Vous êtes sur la page 1sur 16

Energy and Buildings 40 (2008) 1666–1681

Contents lists available at ScienceDirect

Energy and Buildings


journal homepage: www.elsevier.com/locate/enbuild

Natural cross-ventilation in buildings: Building-scale experiments, numerical


simulation and thermal comfort evaluation
G.M. Stavrakakis a, M.K. Koukou b, M.Gr. Vrachopoulos b, N.C. Markatos a,*
a
Computational Fluid Dynamics Unit, School of Chemical Engineering, National Technical University of Athens, Iroon Polytechniou 9, GR-15780 Athens, Greece
b
Environmental Research Laboratory, Department of Mechanical Engineering, Technological Educational Institution of Halkida, Psachna GR-34400, Greece

A R T I C L E I N F O A B S T R A C T

The constantly increasing energy consumption due to the use of mechanical ventilation contributes to
Article history:
Received 29 November 2007 atmospheric pollution and global warming. An alternative method to overcome this problem is natural
Received in revised form 8 February 2008 ventilation. The proper design of natural ventilation must be based on detailed understanding of airflow
Accepted 19 February 2008 within enclosed spaces, governed by pressure differences due to wind and buoyancy forces. In the present
study, natural cross-ventilation with openings at non-symmetrical locations is examined experimentally
Keywords: in a test chamber and numerically using advanced computational fluid dynamics techniques. The
Natural ventilation experimental part consisted of temperature and velocity measurements at strategically selected
Cross-ventilation locations in the chamber, during noon and afternoon hours of typical summer days. External weather
Turbulence conditions were recorded by a weather station at the chamber’s site. The computational part of the study
Buoyancy consisted of the steady-state application of three Reynolds-Averaged Navier-Stokes (RANS) models
CFD
modified to account for both wind and buoyancy effects: the standard k–e, the RNG k–e and the so-called
Thermal comfort
‘‘realizable’’ k–e models. Two computational domains were used, corresponding to each recorded wind
incidence angle. It is concluded that all turbulence models applied agree relatively well with the
experimental measurements. The indoor thermal environment was also studied using two thermal
comfort models found in literature for the estimation of thermal comfort under high-temperature
experimental conditions.
ß 2008 Elsevier B.V. All rights reserved.

1. Introduction and highly glazed facades, often with poor shading. This, combined
with extra heat gains from the electric lighting and office
Mechanical ventilation in buildings is common practice equipment, such as computers and photocopiers, increase the
nowadays, due to the need to provide thermal comfort and good overheating risk that is finally leading to a significant degradation
indoor air quality in enclosed spaces. The energy consumption of indoor thermal comfort [5]. This problem is challenging for
related to the operation of heating, ventilating and air-condition- engineers who must design optimum ventilation systems given
ing (HVAC) systems is considerable. According to recently the above constraints of improvident application of HVAC systems.
published data, nearly 68% of the total energy used in service Natural ventilation uses the freely available resources of wind and
and residential buildings is attributable to HVAC systems [1]. On solar energy and could represent an optimum ventilation
the other hand, natural ventilation replaces indoor air with fresh technique. Although, these resources are free, they are very
outdoor air without any energy consumption, and it also helps to difficult to control [4]. The challenge is to provide the appropriate
overcome common health problems related to insufficient control mechanisms to establish the required indoor air quality
maintenance of mechanical ventilation systems. Typically, the (IAQ). To achieve this, it is necessary to understand the physics of
energy cost of a naturally ventilated building is 40% less than that natural ventilation.
of an air-conditioned building [2,3]. From a design’s point of view, The air movement in a naturally ventilated building is a result of
it is noticeable that modern building designers make imaginative pressure differences produced by wind and buoyancy forces [4,6].
use of glass and space to create well-lit and attractive interiors [4]. The most common models to predict the performance of naturally
However, these buildings are usually characterised by tightness ventilated buildings are network, zonal and field (computational
fluid dynamics, CFD) models. Network models are used for the
airflow rate prediction through the openings of a building. This
* Corresponding author. Tel.: +30 210772 3126; fax: +30 210772 3228. kind of models can predict the airflow rates adequately but have
E-mail addresses: gstavr@mail.ntua.gr, N.Markatos@ntua.gr (N.C. Markatos). the disadvantage of not predicting the airflow field inside the

0378-7788/$ – see front matter ß 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.enbuild.2008.02.022
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1667

