Energy and Buildings 38 (2006) 931–938

www.elsevier.com/locate/enbuild

Experimental and numerical study of a mechanically

ventilated enclosure with thermal effects
Frédéric Kuznik a,*, Gilles Rusaouën a, Raluca Hohotă b
a
Thermal Sciences Center, National Institute of Applied Sciences Lyon, France INSA, Bât FREYSSINET,
20 Av. A. Einstein, 69621 Villeurbanne Cedex, France
b
Technical University of Civil Engineering of Bucharest, 124 Lacul Tei Boulevard, Bucharest 72302, Romania
Received 20 September 2004; received in revised form 17 July 2005; accepted 3 August 2005

Abstract
Full-scale experimental and computational fluid dynamics (CFD) methods are used to investigate the velocity and temperature fields in a
mechanically ventilated enclosure. Detailed airflow characteristics were measured in three cases of ventilation air temperature: an isothermal case,
a hot case and a cold case. The experimental data were used to validate two CFD models: a Reynolds averaged Navier–Stokes equation (RANS)
modelling and a large Eddy simulation (LES) modelling. The RANS model provides results in better agreement with experimental data, excepted
for the cold case. It has been found that the LES model underestimates the expansion of the jet in the three cases, disabling the use of this model for
the prediction of the flow field in ventilated rooms.
# 2006 Elsevier B.V. All rights reserved.

Keywords: LES model; k– e Model; Archimede effect; 3D jet behavior; Building room

1. Introduction low Reynolds number k– e model, a two layers k– e model, a

two-scale k– e model and a renormalization group (RNG) k– e
As we spend most of our time in enclosed spaces, model. He then recommended the RNG k– e model for airflow
environmental control of buildings is becoming very important. simulations. Teodosiu [8] used a realizable k– e model
Among the thermal comfort indices proposed in the literature, developed by Shih et al. [9] to model airflow in enclosures
the Fanger’s empirical one, predicted mean vote (PMV) has and it showed it’s reliability to predict the velocity and
become a standard approach to asses the indoor thermal temperature mean fields.
comfort [1] , but evaluating PMV is only possible if the velocity The LES model should be the next generation to study
and thermal fields are known. These information can be obtain indoor airflow in buildings because of its universality (no many
numerically by using computational fluid dynamics codes adjustable constants are needed) and capability to predict flow
(CFD). Nowadays, the turbulence modelling is the main stake that is difficult to predict with other CFD models. For example,
in airflow modelling. Actually, there are two main types of with a RANS model, the mean airflow through a door will be
approaches used in such types of problems: the Reynolds- closed to zero when a LES model predicts the instantaneous
averaged Navier–Stokes (RANS) and the large Eddy simulation velocity field. However, few LES studies on indoor airflow have
(LES). been reported (Benetsen et al. [10] , Davidson and Nielsen [11] ,
Among the RANS models, the standard k– emodel of Yi and Chen [12]) and no clear response can be deducted from
Launder and Spalding [2] have been extensively used for the literature concerning the comparison between RANS and LES
prediction of air flow in rooms (Niu and Kooi [3] , Youchen [4] , for the prediction of airflow in ventilated enclosures with
Lu et al. [5] , Rouaud and Vavet [6] , . . .), but the model seems thermal effects.
too diffusive for shear layer flows. Chen [7] compared five k– e In order to explore the potential for numerical simulations as
models concerning indoor airflow: the standard k– e model, a a tool in indoor air ventilation design, we compare three-
dimensional CFD results with mean velocity and temperature
measurements of air flow in a mechanically ventilated
* Corresponding author. Tel.: +33 472 438 459; fax: +33 472 438 522. enclosure with thermal effects. The experimental data come
E-mail address: [email protected] (F. Kuznik). from tests carried out in a full scale test room with ventilation
0378-7788/$ – see front matter # 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.enbuild.2005.08.016
932 F. Kuznik et al. / Energy and Buildings 38 (2006) 931–938

