International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

www.elsevier.com/locate/ijhmt

Experimental benchmark data for turbulent natural

convection in an air filled square cavity
F. Ampofo, T.G. Karayiannis *

Division of Environmental, Energy and Building Services Engineering, Faculty of Engineering, Science and Technology,
South Bank University, 103 Borough Road, London SE1 0AA, UK
Received 11 September 2002; received in revised form 19 March 2003

Abstract
An experimental study of low-level turbulence natural convection in an air filled vertical square cavity was con-
ducted. The cavity was 0.75 m high
0.75 m wide
1.5 m deep giving 2D flow. The hot and cold walls of the cavity
were isothermal at 50 and 10 °C respectively giving a Rayleigh number of 1.58
109 . The local velocity and temperature
were simultaneously measured at different locations in the cavity and both mean and fluctuation quantities are pre-
sented, i.e. u, u0rms , v, v0rms , T , Trms
0
, u0 v0 , u0 T 0 and v0 T 0 . The local and average Nusselt numbers, the wall shear stress as well
as the turbulent kinetic energy and the dissipation rate of the temperature variance are also presented. The experiments
were conducted with very high accuracy and as such the results can form experimental benchmark data and will be
useful for validation of computational fluid dynamics codes.
Ó 2003 Elsevier Ltd. All rights reserved.

Keywords: Natural convection; Velocity distribution; Temperature distribution; Heat transfer; Wall shear stress; Turbulence quantities

1. Introduction flow and heat transfer in such a geometry still remains a

challenge. In experimental studies, the flow is very sen-
Natural convection in enclosures is of importance in sitive to the experimental conditions. Further, boundary
many engineering applications. These include energy conditions on the horizontal cavity surfaces, defined in
transfer in rooms and buildings, nuclear reactor cooling, numerical work as adiabatic or perfectly conducting, are
solar collectors and electronic equipment cooling. Nat- not easily realised in experiments, e.g. in a water filled
ural convection in a rectangular cavity is also a very cavity, the thermal boundary conditions on these sur-
good vehicle for both experimental and theoretical faces lie somewhere in between the above two limiting
studies. In experimental terms, the geometry of the cases [1]. Different fluids, e.g. air and water may exhibit
rectangular cavity is simple and its boundary conditions significantly different temperature and fluid flow pat-
are relatively easy to assess so that researchers can focus terns in cavities of similar dimensions [2]. Although,
on the measurements of important quantities such as convection in enclosures has been extensively studied
velocity and temperature profiles. In numerical terms, experimentally, almost all the past researchers measured
the flow phenomena in the cavity are complicated and the velocity and temperature profiles separately. This
plentiful that they intrigue both physicists and engineers. prevented them from studying the velocity–temperature
However, in spite of the developments in the measure- correlations, u0i T 0 , in enclosures which are needed to un-
ment technology and instruments, as well as in numer- derstand better how the heat transfer is affected by flow
ical methods and computers, fully describing the fluid characteristics and physical parameters such as free
stream turbulence, pressure gradients and rough walls.
These experimental data are also needed for the suc-
*
Corresponding author. Tel.: +44-207-815-7628; fax: +44- cessful modelling of turbulent flows. The simultaneous
208-815-7699. measurement of local velocity and temperature is quite
E-mail address: [email protected] (T.G. Karayiannis). difficult. Some of the difficulty arises from the fact that in
0017-9310/03/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0017-9310(03)00147-9
3552 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

Nomenclature

A heat transfer area (m2 ) x, y, z Cartesian co-ordinates

ARx aspect ratio, ARx ¼ H =L X , Y , Z dimensionless co-ordinates
ARz aspect ratio for z-direction ARx ¼ D=L DT temperature difference DT ¼ Th  Tc (K)
Cl empirical constant in turbulence models
Greek symbols
D depth of the cavity (m)
a thermal diffusivity (m2 /s)
g gravitational acceleration (m/s2 )
b thermal expansion coefficient (1/K)
h average convective heat transfer coefficient,
dij Kronecker delta function (dij ¼ 1 if i ¼ j
h ¼ Q=AðTw  Ta Þ (W/m2 K)
and dij ¼ 0 if i 6¼ j)
hj subgrid-scale heat fluxes (m K/s)
e turbulent energy dissipation rate (m2 /s3 )
H height of the cavity (m)
g Kolmogorov length scale (m)
k thermal conductivity (W/m K)
l dynamic viscosity (kg/m s)
k turbulent kinetic energy (m2 /s2 )
m kinematic viscosity (m2 /s)
l turbulence length scale (m)
q fluid density (kg/m3 )
L width of the cavity (m)
sij turbulent stress tensor (m2 /s2 )
m any variable
n any variable Superscripts
Nu local Nusselt number, Nu ¼ hL=k – average value
0
Pr Prandtl number, Pr ¼ m=a fluctuation component
Q rate of heat transfer (W)
Subscripts
Ra Rayleigh number, Ra ¼ gbðTh  Tc ÞL3 =ðamÞ
b bottom wall
RaH Rayleigh number based on height,
c cold wall
RaH ¼ gbðTh  Tc ÞH 3 =ðamÞ
e ambient
S ij large-scale strain rate (s1 )
h hot wall
T temperature (°C, K)
i arbitrary quantity
u fluid velocity component in x-direction (m/s)
o reference condition
u0 v0 Reynolds stress (m2 /s2 )
t top wall
u0i T 0 turbulent heat flux (m K/s)
w wall
v fluid velocity component p inffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y-direction ffi (m/s)
rms root mean square
Vo buoyancy velocity, Vo ¼ gbH DT (m/s)
SGS subgrid-scale
w fluid velocity component in z-direction (m/s)
x displacement in x-direction (m)

most flows, the size of the smallest significant turbulent multaneously measured at different locations in the
eddy is much less than the closest separation at which cavity using a laser Doppler anemometer (LDA) and a
velocity and temperature probes can be used without micro-diameter thermocouple. This helped eliminate the
interfering with one another. In principle, it is possible to above-mentioned problems associated with a multi-
make the measurement using a multi-sensor hot- and sensor hot- and cold-wire anemometry system. The
cold-wire anemometry system since that instrument re- study aimed to provide highly accurate turbulent con-
sponds to both temperature and velocity. In practice, vection data, which can provide further insight into the
however, the use of multi-sensor probes has three distinct turbulent heat transfer in natural convection in enclo-
disadvantages: firstly, interference effects among the sures and be used for the validation of computational
prongs and/or sensors may be present; secondly, the hot- fluid dynamics (CFD) codes.
wire data may need to be corrected when using it in a
heated fluid medium and thirdly, the errors involved in
extracting the values of turbulence heat flux from the 2. Experimental facility and procedure
instrument signals tend to be unacceptably large. Also,
the accuracy of measuring the temperature fluctuations is The experimental rig, see Fig. 1(a), used in the pre-
limited by the frequency response of the sensor. sent study is fully automatically controlled. The major
An experimental study of heat transfer and fluid flow subsystems of the rig are the temperature control sys-
in a standard air filled square cavity was conducted in tem, the cavity, and a facility for measuring the air ve-
this study. The local velocity and temperature were si- locity and temperature simultaneously. The temperature
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3553

Fig. 1. Schematic diagram of the experimental facility and test cavity: (a) experimental facility, (b) three-dimensional schematic di-
agram of the air filled cavity and (c) the test cavity indicating detail of construction (all measurements are in mm).

control system comprised of a PC, a Schlumberger tional–integral–differential (PID) temperature-process

3531F data acquisition system, a multi-loop propor- controller and low noise K-type thermocouples. It
3554 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

