is the typical treatment from the literature (Naylor
et al. [8]). Above and below the channel, two pre-entry plenums
were added to set the outflow conditions with higher accuracy
because the air temperature will approach ambient temperature far
away from the channel. A pressure outlet condition was applied to
the semicircular inlet and outlet boundaries, where the fluid velocity was set normal to the boundary and the pressure defect was set
to zero. A pressure outlet condition was needed in order to allow
air to enter and exit the domain at either end of the channel due to
the opposing buoyancy forces.
A grid sensitivity study was conducted to ensure that the numerical solution was grid independent. Both grid density and farfield boundary conditions were tested. Using the same size domain, three different grid densities were tested. A grid of approximately 50,000 nodes was determined to be sufficient. The average
Nusselt numbers are estimated to be grid independent to better
than 0.7% using the Richardson extrapolation [28]. By adjusting
the radius B shown in Fig. 3, the sensitivity of the results to the
far field boundary location was tested. The radius B 5b was

Experimental Results

4.1 Flow Visualization. Flow visualization was conducted


for two different temperature ratios of RT 0.5 and RT 0.75
at an aspect ratio of A 17.6. Figure 5 shows the streamlines for
RT 0.5 at A 17.6 with Ra(b/L) 67.5. A sketch is included
in Fig. 5(b) to show the observed streamlines. The flow was
observed to be steady. In this case, on the hot wall side of the
channel, air flowed in the bottom of the channel and out the top of
the channel. On the cold wall side of the channel, air entered from
the top and flowed to near the bottom of the channel. A separation
point was located at approximately y/L 0.07 where the air flowing down the cold wall reversed direction with the air flowing up
the hot wall inside the channel. Figure 5(a) shows the separation
point with a line of smoke, which separates the air flowing up the
hot wall and the air flowing up the channel from the cold wall after reversing direction. There was also some smoke separating the
opposing flow from the cold wall in the upper half of the channel
where there was slow air flow. The flow pattern in this case was
similar to what was observed by Sparrow et al. [29] when studying a vertical channel with the hot wall isothermally heated and
the cold wall unheated (RT 0). They observed that in a channel
with an aspect ratio of 15.2 at Ra(b/L) 5270, some ambient air
flowed down about 25% of the cold wall and then re-circulated
with the air flowing up the hot wall.
Figure 5(c) shows the streamlines from the numerical model. It
can be seen that the experimental and numerically predicted
streamline patterns agree qualitatively, with the separation point
located in a similar location. An area of interest is at the top of the
channel, where the air is exiting the channel. The experimental

Fig. 4 Comparison of dimensionless velocity profiles at various temperature ratios for fully developed flow at y/L 5 0.5 and
Ra(b/L) 5 0.5 obtained from the numerical model and the analytical solution by Aung [1]

Journal of Heat Transfer

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

JUNE 2014, Vol. 136 / 062502-5

Fig. 5 (a) Flow visualization, (b) sketch of the observed flow


patterns, and (c) numerical solution streamlines for RT 5 20.5,
A 5 17.6 and Ra(b/L) 5 67.5

visualization using smoke shows that the plume bends abruptly


away from the centerline of the channel towards the hot wall side
as it exits the channel. The numerical results show the plume exiting straight upwards, which differs from the flow visualization.
Figure 6 shows the case of RT 0.75 at A 17.6, with Ra(b/
L) 23.1. Figure 6(b) shows a sketch of the observed flow pattern. Again the flow was observed to be steady. Similar to the previous case, air flowed up the hot wall side of the channel and
down the cold wall side of the channel. But in this case, there was
a closed recirculation cell in the center of the channel. Near the
cold wall, the air flowed out the bottom of the channel and there
was no separation point. It should also be noted that the cell was
off center from the vertical centerline of the channel due to the
upwards buoyancy force from the hot wall being stronger than the
downwards buoyancy force from the cold wall. As the temperature ratio was decreased, less air flowed out the top of the channel,
but more air flowed out the bottom of the channel due to the
increasing negative buoyancy force. The numerical streamlines
are shown in Fig. 6(c). Again, the experimental and numerically
predicted streamline patterns agree qualitatively. Also, the plume
bends towards the hot wall as it exits the channel, unlike the numerical prediction.
4.2 Laser Interferometry. A MZI was used to obtain interferograms for three different temperature ratios (RT 0, 0.5,