Nomenclature

List of symbols
Ar Archimede number
D flow rate
g gravity
k turbulent kinetic energy
p pressure
Pr Prandtl number
Prt turbulent Prandtl number
Re Reynolds number
T temperature
ui velocity
U velocity magnitude
xi direction
Fig. 1. Experimental cell Minibat.
Greek letters
d Kronecker symbol isolates the cell 1 from a climatic chamber whose temperature
e dissipation rate of k is controlled by the means of an air-treatment system. The
m viscosity climatic chamber temperature can vary between
10and 30  C,
mt turbulent viscosity but it is fixed in our cases. A thermal guard allows us to
r density maintain the five other exterior faces walls at a uniform value of
approximatively 20  C.
Our work only deal with tests carried out in cell 1 with an
and temperature controls. Two turbulence models are used in axisymmetric jet coming from an air supply; its exact
this study: the RANS k– e realizable model and a LES model. configuration can be found in Fig. 2. The jet is maintained
Numerical approach was performed using computational fluid at a fixed temperature by an air-treatment system. The
dynamics (CFD) fluent code [13]. ventilation system allows us to impose inlet and outlet flow
rates which are measured with two flowmeters.
2. Description of the experimental set-up The test room have been equipped with thermocouples in
order to measure the wall internal surfaces temperature with a
The experimental full scale test room MINIBAT (CETHIL- resolution of 0:4  C, each face being equipped with nine
INSA de Lyon, France) is represented in Fig. 1. The installation thermocouples. The air temperature is measured with two
consists of two identical enclosures, called cell 1 and cell 2, thermocouples with a resolution of 0:4  C. The air velocity is
whose dimensions are 3:10 m, 3:10 m, 2:50 m according to the measured by an omnidirectional velocity probe with temperature
coordinate directions (x1 , x2 , x3 ). The glazed south façade compensation whose resolution is the worst of 0:25 m/s and

Fig. 2. Cell 1 sensor and mechanical equipment.

 F. Kuznik et al. / Energy and Buildings 38 (2006) 931–938 933

Table 1 depending only with temperature. The filtering equations for

Global conditions during experimentations
continuity, momentum and energy are the following:
Red Ard Tin ( C) Tm ( C) D (m3/h)
@ūi
Isothermal case 11520 0 – – 54.2 ¼0 (1)
@xi
Hot case 10400 0.020 31.0 23.6 53.0
Cold case 13520 0.012 12.6 20.4 66.0  
@ūi @ūi ū j @ @ūi @ p̄ @t i j
r þr ¼ m

r þ rgk dik (2)
@t @x j @x j @x j @xi @x j
3%of the measurement value for a temperature contained  
@T̄ @ū j T̄ @ m @T̄ @h j
between 20 and 26  C adding 0:5%= C outside this range. The r þr ¼
r (3)
velocity measurement rate is 20 samples per second, the mean @t @x j @x j Pr @x j @x j
velocity is measured over 1000 samples. A mobile arm allows us where the bar represent the grid filtering. The subgrid-scale
to move the temperature and air velocity sensor in the room in Reynolds stresses in Eq. (2)t i j ¼ ui u j
ūi ū j is modelled via
order to get complete fields of mean temperature and mean Eddy viscosity model of the following form:
velocity.
The experiment was realized under steady state conditions 1
ti j
tkk di j ¼
2mt S̄i j (4)
and the characteristics of each case are given in Tables 1 and 2. 3
Table 1 presents the Reynolds and Archimede numbers based where mt is the subgrid scale turbulent viscosity and S̄i j is the
on the ventilation inlet diameter. Tm is the value of the mean rate-of-strain tensor for the resolved scale defined by:
temperature in the non-moving air zone (zone where the  
velocity is less than 0:05 m/s). Table 2 shows the internal faces 1 @ūi @ū j
S̄i j ¼ þ (5)
temperatures of the walls during the experimentations and for 2 @x j @xi
the hot and cold cases. These temperatures are used as boundary
conditions for the numerical simulation. The subgrid-scale heat fluxes in Eq. (3) is modelled by:
In conclusion, the experimental methodology has permitted mt @T̄
us to obtain a complete boundary conditions description and the h j ¼ u j T
ū j T̄ ¼
(6)
Prt @x j
detailed dynamic and thermal fields to compare with numerical
data based on our models. In the last equation, the value of Prt is set to 0.85.
The subgrid-scale Eddy viscosity model used has been
3. Numerical methods developed by Yakhot et al. [17] using the renormalization group
(RNG) theory. On one hand, in highly turbulent regions of the
In this section, we present the different computational flow, this model reduces to the Smagorinsky–Lilly model, on
models used to compute the fluid flow. After a brief paragraph the other hand, in low-Reynolds-number regions the turbulent
concerning our RANS modelling aproach, we discuss the LES viscosity becomes null. This enables the RNG based subgrid-
model, the computational scheme, the required computing time scale Eddy viscosity to model the low-Reynolds-number effects
and the boundary conditions. encountered in transitional flows and near-wall regions.
Concerning the RANS modelling, the k– e realizable model The LES model used an unstructured mesh composed of
[9] was used. This model is widely described in [14] and then tetrahedral elements. Since the Reynolds number was around
not detailed here. The only difference concerns the near wall 12,000 and the expected Kolmogorov scale was about 2 mm,
treatment: in our study, an enhanced wall treatment is used and the grid filter length of each cell is included between 4 cm and
it is based on the two-layer model of Wolfstein [15](when it is 4 mm and the total grid elements was 1,599,760. The time step
possible) combined with the wall function approach of Jongen size was 0:02 s and once the flow is statistically reached, 20 s
[16] to give the first boundary cells values. are recorded to compute the mean values. With this grid and the
The governing equations employed in LES approach are time step size, the simulation was required 5-day computing
obtained by filtering the time-dependant Navier–Stokes equa- time on a workstation.
tions in the physical space. The finite-volume discretisation For the RANS, the grid used in LES model was employed
employed in fluent implicitly provides the filtering operation. and the simulation required 2-day computing time on the same
The fluid is considered as incompressible with an ideal-gas law machine as the previous model. Both CFD methods were tested
for grid independence and grid refinement yields to insignif-
icant changes in the numerical results for velocity and
Table 2
temperature fields.
Conditions during experimentation: mean temperature ( C) on internal faces Concerning the boundary conditions, the velocity and
walls temperature values are given as known values using the
South North East west Ceiling Floor
experimental data. The inlet and outlet boundary conditions are
imposed far from the openings in order to be at fully developed
Hot case 19.8 23.6 23.2 23.3 24.3 23.2
flow sections. The classical no-slip boundary conditions are
Cold case 19.7 21.1 21.2 21.2 21.1 21.0
assured at the walls.
934 F. Kuznik et al. / Energy and Buildings 38 (2006) 931–938