building, thus they produce little information about pollutant main conclusion was that indoor thermal comfort in summer can
transport and local thermal discomfort. When experimental data be improved by appropriately controlling window opening, in
do not exist, network models could be used for validation of accordance with the temporal variations of indoor and outdoor
implemented CFD models, such as in reference [7]. On the other climatic conditions.
hand, zonal modelling may predict both the airflow rates and air The objective of the present study is to investigate the potential
distribution with relatively high accuracy, especially when of RANS modelling in natural cross-ventilation for an experimental
temperature variations are concerned. However, applications of chamber with two openings (doors) at non-symmetrical locations.
these models in cases of mixed convection in indoor spaces The research focuses on the numerical analysis of indoor airflow
showed that zonal models did not provide satisfactory predictions using CFD techniques and the validation of the numerical
for velocity distribution concerning indoor airflow [8]. In this last predictions with the experimental results. Both wind and buoy-
investigation, CFD and zonal models were compared with available ancy forces were taken into account to obtain results for velocity
experimental data. The main conclusion was that zonal modelling and temperature distributions. The numerical results were
under-predicted velocity distribution compared to the CFD model validated with the experimental measurements for two typical
which provided higher accuracy concerning the airflow field and summer days. The reliability of the mathematical models applied
thus gave a better prediction of the recirculation region. The same was also investigated using two different computational domains,
conclusion may be found in reference [9], where isothermal indoor according to the wind incidence angle and also accounting for
airflow is investigated. This work shows that airflow predictions in internal–external flow effects. Finally, thermal conditions of the
large spaces are substantially more accurate when obtained by a chamber were examined by integrating two additional mathema-
CFD model, even with coarse grids, than that obtained by various tical models, found in literature [22–26], standing for thermal
zonal models. Finally, for complicated flow fields the specification comfort.
of ‘‘zones’’ is at best speculative. For this reason, field modelling is
considered to be a more accurate method to deal with the problem 2. Method of analysis
of natural cross ventilation. Especially when the airflow is
2.1. General
represented by strong streamline curvature, due to wind forces,
CFD modelling is considered as the most suitable tool for reliable In the present study, natural cross-ventilation is examined
airflow simulation. The more accurate the information of any experimentally and numerically. The experiments took place in a
recirculation region the more advanced the knowledge about local chamber, also presented in reference [27], that was built, among
thermal discomfort (especially due to air draughts) and pollutant others, for indoor air quality examination purposes and consisted
distribution; for example, the possibility of pollutant confinement of a roof covered with roman tiles and a radiant barrier reflective
in certain regions of the indoor space. insulation system (Fig. 1, Section 2.2 for further details). Two doors
Natural ventilation has been widely investigated by many are located at the north and south facing walls and the room is
researchers. Jiang et al. [10] presented an extensive experimental ventilated through these openings. A small internal partition of
and computational study (LES, large eddy simulation) of natural 1 m height is also located adjacent to the north wall. As far as wind
ventilation, driven only by wind forces for simple geometries forces are concerned, both doors were kept wide open (cross-
representing cross-ventilation and single-sided ventilation con- ventilation) to ensure relatively large pressure differences. The
figurations. In order to reduce the computational cost, Evola and reflective insulation of the side walls causes temperature
Popov [11] applied Reynolds Averaged Navier-Stokes (RANS) differences between internal and external surfaces of the walls.
models, the standard and the RNG k–e models, for the experimental This creates temperature difference among internal wall surfaces
set up described in reference [10], and also obtained reasonable
accuracy for the velocity distribution inside the building model and
also for the ventilation flow rates. In most cases, research for wind
forces effects on natural cross-ventilation is focused on wind
tunnel experiments for symmetric building-like models, and is
often associated with advanced turbulence models such as LES
[11–15]. These investigations provide useful information about the
air movement inside buildings, but the geometrical symmetry and
the controlled conditions in wind tunnels mean that the flow takes
place under idealized conditions. Building-scale experimental
results of the wind effect for a single-sided ventilation case were
presented by Dascalaki et al. [16]. A validated LES model for this
case may also be found in reference [15]. Buoyancy-driven single-
sided natural ventilation has also been studied widely for the
assessment of any heat gain by internal heat sources in enclosed
spaces [17,18]. Experimental measurements and numerical
analysis of this type of natural ventilation may be found in
reference [17], where an LES model is validated with experimental
data for a simple geometry. In case of problems with more complex
geometries, for which the demand of computational resources is
high, RANS modelling may produce reasonably accurate results
[18,2]. The first application of such models was back in 1983 by
Markatos [19]. A case of combining wind and buoyancy forces was
presented by Chen [20], for the discussion of the potential of CFD in
order to design buildings that take advantage of the wind, with
internal heat gains taken into account. Thermal comfort provided
by natural ventilation has also been studied in reference [21]. The Fig. 1. (a) Experimental chamber and (b) geometrical details.
1668 G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681

and incoming outdoor thermal masses and consequently genera-


tion of buoyancy forces. The above-mentioned arrangement of the
building establishes an internal air movement governed by both
wind and buoyancy forces.
Referring to the numerical simulation, a mathematical model
was developed for the prediction of indoor air movement, within
the general framework provided by a commercial CFD code
(FLUENT 6.3.26). The model is based on three high Reynolds
number RANS turbulence models, modified for buoyancy effects,
that were tested following the procedure reported in reference
[28]. Boundary conditions were provided by measurements
accomplished for internal flow by thermo-couples or for external
flow by a weather station. The simulated airflow was examined in
terms of two different computational domains, due to various wind
incidence angles, and finally the temperature and velocity
distributions were the outcome of the solution procedure.
Thermal comfort predictions in the occupied zone were also
included using two models found in literature, assessing local and
global (general) discomfort. The first one, used for local discomfort
assessment, is the well-known PMV/PPD model [22–24], extended
to account for non-air-conditioned buildings in warm climates
[25]. The factors predicted mean vote (PMV), predicted percentage
dissatisfied (PPD) and percentage dissatisfied (PD) represent a
quantification of thermal dissatisfaction of occupants under
several activity levels and clothing, with respect to air tempera-
ture, air velocity, turbulence, wall temperatures and relative
humidity. The second thermal comfort model, used for global
discomfort assessment, is the so-called ‘‘adaptive model’’ of
thermal comfort in naturally ventilated buildings found in Fig. 2. Arrangement of probes for: (a) case A and (b) case B.