maintained a constant temperature water flow to The dissipation rate, e, of turbulent kinetic energy, k,
chambers attached to the hot and cold plates, see Fig. was estimated from the relation between the turbulence
1(c). The PC, which controlled the Schlumberger 3531F length scale, l, k and e, i.e.
data acquisition system, recorded the thermocouple
e ¼ Cl3=4 k 3=2 =l ð1Þ
readings and sent the signal to the PID controller. The
PID controller then sent pulses to the burst firing trigger where Cl ¼ 0:09 [7]. The turbulence kinetic energy, k, is
modules to alter the heating power, i.e. the power was given as
either switched on or off at zero voltage. The cold and
hot water streams were pumped through the water k ¼ 12ðu02 þ v02 þ w02 Þ ð2Þ
chambers at a rate of 40 l/min and at predetermined
temperatures. The test cavity, shown in Fig. 1(b) and (c), In this study, the flow is two-dimensional at the mid-
was 0.75 m high
0.75 m wide
1.5 m deep. The hot section (ARz ¼ 2) and as such w  ¼ 0. Cebeci and
and cold walls of the cavity were made of 6 mm mild Bradshaw [8] reported that in isotropic turbulent flow,
steel plate. The hot wall was maintained isothermal at w0 is of the same order as u0 and v0 even if w
 ¼ 0. In this
50 0.15 °C whilst the cold wall at 10 0.15 °C giving a anisotropic turbulent natural convection, it is very dif-
Ra of 1.58
109 . The top and bottom walls were made ficult to estimate w0 without direct measurements. The
from 1.5 mm mild steel sheet and provided highly con- experimental and numerical works of Kreplin and Ec-
ducting boundaries. The temperatures on these isother- kelmann [9] and Spalart [10] respectively suggest that
mal and horizontal walls were measured using K-type along an isothermal vertical wall the following rela-
thermocouples. Two guard cavities surrounded the tionship exists:
passive vertical walls to reduce the heat exchange with u02 6 w02 6 v02 ð3Þ
the ambient. The room temperature was controlled at
30 0.2 °C. This was equal to the cavity average tem- As a first estimation, we can use
perature (average of hot and cold walls temperatures). u02 þ v02
Penot and NÕDame [3] pointed out that the 2D ap- w02 ¼ ð4Þ
2
proximation of experimental natural convection in cav-
ities should be valid if the horizontal aspect ratio (ARz ) i.e. the turbulent fluctuation in the z-direction contrib-
of the cavity is greater than 1.8. In this study, ARz was 2 utes one third of the k. Therefore,
and hence the depth of the cavity resulted in a 2D flow in 1:5ðu02 þ v02 Þ
the mid-plane of the cavity. Also, as stated earlier, we k¼ ð5Þ
2
had two guard cavities on each side of the passive ver-
tical walls. In addition, the average temperature of hot An initial measurement of the velocity distribution at
and cold walls was the same as ambient and the cavity mid-height of the cavity was used to estimate the tur-
was insulated. Hence the heat exchange between the bulent kinetic energy which peaked near the hot wall at
cavity and environment was kept to a minimum. The 2D 4.5
103 m2 /s2 . The velocity distribution was used to
of the flow was also thoroughly examined and verified estimate l. At about cavity mid-height, the experimental
by the earlier work of Tian and Karayiannis [4,5] carried results gave a boundary layer thickness of 5 mm and
out in the same cavity. They measured and compared this value was used for l. Eq. (1) then gives e  102 W/
velocity distribution at three cavity depths, Z ¼ 0, 0.533 kg. The corresponding value of g is 0.4 mm. Therefore,
and 0.8. The three profiles differed by only 4% on the the distance between the temperature and velocity
peak velocity and 0.5 mm on its position, which proved probes in the present study was around 0.4 mm. An E-
that the depth of the cavity, the guard cavities and the type thermocouple of wire diameter 25.4 lm was carried
ambient of 30 °C provided a 2D flow, especially at cavity by a computer controlled two-dimensional displacement
centre. device (accurate to 0.1 mm) and was used to measure the
The simultaneous measurements of local velocity and temperature in the cavity with an accuracy of 0.1 K. The
temperature are quite difficult and as stated earlier, some response time of the 25.4 lm thermocouple is 20 Hz in
of the difficulty arises from the fact that in most flows still air. The choice of thermocouple was based on the
the size of the smallest significant turbulent eddy is much experimental work of Mergui and Penot [11], who stated
less than the closest separation at which velocity and that the highest turbulent frequency is less than 5 Hz in
temperature probes can be used without interfering with an air filled cavity (Ra ¼ 2:33
109 , ARx ¼ 0:9). Thus,
one another. In view of this, it was very important to this thermocouple was sufficiently responsive to the high
estimate the turbulent length scale correctly so as to frequencies in this air cavity flow. A back scatter, two-
reduce this interference between the probes to the min- dimensional LDA with a burst spectrum analyser and a
imum. The Kolmogorov length scale for small eddies, 40 MHz frequency shift Bragg cell was used in the ve-
g ¼ ðm3 =eÞ1=4 [6], was used to estimate the maximum locity measurements in the cavity. The laser source was a
separation between the temperature and velocity probes. 300 mW argon laser. The measured velocity range was
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3555

from )0.5082 to +0.5082 m/s with a resolution of Table 1

6.20e)5 m/s and at a bandwidth of 0.125 MHz. The laser Summary of uncertainties in the measured and estimated pa-
beams entered the cavity through the guard cavity at an rameters
angle of 3.5° to the isothermal wall. A front lens with a Parameter Degree of uncertainty
focal length of 1200 mm was used. The probe volume Wall temperature 0.15 K
dimensions were 0.31 mm (diameter)
9.8 mm (length). Air temperature 0.10 K
Incense smoke was used as seeding, which lasted for Air velocity 0.07%
more than 24 h. The velocity was measured with an Reynolds stress 0.10%
accuracy of 0.07%. The simultaneous velocity and tem- Turbulent heat flux 0.15%
perature measurements were made by placing the LDA Rayleigh number 0.62%
probe volume immediately upstream of the thermo- Nusselt number 0.25–1.13%
Wall shear stress 1.38%
couple wire at a distance of around 0.4 mm. The ana-
logue to digital (A/D) board used to sample the
temperature signal was set to start sampling at the same 0
The root mean square quantities (Trms ; u0rms ; v0rms )
time as the LDA processors and also the A/D data and
" #1=2
the LDA data were synchronised. Preliminary experi- 1 XN
ments were conducted at different times, at mid-height m0rms ¼  Þ2
ðmi  m ð7Þ
N i¼1
(Y ¼ y=L ¼ 0:50) across the cavity, to check among
other things the influence of the measuring volume The cross-correlation quantities (u0 v0 ; u0 T 0 ; v0 T 0 )
(LDA probe) on the thermocouple probe for the tem-
perature measurements. The results showed that the X
N
m0 n0 ¼ ðmi  m
 Þðni  
nÞ ð8Þ
influence of the measuring volume (LDA probe) on the
i¼1
thermocouple probe for the temperature measurements
was negligible and that the repeatability for the various N is the number of readings taken in the experiments––
parameters measured was very good. The maximum 10 000 in this study. The choice of the total number of
deviation between readings obtained during experiments samples acquired was based on the experimental work of
performed at different times was 0.5 K for the temper- Wardana et al. [19] who proved that the uncertainty of a
ature and 2 mm/s for the velocity, see Ampofo [12] for statistical value decreases with increasing number of
further details. Also, the velocity and temperature re- samples, and the uncertainty becomes less than 5%
sults from the preliminary experiments were compared, when the number of samples is over 5000. All the mea-
with excellent agreement, to the earlier work of Tian and surements were taken at steady conditions at the mid-
Karayiannis [4,5] carried out in the same cavity but with plane of the cavity on a fine non-uniform mesh. The
a different approach. The velocity and temperature authors performed an energy balance for the cavity. On
profiles were measured separately in that study and the whole, heat transfer into the cavity from the hot and
hence they did not report any heat flux measurements. bottom walls was 98.12 and 21.67 W respectively. The
Tian and Karayiannis compared their results with the net loss from the cold and top walls was 97.77 and 22.53
work of past researchers like Mergui et al. [13], Lank- W respectively. The percentage error of heat input and
horst [14], Beghein et al. [15], Ziai [16], King [17] and output in the whole cavity was less than 0.5%. Table 1
Paolucci [18]. The authors found that compared with gives a summary of the uncertainties in the measured
earlier results, the agreement on the Nusselt number was and estimated parameters in this study. For further de-
good. Acceptable agreement was also found when tails on error analysis (bias and precision errors) and
comparing the temperature and velocity profiles at mid- propagation see Ampofo [12].
height. Differences were found along the mid-width and
the change rates of velocity and temperature along the
walls. These comparisons will not be repeated in this 3. Experimental results
paper since, as mentioned above, the present data agree
with [4,5]. In the present work, 10 000 velocity and One of the objectives of the study was to provide
temperature samples were acquired simultaneously at highly accurate turbulent convection data, which can be
the mid-plane of the cavity, see Fig. 1(b), for every point used for the validation of CFD codes. In view of this, the
and the mean, root mean square and cross-correlation numerical values of the experimental results, v, v0rms ,  u,
quantities were calculated as follows: u0rms , T , Trms
0
, u0 v0 , u0 T 0 , v0 T 0 , k, at the mid-height of the
cavity are included in this paper, see Table 2, for easy
The mean quantities (T , u, v) assess and comparison with CFD results. Tables 3–6
also present the experimental results for wall shear
1 XN
¼
m mi ð6Þ stress, mean temperature distribution on the horizontal
N i¼1
cavity walls, mean temperature distribution along the
 3556
F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572
Table 2
Experimental results at Y ¼ 0:5 in the cavity
v u
 v0rms u0rms T Tc 0
Trms u0 v0 v0 T 0 u0 T 0 k
X V0 V0 V0 V0 DT DT V02 V0 DT V0 DT V02