Fig. 3. Measurement positions in MINIBAT.

4. Computational results and discussions ventilation air inlet), a profile at x2 ¼ 1 m and a profile at
x2 ¼ 2 m.
This section presents the numerical results for the isothermal, Fig. 4 presents the comparison between the models and the
hot and cold cases. The comparisons concern the mean velocity experimental data. Close to the ventilation inlet, the experi-
magnitude and the mean temperature fields. All the experimental mental and numerical velocity profiles are in good agreement
data are presented with the uncertainty defined in Section 2. Two which validate the model for this ventilation system. But the
plans have been chosen for the comparisons: the median plan (or more the distance from the inlet increases, more the LES
longitudinal according to the flow; coordinate x1 ¼ 1:55 m) and modelling used predicts a value for the maximum velocity
the transversal plan (coordinate x3 ¼ 2:32 m). These two plans higher than the experimental one. Concerning the k– e model,
are represented in Fig. 3. even if it seems to predict a value for the velocity maximum in
good agreement with experiment, there is a poor match between
4.1. The isothermal case the exact experimental data and numerical profiles.

4.1.1. Longitudinal profiles 4.1.2. Transversal profiles

In the ventilated enclosure, the dynamic of the flow is quite In the previous section, we compared the velocity profiles in
simple because there are two principal zones: the jet zone and the median plan and we notice differences between numerical
the non-moving fluid zone where the fluid velocity is less than models and experimentation. In order to have a better
0:05 m/s and then not measurable by our means. In order to understanding of these phenomena, we now have a look at
evaluate the accuracy of our models, we then compare the profiles in the transversal plan defined in Fig. 3. The profiles
profiles of the velocity in the jet zone. We chose three profiles in discussed here are placed at x2 ¼ 0:6 m (at 3 cm from the
the median plan: a profile at x2 ¼ 0:6 m (at 3 cm from the ventilation air inlet), at x2 ¼ 1:1 m and at x2 ¼ 1:6 m.

Fig. 4. Velocity profiles in median plan at x2 ¼ 0:6 m (left), x2 ¼ 1 m (middle) and x2 ¼ 2 m (right)—isothermal case.
 F. Kuznik et al. / Energy and Buildings 38 (2006) 931–938 935

Fig. 5. Velocity profiles in transversal plan at x2 ¼ 0:6 m (left), x2 ¼ 1:1 m (middle) and x2 ¼ 1:6 m (right)—isothermal case.