references [26] and [29], where it is presented as an optional


model. It is based on an extensive field study by de Dear et al. [30]
and determines a range of indoor comfort temperatures that (one-dimensional) velocity measurements in the test room were
correspond to a percentage of thermal acceptability, as a function obtained using the KIMO thermo-anemometer multi-probes VT
of the mean monthly outdoor temperature. Details for both models 200F [31] with accuracy of 3% for readings of 0–3 m/s and 2% for
are given in Section 2.3 below. reading of 0.1 8C.
Temperature of internal wall surfaces of the walls was
2.2. Experimental equipment, methodology and results measured using UTECO thermo-couples DIN 43732, connected
to a data logging system [32] for the collection of the experimental
The test room is of dimensions 6 m  4 m  5.5 m (Fig. 1a and results.
b). The side walls are a two series brick construction with a bubble Sample experimental results obtained (also presented in
material lamination among layers of aluminum foil placed in the reference [33]), for example, for case A, referring to velocity and
20 mm gap of the brick layers. The total wall thickness consists of temperature at the selected locations and the temperature of the
90 mm brick, 10 mm air gap, 1 or 2 mm reflective insulation, building walls are presented in Figs. 3 and 4, respectively.
10 mm air gap and 90 mm brick. The walls can be coated internally Bi-directional flow may occur through the openings due to
and externally resulting to a total width ranging from 200 to buoyancy forces. In this case, information about the infiltrating air
240 mm (including a 20 mm thickness of the wall sheathing- may be lost by placing the air speed sensor at the middle of the
plaster on each side). The experimental measurements referring to inlet door. Thus, it is important to investigate this possibility under
velocity and temperature, obtained at various locations of the test the current experimental conditions for both cases. The measured
room were selected according to regulations found in reference temperature differences between indoor and outdoor airflows
[24]. The schedule of the experiments included 3 h total sampling were 1.9 and 1.0 8C for cases A and B (see Table 2; Fig. 3),
period for two typical summer days and the time-interval respectively. According to the above temperature differences, the
sampling between consecutive measurements was 60 s. Under comparison of the importance of wind and buoyancy forces is
those imposed conditions, there were simultaneous recordings of obtained using the Archimedes number as follows [15]:
temperature and velocity at the middle of the inlet door A1 or A2
(being south or north doors, respectively, according to wind data Grashof bgH3 DT
Ar ¼ 2
¼ (1)
taken by a weather station), at positions B1, B2, C1, C2, C3 and of Reynolds U 2 D2
temperature at point D (middle of the room) (Fig. 2a). That was for
the first day’s arrangement (case A). For the second day’s where b = 1/Tref is the thermal expansion coefficient, Tref the
experimental arrangement (case B), the sampling points were reference temperature taken as T ref ¼ ðT outdoor þ T walls average Þ=2,
 
different and measurements were obtained as follows: velocity 1:9  C for case A
g = 9.81 m/s2 the gravitational force, DT ¼  , U
and temperature at A1 or A2, B1, B2, B3, C1, C2, C3 (Fig. 2b). For 1:0 C for case B
both experiments, temperatures at the middle of the internal the wind speed at the building height. Its value is taken by the
surface of each wall were also recorded. The exact location, power-law profile of the incoming wind using Eq. (3). H is the
measured flow property at each sampling point and which door height of the inlet door and D is the chamber’s depth equal to 4 m
plays the role of inlet are given in Table 1. Temperature and (see Fig. 1b).
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1669

Table 1
x, y, z coordinates of experimental locations

Case A A1 A2 (inlet) B1 B2 C1 C2 C3 D

Coordinates (x, y, z) 5.25, 0, 1.1 0.75, 4, 1.1 4.7, 0.7, 0.5 4.7, 0.7, 1.25 1.8, 1.6, 0.77 1.8, 1.6, 1.4 1.8, 1.6, 1.93 3, 2, 2.75
Temperature O X X X X X X X
Velocity O X X X X X X O

Case B A1 (inlet) A2 B1 B2 B3 C1 C2 C3

Coordinates (x, y, z) 5.25, 0, 1.1 0.75, 4, 1.1 3, 2, 0.2 3, 2, 2 3, 2, 3 0.7, 2, 0.5 0.7, 2, 1 0.7, 2, 2
Temperature X O X X X X X X
Velocity X O X X X X X X

Symbols: X for measured property, O for non-measured property.

