ARTICLE IN PRESS

Building and Environment 41 (2006) 11711181

www.elsevier.com/locate/buildenv

Verication of a CFD model for indoor airow and heat transfer

A. Stamou, I. Katsiris
School of Civil Engineering, National Technical University of Athens (NTUA), Heroon Polytechniou 5, 15780 Athens, Greece
Received 20 April 2004; received in revised form 17 June 2005; accepted 23 June 2005

Abstract
The SST ko based model is applied to calculate air-ow velocities and temperatures in a model ofce room. Calculations are
compared with experiments and with the results of the standard ke, the RNG ke model and the laminar model. It is concluded that (a)
all the three tested turbulent models predict satisfactorily the main qualitative features of the ow and the layered type of temperature
elds and (b) computations with the SST ko based model show the best agreement with measurements. The use of this model is
proposed combined with a suitable grid.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Computational uid dynamics (CFD); Mathematical models; Indoor environment; Ofce spaces; Thermal comfort; CFX

1. Introduction
In ventilated interior environments of buildings, the
determination of air-ow velocities, temperatures and
concentrations of pollutants is required to evaluate
comfort conditions (thermal and draught) and indoor air
quality. This determination can be performed with
computational uid dynamics (CFD) methods. In the last
15 years, a signicant number of papers has been published
on the application of CFD methods in pilot, experimental
or real scale interior environments with considerable
success.
In general, the main types of CFD methods are the
following: direct numerical simulation (DNS), large eddy
simulation (LES) and Reynolds averaged NavierStokes
(RANS).
1.1. Direct numerical simulation
Most ows encountered indoors are turbulent, characterized by eddies with a wide range of length and time
scales. The largest eddies are typically comparable in size to
the characteristic length of the mean ow (such as the
dimensions of the interior spaces) and the smallest scales
Corresponding author. Tel.: +30 210 7722809; fax: +30 210 7722814.

E-mail address: stamou@central.ntua.gr (A. Stamou).

0360-1323/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.buildenv.2005.06.029

(which are responsible for the dissipation of turbulence

kinetic energy) are of the order of the Kolmogorov microscale. It is theoretically possible to directly (without any
approximations) resolve the whole spectrum of turbulent
scales by solving the exact NavierStokes equations, using
the approach of DNS. However, DNS is not feasible for
indoor environment spaces, because it requires a very ne
grid resolution, which is prohibitive for the current
computers.
Two alternative methods can be employed to transform
the exact NavierStokes equations in such a way that the
small-scale turbulent uctuations do not have to be directly
simulated: Filtering and Reynolds averaging.
1.2. Filteringlarge eddy simulation
Filtering is a manipulation of the exact NavierStokes
equations to remove the eddies, which are smaller than the
size of the lter (usually taken equal to the grid size). The
ltered equations are used to compute the large eddies
(large eddy simulation, LES) and the small eddies are
modeled independently of the ow geometry. This results
in grid sizes that are at least one order of magnitude smaller
than with DNS, but extremely ne grids are still required.
The application of LES to indoor environments is in its
infancy. Only a few applications exist to date, which are
restricted to very simple geometries (due to the large

ARTICLE IN PRESS
1172

A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

computer resources required to resolve the energy-containing turbulent eddies) and use high-order spatial discretisation schemes (with great care being taken to resolve all
scales larger than the inertial sub-range). Furthermore, the
use of wall functions with LES is an approximation that
requires further validation. Zhang and Chen [1] used a LES
model with a ltered dynamic sub-grid scale model and a
second-order explicit differencing scheme to calculate
natural, forced and mixed convection ows in rooms
having the simple geometry of a cavity. It was concluded
that LES has a good potential to simulate indoor airow
in the near future, due to the explosive increases in
computer hardware performance coupled with the availability of parallel processing.