0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000
3.3333e)4 0.0403 )3.6440e)4 3.9900e)4 2.6400e)4 0.9525 0.0164 )2.0000e)6 8.1925e)7 1.8250e)6 1.7167e)7
6.6667e)4 0.0611 )4.0890e)4 2.3880e)3 3.8600e)4 0.9300 0.0204 )4.0000e)6 2.0693e)6 4.3250e)6 4.3887e)6
1.0000e)3 0.0843 )4.1192e)4 7.5980e)3 1.0700e)3 0.9200 0.0288 )7.0000e)6 2.6700e)6 8.6000e)6 4.4156e)5
1.3333e)3 0.1001 )6.9210e)4 0.0205 4.9570e)3 0.9125 0.0345 )6.0300e)6 5.1933e)5 1.5310e)5 3.3442e)4
1.6667e)3 0.1158 )7.5100e)4 0.0201 5.2520e)3 0.8750 0.0391 )7.0000e)6 7.2950e)5 2.7483e)5 3.2409e)4
2.0000e)3 0.1325 )8.1300e)4 0.0283 6.4380e)3 0.8685 0.0453 )7.5000e)6 8.8798e)5 9.1600e)6 6.3269e)4
2.3333e)3 0.1457 )8.6300e)4 0.0296 6.8920e)3 0.8373 0.0548 )8.0000e)6 1.1604e)4 3.6800e)5 6.9266e)4
2.6667e)3 0.1613 )6.3400e)4 0.0359 7.7500e)3 0.8298 0.0586 )1.0000e)5 2.3725e)4 4.9525e)5 1.0098e)3
3.0000e)3 0.1655 )7.4700e)4 0.0378 8.8760e)3 0.8071 0.0585 )1.3200e)5 5.2205e)4 7.6475e)5 1.1330e)3
3.3333e)3 0.1739 )6.6700e)4 0.0396 9.0360e)3 0.7830 0.0620 )8.0000e)6 1.3910e)3 1.2780e)4 1.2349e)3
3.6667e)3 0.1831 5.7000e)4 0.0406 0.0109 0.7701 0.0629 )2.0000e)6 2.1665e)3 1.8410e)4 1.3236e)3
4.0000e)3 0.2000 6.4600e)4 0.0457 0.0116 0.7635 0.0669 1.4000e)5 2.8213e)3 2.2600e)4 1.6656e)3
6.6667e)3 0.2127 1.7600e)4 0.0651 0.0178 0.6793 0.0753 2.0000e)4 3.0438e)3 2.5590e)4 3.4158e)3
9.3333e)3 0.2081 3.2140e)3 0.0705 0.0270 0.6186 0.0658 4.0000e)4 2.7488e)3 2.7545e)4 4.2732e)3
0.0133 0.1745 4.6790e)3 0.0717 0.0281 0.5655 0.0527 6.0000e)4 2.5878e)3 2.4818e)4 4.4506e)3
0.0200 0.1308 4.7330e)3 0.0681 0.0306 0.5270 0.0327 9.7300e)4 1.9623e)3 1.7593e)4 4.1770e)3
0.0267 0.0918 3.8240e)3 0.0624 0.0323 0.5171 0.0256 1.0550e)3 8.5705e)4 1.2693e)4 3.6992e)3
0.0333 0.0620 3.2180e)3 0.0560 0.0343 0.5098 0.0186 1.0870e)3 2.5128e)4 7.0375e)5 3.2332e)3
0.0400 0.0374 )2.6610e)3 0.0443 0.0298 0.5045 0.0172 8.0000e)4 )8.4875e)5 8.9250e)6 2.1383e)3
0.0533 0.0130 )2.4680e)3 0.0371 0.0263 0.5149 0.0167 5.0000e)4 )1.6765e)4 )1.7425e)5 1.5471e)3
0.0667 2.8710e)3 )3.2530e)3 0.0230 0.0208 0.5175 0.0131 3.0000e)4 )5.2150e)5 )3.8650e)5 3.3113e)4
0.0800 2.1470e)3 )3.6510e)3 0.0165 0.0106 0.5313 9.1302e)3 3.8000e)5 )4.7125e)6 )5.4675e)5 2.8862e)4
0.1067 2.6300e)4 )3.3230e)3 0.0108 8.3850e)3 0.5375 7.8789e)3 )1.1000e)5 2.3270e)5 )1.4550e)5 1.4060e)4
0.1333 3.7280e)4 )2.3600e)3 9.8790e)3 7.3030e)3 0.5400 5.3791e)3 )1.0000e)6 1.9900e)6 )1.2750e)6 1.1320e)4
0.1600 3.7200e)4 )1.6450e)3 7.7750e)3 6.6490e)3 0.5236 3.3575e)3 3.0000e)6 )1.5575e)6 )8.5000e)7 7.8495e)5
0.2000 3.9700e)4 )1.1380e)3 7.7490e)3 4.9050e)3 0.5270 3.0334e)3 1.3000e)6 4.9950e)6 )9.0000e)7 6.3080e)5
0.2400 3.3100e)3 )1.0910e)3 7.5590e)3 4.8580e)3 0.5251 2.8629e)3 1.50000)6 4.0150e)7 )9.0000e)7 6.0554e)5
0.2800 2.7900e)4 )1.2860e)3 7.4410e)3 4.0490e)3 0.5277 2.9035e)3 1.6000e)6 1.0933e)6 )2.2500e)7 5.3822e)5
0.3200 1.7200e)4 )9.7800e)4 7.1790e)3 4.1550e)3 0.5308 3.2389e)3 2.0000e)6 )1.5413e)6 1.7000e)6 5.1602e)5
0.3600 1.1100e)4 )5.8400e)4 6.5000e)3 4.3470e)3 0.5280 3.1980e)3 1.0000e)6 )2.2993e)6 )6.2500e)7 4.5860e)5
0.4000 1.8100e)4 5.1000e)4 6.1530e)3 3.9160e)3 0.5217 2.8874e)3 )1.0000e)6 )2.1257e)7 )2.0000e)7 3.9896e)5
0.4400 )1.2700e)4 6.3800e)4 5.4530e)3 4.0150e)3 0.5200 2.9203e)3 )2.0000e)6 )3.4800e)7 1.0000e)7 3.4392e)5
0.5000 )6.9000e)5 5.7900e)4 5.0290e)3 5.1590e)3 0.5174 3.1881e)3 1.1000e)6 8.7250e)7 1.0325e)6 3.8930e)5
0.5600 )4.8700e)4 1.0660e)3 4.4400e)3 5.0040e)3 0.5188 3.1007e)3 )2.8000e)6 1.4750e)5 1.1250e)6 3.3565e)5
 0.6000 )1.8100e)4 1.7850e)3 4.4200e)3 4.1250e)3 0.5154 4.8337e)3 )2.0000e)6 )2.3500e)6 1.1500e)6 2.7414e)5
0.6400 )6.0600e)4 2.1460e)3 4.5170e)3 4.1350e)3 0.5165 4.0352e)3 )2.0000e)6 5.0000e)7 )4.2500e)7 2.8126e)5
0.6800 )4.9400e)4 2.3670e)3 4.4180e)3 3.4520e)3 0.5190 3.0430e)3 )3.0000e)6 7.1250e)6 2.0000e)7 2.3576e)5
0.7200 )4.9900e)4 2.4850e)3 4.8540e)3 3.3290e)3 0.5202 3.3391e)3 )2.0000e)6 6.5500e)7 )7.5000e)8 2.5983e)5
0.7600 )4.5800e)4 2.5830e)3 4.7060e)3 3.4520e)3 0.5160 4.3246e)3 )1.0000e)6 6.3500e)6 )9.3250e)6 2.5547e)5
)2.2700e)4 )1.0000e)6 )5.0000e)7