Fig. 6 (a) Flow visualization, (b) sketch of the observed flow


patterns, and (c) numerical solution streamlines for RT 5 20.75,
A 5 17.6 and Ra(b/L) 5 23.1

and 0.75) at four different aspect ratios (A 26.4, 17.6, 13.2,


and 8.8). This provided data over a modified Rayleigh number
range of 4.7 < Ra(b/L) < 1084. Figure 7 shows three infinite
fringe interferograms at a channel aspect ratio of A 17.6 at the
three different RT values. There are six horizontal pins that are
visible on each channel wall. These small pins were used for laser
alignment and image locating purposes and have no significant
impact on the convection. All of these three cases were observed
to be steady. At RT 0, a developing thermal boundary layer can
be seen on the hot wall over the full length of the channel. As RT
becomes negative, it can be seen that the boundary layer nature of
the temperature field quickly diminishes and the heat transfer in
the center region of the channel becomes increasingly conduction
dominated. This is evident from the more evenly spaced isotherms
in the interferograms. Similar to what was observed with flow

062502-6 / Vol. 136, JUNE 2014

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

Transactions of the ASME

Fig. 7 Infinite fringe interferograms at A 5 17.6 for different


temperature ratios

visualization, Fig. 7 shows that at the top of the channel, the


plume of warm air bends abruptly away from the centerline of the
channel.
Figure 8 shows two interferograms at RT 0.5 at two different aspect ratios. These two interferograms along with Fig. 7(b)
show the effect of modified Rayleigh number on the temperature
field. In Fig. 8(a), the modified Rayleigh number is low (near the
fully developed limit) and the temperature field appears to be
almost pure conduction over much of the channel length. In Fig.
8(c), the modified Rayleigh number is much higher and thermal
boundary layers can be seen on both walls. It can be seen that the
fringes are wavy in the center region near the top of the channel.
In fact, the temperature field was observed to be unsteady at this
location for this case. It is interesting to note that compared to a
channel with unidirectional buoyancy, the flow in a channel with
strongly opposing buoyancy forces becomes unsteady at a much
lower modified Rayleigh number. The experimental results of
Aung et al. [3] and Sparrow et al. [29] show that a vertical channel
with positive temperature ratios (RT > 0) remains steady and laminar up to Ra(b/L) 104, whereas the current flows are unstable at
Ra(b/L) 214 for RT 0.5 and Ra(b/L) 76.6 for RT 0.75.
Figures 8(b) and 8(d) provide the ability to compare between
the numerically predicted isotherms and the interferograms for
these two cases. Overall, there is good agreement between the
steady laminar numerical isotherms and the interferograms. However, some disagreement can be seen at the top of the channel.
Once again, the plume of air leaving the channel bends towards
the hot wall side in the interferograms, which is not predicted

Fig. 8 Infinite fringe interferograms and numerically predicted


isotherms at RT 5 20.5 for different aspect ratios

numerically. The interferograms in Figs. 7 and 8 show that the


horizontal surfaces at the entrance and exit region of the channel
are not perfectly adiabatic, as they were assumed to be in the numerical model. A test was conducted to check if this difference in
the temperature boundary condition on the horizontal surfaces
was the cause of the poorly predicted outlet flows. Using a user
defined function in ANSYS FLUENT, numerical solutions were
obtained with a variable temperature boundary condition on these
surfaces for comparison. This variable temperature boundary condition was based on the measured surface temperature distribution
from the interferograms. When comparing the numerical results
with adiabatic and variable temperature boundary conditions, the
plume still exited the channel straight vertically and the difference
between the heat transfer rates was about 1%. Based on this testing, it appears that other issues associated with the imperfect nature of the outflow conditions on the semicircular boundaries are
producing the discrepancy.
Figure 9 shows a comparison between the experimentally measured and numerically predicted local Nusselt number distributions
for RT 0.5 with A 26.4 and Ra(b/L) 12.3. These results
correspond to the interferogram in Fig. 8(a). For comparison, the
Nusselt numbers corresponding to one-dimensional pure conduction between the walls were added to this graph. Figure 9 shows
that the local heat transfer rate corresponds to essentially pure
conduction for the top three quarters of the channel. It can also be
seen that the laminar CFD solution shows good agreement with
the experimental data except for on the hot wall near the bottom
of the channel. For the bottom 25% of the hot wall the CFD prediction was 15% higher than the experimental data. This slight
difference is due to the nonadiabatic horizontal walls in the