1.3. Reynolds averaged NavierStokes

RANS equations represent transport equations only for
the mean ow quantities with all the scales of the
turbulence being modeled. The RANS approach has been
applied in the majority of the existing indoor airow CFD
calculations using turbulence models, such as the standard
ke model and its variants, the standard ko and its
variants, and the Reynolds-stress model (RSM).
The standard ke model is a semi-empirical model, which
is valid only for fully turbulent ows. The model is based
on model transport equations for the turbulent kinetic
energy, k, and its dissipation rate, e. The turbulent viscosity
is computed from these scalars. The ke formulation is
derived using a high Reynolds number hypothesis; also
near wall treatment is based on the application of wall
functions, rather than solving the governing equations
inside the boundary layer. This model has been extensively
used in the rst CFD applications in indoor environments
performed in the period 19781998; see [2] for a brief
literature review.
The standard ke model has proven very successful for
numerous engineering applications. However, certain
characteristics of indoor airow, such as the creation of
regions with very low velocities and thus low Reynolds
numbers, especially near the wall boundaries, require the
use of more effective models. This requirement led to the
formulation of modied ke turbulence models, which are
expected to be more effective for such regions. These
models are the low-Reynolds number ke model (LR ke)
and the RNG ke model.
The LR ke model differs from the standard ke model
in the values of the empirical coefcients. A disadvantage
of the model is that one of the damping functions requires
the calculation of the local distance to the nearest wall.
Costa et al. [3] have tested eight LR ke models to simulate
the mixed convection airow generated by two nonisothermal plane wall jets. It was concluded that the model
of Nagano and Hishida [4] provided the best overall performance, although suffering from singular defects occurring near separation/reattachment points of the ow.

The RNG ke model is derived using a rigorous

statistical method called renormalization group (RNG).
It involves a theory for the large scales, in which the effects
of the small scales are represented by modied transport
coefcients. A basic characteristic of the RNG ke model is
that it involves an analytically derived differential formula
for effective viscosity that accounts for low-Reynoldsnumber effects. This feature combined with an appropriate
treatment of the near-wall region makes the RNG ke
model more accurate and reliable for a wider class of
indoor airows than the standard ke model. Chen [5]
compared ve different ke models, including the standard
ke, the LR ke and the RNG ke model. He recommended
only the RNG ke model for simulations of indoor air-ow
and noted that the performance of the other models was
not stable. Rouaud and Havet [6] showed that both the
standard ke and the RNG ke model predict well the main
features of the ow in clean rooms. They also claimed that
the RNG ke seems to be more suitable, while the standard
ke model overestimates turbulent diffusion. Gebremedhin
and Wu [2] have evaluated ve RANS models (the ke, the
RNG ke, the LR ke, the ko and the RSM) with the
code PHOENICS [7]. Based on convergence and computational stability criteria, they concluded that the RNG ke
model is the most appropriate model that characterizes the
ow eld in a ventilated space. Cheong et al. [8] evaluated
the current thermal comfort conditions of an air-conditioned lecture theatre, using the code Fluent [9] and the
RNG ke model. Calculations of airow characteristics
and temperature gradients were in fair agreement with
empirical measurements. Posner et al. [10] have evaluated
the laminar, the standard ke and the RNG ke models
with respect to their performance in simulating the ow in
a model room. Their simulations using the code Fluent [9]
with the laminar and the RNG ke models agreed better
with experimental data than calculations with the standard
ke model.
The standard ko model [11] is based on model
transport equations for k and the turbulence frequency,
o. The turbulent viscosity is computed from these scalars.
The model is numerically stable, especially the lowReynolds number version, as it tends to produce converged
solutions more rapidly than the ke model; and the lowReynolds number version is more efcient than the LR ke
model in that it does not require calculations of wall
distances, additional source term and/or damping functions based on the friction velocity. The main weakness of
the ko model is its strong sensitivity to free-stream
conditions. The application of the ko model in indoor
environments was not successful [2].
The SST model [12] combines the ke and the ko
models using a blending function. The SST model activates
the ko model in the near-wall region and the ke model
for the rest of the ow. By this approach, the attractive near-wall performance of the ko model is utilized
without the potential errors resulting from the free
stream sensitivity of that model. No work was found in