F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572
0.8000 3.4510e)3 4.6640e)3 3.6450e)3 0.5135 7.0418e)3 7.0000e)7 2.6279e)5
0.8400 )2.4900e)4 3.1830e)3 4.9470e)3 4.0910e)3 0.5138 5.0101e)3 )1.0000e)6 )5.3500e)6 )3.6750e)6 3.0907e)5
0.8667 )3.1700e)4 3.3130e)3 5.1190e)3 4.3730e)3 0.5079 6.5627e)3 )6.0000e)7 3.5250e)6 )3.9000e)6 3.3995e)5
0.8933 1.2600e)3 2.9930e)3 8.0740e)3 6.6590e)3 0.5076 5.8379e)3 )2.7000e)5 )3.9250e)6 )1.2450e)5 8.2149e)5
0.9200 4.4080e)3 2.8370e)3 0.0156 0.0151 0.5061 7.7054e)3 3.5000e)5 )3.7525e)5 )3.6425e)5 3.5210e)4
0.9333 2.7420e)3 1.9950e)3 0.0245 0.0156 0.5151 0.0115 1.0000e)5 )6.5450e)5 3.7775e)5 6.3357e)4
0.9467 )0.0108 1.3280e)3 0.0352 0.0223 0.5257 0.0147 3.5700e)4 )8.3900e)5 6.7250e)5 1.2997e)3
0.9600 )0.0403 )2.2620e)3 0.0465 0.0254 0.5198 0.0166 5.9600e)4 )9.9100e)5 7.0725e)5 2.1030e)3
0.9667 )0.0684 )4.3800e)3 0.0559 0.0286 0.5313 0.0166 8.7000e)4 3.5020e)4 8.0750e)5 2.9600e)3
0.9733 )0.0916 )5.3000e)3 0.0639 0.0310 0.5199 0.0248 9.9300e)4 5.8808e)4 1.3713e)4 3.7844e)3
0.9800 )0.1252 )5.1300e)3 0.0661 0.0272 0.5047 0.0353 9.4400e)4 1.0818e)3 1.87780)4 3.8323e)3
0.9867 )0.1828 )4.9900e)3 0.0652 0.0224 0.4556 0.0550 6.4700e)4 2.0298e)3 2.4098e)4 3.5665e)3
0.9907 )0.2185 )4.6620e)3 0.0659 0.0183 0.3834 0.0703 )3.3000e)4 2.7414e)3 1.5248e)4 3.5079e)3
0.9933 )0.2257 )2.7390e)3 0.0586 0.0158 0.3137 0.0565 2.0900e)4 2.4064e)3 1.18000)4 2.7624e)3
0.9960 )0.1884 )1.5920e)3 0.0429 0.0106 0.2434 0.0572 1.0000e)6 1.9478e)3 7.0950e)5 1.4673e)3
0.9963 )0.1803 )1.5680e)3 0.0404 9.8470e)3 0.2318 0.0519 8.0000e)7 1.6570e)3 6.11500)5 1.2987e)3
0.9967 )0.1736 )1.4050e)3 0.0382 8.6060e)3 0.2136 0.0465 )4.0000e)6 8.2023e)4 5.8625e)5 1.1511e)3
0.9970 )0.1604 )1.4560e)3 0.0355 7.4360e)3 0.2050 0.0423 )6.0000e)6 5.0158e)4 3.6425e)5 9.8586e)4
0.9973 )0.1488 )8.6500e)4 0.0326 6.6200e)3 0.1865 0.0361 )2.3000e)5 4.1465e)4 3.4300e)5 8.3190e)4
0.9977 )0.1268 )6.8500e)4 0.0309 5.8210e)3 0.1608 0.0308 )1.4000e)5 3.7715e)4 3.0850e)5 7.4138e)4
0.9980 )0.1056 )5.1000e)4 0.0277 5.0320e)3 0.1324 0.0297 )6.0000e)6 2.7485e)4 1.6625e)5 5.9591e)4
0.9983 )0.0835 )4.3900e)4 0.0230 4.7780e)3 0.1183 0.0230 )2.0000e)6 1.3040e)4 1.5750e)5 4.1470e)4
0.9987 )0.0588 )3.9700e)4 0.0211 4.0410e)3 0.0935 0.0171 )1.0000e)6 3.6300e)5 1.4000e)5 3.4555e)4
0.9990 )0.0314 )1.9100e)4 0.0158 3.3280e)3 0.0777 0.0174 )1.0000e)6 )9.9750e)6 1.1500e)5 1.9592e)4
0.9993 )3.3890e)3 )2.2000e)5 0.0101 2.7850e)3 0.0657 0.0115 )1.0000e)6 )9.5250e)6 1.0200e)5 8.2340e)5
0.9997 )4.1200e)4 )1.2000e)5 8.1010e)3 7.8500e)4 0.0309 0.0105 )1.0000e)6 )9.4750e)6 8.0250e)6 4.9682e)5
1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X ¼ x=L, Y ¼ y=L, V0 ¼ gbH DT ¼ 1 m/s, DT ¼ Th  Tc ¼ 40 K.

3557
3558 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

Table 3 Table 5
Wall shear stress Mean temperature distribution along mid–width
Hot wall Cold wall ðT  Tc Þ=DT Y
3 3
Distance sw
10 Distance sw
10 0.3553 0.0000
from bottom (N/m2 ) from top (N/m2 ) 0.3328 1.3333e)3
wall, Y wall, 1  Y 0.3253 2.6667e)3
0.0000 0.0000 1.0000 0.0000 0.3178 4.0000e)3
0.0500 1.0600 0.9500 1.0800 0.2978 6.6667e)3
0.1000 1.4400 0.9000 1.4200 0.2828 9.3333e)3
0.2000 1.5600 0.8000 1.6500 0.2778 0.0133
0.3000 1.6200 0.7000 1.7000 0.2737 0.0200
0.4000 1.6600 0.6000 1.7600 0.2790 0.0267
0.5000 1.6400 0.5000 1.7200 0.2814 0.0360
0.6000 1.5200 0.4000 1.6000 0.2906 0.0500
0.7000 1.3000 0.3000 1.4000 0.3086 0.0750
0.8000 1.2000 0.2000 1.3000 0.3250 0.1000
0.9000 1.0800 0.1000 1.1000 0.3571 0.1500
0.9500 0.7000 0.0500 0.7500 0.3804 0.2000
1.0000 0.0000 0.0000 0.0000 0.4001 0.2500
0.4213 0.3000
Y ¼ y=L. 0.4395 0.3500
0.4564 0.4000
Table 4 0.4854 0.4500
Mean temperature distribution on horizontal walls 0.5174 0.5000
ðT  Tc Þ=DT 0.5463 0.5500
X
0.5662 0.6000
Top wall Bottom wall 0.5800 0.6500
0.0000 1.0000 1.0000 0.6068 0.7000
2.0000e)3 0.9490 0.9184 0.6234 0.7500
6.6700e)3 0.9333 0.9038 0.6356 0.8000
0.0133 0.9342 0.8862 0.6644 0.8500
0.0267 0.9183 0.8639 0.6825 0.9000
0.0533 0.8782 0.7733 0.6991 0.9250
0.1000 0.8210 0.6608 0.7182 0.9500
0.2000 0.7597 0.5263 0.7250 0.9640
0.3000 0.7107 0.4520 0.7300 0.9733
0.4000 0.6779 0.3960 0.7375 0.9800
0.5000 0.6393 0.3503 0.7350 0.9867
0.6000 0.6135 0.3116 0.7250 0.9907
0.7000 0.5578 0.2722 0.7125 0.9933
0.8000 0.4880 0.2214 0.7000 0.9960
0.9000 0.3372 0.1490 0.6850 0.9973
0.9467 0.2338 0.0938 0.6635 0.9987
0.9733 0.1409 0.0445 0.6443 1.0000
0.9867 0.1334 0.0352 Y ¼ y=L, DT ¼ Th  Tc ¼ 40 K.
0.9933 0.1234 0.0272
0.9980 0.0967 0.0185
1.0000 0.0000 0.0000
isothermal vertical walls will be used to study the near-
X ¼ x=L, DT ¼ Th  Tc ¼ 40 K. wall turbulence in the cavity. Also, the distribution of
the turbulent heat flux in the cavity will be presented and
cavity mid-width and local Nusselt number distribution discussed in detail being the first time it has been directly
along the walls of the cavity respectively. The experi- measured in an air filled square cavity at low-level tur-
mental results at different heights of the cavity are also bulence.
available from the authors. The general flow and ther-
mal behaviour of the fluid in the cavity would not be 3.1. Inner and outer boundary layer structures
presented in graphical form in this paper. Interested
readers can consult Tian and Karayiannis [4,5] where A turbulent boundary layer can be described as a
such information is given. Instead, in this section, the two-layer structure, namely an inner layer and an outer
experimental results in the immediate vicinity to the layer [20], separated by the position of the velocity
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3559

Table 6
Local Nusselt number distribution along the walls
Y Nu X Nu
Hot wall Cold wall Bottom wall Top wall
0.0200 136.0000 21.0000 0.0133 75.0000 22.0000
0.0493 122.0000 33.0000 0.0400 58.0000 18.0000
0.1000 95.0000 42.0000 0.0800 40.0000 8.0000
0.2000 84.0000 44.0000 0.1333 38.0000 5.0000
0.3000 72.0000 47.0000 0.2000 36.0000 2.0000
0.4000 65.0000 50.0000 0.2800 20.0000 )4.0000
0.5000 58.0000 60.0000 0.3600 16.0000 )8.0000
0.6000 52.0000 62.0000 0.5000 10.0000 )11.0000
0.7000 47.0000 69.0000 0.6400 8.0000 )18.0000
0.8000 40.0000 80.0000 0.7200 4.0000 )23.0000
0.9000 36.0000 87.0000 0.8000 1.0000 )31.0000
0.9493 28.0000 122.0000 0.8667 )12.0000 )35.0000
0.9867 17.0000 138.0000 0.9200 )15.0000 )42.0000
0.9600 )19.0000 )55.0000
0.9867 )25.0000 )70.0000
X ¼ x=L, Y ¼ y=L.