Using the above data, Eq. (1) gives an Ar number equal to 0.02 2.3. Mathematical modelling
and 0.0029 for cases A and B, respectively. Thus, because of Ar  1
for both cases, natural convection due to the buoyancy effect is 2.3.1. The governing equations
much smaller than forced convection due to the wind effect. Natural ventilation is a phenomenon of random nature due to
Furthermore, the chamber is free of internal thermal sources and the constant changes of external weather conditions. Thus, any
also, even though the height at which wind speed is recorded is mathematical model applied for the prediction of natural
relatively low (7.5 m), the chamber is built in a rural environment ventilation should include the dynamic nature of the external
free of thermal sources. All the above advocate that the airflow conditions. In applying a CFD method one should ideally use a
inside and outside the chamber is wind dominant. However, the time-dependent approach, which, however, would require knowl-
buoyancy effect is not neglected in the CFD model due to the edge of the time-dependent variations of the boundary conditions
presence of experimental results for temperature, used for further used. This technique would provide very detailed and useful
verification of the numerical results. information about natural ventilation but it requires excessive
Under the action of the wind, and taking into account that both computational resources for practical applications. An engineer’s
openings have the same dimensions and also that the internal approach to overcome such technical restrictions is the steady-
space has no obstacles except the small partition, the air flows in state assumption, as most phenomena take place at almost steady-
through the one opening leading to high pressures at this area. The state conditions over long periods of time. Furthermore, during
pressure at the opposite door is lower and due to this pressure day-time cycles in real buildings, temperature changes occur but
difference between the two doors, the air is finally leaving the the steady-state assumption is considered to be valid over long
chamber through the opposite door. According to the above periods, because the time needed for the development of the
analysis, the flow through the openings is not bi-directional (if it airflow pattern is short compared to the duration of the day-time
were it would have been predicted by the CFD model, which is cycle and the values of the time constants of the massive building
described below, as, for example, in reference [2]). Thus, the elements required for passive ventilation. Thus, in order to
velocity at the middle of the opening is representative of the air recognize potential time-averaged values of temperatures and
entering the chamber. velocities, two temporal sub-domains were chosen, representative
of minimum fluctuations of monitored flow properties for noon
and afternoon hours of cases A and B, respectively. The selected
time periods were 13:00–13:30 h (case A) and 19:00–19:30 h (case
B). In any case, the mathematical model developed is general and
may be applied to both steady-state and transient problems, as
required.
The ensemble-averaged values of the external wind were
selected for time periods not exceeding the recommended time
period limit of 30 min, as found in reference [34] for field
measurements. Following this, the selected 30 min time periods of
the experimental cases studied were identified according to – as

Fig. 3. Typical experimental results for velocity and temperature, case A. Fig. 4. Temperature at the middle of the internal surface of each wall, case A.
1670 G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681

Table 2
Experimental conditions

External conditions

Cases Inflow planes (Fig. 5) Relative humidity (%) Outdoor temperature (K) Wind speed at 7.5 m (m/s) Incidence angle (u)

Case A A1, A2 19.2 308.90 1.48 358


Case B B1 24.8 305.50 2.85 908

Internal conditions

Cases Tnorth wall Tsouth wall Twest wall Teast wall

Case A 300.9 303.0 301.4 302.0


Case B 302.0 302.1 302.0 302.1

stable as possible – recordings of external and internal conditions vertical wind direction (case B) the computational domain, for
provided by the weather station and the hot-wires, respectively. A which grid B was constructed, had a downstream length of 10H (H:
similar technique can also be found in reference [35]. Significant chamber’s height), an upstream length of 5H, a lateral length of 5H
information is provided by the steady-state solution concerning on both sides of the chamber and a height of 18.2H, in order to
the effects of the prevailing ensemble-averaged values of winds of impose, as inflow boundary conditions, the appropriate velocity
any building’s site on the internal structure of the flow. Since CFD is and turbulence distribution, corresponding to the atmospheric
based on mean representative wind data, it could lead to optimal boundary layer. In case of non-vertical wind direction (case A), two
designs even in the pre-construction phase of a building, with the inflow boundaries are needed to account for the incoming wind. In
external ‘‘fluid mechanics’’ taken into account. order to capture the reattachment point downwind to the chamber
The mathematical model applies numerical techniques to solve correctly, two outflow boundaries were imposed being the
the Navier-Stokes (N-S), continuity and energy equations for 3D, downwind faces of the flow domain. This leads to a computational
turbulent fluid flow. All the governing conservation equations, for domain, resulting to grid A, which consisted of 10H length among
steady-state conditions, can be written in the following general the chamber’s external walls and all lateral boundaries with the
form [36]: high boundary being at the same height as in the case of vertical
wind direction. The computational domains constructed for each
divðr~
u’  G ’ grad’Þ ¼ S’ (2)
experimental case and the corresponding inlet door are presented
where w is the dependent variable, i.e. 1 for mass continuity, ~u for in Fig. 5.
velocity, T for temperature, Ci for the concentration of various Since grid A is simply a result of dimensional expansion of grid B
chemical species i, Gw the ‘‘effective’’ exchange coefficient of in regions where no flow obstructions occur, a grid-independency
variable w and Sw is the source/sink term of variable w [36]. The study was only performed for grid B. The optimum mesh produced
transferred quantity w also stands for turbulence dependent for grid B led to the optimum grid A with additional meshing in the
variables, here k the kinetic energy of turbulence and e the eddy extra regions of the computational domain in the case of non-
dissipation rate. In this study, three RANS models were applied, the vertical flow direction. Spatial discretization of the flow domain
standard k–e, the RNG k–e [37] and the so-called ‘‘realizable’’ k–e was examined by repeating runs for grids with continuously
[38] model, all modified to account for buoyancy effects due to increased grid-nodes density. A comparison for the vertical
density differences [28]. All models treat flow effects close to walls velocity distribution at the middle of the test room, obtained
using standard wall functions [39,40]. The assumptions made for the using various grids is presented in Fig. 6.
problem were: (a) single phase, steady-state flow for a Newtonian The solution obtained using each grid was also examined for
fluid and (b) heat transfer at the walls by either conduction or other independent variables at various physical locations of the
radiation was neglected due to the already measured internal-wall- domain and, as demonstrated in Fig. 6, a grid-independent solution
surface temperature and the use of reflective insulation. was achieved using a grid consisting of 636,456 hexahedral cells
The CFD code used for the numerical simulations employs a for grid B and, consequently, a grid of 730,488 hexahedral cells for
standard finite-volume method and a body-fitted structured grid grid A, corresponding to experimental cases B and A, respectively.
[41], which is adopted to account for two different computational
domains, corresponding to 358 (grid A) and 908 (grid B) wind
incidence angle for cases A and B, respectively (Section 2.3.2 for
further details). The first-order upwind discretization scheme and
the SIMPLE solution algorithm for handling pressure were used
[42]. Simulations were performed on a Windows PC with one
2.4 GHz CPUs and 1 GB of RAM and required approximately 36 and
24 h for the optimum grids A and B, respectively.