ARTICLE IN PRESS
A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

the literature on the use of the SST model in indoor

environments.
The RSM allows the development of transport and
individual Reynolds stresses. It has the advantage of
automatically introducing terms accounting for anisotropic
effects into the stress transport equations, which play a
very important role in ows with signicant buoyancy,
streamline curvature, swirl or strong circulation. A full
RSM closure generally consists of six transport equations
for the Reynolds stressesthree transport equations for
the turbulent uxes of each scalar property and one
transport equation for the dissipation rate of turbulence
energy. Chen [13] compared three different RSM models
and the standard ke for predicting typical airow in a
room. It was shown that the computations and the
experimental data are in agreement for the mean air
velocities, but less satisfactory for the turbulent quantities.
The additional computing effort required by the RSM was
noted. Moureh and Flick [14] studied the airow generated
by a wall jet within a long and empty slot-ventilated
enclosure on a scale model using the code Fluent [9] and
the turbulence models ke, RNG ke and RSM. It was
demonstrated that only the RSM model enabled detection
of the presence and the localization of separated ow and
correctly predicts airow patterns. Generally, RSM can
provide a more realistic and rigorous approach to solving
complex engineering ows, but storage and execution time
are very expensive especially for real scale 3D indoor ows.
In the present work the SST ko [12] based model is used
to predict air-ow velocities and temperatures in a model
ofce room. Calculations are compared with experiments
[15] and with the results of the standard ke, the RNG ke
model and the laminar model.
2. Experimental data
The experimental data (case 6a) of Loomans [15] were
used for the verication of the model. The experiments
have been performed in a model ofce room with the
rectangular geometry shown in Fig. 1(a).
The main components of the model ofce room are the
following:
(1) Air supply and exhaust: The air was introduced into the
room via a supply unit, placed under the desk table and
exited via an exhaust outlet, located in the South-East
upper corner of the room. The supply set-point
temperature was controlled and the ow rate has been
determined (with 0.5% accuracy) with a differential
pressure transducer. Two positive temperature coefcient resistors (PTCs) have been used to measure the
supply and the exhaust air temperatures.
(2) Heat sources: Four types of heat sources were used in
the experimental conguration to simulate the heat
load in the model ofce room; a thermal mannequin,
two PC-simulators, a lighting simulator and three
uorescent lamps. For each heat source the radiant

1173

part of the total heat input has been determined via

infrared thermography and then the convective part
has been calculated using the radiant heat input data
and other information, such as uorescent lamps
manufacturer data.
(3) Walls: The walls of the model ofce room were at and
temperature controlled. Ceiling, oor and wall temperatures could be controlled within a standard
deviation of 0.10, 0.10 and 0.15 1C, respectively.
The basic characteristics of the measurements are the
following:
(1) Air velocities: Steady-state velocity measurements have
been obtained at 132 locations using 12 hot sphere
anemometers, which have been calibrated in vertical
and horizontal position. The accuracy of the measurements was determined at 70.025 m/s for velocities
ranging between 0.05 and 0.50 m/s.
(2) Air and wall temperatures: Steady-state air temperature
measurements have been measured at 520 locations with
40 thermocouples and wall temperatures with 12 positive
temperature coefcient resistors (PTCs). Calibration of
the thermocouples and PTC has been performed using
the reference principle. The accuracy of the temperature
measurements was determined at 0.1 and 0.125 1C for
the thermocouples and the PTCs, respectively.

3. Calculations
3.1. The computer code
There are various efcient computer codes [7,9,16],
which are frequently used for indoor CFD calculations.
In the present work, the latest version of the computer code
CFX [16] is used.
The code calculates the 3D ow eld and heat transfer
using the continuity, momentum and energy equations.
The nite control-volume method is implemented for the
spatial discretisation of the domain. The continuity,
momentum and energy equations are integrated over each
control volume, such that the relevant quantity (mass,
momentum, energy, etc.) is conserved, in a discrete sense,
for each control volume. For the continuity equation
(pressurevelocity coupling) a second order central difference approximation is used, modied by a fourth order
derivative in pressure, which redistributes the inuence of
pressure. The second-order upwind Euler scheme approximates the transient term CFX [16]. A scalable and fully
implicit coupled solver is used for the solution of the
equations, which is one of the basic advantages of the code.
3.2. Boundary conditions and the grid
The geometry used in the computations (see Fig. 1(b))
is approximately the same with the original geometry

ARTICLE IN PRESS
1174

A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

Fig. 1. Geometry of the model ofce room: (a) original geometry (obtained from Loomans [15]) and (b) geometry used in the simulations.