maximum. Fig. 2(a) shows the boundary layer structure walls to a peak at X ¼ 0:007 (5 mm) and then decreases
(vertical velocity profile) along the hot wall in the cavity rapidly to zero at about X ¼ 0:107 (80 mm) from the hot
at Y ¼ 0:5. The boundary layer on the cold wall is al- wall, see Fig. 3(a). The profile of v near the cold wall is
most exactly anti-symmetrical. The inner layer is 5 mm almost anti-symmetrical at mid-height of the cavity. The
wide (at Y ¼ 0:5) and the outer layer is 75 mm wide, i.e. horizontal velocity profile, u, is almost negligible in the
the inner layer is less than 7% of the outer layer. The boundary layer, see Fig. 3(a). In general,  u is one order
thickness of the inner layer along the isothermal walls is of magnitude smaller than v. Fig. 3(a) also shows a
not constant. It varies between 4 and 7 mm depending comparison of the mean component of the vertical ve-
on the height along the isothermal walls. It is narrow at locity at the cavity mid-height with the numerical sim-
the bottom of the hot wall (and at the top of the cold ulation of Barozzi et al. [22] in an air filled cavity with
wall) and wider at the top of the hot wall (and at the ARx ¼ 1 and Ra ¼ 1010 . Also shown in the figure are
bottom of the cold wall). The inner layer can be divided the large-eddy simulation (LES) results of Peng and
further into a viscous layer next to the wall where the Davidson [23] in a similar cavity to the present study and
Reynolds stress is negligible and a buoyant sublayer for under the same parameters. For the vertical velocity
the rest of the inner layer, see Fig. 2(b). In the viscous profile, the results of Barozzi et al. [22], which were
layer, very close to the wall, the temperature profile is based on the k–e model, and that of Peng and Davidson
nearly linear and this region is referred to as conductive [23] are generally in good agreement with the present
layer. The linearity of the temperature profile is shown results. However, from Fig. 3(a), it can be seen that the
in Fig. 2(b) with a line drawn through the measured LES results of Peng and Davidson [23] marginally
points. In the conductive layer, the heat flux is constant overestimated the peak value of the vertical velocity.
[21]. The viscous layer is about 3 mm in this study with 2 Also, the results of Barozzi et al. [22] have a steeper
mm of it being the conductive layer. The velocity profile velocity gradient near the wall and the location of the
in the viscous layer is a cubic function of distance. peak velocity is much closer to the wall than in the
Figs. 3–10 show the profiles of all the measured pa- present study. Thus near the wall, the k–e model predicts
rameters, v, v0rms , u, u0rms , T , Trms
0
, u0 v0 , u0 T 0 , v0 T 0 , k, in the a higher rate of change of velocity which can lead to
boundary layer along the isothermal walls at Y ¼ 0:5. In overestimation of the wall shear stress.
some of these figures, the present results are compared Fig. 3(b) shows that the vertical velocity fluctuations,
with past numerical results. The parameters are divided v0rms , and the horizontal velocity fluctuation, u0rms , are
into three main groups (velocity field, temperature field concentrated in the boundary layer and decrease to al-
and turbulence quantities). most nothing outside the boundary layer. One feature
shown in Fig. 3(b) is the difference between u0rms and v0rms .
3.2. Velocity field The value of u0rms is less than half that of v0rms in the
boundary layer at the cavity mid-height. On the whole,
The vertical velocity profile, v, which defines the the profile of fluctuation quantities is anti-symmetric.
boundary layer increases steeply from the isothermal Fig. 3(b) also shows a comparison of the fluctuation
3560 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

Inner layer
Outer layer

250

Vertical velocity, vx103 (m/s) 200

150

100

50

-50
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80

(a) Distance from hot wall (mm)

Inner layer

Viscous layer Buoyant sub-layer

Conductive layer
250

200

150
vx103 (m/s)
v, T, u'v'

T (oC)
100 u'v' x106 (m2/s2)

50

-50
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
(b) Distance from hot wall (mm)

Fig. 2. Boundary layer structure at Y ¼ 0:5: (a) vertical velocity profile near the hot wall and (b) inner boundary layer near the hot
wall.

component of the vertical velocity at the mid-height of the same and were estimated by the following relation-
the cavity with the numerical simulation of Barozzi et al. ship:
[22] and the LES results of Peng and Davidson [23]. As rffiffiffiffiffi
0 2k
noted above, the numerical simulation of Barozzi et al. vrms ¼ ð9Þ
3
[22] was based on the k–e model, which assumed the
local isotropy hypothesis, i.e. both the vertical and From Fig. 3(b), it can be seen that the experimental
horizontal velocity fluctuations were assumed to be results are higher than the predicted values from the k–e
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3561

0.25
v, this study
u, this study
0.20 v, [22]
v, [23]

0.15
v/Vo, u/Vo

0.10

0.05

0.00

-0.05
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200

(a) Distance from hot wall, X=x/L

0.10
v'rms, this study
u'rms, this study
0.08 v'rms, [22]
v'rms, [23]

0.06
v'rms/Vo, u'rms/Vo

0.04

0.02

0.00

-0.02
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200

(b) Distance from hot wall, X=x/L

Fig. 3. Velocity field: (a) mean velocity at Y ¼ 0:5 and (b) velocity fluctuation at Y ¼ 0:5.

model of Barozzi et al. [22]. This is expected because the oT 0

hj ¼ T 0 u0j  T 0 u0 j ¼ at ð11Þ
flow in the cavity is a wall shear flow and not isotropic oxj
and as such the vertical and horizontal velocity fluctu-
ations are not expected to be the same. In the LES of The SGS time scaling, TSGS , took the following form:
Peng and Davidson [23], the subgrid-scale (SGS) stresses
!1=2
appearing in the filtered Navier–Stokes equations were  2 gb oT 0
responsible for the energy occurrence between the large- TSGS ¼ S   d2j ð12Þ
Prt oxj
scale and the SGS eddies. They were modelled in
alignment
  with the large-scale strain rate, S ij ¼ where the
1 o
ui o
u
þ oxji , using the SGS eddy viscosity, mt , i.e.   magnitude
qffiffiffiffiffiffiffiffiffiffiffiffiffi of the resolved local strain rate
2 oxj tensor, S  ¼ 2S ij S ij . The present results agree rea-
dij sonably well with the LES results of Peng and Davidson
sij ¼ u0i u0j  u0 i u0 j ¼ 2mt S ij þ skk ð10Þ
3 [23] except that the predicted values have a marginally
By means of the gradient diffusion hypothesis, the SGS higher peak and a thinner boundary layer structure. Fig.
heat fluxes, hj , in the filtered thermal energy equation 3(b) confirms that on the whole, LES predicts turbulent
were modelled as quantities much better than k–e models.
3562 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

0.65

Dimensionless temperature, (T-Tc)/(Th-Tc)

0.60
0.55
0.50
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
This study
0.05
[23]
0.00
0.80 0.82 0.84 0.86 0.88 0.90 0.92 0.94 0.96 0.98 1.00

(a) Distance from the hot wall, X=x/L

4.5
This study
4.0 [23]
Temperature fluctuation, T'rms (K)

3.5

3.0

2.5

2.0

1.5

1.0

0.5

0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

(b) Distance from the hot wall, X=x/L

Fig. 4. Temperature field: (a) mean temperature at Y ¼ 0:5 and (b) temperature fluctuation at Y ¼ 0:5.

The flow in the present cavity was limited in the tion was obtained for the velocity profile. The equation
boundary layers along the walls. The velocity profile in was differentiated at X ¼ 0 and 1 to obtain the velocity
the viscous layer was used to determine the wall shear gradient and then the wall shear stress. The agreement
stress. As stated earlier, in the viscous layer the velocity of the shear stress values between the two isothermal
profile is a cubic function of distance from the wall, i.e. walls is good, the largest difference being 0.08%, see
Table 3.
v ¼ ax þ bx2 þ cx3 ð13Þ

where a, b and c can vary depending on height. Then the 3.3. Temperature field
wall shear stress can be expressed as
 Fig. 4(a) shows the temperature profile, T , in the cold
ov  wall boundary layer at mid-height. The profile changes
sw ¼ l  ¼ a ð14Þ
ox  steeply near the isothermal wall. The temperature profile
x¼0
is nearly anti-symmetric in the hot wall boundary layer
Using the measured results in the viscous layer and a at mid-height. The authors present the cold wall data in
cubic polynomial least squares fitting process, an equa- Fig. 4(a) to allow comparison with the numerical work
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3563

180
This study, hot wall
160 This study, cold wall
[23], hot wall
140 [23], cold wall

Local Nusselt number, Nu 120

100

80

60

40

20

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Distance from the bottom wall, Y=y/L for the hot wall
(a) and distance from the top wall, 1-Y for the cold wall

100
This study, bottom wall
This study, top wall
80 [23], bottom wall
[23], top wall
Local Nusselt number, Nu

60

40

20

-20

-40
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Distance from the hot wall, X=x/L for the bottom wall
(b) and distance from the cold wall, 1-X for the top wall

Fig. 5. Comparison of local Nusselt number: (a) local Nusselt number along the isothermal walls and (b) local Nusselt number along
the horizontal walls.