Journal of Heat Transfer

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

JUNE 2014, Vol. 136 / 062502-7

Fig. 9 Graph of local Nusselt number versus nondimensional


vertical distance for RT 5 20.5, A 5 26.4, and Ra(b/L) 5 12.3

Fig. 11 Graph of local Nusselt number versus nondimensional


vertical distance for RT 5 20.5, A 5 8.8, and Ra(b/L) 5 1084

Table 2 Comparison of the overall channel average Nusselt


numbers from the experiment, numerical predictions and the
Raithby and Hollands [5] correlation for RT 5 0
Modified
Aspect Rayleigh Experimental
ratio
number
data
A
26.4
17.6

Fig. 10 Instantaneous and running time-averaged local heat


fluxes for RT 5 20.5 with A 5 8.8 at y/L 5 0.5

experimental model. Air was being preheated as it flowed into the


channel causing the measured heat flux to be slightly lower on the
hot wall near the entrance.
The flow was observed to be unsteady for RT 0.5 at A 8.8
and Ra(b/L) 1085, so an unsteady interferometric analysis was
performed with a high speed digital movie camera. Figure 10
shows typical instantaneous heat fluxes for this case. These measurements were made at y/L 0.5 on both the hot and cold walls
over a 20 s interval at a frame rate of 30 fps. The RMS fluctuation
is 10% or less for these cases and the peak to peak fluctuation is
between 20 and 30% about the mean. This graph also shows the
running time-averaged heat flux becomes nearly stationary after
20 s. The local Nusselt numbers for RT 0.5 at A 8.8 and
Ra(b/L) 1084 are shown in Fig. 11. Two CFD solutions, one
assuming laminar steady conditions and the other using a turbulence model, were added to the graph for comparison. It can be
seen that the turbulent CFD solution shows better agreement than
the steady laminar solution with the experimental data near the
top of the channel. Again, there was preheating of the air at the
bottom of the channel. An area of interest is near the top of the
channel, where both the hot and cold walls show some discrepancy with the numerical solutions. As previously discussed, this is
most likely due to the plume bending to the hot wall side of the
channel when it exits the top of the channel. This draws air
toward the hot wall as it leaves the top of the channel, increasing
the heat transfer, while drawing air away from the cold wall,
decreasing its heat transfer. Some of the error of the cold wall
could also be from some precooling of the air as it enters the top
of the channel.

Ra(b/L)
24.8
128

(Nu)EXP
0.82
1.93

Numerical
solution

Raithby and
Hollands [5]

(Nu)LAM Difference (Nu)COR Difference


0.79
3.7%
0.85
3.4%
1.99
3.1%
1.93
0%

Table 2 shows a comparison of the overall channel average


Nusselt numbers of the RT 0 case for two different aspect ratios.
The numerically predicted overall channel average Nusselt numbers are within 4% of the experimental data. Since the RT 0
case has been previously studied in the literature, the correlation
of Raithby and Hollands [5], Eq. (7), was also included in Table
2. The correlation is within about 3% of the experimental data.
The overall channel average Nusselt numbers for both the experimental data and numerical solution are compared in Table 3 for
RT 0.5 and RT 0.75. The numerical predictions presented
in this table were all solved with the steady laminar model. The
Nusselt numbers agree quite well for A 13.2, even though these
cases were observed to be unsteady experimentally. Overall, the
numerically predicted overall channel average Nusselt numbers
are within 7% of the experimental data. Table 4 shows a comparison between the results of the experimental data, the steady laminar numerical model and the steady turbulence model with the
enhanced wall functions at the higher modified Rayleigh numbers.
It should be noted that these three cases were all observed to be
unsteady experimentally. This table shows that at A 13.2, for
RT 0.5, the laminar and turbulent predictions give similar
results. But for RT 0.75, the turbulent solution shows better
agreement with the experimental data. At A 8.8, Table 4 shows
that the turbulence model has better agreement with the experimental data as the modified Rayleigh number was increased.