2.3.2. Computational domains and spatial discretization


As mentioned earlier, two grid methodologies were applied for
the simulation according to the wind incidence angle. The main
purpose was to simulate the air movement in the chamber as a
result of the interaction among internal and external flow effects,
within reasonable computational resources. The problem follows
the general theoretical aspects of the flow around a bluff body.
Consequently, specific techniques found in literature [43,44] were Fig. 5. Computational domains for each case: A1 and A2, inflow boundaries for case
applied concerning the computational domain. Thus, in case of A; B1, inflow boundary for case B.
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1671

near each wall was 0.05 m, leading to y+ values less than 150 in
order to apply properly ‘‘wall-function’’ boundary conditions
[39,46].

2.3.3. Boundary conditions and special sources


For some of the imposed boundary conditions measured data
representing external and internal conditions have been used,
while the rest are set according to well-known practices
[2,11,21]. For the selected time periods (Section 2.2) of
experimental monitoring, time-averaged values were produced
referring to outdoor temperature, relative humidity, wind speed
and direction at 7.5 m from the ground, as provided by a weather
station. At the interior, wall temperatures were measured at the
Fig. 6. Velocity distribution with height at the middle of the test room for various middle of each internal wall surface. The measured boundary
grids (case of grid B). values are summarized in Table 2. According to these measured
flow variables, the boundary conditions were categorized as
follows:
The optimum discretization for both grids A and B is presented in
Fig. 7. The selection of a structured grid, instead of a more (1) Solid planes: No-slip, no-penetration condition for momentum
convenient unstructured was due to more accurate results that and fixed temperatures.
may be obtained, at least for the case of the flow around a bluff (2) Symmetry planes: Zero normal velocity and zero normal
body, using a structured grid, as found in reference [45]. gradients of all variables.
The grids used were highly non-uniform characterised by high (3) Outflow planes: Zero diffusion flux for all variables and overall
grid-nodes density near solid surfaces. The minimum cell size mass balance correction [47].

Fig. 7. Optimum mesh for both grids A and B: (a) three-dimensional meshing, (b) longitudinal meshing, (c) partition’s height plane meshing and (d) traverse plane meshing.
1672 G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681

(4) Inflow planes: Specified external temperature and moisture recorded velocity and the calculated one (Eq. (3)) at 7.5 and at
mass fraction as given by the weather station. Modified 300 m, respectively, by the weather station using Eq. (3). l is the
equations of the atmospheric boundary layer [48] were mixing length, l = min[kz, 0.085h]. k = 0.41 stands for the Von
used to account also for incidence angles other than Karman constant and Cm = 0.09.
vertical: The main difference of the ‘‘realizable’’ k–e model against the
other two models is that it is based on the dynamic equation for
 z 1=7 fluctuating vorticity. This leads to a Cm value expressed as function
u ¼ uref (3) of vorticity and not just as a fixed value [38]. However, the imposed
h
Cm at the inflow boundaries is considered acceptable as the
 z 1=7 incoming flow is simulated as a flat boundary layer, for which
v ¼ vref (4) experiments have shown that Cm  0.09 in the inertial sublayer
h
[38].
u2  z 2 The experimental chamber is built in a flat rural environment
kinflow ¼ pffiffiffiffiffiffi 1 (5) (Psachna area, Evia, Greece) surrounded by low-rise trees and low-
Cm h
rise residences at a sufficiently away distance. Thus, any external
3=2 disturbance of the incoming wind is assumed to be eliminated and
Cm3=4 k
einflow ¼ (6) the flow is re-established at its form of the atmospheric boundary
l
layer equations. Therefore, for full-scale 3D modelling, a stand-
where u, v, kinflow, einflow are the inflow velocity and turbulence alone rural-type chamber is assumed. The exponent of the power-
properties; uref ¼ velref sin u and vref ¼ velref cos u, h = 300 m. u stands law for a surface terrain that represents a flat rural site is taken as
for wind incidence angle. If u = 908 then Eqs. (3)–(6) were imposed 1/7 (Eqs. (3) and (4)). Two additional runs have been performed for
only at the upwind boundary (plane B1, see Fig. 5). If u < 908 then the investigation of the terrain’s roughness impact on the
Eqs. (3)–(6) were imposed at the upwind (west) and at the one numerical results, using the standard k–e model. For this reason,
lateral (north) boundary (planes
pffiffiffiffiffiffiffiffiffiffi
ffi A1 and A2, see Fig. 5). u* is the 1/7 and 1/6 power-law profiles were applied, corresponding to flat
friction velocity, u ¼ tw =r. tw is the shear stress and tw ¼ and typical rural environment, respectively [49]. The results
f rvel2ref =2 is the density of the mixture of air and moisture. f is the obtained are presented, for example, along the vertical direction of
1=4
friction coefficient f ¼ 0:045ðvelref h=vÞ . v is the kinematic C1, C2 and C3 experimental locations, in Fig. 8. It is observed that
viscosity of air. velref is the wind velocity at 100 m provided by the no significant differences occur and thus both profiles could be