of Peng and Davidson [23] who reported cold wall val- parameters as the present study. Overall, the agreement
ues. After an initial sharp decrease (from hot wall) and between the present results for the mean temperature
increase (from cold wall), the temperature remained and the LES results of Peng and Davidson [23] is good
fairly constant (from X  0:1 to X  0:9), indicating in the cold wall boundary layer except that the temper-
that the fluid in this region of the cavity was nearly ature over-shoot is more pronounced in their case.
stationary. The dimensionless temperature at the core of As seen in Fig. 4(b), the temperature fluctuation in-
the cavity was 0.52 (30.7 °C), which is nearly equal to the tensity increases with distance from the hot wall to a
mean temperature (and the ambient). This indicates that peak and then decreases to almost nothing. At mid-
there was no heat loss through the passive vertical walls, height the profile is almost symmetrical with a maximum
0
i.e. through the guard cavities. The numerical results of value of around Trms ¼ 3:0 K near both the hot and cold
Peng and Davidson [23] were obtained under the same walls. In general, the temperature fluctuation profile in
3564 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

1.50
This study
[23]
1.25

1.00

u'v' x103 (m2/s2)

0.75

0.50

0.25

0.00

-0.25
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
Distance from the hot wall, X=x/L

Fig. 6. Reynolds stress at Y ¼ 0:5.

the cavity is nearly anti-symmetrical about the cavity number on the hot, cold, bottom and top walls in the
centre. The present results are also compared with the cavity is compared with the numerical results of Peng
numerical results of Peng and Davidson [23] in Fig. 4(b). and Davidson [23] at the same parameters as the present
The distribution of Peng and Davidson [23] has a higher study in Fig. 5. On the lower part of the hot and cold
peak value than in the present study but on the whole, walls, the simulation from Peng and Davidson [23] yields
the agreement between the present results and their re- lower heat transfer rates than the present results (Fig.
sults is good. 5(a)). The agreement between the present results and
The local and average Nusselt numbers on the hot, that of Peng and Davidson [23] for the local Nusselt
cold, bottom and top walls can be determined using the number along both the bottom and top walls of the
measured temperature data in the thermal conductive cavity is good (Fig. 5(b)).
boundary layer (see Fig. 2(b)) where the heat flux is
constant as follows: 3.4. Turbulence quantities

L oT 
Nulocal ¼  ð15Þ
Th  Tc oxi w The turbulence quantities obtained in this study are
presented in Figs. 6–10. The Reynolds stress, u0 v0 , near
From this definition, the heat flux is positive with axis the hot wall is depicted in Fig. 6. For the first 3 mm
direction, i.e. the heat transfer from the hot wall into the form the wall the Reynolds stress in zero (see also Fig.
cavity and from the cavity into the cold wall is positive. 10(b)). The maximum positive value of u0 v0 is about
The heat transfer from the bottom wall into the cavity 1.08
103 m2 /s2 and 0.99
103 m2 /s2 near the hot and
and from the cavity into the top wall is positive. The cold walls respectively and is located well outside the
temperature gradient with respect to x or y was esti- maximum velocity point (5 mm or X ¼ 0:007) at about
mated by a linear best fitting from the first 6–9 mea- 25 mm (X ¼ 0:033) from the walls, see Fig. 6. This lo-
suring points near the wall. The local Nusselt numbers cation coincides with the point where the intensities of
on the hot, cold, bottom and top walls of the cavity are velocity fluctuations become maximum in the outer
presented in Table 6. The local Nusselt number reaches layer. Thus, the outer layer has the same characteris-
a maximum at the bottom of the hot wall and at the top tics as the forced convection boundary layer because
of cold wall because of a thinner thermal boundary oV =ox < 0 and u0 v0 > 0. The Reynolds stress is near zero
there. The measured maximum Nu number is about 138. between X ¼ 0:1 and 0.9 at mid-height because outside
At mid-height, the local Nusselt number is about 59. At the boundary layer the budget of u0 v0 are counter-bal-
the other end, because the fluid flow meets the wall, the anced. Also, at mid-height the values of the Reynolds
heat transfer diminishes significantly and the local stress, u0 v0 , near the two isothermal walls differ by about
Nusselt number drops to about 20. The average Nusselt 8%. The dynamic viscosity of air is 17.62
106 N s/m2
number was 62.9, 62.6, 13.9 and 14.4 for the hot, cold, at 10 °C and 19.55
106 N s/m2 at 50 °C, i.e. the 40 K
bottom and top walls respectively. The local Nusselt temperature difference causes about 11% density differ-
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3565

220
This study
200
[23]
180

160

140
v'T' x10 (mK/s)
120

100
3

80

60

40

20

-20
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

(a) Distance from hot wall, X=x/L

35
This study
[23]
30

25
u'T' x103 (mK/s)

20

15

10

-5
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

(b) Distance from hot wall, X=x/L

Fig. 7. Turbulent heat flux: (a) Y ¼ 0:5 (vertical) and (b) Y ¼ 0:5 (horizontal).

ence in the cavity and as such the Reynolds stress at mid- u0 T 0 differ by about 12%. This is due to the change in
height can be considered to be symmetric. In Fig. 6, the value of the dynamic viscosity of air near the isothermal
Reynolds stress, u0 v0 , at mid-height in the hot wall walls as explained above. Both v0 T 0 and u0 T 0 increase
boundary layer of the cavity is also compared with the rapidly from the isothermal walls to a peak; v0 T 0 peaks at
numerical results of Peng and Davidson [23] at the same about 0.121 and 0.110 m K/s near the hot and cold walls
Ra and aspect ratio. Overall, the agreement between the respectively. The corresponding values for u0 T 0 are about
present results and that of Peng and Davidson [23] is 0.011 and 0.0097 m K/s. The peak value of the turbulent
poor. The numerical results of Peng and Davidson [23] heat fluxes, u0i T 0 , is located in the vicinity of the maxi-
have a relatively lower peak value, a narrow boundary mum velocity location, which is about 5–7 mm from the
layer structure and non-zero u0 v0 beyond X ¼ 0:1. walls, see Fig. 7. The turbulent heat fluxes are near zero
The experimental results of v0 T 0 and u0 T 0 , the turbu- between X ¼ 0:1 and 0.9 at mid-height. In the overall
lent heat fluxes in the y- and x-directions, at mid-height boundary layer region, the fluctuating amplitude of v0 T 0
are shown in Fig. 7. At mid-height the values of v0 T 0 near far exceeds that of u0 T 0 because of the large amplitude of
the two isothermal walls differ by about 9% and that of the v0 fluctuation, see Fig. 3(b). In general, the horizontal
3566 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

1.0

0.9

0.8

Distance from bottom wall, Y=y/L

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Distance from hot wall, X=x/L

Fig. 8. Vector plot of the turbulent heat flux (scale: 0.0168 m K/s/mm).

5.0
This study
4.5 [24]

4.0

3.5
k x103 (m2/s2)

3.0

2.5

2.0

1.5

1.0

0.5

0.0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10

Distance from hot wall, X=x/L

Fig. 9. Turbulent kinetic energy at Y ¼ 0:5.

turbulent heat flux is one order of magnitude smaller [23] at the same Ra and aspect ratio. For the vertical
than the vertical turbulent heat flux. The turbulent heat turbulent heat flux, the agreement between the two sets
flux at mid-height in the cavity is compared in Fig. 7 of results in terms of the boundary layer thickness is
with the numerical simulation of Peng and Davidson good but the simulation over-predicted the peak value
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3567

3
v x10 (m/s)
3
v' x10 (m/s)
u x103 (m/s)
u' x103 (m/s)
T (oC)
o
T' ( C)
220 50 4

200

180

Temperature fluctuation, T' (oC)

160 45 3

Mean temperature, T (oC)

Velocity quantities

140

120

100 40 2

80

60

40 35 1

20

-20 30 0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

(a) Distance from hot wall (mm)

140
u'v' x105 (m2/s2)
120 u'T' x103 (mK/s)
3
v'T' x10 (mK/s)
4 2 2
k x10 (m /s )
100
Turbulent quantities

80

60

40

20

-20
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

(b) Distance from hot wall (mm)

Fig. 10. Profiles of experimental parameters near the hot wall at Y ¼ 0:5: (a) velocity and temperature quantities and (b) turbulent
quantities.