Numerical Results

The numerical model was used to conduct a parametric study


for the following range of parameters:
A 50;

RT  0:75;  0:5;

0:1 < Rab=L < 104 (16)

The overall channel average Nusselt numbers were determined


using the numerical solutions. Both laminar and turbulent CFD

062502-8 / Vol. 136, JUNE 2014

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

Transactions of the ASME

Table 3 Comparison of the overall channel average Nusselt numbers from the experiment and numerical predictions for RT 5 20.5
and 20.75
RT 0.5

RT 0.75

Aspect Ratio Modified Rayleigh number Experimental data Numerical solution Modified Rayleigh number Experimental data Numerical solution
A
26.4
17.6
13.2

Ra(b/L)
12.3
66.6
214

(Nu)EXP
0.67
2.19
3.51

(Nu)LAM Difference
0.70
3.6%
2.27
3.7%
3.53
0.6%

Ra(b/L)
4.7
22.9
76.6

(Nu)EXP
0.68
2.17
3.62

(Nu)LAM Difference
0.66
3.2%
2.27
4.6%
3.85
6.4%

Table 4 Comparison of the laminar and turbulent numerical models with the experimental data at aspect ratios of 13.2 and 8.8 for
RT 520.5 and 20.75
Temperature ratio
RT
0.75
0.5
0.5

Aspect ratio

Modified Rayleigh number

Experimental data

A
13.2
13.2
8.8

Ra(b/L)
76.6
214
1084

(Nu)EXP
3.62
3.51
5.42

Fig. 12 Overall channel average Nusselt number variation with


modified Rayleigh number for RT 5 20.5

models were used. Figures 12 and 13 show numerical solutions


for RT 0.5 and RT 0.75, respectively, at an aspect ratio of
50. Figure 12 shows that laminar results were obtained over the
range 0.1 < Ra(b/L) < 100 and turbulent results over the range
50 < Ra(b/L) < 104. Figure 13 shows laminar results over
0.1 < Ra(b/L) < 20 and turbulent results over 10 < Ra(b/L) < 104.
The air flowing down the cold wall mixes with the air flowing up
the hot wall in the center region of the channel. This shear layer
caused by the opposing buoyancy forces creates instability inside
the vertical channel which requires the use of a turbulence model
to obtain a solution at higher modified Rayleigh numbers. Based

(Nu)LAM
3.85
3.53
4.31

Turbulent CFD solution with EWF

Difference
6.4%
0.6%
20.5%

(Nu)TUR
3.66
3.48
5.69

Difference
1.1%
0.9%
5.0%

on the inability to obtain convergence of a steady laminar model,


it seems that instability inside the channel occurs at a much lower
modified Rayleigh number than when studying unidirectional
buoyancy forces. This is compared to other studies in the literature, where the flow remains steady and laminar to Ra 104 for
positive temperature ratios [3, 29].
Four experimental data points are shown in Fig. 12 and three
experimental data points are shown in Fig. 13 for comparison.
Both the laminar and turbulent solutions show good agreement
with the experimental data points. The average Nusselt number
for fully developed flow that was developed by Aung [1] given in
Eq. (6) was added to these graphs for comparison. This analytical
asymptote was developed for positive RT values. But as can be
seen in Figs. 12 and 13, the numerical solutions for negative RT
values also follow this asymptote. This asymptote will be used as
a starting point for developing a correlation for the average Nusselt number data.

Fig. 13 Overall channel average Nusselt number variation with


modified Rayleigh number for RT 5 20.75

Laminar CFD solution

Semi-Empirical Correlation

A semi-empirical correlation was developed using the experimental Nusselt number data and applying the correlation method
of Churchill and Usagi [30]. A new correlation was developed
because the existing correlations in the literature were unable to
accurately predict the heat transfer of the negative temperature ratio cases. It can be seen in Figs. 12 and 13 that the Nusselt number
approaches asymptotes at low and high modified Rayleigh numbers. As discussed Sec. 5, at low modified Rayleigh numbers the
data follows the fully developed flow Nusselt number asymptote
that was developed by Aung [1]. This asymptote is given in Eq.
(6). At high modified Rayleigh numbers, the Nusselt number follows the vertical isothermal flat plate asymptote. An expression
for the vertical isothermal flat plate Nusselt number asymptote
was determined to be

NuIP C

2
1 RT

n1

1 jRT jn RT Rab=Ln

(17)

This expression was obtained by combining the standard expression for the average Nusselt number for free convection from two
vertical isothermal plates into an overall Nusselt number for the
channel (see Appendix). These two asymptotes are then used with
the correlation method of Churchill and Usagi [30] as follows:

Journal of Heat Transfer

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

Nu

Nufd

m

NuIP m

(18)

JUNE 2014, Vol. 136 / 062502-9

G
I
Iavg
k
ks

Fig. 14 Comparison of the semi-empirical correlation for


RT 5 20.5 and RT 5 20.75 with the experimental and numerical
data

It was determined that C 0.311, n 0.249, and m 1.51.