(see Fig. 1(a)). Boundary conditions are dened at the main

components of the model room (see Section 2). At the inlet
the air supply unit is divided in six segments, where the
values of the air velocities are set equal to the experimental
[15], resulting to a total airow rate equal to 0.047 m3/s.
The air supply temperature is set equal to 19.8 1C. The
distributions of k, e and o are assumed to be uniform. At
the outlet the average static pressure is set equal to the
atmospheric pressure and the vertical gradients of all other
variables are set equal to zero.
The boundary temperatures are specied to represent the
actual temperatures of the room surfaces. The temperatures of the oor, the ceiling, the northern, the southern,
the western and the eastern walls are set equal to 22.2, 22.3,
23.2, 22.8, 22.6 and 22.7 1C, respectively. Thus, no
radiation model is used. Heat uxes are modeled to
represent the actual amount of heat generated by the four
types of heat sources in the room. The thermal mannequin
has a surface area of 1.6 m2 and is modeled to generate
convective heat equal to 59.8 W. This value corresponds to
37.4 W/m2. Heat emitted by the PC-simulator and the
lighting simulator is set equal to 61.5 and 10.9 W,
respectively. Each uorescent lamp generates 18.1 W.

An automatic, unstructured hybrid element mesh generator

with an adaptive mesh renement algorithm is employed,
which permits a very accurate representation of the boundaries. The advantage of using unstructured mesh is the
minimization of numerical errors and the consistency of the
solution throughout the domain. The numerical grid consists
of approximately 300 000 unstructured elements (tetrahedrals,
prisms and pyramids) with grid renement on the surface of
the heat sources and in the inlet and outlet regions.
3.3. Results and discussion
In Figs. 2(a)(c) the calculated ow eld using the SST
ko model is shown in the central yz plane (at x 1:80 m),
in a horizontal xy plane (at z 0:10 m above oor level)
and in a xy plane (at z 2:40 m, i.e. 0.10 m beneath ceiling
level). In Figs. 2(d) and (e) the experimental ow eld,
derived using video recordings of the diffusion of smoke
[15] is shown, at the same planes as in Figs. 2(a) and (b).
In Figs. 3 and 4 both velocity vectors (arrows on
separate equidistant grids) and temperature contours are
shown, respectively, at two planes; the central yz plane and
plane xz at y 2:4 m.

ARTICLE IN PRESS
A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

1175

Fig. 2. Qualitative comparison of calculated and experimental ow eld: (a) calculations: central yz plane, (b) calculations: xy plane (z 0:10 m),
(c) calculations: xy plane (z 2:40 m), (d) measurements: central yz plane, and (e) measurements: xy plane (z 0:10 m).

In Figs. 5 and 6 calculated magnitudes of velocities and

temperature values, respectively, are compared with
experiments at various locations.
3.3.1. Flow field calculations
The calculated ow eld of Figs. 2(a)(c) shows the
characteristics of typical displacement ventilation ows and
is characterized by the following four main regions:
1. The region of the incoming ow and the oor layers.
The incoming ow (19.8 1C) is colder than the room air
temperature (approximately 2224 1C). The cool airow
enters the room via the inlet, bends downwards due to
buoyancy and forms a oor layer (see Figs. 2(a), (d) and
3), which ows along the oor with velocities of ranging
from 0.05 to 0.13 m/s (measured values 0.070.11 m/
s). The oor layer strikes the western wall and creates a
rebound effect, which results in the formation of a
second layer owing in the opposite direction. This type
of ow is similar to buoyant water ows [17].