by about 40% (Fig. 7(a)). In the case of the horizontal which is coarser than the measurement mesh. This figure
turbulent heat flux, the numerical simulation over-pre- provides a very important insight into the structure of
dicted the peak value by more than 67% (Fig. 7(b)). The flow within the cavity. At the bottom of the hot wall (top
whole turbulent heat flux vector plot in the cavity is of the cold wall) the turbulent heat flux is effectively
given in Fig. 8. The turbulent heat flux vector is based zero. The absence of a turbulent heat flux in these re-
on all the experimental data but is plotted on a mesh, gions suggests that the lower part of the hot wall (upper
3568 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

part of the cold wall) boundary layer has many of the temperature) is aligned with the mean scalar gradient
characteristics of a laminar flow. The figure also con- vector. Even in simple turbulent flows this is found not
firms that the majority of the turbulent heat flux is to be the case. For example, in an experiment on homo-
confined in a narrow strip along the isothermal walls. geneous turbulent shear flow [25] the angle between rT
The turbulent kinetic energy was estimated by Eq. (5) and u0i T 0 was measured to be 65°. Similarly, the tur-
and the results at mid-height in the cavity are compared bulent viscosity hypothesis (Eq. (17)) implies that the
with the numerical results of Vasic and Hanjalic [24] anisotropy tensor, aij , is aligned with the mean rate of
using a low-Re number k–e model at Ra ¼ 5
1010 and strain tensor, i.e.
ARx ¼ 1, see Fig. 9. Very close to the wall, the numerical 
predicted turbulent kinetic energy results of Vasic and 2 oui ouj
aij  u0i u0j  dij k ¼ mt þ ð18Þ
Hanjalic [24] agree reasonably well with the present re- 3 oxj oxi
sults. However, the predicted peak value in their case is
much smaller and the boundary layer is also thinner Being symmetric and deviatoric, both aij and the mean
than the measured one in the present study. rate of strain have five independent components. Ac-
Fig. 10 shows the profiles of all the experimental cording to the turbulent viscosity hypothesis, these five
parameters in the inner boundary layer of the hot wall at components are related to each other through the scalar
mid-height. The figure confirms what have been stated coefficient mt . Again, even in simple shear flow, it is
already in the above paragraphs, i.e. the horizontal ve- found that this alignment does not occur [6]. An im-
locity profile, u, is one order of magnitude smaller than portant class of flows consists of those that can be de-
the vertical velocity profile, v; the temperature profile, T , scribed by the two-dimensional turbulent boundary
changes steeply near the isothermal wall; the Reynolds layer equations (of which the present study is an
stress, u0 v0 , is near zero for the first 3 mm from the iso- example). In the present study, the mean velocity is
thermal wall; etc. In addition to these findings, the figure predominantly in the y-coordinate direction, while
shows that the horizontal velocity profile, u , is almost variations in mean quantities are predominantly in the x-
negligible in the inner boundary layer. The value of the coordinate direction. As a result of this, only one com-
horizontal velocity fluctuation, u0 , is less than half that ponent (v0 T 0 ) of the scalar flux and one Reynolds stress
of the vertical velocity fluctuation, v0 . Near the isother- (u0 v0 ) appear in the boundary layer equations. Conse-
mal hot wall, the Reynolds stress, u0 v0 , decreases to quently, the turbulent diffusivity hypothesis reduces to
negligible values faster than its individual factors, u0 and 
oT
v0 . Fig. 10(b) shows that in the conductive layer (X < 2 v0 T 0 ¼ at ð19Þ
ox
mm) the turbulent quantities are almost negligible,
which confirms that the heat transfer in this region of the and the turbulent viscosity hypothesis to
boundary layer is by pure conduction. Beyond this re- 
gion to a greater part of the outer region, the heat ov
u0 v0 ¼ mt ð20Þ
transfer is by convection. ox

3.5. Turbulent diffusivity (at ), turbulent viscosity (mt ) and Both of these equations relate a single covariance to a
turbulent Prandtl number (Prt ) single gradient. Providing that the covariance and the
gradient have opposite signs––which is almost the
In the eddy viscosity model, one gives up the mode- case––then, rather than being hypotheses or assump-
ling of turbulent heat flux, u0i T 0 , and Reynolds stress, tions, these equations can be taken as definitions of at
u0i u0j , equations and adopts the generalized Boussinesq and mt [6]. The distributions of the turbulent diffusivity
eddy viscosity model. Thus, the turbulent heat flux and the turbulent viscosity close to the hot wall
equation is given by (X ¼ 0–0:0333) at mid-height of the cavity is shown in
 Fig. 11. On the basis of the measured temperature and
oT velocity results, these distributions were calculated using
u0i T 0 ¼ at ð16Þ
oxi Eqs. (19) and (20). Using regression analysis, a poly-
nomial equation was fitted to the measured mean data.
whereas the Reynolds stress equation is given by The resulting best fit equation for the respective profile
 (temperature or velocity) was differentiated at the de-
oui ouj 2
u0i u0j ¼ mt þ  dij k ð17Þ sired positions to obtain the temperature and velocity
oxj oxi 3
gradients. The turbulent diffusivity and viscosity data in
These hypotheses (Eqs. (16) and (17)) are valuable Fig. 11 were made dimensionless by the kinematic vis-
concepts, whose limitations should always be borne in cosity of air.
mind. The turbulent diffusion hypothesis (Eq. (16)) im- The distribution of the turbulent Prandtl number,
plies that the scalar flux vector (in the present study Prt , which is also shown in Fig. 11, is given by
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3569

10 2
αt/ν
νt/ν
8
Prt

6 1
αt/ν, νt/ν

Pr t
4

2 0

-2 -1
0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035

Distance from hot wall, X=x/L

Fig. 11. Turbulent diffusivity, turbulent viscosity and turbulent Prandtl number.

mt 3.6. Dissipation rate of the temperature variance, eh

Prt ¼ ð21Þ
at
In numerical studies, the strong coupling of the
When considering the structure of turbulent heat
boundary layer and the cavity core flow makes compu-
transfer, these quantities are essentially important but
tation very difficult for this wall bounded shear flow.
have not been shown until now for a low level turbulent
Direct simulation of turbulent natural convection in a
natural convection boundary layer along the isothermal
cavity is still too costly and numerical results from
walls of an air filled square cavity. As pointed out above,
various k–e models are not identical because the two-
the concept of turbulent diffusivity and turbulent vis-
equation models cannot account correctly for the non-
cosity defined by Eqs. (19) and (20) is introduced under
homogenous flow characteristics. Hanjalic et al. [26]
the condition that turbulent heat flux, v0 T 0 , or Reynolds
proposed three- and four-equation models, k–e–T 02 and
stress, u0 v0 , has a close relation with the mean tempera-
k–e–T 02 –eh , to solve this problem. Their results were
ture gradient or the mean velocity gradient. In the nat-
compared with experimental data for Nusselt number,
ural convection boundary layer, as shown in Fig. 10,
mean velocity and temperature. However, no compari-
Reynolds stress takes a very small value of almost zero
son was made for the turbulent quantities, possibly due
near the wall, in the presence of a large mean velocity
to the limitations of experimental data. The dissipation
gradient and has a positive value at the maximum ve-
rate of the temperature variance, eh , has the form
locity location (ov=ox ¼ 0) without u0 v0 becoming zero.
Therefore, the distribution of turbulent viscosity for  0 2
oT
momentum has a discontinuity at the maximum velocity eh ¼ 2a
oxi
location. On the other hand, turbulent diffusivity for " 2  0 2  0 2 #
heat shows a continuous profile. As a consequence, the oT 0 oT oT
¼ 2a þ þ ð22Þ
distribution of turbulent Prandtl number defined as a ox oy oz
ratio of turbulent viscosity and turbulent diffusivity also
has a discontinuity at the maximum velocity location. where a is the fluid thermal diffusivity. This equation
The significance of introducing turbulent diffusivity, indicates that at least two-point measurements are nee-
turbulent viscosity and turbulent Prandtl number is ded for a direct experimental estimation. The dissipation
poor for the inner layer from the wall to the maximum rate of the temperature variance, eh , in the cavity could
velocity location. In the comparatively wide region not be measured directly in this study due to time con-
(X ¼ 0:015–0:03 of Fig. 11), turbulent viscosity and straint and as such, based on the measured temperature
turbulent diffusivity have a similar profile and so the profiles in the cavity, an estimate would be made for this
turbulent Prandtl number takes a value of about unity. quantity near the isothermal walls (ehw ) in this section.
3570 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

Isotropic turbulent scalar fields have direction-indepen- temperature fluctuation profile very close to a wall of a
dent mean-square scalar gradients due to the rotational 2D flow in cavities has no significant variation in the
symmetry. It is therefore sufficient for isotropic fields to direction along the wall, see Fig. 12(a) and as such it can
measure only one of the three gradients (usually the be expressed as follows (using TaylorÕs series):
streamwise gradient using TaylorÕs hypothesis), from
which the scalar dissipation, eh , is then given by the T 0 ¼ ai x þ bi x2 þ    ð24Þ
isotropic relation
 0 2 where the coefficients ai and bi are functions of time and
oT
eh ¼ 6a ð23Þ their time averages are zero and x is distance from the
ox
wall. The dissipation rate of the temperature variance at
In this study, the flow is not 100% isotropic and as such the wall, ehw , can then be calculated from the experi-
a different approach will be used to estimate eh . The mental data using

27

24
Temperature fluctuation,T'rms(K)

21 Y=0.1, T'rms+0
Y=0.2, T'rms+3
18
Y=0.3, T'rms+6
15 Y=0.4, T'rms+9
Y=0.5, T'rms+12
12 Y=0.6, T'rms+15
Y=0.7, T'rms+18
9 Y=0.8, T'rms+21
Y=0.9, T'rms+24
6

0
0.000 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.010

(a) Distance from hot wall, X=x/L

8
Hot wall
Cold wall
Dissipation rate of temperature variance (K /s)
2

0
0.0 0.2 0.4 0.6 0.8 1.0

(b) Distance from bottom wall, Y=y/L

Fig. 12. Temperature fluctuation profiles with estimated dissipation rate of the temperature variance near the walls: (a) temperature
fluctuation profiles at different heights near the hot wall and (b) estimated dissipation rate of the temperature variance near the walls.
 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572 3571