These constants have been adjusted to minimize the sum-squared
percent error between the experimental data and the semiempirical correlation. Figure 14 shows the correlation and the experimental data plotted on the same graph. Since there was no experimental data available at high modified Rayleigh numbers,
some numerical predictions were used as can be seen in Fig. 14.
This was necessary in order to adjust the constants C and n properly for the upper asymptote. The maximum error between the
data and the correlation is 610% and the standard deviation is
64%. It should be noted that this correlation has been developed
to be applicable for temperature ratios in the range 1 < RT  1.
For positive temperature ratios, it is comparable to the correlation
of Raithby and Hollands [5]. Note that this correlation does not
apply for the antisymmetrical case RT 1. This special case has
been recently studied by Roeleveld et al. [13].

L
Nu
Nufd
NuIP
Nuy
P
Pr
q
qy
R
Ra(b/L)
RT
T
DT
W
u, v
V
x, y
x1

Gladstone Dale constant, m3/kg


pixel intensity
average pixel intensity
fluid thermal conductivity at film temperature, W/mK
fluid thermal conductivity at surface temperature, W/
mK
channel height, m
overall channel average Nusselt number
fully developed flow Nusselt number
vertical isothermal flat plate Nusselt number
local Nusselt number
pressure, Pa
Prandtl number
convective heat flux, W/m2
local convective heat flux, W/m2
gas constant, J/kgK
modified Rayleigh number
temperature ratio
temperature, K
characteristic temperature difference, K
channel width, m
fluid velocity in x, y-direction, m/s
dimensionless y-velocity, b2 vq Pr=LlRab=L
Cartesian coordinate system, m
location of first pixel, m

Greek Symbols
b
e
k0
l
q
/

fluid thermal expansion coefficient, K1


fringe shift
laser light wavelength in a vacuum, m
fluid dynamic viscosity, Ns/m2
fluid density, kg/m3
phase shift

Subscripts
7

Conclusions

Free convective heat transfer rates in an asymmetrically, isothermally heated vertical channel with opposing buoyancy forces were
determined using laser interferometry. Some flow visualization was
also conducted to determine the streamlines inside the channel. As
the temperature ratio was decreased, the opposing buoyancy forces
caused the flow to become unstable at lower modified Rayleigh
numbers. Both steady and turbulent numerical models were developed to obtain solutions for a wide range of modified Rayleigh
numbers at various temperature ratios. The opposing buoyancy
forces produced instability inside the channel and in general, a k-e
turbulence model with enhanced wall functions showed good
agreement for this type of flow at higher modified Rayleigh numbers. A correlation for the overall channel average Nusselt number
has been obtained based on the asymptotic behavior at high and
low values of modified Rayleigh number.

C
H
s
1

Appendix
An isolated flat plate asymptote was developed for use in the
semi-empirical correlation at high Rayleigh numbers. If the channel
is comprised of two isolated flat plates, the heat flux of each plate is
qH hH TH  T1 and qC hC TC  T1

(A1)

where the Nusselt numbers are of the form


NuH

Acknowledgment
This work was funded in part by the Canadian Solar Buildings
Research Network under the Strategic Network Grants Program
of the Natural Sciences and Engineering Research Council
of Canada.

cold wall
hot wall
surface
ambient

hH L
hC L
CRanL;jTH T1 j and NuC
CRanL;jTC T1 j
k
k
(A2)

Note that the channel height L is the characteristic length in Eq.