2. The region of the buoyant plumes. Figs. 2(a), (b), 3 and

4 show that in the region of the mannequin a large
buoyant vertical plume is formed, originating from three
heat sources; the thermal mannequin, the PC-simulator
and the lighting simulator. The vertical velocities of the
plume range from 0.15 m/s to 0.18 m/s (measured
values 0.150.22 m/s). A smaller plume originates
from the second PC-simulator.
3. The region of the ceiling layer and the outlet ow. The
two plumes are directed upwards and strike on the
ceiling, where they form a layer owing along the ceiling
towards all directions (see Figs. 2(c), 3 and 4) with
velocities ranging from 0.06 to 0.12 m/s. The main part
of this ow leaves the room via the outlet.
4. The region of the entrained ow and the re-circulation
regions. The ow of the two oor layers remains near
oor level and is carried towards the ceiling by entrainment into the two buoyant plumes (see Figs. 2(a), (b), 3
and 4). The two buoyant plumes, mainly the one
originating from the mannequin, can entrain only a

ARTICLE IN PRESS
1176

A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

Fig. 3. Temperature contours and velocity vectors at the central yz plane.

Fig. 4. Temperature contours and velocity vectors at a xz plane (y 2:40 m).

part of the incoming fresh air. The remaining air moves

following paths, which are determined by the created recirculation regions with velocities smaller than 0.05 m/s
(experimental values p0.05 m/s).
Assuming that no air can entrain the ow near the oor,
an interface is found at the height, where the supply ow
rate equals the convection ow rate of the plumes. The

interface height was calculated equal to 1.2 m. This value

coincides with the estimation of Loomans [15].
3.3.2. Temperature calculations
Figs. 3 and 4 show the strong dependence of the
temperature eld on the ow eld and vice-versa. The ow
pattern described in the previous section results in the
formation of three main temperature layers, as depicted by

ARTICLE IN PRESS
A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

y = 1.50m

y = 2.40m

y = 3.00m

1177

y = 4.50m

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

2.5

0.0
0.0

0.1

0.2

0.0
0.0

0.1

0.2
x = 0.93 m

0.0
0.0

0.1

0.2

0.0
0.0

2.5

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.0
0.0

0.0
0.0

0.0
0.0

0.1

0.2

0.1

0.2
x = 1.80 m

0.1

0.2

0.0
0.0

2.5

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.0
0.0

0.1

0.2

0.0
0.0

0.1

0.2
x = 2.05 m

0.0
0.0

0.1

0.2

0.0
0.0

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.2

0.1

0.2

0.1

0.2

0.1

0.2

0.1

0.2

2.5

0.1

0.0
0.0

0.1

0.2

0.0
0.0

0.1

0.2
x = 2.30 m

0.0
0.0

0.1

0.2

0.0
0.0

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

2.5

0.0
0.0

0.1

0.2

0.0
0.0

0.1

0.2
x = 2.68 m

0.0
0.0

0.1

0.2

0.0
0.0

Fig. 5. Calculated air velocity magnitudes vs. measurements (m/s).

the iso-temperature lines of Figs. 3 and 4; the rst up to a

height of 0.60 m with temperatures ranging from 21 to 22 1C,
the second from a height of 0.60 m up to 1.00 m with
temperatures ranging from 22 to 23 1C and the third layer
from a height of 1.0 m up to 2.5 m with temperatures ranging

from 23 to 24 1C. Higher temperatures (2425 1C) are

observed in the region of the buoyant jets and the ceiling
layer, while low temperatures (2021 1C) are found in the
region of the incoming ow. The exhaust temperature was
calculated equal to 23.2 1C (measured value 23.4 1C).

ARTICLE IN PRESS
A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

1178

y = 1.50 m

y = 2.25 m

y = 3.00 m

y = 3.75 m

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.0
20 21 22 23 24 25

0.0
20 21 22 23 24 25

0.0
20 21 22 23 24 25

0.0
20 21 22 23 24 25

2.5

x = 0.68 m
2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.0
20 21 22 23 24 25

0.0
20 21 22 23 24 25

0.0
20 21 22 23 24 25

0.0
20 21 22 23 24 25

2.5

x = 1.43 m

z
z
z

2.5

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.0
20 21 22 23 24 25

0.0
0.0
20 21 22 23 24 25
20 21 22 23 24 25
x = 1.93 m

0.0
20 21 22 23 24 25

2.5

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.0
20 21 22 23 24 25

0.0
0.0
20 21 22 23 24 25
20 21 22 23 24 25
x = 2.43 m

2.5

2.5

2.5

2.5

2.0

2.0

2.0

2.0

1.5

1.5

1.5

1.5

1.0

1.0

1.0

1.0

0.5

0.5

0.5

0.5

0.0
20 21 22 23 24 25

0.0
0.0
20 21 22 23 24 25
20 21 22 23 24 25
x = 2.93 m

0.0
20 21 22 23 24 25

0.0
20 21 22 23 24 25

Fig. 6. Calculated temperature proles vs. measurements (1C).