Fig. 5 shows the experimental and numerical velocity profile Fig. 7 presents the temperature profiles in the median plan
in the transversal plan. Both the numerical models and the and for the same positions as in Fig. 6. The two numerical
experimental data are, once again, in good agreement near the models overestimate the maximum of the temperature at
air inlet. The more we go away from the inlet, the more the two x2 ¼ 1 m and x2 ¼ 2 m, even if at 3 cm from the air inlet the
numerical models present an expansion smaller than the profiles are in good agreement. The LES model overestimates
experiment. That is explaining why the maximum velocity is the maximum temperature more than the k– emodel.
overestimated by the numerical models. Concerning the The transversal profiles, which are not presented in this
numerical modelling, the k– e model is, obviously in Fig. 5, paper, show that, as for the isothermal case, the expansion of the
in better agreement with experimental data than the large Eddy jet is underestimated by the two models. Moreover, the LES
simulation model. model underestimated more the jet expansion than the k–
emodel.
4.2. The hot case
4.3. The cold case
The analyze of the hot case is made by the comparisons of
the mean velocity and mean temperature profiles in the median This section deals with the comparisons of the profiles in the
plan: profiles at x2 ¼ 0:6 m (at 3 cm from the ventilation air median plan for the cold case. The velocity and temperature
inlet), at x2 ¼ 1 m and at x2 ¼ 2 m. profiles are taken at the same positions as for the hot case.
Fig. 6 shows the three velocity profiles for the numerical Fig. 8 shows the mean velocity profiles for the numerical
models and the experiment. The k– e model predicts well the models and the experiment. Identically to the hot and
velocity, but some differences exists mainly on the part of the isothermal cases, the velocity profiles are in good agreement
jet which is near the ceiling. Concerning the LES model, more near the ventilation inlet. But from x2 ¼ 1 m, the numerical
we are far from the inlet, more the velocity maximum is models overestimates the maximum of the velocity and the
overestimated and more the experimental and model profiles position of this maximum is far from the experimental data.
are different. Even if the k– e model is closer to the experiment, the two

Fig. 6. Velocity profiles in median plan at x2 ¼ 0:6 m (left), x2 ¼ 1 m (middle) and x2 ¼ 2 m (right)—hot case.
936 F. Kuznik et al. / Energy and Buildings 38 (2006) 931–938

Fig. 7. Temperature profiles in median plan at x2 ¼ 0:6 m (left), x2 ¼ 1 m (middle) and x2 ¼ 2 m (right)—hot case.

Fig. 8. Velocity profiles in median plan at x2 ¼ 0:6 m (left), x2 ¼ 1 m (middle) and x2 ¼ 2 m (right)—cold case.

numerical models are not able to predict the correct dynamic of 4.4. Maximum values analysis
the jet.
The temperature profiles in the median plan are presented in In the former sections, we analyzed the dynamic of the jet
Fig. 9. The first conclusions are nearly similar to those made for and the heat transfer (by the means of the velocity and
the velocity profiles: near the inlet the profiles are in good temperature profiles, respectively). In all cases, the large
agreement but the numerical models are less reliable as the Eddy simulation model cannot correctly predict the expansion
profile is far from the air inlet. For the temperature profiles, the of the jet and then the dynamic of the fluid. Concerning the k–
two numerical models are not reliable. emodel, the hot and isothermal cases present a quite good
The transversal profiles (not showed in this article) confirm agreement with the experimental data, on the contrary the
that the numerical models underestimate again the expansion cold case is not well predicted. In this section, we analyze
of the jet. the global accuracy of the numerical models by the means of

Fig. 9. Temperature profiles in median plan at x2 ¼ 0:6 m (left), x2 ¼ 1 m (middle) and x2 ¼ 2 m (right)—cold case.
 F. Kuznik et al. / Energy and Buildings 38 (2006) 931–938 937

Fig. 10. Maximums of velocity at different x2 sections in median plan for isothermal (left), hot (middle) and cold (right) cases.

Fig. 11. Maximums of velocity at different x2 sections in median plan for isothermal (left), hot (middle) and cold (right) cases.