(A2). Since TH  T1 is always positive, the absolute value can be
dropped. This is substituted into Eq. (4):
hH TH  T1 hC TC  T1 b
2kDT





1
TH  T1
TC  T1 b
n
CRanL;jTC T1 j
CRaL;TH T1
2
L
DT
DT
(A3)

Nu

Nomenclature
A
B
b
D
F
g

aspect ratio
channel inlet and outlet domain size, m
channel spacing, m
amplitude of spatial fringe intensity
rate of change of phase shift, m1
gravity, m/s2

The Rayleigh numbers need to be converted to the common characteristic temperature difference, DT:

062502-10 / Vol. 136, JUNE 2014

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

Transactions of the ASME

RanL;TH T1 RanL;DT


RanL;DT




TH  T1
DT

n

jT C  T 1 j
DT

and RanL;jTC T1 j

n
(A4)

This Rayleigh number can then be converted to the modified Rayleigh number based on the characteristic length, b:
RanL;DT Ranb;DT

 n  4n
b
L
L
b

(A5)

Substituting these into Eq. (A3)



n 
n
n1 
b
TH  T1
b
Nu C Rab;DT
Rab;DT
L
L
DT

n 
 4n1
jTC  T1 j
TC  T1
L

b
DT
DT

(A6)

where
n1


n1 
TH  T1 n1
2TH  T1
2

1 RT
TH  T1 TC  T1
DT
and

 
 
n
2jTC  T1 j
jTC  T1 j n TC  T1

TH  T1 TC  T1
DT
DT
n1

 
2TC  T1
2

jRT jn RT
(A7)

1 RT
TH  T1 TC  T1


Substituting these results back into Eq. (A6) and rearranging


NuIP C A4n1

2
1 RT

n1

1 jRT jn RT Rab=Ln

(A8)

It was determined that the effect of aspect ratio on the upper asymptote is small (since n 0.25, (A)4n1 1), so the term (A)4n1
can be neglected. Therefore the upper asymptote becomes

NuIP C

2
1 RT

n1

1 jRT jn RT Rab=Ln

(17)

References
[1] Aung, W., 1972, Fully Developed Laminar Free Convection Between Vertical
Plates Heated Asymmetrically, Int. J. Heat Mass Transfer, 15(8), pp.
15771580.
[2] Elenbaas, W., 1942, Heat Dissipation of Parallel Plates by Free Convection,
Physica, 9, pp. 128.
[3] Aung, W., Fletcher, L. S., and Sernas, V., 1972, Developing Laminar Free
Convection Between Vertical Flat Plates With Asymmetric Heating, Int.
J. Heat Mass Transfer, 15(11), pp. 22932308.
[4] Bar-Cohen, A., and Rohsenow, W. M., 1984, Thermally Optimum Spacing of
Vertical, Natural Convection Cooled, Parallel Plates, ASME J. Heat Transfer,
106(1), pp. 116123.