In the whole room, with the exception of the region of

the buoyant plumes, the iso-temperature lines are almost
horizontal, thus showing that the horizontal temperature
differences are small compared to the vertical ones.

3.3.3. Comparison of computations with measurements

The calculated ow eld of Figs. 2(a) and (b) agrees
qualitatively with the experimental, shown in Figs. 2(d)
and (e).

ARTICLE IN PRESS
A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

1179

Figs. 5 and 6 depict a satisfactory agreement between

calculated proles for air velocity magnitudes and temperatures, respectively, in all regions of the ow.
For the total of 142 velocity and 520 temperature data,
the average velocity and temperature prediction errors
using the SST ko model and the typical grid (300 000
control volumes) were calculated equal to 70.03 m/s and
70.33 1C (71.4%), respectively. These values verify the
satisfactory performance of the model, taking into account
the accuracy of the measurements.
3.3.4. Effect of the turbulence model
Calculations have been repeated using the standard ke,
the RNG ke and the laminar model. In Fig. 7 temperature
contours and velocity vectors are shown at the central yz
plane for all the four models tested. In Fig. 8 the temporal
variation of the value of the main (vertical) velocity at a
point, which located in the region of the buoyant plume
(x 1:80 m, y 2:58 m and z 1:80 m), is shown. This
gure provides an indication on the stability of the
calculations.
From the comparison of the results and Figs. 7 and 8,
the following remarks can be made:
(1) All three turbulent models predict satisfactorily the
main qualitative features of the ow and the layered
type of temperature elds, as shown in Fig. 7.
(2) The SST ko based model shows the best agreement
with measurements and the laminar model the worst.
The average prediction errors for temperatures with the
standard ke, the RNG ke and the laminar model were
equal to 70.56, 70.72 and 70.94 1C, respectively. The
corresponding exhaust temperatures were calculated
equal to 24.3, 24.5 and 24.6 1C, respectively.
(3) Convergence has been achieved with all models. For
the standard ke, the RNG ke and the laminar model
convergence was achieved only after the modication
of the outlet boundary condition, by setting the outlet
ow equal to the inlet ow. Furthermore, for these
models a completely steady state solution could not be
attained, as shown in Fig. 8.
It should be mentioned that universality cannot be
claimed for these remarks, since the performance of the
tested models is problem dependent.
3.3.5. Effect of the grid
Calculations have been performed with various sizes of
grids, ranging from 120 000 to 1 200 000 control volumes.
Grid independence, within a relative variation of less than
3% in temperature values, has been achieved by using
approximately 300 000 control volumes. This size is
signicantly larger than the common values used in the
majority of the previous cases, which are reported in
Section 1.
Calculations with the SST ko based model required a
resolution of a minimum number of 30 layers in the wall

Fig. 7. Temperature contours and velocity vectors at the central yz plane:

(a) SST ko model, (b) ke model, (c) RNG ke model, and (d) laminar
model.

region up to y 15 to capture the rapid variations of the

variables (velocities and temperatures) and thus to predict
satisfactorily the proles near the walls.

ARTICLE IN PRESS
A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181

1180

900 Mb and 2 Gb for the typical grid (300 000 control

volumes) and the nest grid (1 200 000 control volumes),
respectively. For the typical grid the average time-step used
in the calculations was 1.5 s. 1000 iterations were required
to achieve steady-state conditions, corresponding to a total
simulation time of 25 min. The total computer time was
approximately 23 h.