Fig. 8. Numerical results for 1/7 and 1/6 power-law inlet profiles for: (a) incoming wind velocity, (b) x-velocity and (c) temperature along the vertical direction of C1, C2 and
C3 experimental locations (case B, u = 908).
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1673

used to predict natural ventilation of the chamber with the same PPD factor. As far as discomfort due to air draughts is concerned, it
accuracy. can be quantified with respect to local air velocity, temperature and
turbulence intensity using the PD factor. The algebraic expressions
2.3.4. Water vapour transportation modelling of this integrated model are as follows.
The basic assumption of moisture transport modelling is that
air is considered as a mixture of dry air and water vapour. Water 2.3.5.1. PMV for natural ventilation.
vapour represents a scalar transported by the airflow (dry air:
PMVtrad ¼ ½0:303expð0:036MÞ þ 0:028 L (9)
carrier gas), according to the general Eq. (2), that becomes:
where PMVtrad is the traditional PMV (for HVAC), M the metabolic
uY H2 O þ ~
divðr~ J H2 O Þ ¼ S H2 O (7) rate (W/m2), applied for standing sedentary activity in the present
where Y H2 O is the water vapour mass fraction in the mixture, ~
J H2 O study [24] and L is the thermal load on the body expressed as
the diffusion flux and SH2 O is the source (water vapour production) follows:
imposed as a fixed mass fraction at the inflow boundary as a L ¼ internal heat production
function of the measured relative humidity, temperature and
 heat loss to the actual environment
vapour pressure at the chamber’s site.
The diffusion flux, ~
JH2 O , can be written as:
  L ¼ M  W  f3:96  108 f cl ½ðT cl þ 273Þ4  ðT r þ 273Þ4
m rT
J H2 O ¼  rDH2 O;m þ t rY H2 O  DT;H2 O
~ (8) þ f cl hc ðT cl  TÞ þ 3:05  103 ½5733  6:99ðM  WÞ
Sct T
 pv þ 0:42ðM  W  58:15Þ þ 1:7  105 Mð5867
where DH2 O;m is the mass diffusion coefficient for water vapour in the
mixture and DT;H2 O is the thermal diffusion coefficient. The first one  pv Þ þ 0:0014Mð34  TÞ g (10)
found in literature [50] with respect to the reference temperature 2
where W stands for active work or shivering (W/m ) and fcl is the
(defined as an averaged value among internal and external tempera- garment insulation factor (1 clo = 0.155 m2 K/W) expressed as:
ture), while the second one was calculated using the kinetic-theory  
method [47]. As far as the other factors are concerned, Sct is the 1:05 þ 0:645Icl ; Icl
0:078
f cl ¼ (11)
turbulence Schmidt number and mt is the turbulence viscosity. 1 þ 1:29Icl ; Icl < 0:078
Referring to the physical properties of the mixture, they were
The term Icl stands for the resistance to sensible heat transfer
imposed using the ideal gas law for an incompressible flow for
provided by a clothing ensemble (clo) and its value was taken for
density, the ideal gas mixing law for both thermal conductivity and
typical clothing insulation under summer conditions (0.5 clo). The
viscosity, while the heat capacity is expressed as a mass-fraction-
Tcl (8C) term is defined as the cloth temperature and is determined
average of moisture and air’s heat capacities [47]. The above
below as:
methods use constant values of the species properties at the
reference temperature. T cl ¼ 35:7  0:028ðM  WÞ  Icl f3:96

2.3.5. Thermal comfort modelling  108 f cl ½ðT cl þ 273Þ4  ðT r þ 273Þ4 þ f cl hc ðT cl  TÞg (12)
For thermal comfort evaluation, under the already described In Eqs. (10) and (12), T (8C) is the calculated local air
extreme summer experimental conditions, the extended PMV temperature by the CFD model, hc is the heat-transfer coefficient
model was firstly implemented in the mathematical CFD model. between the cloth and air (W/m2 K) and Tr (8C) is the mean radiant
The extended PMV model for non-air-conditioned buildings is temperature (Eq. (14)). The heat-transfer coefficient is given by:
based on the inclusion of the expectancy factor, e, which is  
assumed to depend on the duration of the warm weather over the 2:38ðT cl  TÞ0:25 for 2:38ðT cl  TÞ0:25
12:1u0:5
hc ¼ 0:5 0:25 0:5
(13)
year and whether non-air-conditioned buildings can be compared 12:1u for 2:38ðT cl  TÞ < 12:1u
with many others in the region that are air-conditioned. For
where u is the local velocity calculated by the CFD model.
example, if the weather is warm all year or most of the year and
The mean radiant temperature is computed for an averaged
there are no or few other air-conditioned buildings, e may be 0.5,
wall temperature, since there were no significant differences
while it may be 0.7 if there are many other buildings with air-
between wall temperatures during the experiments, i.e.:
conditioning. In regions with only brief periods of warm weather
during the summer, the expectancy factor may be 0.9–1. Another X
4
critical factor which contributes to the reported difference Tr ¼ T i F pi (14)
i¼1
between the calculated PMV and actual thermal sensation in
non-air-conditioned buildings is the estimated activity. People, where Ti is the temperature value at the i wall and Fpi stands for
unconsciously, tend to slow down their activity and thus they the radiation shape factor from face p of a grid cell to the visible
adapt to the warm environment by decreasing their metabolic rate. room surface i.
It has been found that the metabolic rate is reduced by 6.7% for The water vapour pressure, which participates in Eq. (10), was
every scale unit of PMV above neutral. Further details of the calculated using the following equation:
extended PMV can be found in reference [25]. The numerical
PY H2 O =ð1  Y H2 O Þ
procedure includes the calculation of the traditional PMV (for air- pv ¼ (15)
0:622 þ Y H2 O =ð1  Y H2 O Þ
conditioned spaces) and the reduction of the metabolic rate by 6.7%,
using linear interpolation techniques to account for intermediate where P is the local absolute pressure calculated by the CFD model.
values of PMV in the intervals {0,1}, {1,2}, {2,3} (Eq. (16)). After that, The PMV is then recalculated using Eqs. (9)–(15) with a reduced
the PMV is recalculated and the emerged value is multiplied by the metabolic rate (Mred) according to Eq. (16) below:
expectancy factor, e. PMV ranges from 3 to +3 for cold and hot 8 9
sensations, respectively, with mid values representing intermediate < 0:067M  PMVtrad þ M; PMVtrad 2 ½0; 1 =
Mred ¼ 0:067M  PMVtrad þ 1:004M; PMVtrad 2 ð1; 2 (16)
thermal perception states. On the other hand, this factor can be : ;
0:067M  PMVtrad þ 1:013M; PMVtrad 2 ð2; 3
expressed in terms of dissatisfied occupants’ percentage using the
1674 G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681