the comparisons of the maximum values of mean temperature models for the prediction of airflow in rooms: a k– e realizable
and velocity. model and a large Eddy simulation model. Three configurations
Fig. 10 shows the values of the maximum velocity in of jet airflow temperature were tested: an isothermal case, a hot
different sections of the median plan. The velocity Um is scaled case and a cold case.
by the maximum of the experimental velocity in the median The LES model predicts, in all cases, a fluid flow which
plan, called Um;exp . For all the cases, the k– e model fits the underestimates the expansion of the jet and overestimates the
experimental data better than the LES model. Even if the maximum of velocity and temperature. This model is not
analyze of the dynamic of the cold jet (Section) shows adapted to the determination of the fluid flow in a
differences between the profiles of the experimental data and k– mechanically ventilated enclosure. To overcome the problems
e model, the maximum of the velocity curves fit well until of the LES simulation presented here, new LES models are in
x2 ¼ 2:1 m. development as the dynamic subgrid-scale model [18] for
Concerning the temperature, Fig. 11 presents the maximum example.
(for the hot case) and the minimum (for the cold case) of the The k– e realizable model seems to predict well the velocity
temperature in different sections of the median plan. The and temperature fields in the isothermal and hot cases. The cold
temperature Tm is scaled by the maximum of the experimental case is not as well predicted as the other cases, even if the global
temperature in the median plan noted Tm;exp . In all cases, the k– comportment of the fluid is correctly solved. However, the k– e
e realizable model fits better the experimental data. For the hot models is not appropriate to the unsteady situations and then it
case, the k– e model is in good agreement with the experiment. limits their use.
Nevertheless, the cold case is not well predicted by the k– e The prediction of the airflow in the cold case is difficult
model. for our models and the exact slope of the jet is not well
predicted. Nevertheless, this parameter is very important for
5. Concluding remarks the thermal comfort of the occupants of a building room.
Further investigations are needed to comprehend the
In this study, experiments and numerical simulations were dynamic of jets by the means of velocity turbulence
carried out in order to evaluate the accuracy of two turbulence measurements. Further investigations are needed too to
938 F. Kuznik et al. / Energy and Buildings 38 (2006) 931–938

develop numerical models for the exact prediction of such [9] T. Shih, W.W. Liou, A. Shabbir, Z. Yang, J. Zhu, A new k
e Eddy
viscosity model for high Reynolds turbulent flows, Compuer Fluids 24 (3)
airflows in rooms.
(1995) 227–238.
[10] J.C. Benetsen, J.N. Sørensen, H.T. Søgaard, P.L. Christiansen, Numerical
References simulation of turbulent airflow in a livestock building, Proceeding of
Roomvent 96 (1996) 169–176.
[1] ISO 7730, Ergonomics of the thermal environment—Analytical determi- [11] L. Davidson, P.V. Nielsen, Large eddy simulation of the flow in a three
nation and interpretation of thermal comfort using calculation of PMVand dimensionnal ventilated room, Proceeding of Roomvent 96 (1996) 161–
PPD indices and thermal comfort criteria, International Standards Orga- 165.
nisation, Geneva, 2005. [12] J. Yi, Q. Chen, Buoyancy-driven single-sided natural ventilation in
[2] B.E. Launder, D.B. Splading, The numerical computation of turbulent flow, buildings with large openings, International Journal of Heat and Mass
Computer Methods in Applied Mechanics and Energy 3 (1974) 269–289. Transfer 46 (2003) 973–988.
[3] J. Niu, J.V.D. Kooi, Indoor climate in rooms with cooled ceiling systems, [13] Fluent Inc., Fluent User’s Guide, Version 6.0, Fluent Inc., Lebanon - NH
Building and Environment 29 (3) (1994) 283–290. (USA), 2002.
[4] F. Youchen, CFD modelling of the air contaminant distribution in rooms, [14] C. Teodosiu, R. Hohota, G. Rusaouën, M. Woloszyn, Numerical predic-
Energy and Buildings 23 (1995) 33–39. tion of indoor air humidity and its effect on indoor environment, Building
[5] W. Lu, A.T. Howarth, A.P. Jeary, Prediction of airflow and temperature and Environment 38 (2003) 655–664.
field in a room with convective heat source, Building and Environment 32 [15] M. Wolfstein, The velocity and temperature distribution of one dimen-
(6) (1997) 541–550. sionnal flow with turbulence augmentation and pressure gradient, Inter-
[6] O. Rouaud, M. Vavet, Computation of the airflow in apilot scale clean national Journal of Heat and Mass Transfer 12 (1969) 301–318.
room using k– e turbulence models, Intenational Journal of Refrigeration [16] Jongen T. Simulation and Modelling of Turbulent Incompressible Flows.
25 (2002) 351–361. Ph.D. Thesis, EPF Lausanne, Lausanne, Switzerland, 1992.
[7] Q. Chen, Comparison of different k– e models for indoor airflow compu- [17] A. Yakhot, S.A. Orszag, V. Yakhot, M. Israeli, Renormalization group
tation, Numerical Heat Transfer, part B 28 (1995) 253–369. formulation of large-eddy simulation, Journal of Scientific Computing 4
[8] Teodosiu C. Modelisation des systèmes techniques dans le domaine des (1989) 139–158.
equipements des Bâtiments à l’aide des codes CFD, Ph.D. Thesis pre- [18] M. Germano, U. Piomelli, P. Moin, W. Cabot, A dynamic subgrid-scale
sented at the INSA of Lyon, 2001. eddy viscosity model, Phys. Fluids A 3 (1991) 1760–1765.