 0 2   0 2 
oT  2a X
N
2 oTrms  also has a discontinuity at this location. However, be-
ehw ¼ 2a   a  2a  ð25Þ yond the maximum velocity location, turbulent viscosity
ox  N i¼1 i ox 
w w and turbulent diffusivity have a similar profile and so the
and turbulent Prandtl number takes a value of about unity.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi From the above comparison and discussion, it can be
u stated that the k–e model and LES can predict the mean
u1 X M
0
Trms ¼ ðT 02 Þ1=2 ¼t T 02 ð26Þ quantities such as velocity and temperature in the pre-
N i¼1 i
sent cavity reasonably well but cannot predict the fluc-
tuation and turbulence quantities. LES predicts the flow
where N in the number of readings taken in the experi-
quantities much better than the k–e model. The results
ments––10 000 in this study. The results based on
described in this paper were obtained with high precision
the present measurements are shown in Fig. 12(b). The
and can be useful as benchmark data for comparison
difference between the hot and cold walls is obvious; the
with CFD codes. Such numerical simulation should in-
average value for ehw is 5.47 at the hot wall and 3.85 at
clude realistic boundary conditions at the top and bot-
the cold wall, i.e. 30% higher. It can be seen from this
tom horizontal walls as given in this study.
figure that for a major part of the walls the dissipation is
nearly constant along the boundary layer. To the best of
the authorsÕ knowledge, this is the first experimental
evidence relating to the dissipation rate of the temper- References
ature variance at a wall and could be useful for nu-
merical modelling where T 02 –eh equations are involved. [1] M.W. Nansteel, R. Grief, Natural convection in enclosures
with two- and three-dimensional partitions, Int. J. Heat
Mass Transfer 27 (4) (1984) 561–571.
[2] D.A. Olson, L.R. Glicksman, H.M. Ferm, Steady-state
4. Conclusions natural convection in empty and partitioned enclosures at
high Rayleigh numbers, ASME J. HT 112 (1990) 640–647.
The present study shows that for this kind of cavity [3] F. Penot, A. NÕDame, Successive bifurcations of natural
the inner boundary layer along the isothermal walls is convection in a vertical enclosure heated from the side, in:
less than 7% of the outer boundary layer. Also, the Heat Transfer, Third UK National Conference Incorpo-
viscous layer, which next to the isothermal walls, is rating First European Conference on Thermal Sciences,
Birmingham, UK, vol. 1, 1992, pp. 507–513.
about 3 mm (60% of the inner layer) with 2 mm of it
[4] Y.S. Tian, T.G. Karayiannis, Low turbulence natural
being the conductive layer. The turbulent quantities are convection in an air filled square cavity, Part I: Thermal
almost negligible in the conductive layer, which confirms and fluid flow fields, Int. J. Heat Mass Transfer 43 (2000)
that the heat transfer in this region of the boundary 849–866.
layer is by pure conduction. Overall, the local Nusselt [5] Y.S. Tian, T.G. Karayiannis, Low turbulence natural
number reaches a maximum at the bottom of the hot convection in an air filled square cavity, Part II: The
wall and at the top of the cold wall because of a thinner turbulence quantities, Int. J. Heat Mass Transfer 43 (2000)
thermal boundary there. 867–884.
The study also shows that at the bottom of the hot [6] S.B. Pope, Turbulent Flows, Cambridge University Press,
wall (top of the cold wall) the turbulent heat flux is ef- 2000.
[7] B.E. Launder, B.I. Sharma, Application of the energy
fectively zero. The absence of a turbulent heat flux in
dissipation model of turbulence to the calculation of flow
these regions suggests that the lower part of the hot wall near a spinning disc, Letters in Heat and Mass Transfer 1
(upper part of the cold wall) boundary layer has many of (2) (1974) 131–138.
the characteristics of a laminar flow. In general, the [8] T. Cebeci, P. Bradshaw, Physical and Computational
horizontal turbulent heat flux is one order of magnitude Aspects of Convective Heat Transfer, Springer-Verlag,
smaller than the vertical turbulent heat flux. Berlin, 1988.
The turbulent diffusivity, turbulent viscosity and [9] H.P. Kreplin, H. Eckelmann, Behaviour of the three
turbulent Prandtl number are essentially important fluctuating velocity components in the wall region of a
quantities when considering the structure of turbulent turbulent channel flow, Fluid 22 (1979) 1233–1239.
heat transfer. However, as the present study has shown, [10] P.R. Spalart, Direct simulation of a turbulent boundary
layer up to Rh ¼ 1410, J. Fluid Mech. 187 (1988) 61–98.
the significance of introducing these quantities is limited
[11] S. Mergui, F. Penot, Natural convection in a differentially
for the inner layer, i.e. from the wall to the maximum
heated square cavity: experimental investigation at
velocity location. This is due to the fact that the distri- RaH ¼ 1:69
109 , Int. J. Heat Mass Transfer 39 (3)
bution of turbulent viscosity for momentum has a dis- (1996) 563–574.
continuity at the maximum velocity location and as such [12] F. Ampofo, Turbulent natural convection in an air filled
the distribution of turbulent Prandtl number defined as standard or partitioned square cavity, PhD Thesis, South
a ratio of turbulent viscosity and turbulent diffusivity Bank University, December 2001.
3572 F. Ampofo, T.G. Karayiannis / International Journal of Heat and Mass Transfer 46 (2003) 3551–3572

[13] S. Mergui, F. Penot, J.L. Tuhault, Experimental natural [20] W.K. George, S.P. Capp, A theory for natural convection
convection in an air-filled square cavity at RaH ¼ 1:7
109 , turbulent boundary layer next to heated vertical surfaces,
in: R.A.W.M. Henkes, C.J. Hoogendoorn (Eds.), Turbu- Int. J. Heat Mass Transfer 22 (1979) 813–826.
lent Natural Convection in Enclosures––A Computational [21] J.T. Davies, Turbulence Phenomena, Academic Press, New
and Experimental Benchmark Study, Proceedings of the York, 1972.
Eurotherm Seminar No. 22, Delft, The Netherlands, [22] G.S. Barozzi, E. Nobile, A.C.M. Sousa, Contribution to
March 25–27, 1992, pp. 97–108. the numerical simulation of turbulent natural convection in
[14] A.M. Lankhorst, Laminar and turbulent natural convec- rectangular enclosures, in: R.A.W.M. Henkes, C.J. Hoog-
tion in cavities––numerical modelling and experimental endoorn (Eds.), Turbulent Natural Convection in Enclo-
validation, PhD Thesis, Technology University of Delft, sures––A Computational and Experimental Benchmark
1991. Study, Proceedings of the Eurotherm Seminar No. 22,
[15] C. Beghein, F. Allard, A. Draoui, Numerical modelling of Delft, The Netherlands, March 25–27, 1992, pp. 19–30.
turbulent convection in a thermally-driven square cavity, [23] S.H. Peng, L. Davidson, Numerical investigation of
in: R.A.W.M. Henkes, C.J. Hoogendoorn (Eds.), Turbu- turbulent buoyant cavity flow using LES, in: 3rd Interna-
lent Natural Convection in Enclosures––A Computational tional Symposium on Turbulence, Int. J. Heat Mass
and Experimental Benchmark Study, Proceedings of the Transfer (2000) 1–8.
Eurotherm Seminar No. 22, Delft, The Netherlands, [24] S. Vasic, K. Hanjalic, Turbulent natural convection in a
March 25–27, 1992, pp. 31–42. square cavity––benchmark computations, in: R.A.W.M.
[16] S. Ziai, Turbulent natural convection in a large rectangular Henkes, C.J. Hoogendoorn (Eds.), Turbulent Natural
air cavity, PhD Thesis, Queen Mary College, University of Convection in Enclosures––A Computational and Exper-
London, 1983. imental Benchmark Study, Editions Europeennes Thermi-
[17] K.J. King, Turbulent natural convection in rectangular air que et Industrie, 1993, pp. 133–144.
cavities. PhD Thesis, Queen Mary College, University of [25] S. Tavoularis, S. Corrsin, Experiments in nearly homo-
London, 1989. geneous turbulent shear flow with a uniform mean
[18] S. Paolucci, Direct numerical simulation of two-dimen- temperature gradient, Part 1, Fluid Mech. 104 (1981) 311–
sional turbulent natural convection in an enclosed cavity, J. 347.
Fluid Mech. 215 (1990) 229–262. [26] K. Hanjalic, S. Kenjeres, F. Durst, Natural convection in
[19] I.N.G. Wardana, T. Ueda, M. Mizomoto, Velocity–tem- partitioned two-dimensional enclosures at higher Rayleigh
perature correlation in strongly heated channel flow, Exp. number, Int. J. Heat Mass Transfer 39 (7) (1996) 1407–
Fluids 18 (1995) 454–461. 1427.