Consequently, PMV for natural ventilation (PMVNV) is deter- both centered around the optimum comfort temperature calcu-
mined as follows: lated as follows [26,29]:
PMVNV ¼ e½0:303expð0:036M red Þ þ 0:028 L (17) T opt:comf: ¼ 0:31T a;outdoor þ 17:8 ð CÞ (20)
where e is the expectancy factor (e = 0.7 for Athens [25]). where Topt.comf. is the optimum indoor temperature for thermal
neutrality and Ta,outdoor is the mean monthly outdoor temperature.
2.3.5.2. PPD and PD. Finally, the factors PPD and PD are computed The adaptive model can be applied within a mean outdoor
as follows: temperature range from 10 to 33 8C and thus it cannot provide
information for more severe weather conditions, such as those
PPD described in case A. On the other hand, the PMV model [25] is
PPDð%Þ ¼ 100  95exp½0:03353ðPMVNV Þ4 acceptable for all conditions. According to this restriction both
thermal comfort models were applied for case B. For case A, only
 0:2179ðPMVNV Þ2 (18)
the extended PMV is used since the recorded outdoor temperature
PD (35.9 8C) was out of the range of the adaptive model’s applicability.

PDð%Þ ¼ ð34  TÞðu  0:05Þ0:62 ð3:14 þ 0:37uTu Þ (19) 3. Results and discussion
for u < 0.05, use u = 0.05 m/s and for PD > 100%, use PD = 100%, 3.1. Airflow patterns
where Tu is the turbulence intensity Tu(%) = 100(2k)0.5/u.
Due to the dynamic nature of the external wind, speed and
The adaptive comfort standard has a mean comfort zone band direction, the specification of the incoming wind’s boundary
of 5 K for 90% acceptance, and another of 7 K for 80% acceptance, conditions is highly uncertain. Thus, any mathematical approach

Fig. 9. Numerical results by various wind velocities at 7.5 m height for: (a) incoming wind velocity, (b) x-velocity and (c) temperature along the vertical direction of
experimental locations C1, C2 and C3 (case B, u = 908).
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1675

requires a parametric study to quantify the effect of boundary maximum divergence among the numerical results obtained for
conditions (bc’s) on the interior results. For this reason, such 2.85 m/s wind speed with those obtained for the other wind speeds is
sensitivity study was performed concerning vertical incidence ranging from 0.5% to 4% and 0.16% to 0.3% for x-velocity and
angle of the incoming wind (case B), using the standard k–e model. temperature, respectively.
This study produced results obtained for different wind speeds The previous analysis verifies the assumption of using just one
measured at 7.5 m height, corresponding to possible fluctuating mean wind-speed value as a time-averaged value and thus makes
recordings by the weather station. A 10% fluctuation factor to the the steady-state assumption to be valid, at least for practical
mean value of 2.85 m/s is applied to investigate the response of the engineering purposes. Furthermore, the solution is not substan-
numerical results. According to this, recorded wind speeds of 2.565, tially affected by the turbulence distributions at the inlet
2.7075, 2.85, 2.9925 and 3.135 m/s were used. The corresponding bc boundaries as presented in reference [33]. In this last investigation,
velocity profiles of the incoming wind are presented in Fig. 9(a), fitted the problem was solved using the equations of atmospheric
by the power-law equation (Eq. (3)). Runs were performed using boundary layer for the extended domain and also for a limited,
these profiles restarting the CFD program by the solution obtained for between the two doors, domain applying a uniform velocity at the
u7.5m = 2.8 m/s to accelerate convergence. The numerical results inlet door. No significant qualitative difference was observed
along the vertical direction of C1, C2 and C3 experimental locations among the two approaches with the first one, adopted in the
are presented in Fig. 9(b) and (c) for the x-component of velocity and present study, leading to more accurate results.
for temperature, respectively. It is observed that the solution is little The problem is also solved for different wind incidence angles,
sensitive to a 10% variation of external wind speeds. Specifically, the using the standard k–e model, to evaluate the sensitivity of the