[5] Raithby, G. D., and Hollands, K. G. T., 1998, Natural Convection, Handbook
of Heat Transfer, Rohsenow, W. M., Harnett, J. P., and Cho, Y. I., eds.,
McGraw-Hill, New York, pp. 4.324.35.
[6] Bodoia, J. R., and Osterle, J. F., 1963, The Development of Free Convection
Between Heated Vertical Plates, ASME J. Heat Transfer, 84(1), pp. 4044.
[7] Burch, T., Rhodes, T., and Acharya, S., 1985, Laminar Natural Convection
Between Finitely Conducting Vertical Plates, Int. J. Heat Mass Transfer,
28(6), pp. 11731186.
[8] Naylor, D., Floryan, J. M., and Tarasuk, J. D., 1991, Numerical Study of
Developing Free Convection Between Isothermal Vertical Plates, ASME
J. Heat Transfer, 113(3), pp. 620626.
[9] Kazeminejad, H., 2005, Laminar Free Convection in a Vertical Channel With
Asymmetrical Heating, Proceedings of the ASME Summer Heat Transfer
Conference, San Francisco, CA, Vol. 3, pp. 521530.
[10] Roeleveld, D., Naylor, D., and Oosthuizen, P. H., 2009, Empirical Correlations
for Free Convection in an Isothermal Asymmetrically Heated Vertical
Channel, Heat Transfer Eng., 30(3), pp. 189196.
[11] Habib, H. A., Said, S. A. M., Ahmed, S. A., and Asghar, A., 2002, Velocity
Characteristics of Turbulent Natural Convection in Symmetrically and Asymmetrically Heated Vertical Channels, Exp. Therm. Fluid Sci., 26(1), pp.
7787.
[12] Ayinde, T. F., Said, S. A. M., and Habib, M. A., 2008, Turbulent Natural Convection Flow in a Vertical Channel With Anti-Symmetric Heating, Heat Mass
Transfer, 44(10), pp. 12071216.
[13] Roeleveld, D., Naylor, D., and Leong, W. H., 2014, Free Convection in Antisymmetrically Heated Vertical Channels, J. Heat Transfer, 136(1), p. 012502.
[14] ANSYS, Inc., 2010, ANSYS FLUENT (Version 13.0).
[15] Goldstein, R. J., 1976, Optical Temperature Measurement, Measurement
Techniques in Heat Transfer, E. R. G. Eckert and R. J. Goldstein, eds., Technivision Services, Slough, UK, pp. 177228.
[16] Poulad, M. E., Naylor, D., and Oosthuizen, P. H., 2011, Measurement of
Time-Averaged Turbulent Free Convection in a Tall Enclosure Using Interferometry, ASME J. Heat Transfer, 133(4), p. 042501.
[17] The Mathworks, Inc., 2010, MATLAB (Version R2010a), Natick, MA.
[18] Slepicka, S. S., and Cha, S. S., 1995, Stabilized Nonlinear Regression for
Interferogram Analysis, Appl. Opt., 34(23), pp. 50395044.
[19] Kline, S. J., and McClintock, F. A., 1953, Describing Uncertainties in SingleSample Experiments, Mech. Eng., 75(1), pp. 38.
[20] Touloukian, Y. S., Liley, P. E., and Saxena, S. C., 1970, Thermal Conductivity: Nonmetallic Liquids and Gases, Thermophysical Properties of Matter,
Vol. 3, Plenum Publishing Corporation, New York.
[21] Touloukian, Y. S., and Makita, T., 1970, Specific Heat: Nonmetallic Liquids
and Gases, Thermophysical Properties of Matter, Vol. 6, Plenum Publishing
Corporation, New York.
[22] Touloukian, Y. S., Saxena, S. C., and Hestermans, P., 1975, Viscosity: Nonmetallic Liquids and Gases, Thermophysical Properties of Matter, Vol. 11,
Plenum Publishing Corporation, New York.
[23] Roeleveld, D., 2013, Experimental and Numerical Study of Free Convection
in a Vertical Channel With Opposing Buoyancy Forces, Ph.D. thesis, Ryerson
University, Toronto, Ontario.
[24] Walsh, P. C., and Leong, W. H., 2004, Effectiveness of Several Turbulence
Models in Natural Convection, Int. J. Numer. Methods Heat Fluid Flow, 14(5),
pp. 633648.
[25] Versteeg, H. K., and Malalasekera, W., 1995, An Introduction to Computational
Fluid Dynamics: The Finite Volume Method, Longman Group Ltd., Essex, UK.
[26] Patankar, S. V., 1980, Numerical Heat Transfer and Fluid Flow, Hemisphere
Publishing Corporation, New York.
[27] Van Doormaal, J. P., and Raithby, G. D., 1984, Enhancements of the SIMPLE
Method for Predicting Incompressible Fluid Flows, Numer. Heat Transfer,
7(2), pp. 147163.
[28] Celik, I. B., Ghia, U., Roache, P. J., Freitas, C. J., Coleman, H., and Raad, P. E.,
2008, Procedure for Estimation and Reporting of Uncertainty Due to Discretization in CFD Applications, ASME J. Fluids Eng. 130(7), p. 078001.
[29] Sparrow, E. M., Chrysler, G. M., and Azevedo, L. F., 1984, Observed Flow
Reversals and Measured-Predicted Nusselt Numbers for Natural Convection in
an One-Sided Heated Vertical Channel, ASME J. Heat Transfer, 106(2), pp.
325332.
[30] Churchill, S. W., and Usagi, R., 1972, A General Expression for the Correlation of Rates of Transfer and Other Phenomena, AIChE J., 18(6), pp.
11211128.

Journal of Heat Transfer

Downloaded From: http://asmedigitalcollection.asme.org/ on 06/09/2015 Terms of Use: http://asme.org/terms

JUNE 2014, Vol. 136 / 062502-11

Vous aimerez peut-être aussi