0.25
0.2
0.15
0.1
0.05
0

4. Conclusions
0

200

400

600

800

1000

The SST ko based model is used to predict air-ow

velocities and temperatures distributions in a model ofce
room. From the comparison of the calculations with
experiments [15] and with the results using the standard
ke, the RNG ke and the laminar model the following
conclusions are drawn:

-0.05
(a)
0.25
0.2
0.15
0.1
0.05
0

200

400

600

800

1000

-0.05
(b)
0.25

(1) Calculations show that all the tested turbulent models

predict satisfactorily the main qualitative features of
the ow and the layered type of temperature elds.
Thus, all these models can be used for practical
purposes.
(2) Calculations with the SST ko based model show the
best agreement with measurements and the laminar
model the worst. The use of SST ko based model is
proposed combined with a suitable grid.

0.2

References

0.15
0.1
0.05
0

200

400

600

800

1000

200

400

600

800

1000

-0.05
(c)
0.25
0.2
0.15
0.1
0.05
0
-0.05
(d)
Fig. 8. History of vertical velocity at a monitoring point: (a) SST ko
based model, (b) ke model, (c) RNG ke model, and (d) laminar model.

3.3.6. Characteristics of the computations

All calculations have been performed with a single PC
with a 2.4 GHz P4 processor. The required RAM was

[1] Zhang W, Chen Q. Large eddy simulation of indoor airow with a

ltered dynamic sub-grid scale model. International Journal of Heat
and Mass Transfer 2000;43(17):321931.
[2] Gebremedhin KG, Wu BX. Characterization of ow eld in a
ventilated space and simulation of heat exchange between cows and
their environment. Journal of Thermal Biology 2003;28(4):30119.
[3] Costa JJ, Oliveira LA, Blay D. Test of several versions for the ke type
turbulence modeling of internal mixed convection ows. International
Journal of Heat and Mass Transfer 1999;42(23):4391409.
[4] Nagano Y, Hishida M. Improved form of the ke model for wall
turbulent shear ows. Journal of Fluids Engineering 1987;109:15660.
[5] Chen Q. Comparison of different ke models for indoor airow
computations. Part B, Fundamentals. Numerical Heat Transfer
1999;28(3):35369.
[6] Rouaud O, Havet M. Computation of the airow in a pilot scale
clean room using ke turbulence models. International Journal of
Refrigeration 2002;25(3):35161.
[7] PHOENICS 3.2 (software package). London, UK: CHAM; 1999.
[8] Cheong KWD, Djunaedy E, Chua YL, Tham KW, Sekhar SC, Wong
NH, Ullah MB. Thermal comfort study of an air-conditioned lecture
theatre in the tropics. Building and Environment 2003;38(1):6373.
[9] Fluent Inc. Fluent users guide. Version 5.0. Lebanon, NH, USA:
Fluent Inc.; 1998.
[10] Posner JD, Buchanan CR, Dunn-Rankin D. Measurement and
prediction of indoor air ow in a model room. Energy and Building
2003;35(5):51526.
[11] Wilcox DC. Reassessment of the scale determining equation for
advanced turbulence models. AIAA Journal 1988;26(11):1299311.
[12] Menter FR. Two-equation eddy viscosity turbulence models for
engineering applications. AIAA Journal 32(8):26989.
[13] Chen Q. Prediction of room air motion by Reynolds-stress models.
Building and Environment 1996;31(3):23344.

ARTICLE IN PRESS
A. Stamou, I. Katsiris / Building and Environment 41 (2006) 11711181
[14] Moureh J, Flick D. Wall airjet characteristics and airow patterns
within a slot ventilated enclosure. International Journal of Thermal
Sciences 2003;42(7):70311.
[15] Loomans M. The measurement and simulation of indoor air ow.
PhD. thesis. The Netherlands: University of Eindhoven; 1998.

1181

[16] AEA Technology, 2002; CFX-5.5.1 User Guide, 8.19, Harwell,

Didcot, Oxfordshire, OX110RA, UK.
[17] Lyn DA, Stamou AI, Rodi W. Density Currents and Shear Induced
Flocculation in Sedimentation Tanks. Journal of Hydraulic EngineeringASCE 1992;118(6):84967.