Fig. 10. Impact of the incoming wind’s incidence angle on the: (a) incoming wind x-velocity, (b) incoming wind y-velocity, (c) x-velocity and (d) temperature along the vertical
direction of experimental locations B1 and B2 (case A, u = 358).
1676 G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681

Fig. 11. Experimental and numerical results for: (a) x-velocity (case A, u = 358), (b) temperature (case A, u = 358), (c) x-velocity (case B, u = 908) and (d) temperature (case B, u = 908).

results to this parameter. The mean recorded value of the wind experimental results. In Fig. 10(a) and (b), the velocity
speed for the case A is 1.48 m/s (Table 2). Three incidence angles components of the incoming wind (Eqs. (3) and (4)) that
were tested numerically: 308, 358 and 408 corresponding to a represent the tested wind directions are presented. The impact of
time-averaged value of 358, used for comparisons with the the incoming wind’s incidence angle on the x-velocity and

Fig. 12. Velocity vectors at partition’s height for case A: (a) standard k–e; (b) realizable k–e and (c) RNG k–e.
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1677

temperature is presented in Fig. 10(c) and (d) along the vertical may be partly due to experimental errors but it may also occur
direction of experimental locations B1 and B2. It is seen that the because of the possibility of air infiltration through the opposite
results are little sensitive to a 15% change of the mean incidence door (A1, Fig. 2a), considered as outlet in the mathematical model,
angle. due to random direction variations, since B2 was placed next to the
It should be emphasized that the usefulness of the developed outlet door. That is why the error at location B1, which was placed at
model lies in the fact that the same uncertainty in the results exists the same vertical direction, was also high. Specifically, it was around
for all designs that may be studied. In other words, although the 4% using the RNG model, which presents the lower discrepancy
error in predicting a variable may be small or large it is going to be value compared to the other models. Referring to the rest of the
also the same for all alternative designs considered. Thus, the experimental locations, the difference varied from 2% to 4% for all
relative change to the results introduced by a design change will be models, while the ‘‘realizable’’ model was found to lead to slightly
certainly predicted correctly. higher differences. The x-velocity discrepancies were higher but
Results obtained by both experimental measurements and could be considered acceptable because of the inevitable uncer-
numerical predictions are presented in Fig. 11. Absolute values of tainties during the experiments, due to uncontrolled weather
x-velocities are used for validation, i.e. absolute values of velocities conditions such as external wind speed and turbulence. Particularly,
presented in Figs. 9(b) and 10(c) at each experimental location, as the predicted inlet x-velocity (point A2, Fig. 2a) differed from
only one-dimensional velocity magnitude measurements were measurements about 11–12% using the standard and the ‘‘realiz-
performed. able’’ models, respectively, being much lower using the RNG model
There are significant points to be noted resulting from the (around 2%). On the contrary, the standard k–e model provided a
comparison among experimental and numerical results. First of all, better solution at B1 and B2 (close to the outlet door) with the
it can be noticed that for case A, which represents noon hours with discrepancy being 16% and 9.5%, respectively. However, using the
358 incidence angle, the discrepancy for temperature at the inlet ‘‘realizable’’ model, a better prediction was obtained for the
door (A2) was only 0.8% using the standard k–e model, being locations in front of the inlet door (C1, C2, C3), i.e. the discrepancy
slightly higher using the other two models but in any case not at the location C1 was approximately 8% rather than 25% and 18%
exceeding 1.5% when applying the RNG model. Low differences using the other two models. Much lower discrepancies can be
were also obtained referring to the middle location of the chamber observed concerning experimental case B which represents the
(point D), especially by the RNG k–e model for which the calculated afternoon hours. For example, the difference of the computed x-
minimum error was only 0.9%. As far as any other location is velocity decreased and it was about the same for all models leading
concerned, it is observed that the discrepancy for temperature is a to an adequate prediction, i.e. the discrepancy at the inlet (A1), B3,
little higher and obtains its maximum value at the location B2, C1 and C2 does not exceed 6%. The same conclusion may be stated
approximately 4.1% using the RNG model, which represents the for the temperature prediction with all models performing well
lower discrepancy among all the applied models. This difference enough, compared to the experimental results.

Fig. 13. Velocity vectors at partition’s height for case B: (a) standard k–e; (b) realizable k–e and (c) RNG k–e.
1678 G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681

It is obvious that at the afternoon hours the proposed models also acceptable quantitative predictions in the sense of relative
performed better for both velocity and temperature predictions. designs. Due to the existing uncertainties depending on the
This may be due to the omission of heat conduction inside the walls constant variation of the external weather conditions, the
and of solar radiation. According to the assumptions used, the proposed approach is considered sufficient enough to represent
imposed boundary conditions at internal wall surfaces were the natural cross-ventilation.
experimental results at just one point of each wall, so in reality a The distributions of velocity magnitude vectors, obtained by all
mean value of temperature applied at each wall, rather than a models for each case studied, are presented in Figs. 12 and 13. It
distribution of temperature (which would be an outcome of the can be noticed that all models lead to almost identical internal flow
solution procedure if conduction had been taken into considera- behaviour and, at least qualitatively, the recirculations occur at the
tion). Furthermore, solar radiation may have a significant impact same place. On the contrary, interesting differences may be
on the physical phenomenon, especially during the noon hours, observed for the external flow. The main difference is that the
thus providing a different temperature distribution than that standard k–e leads to larger vortices downwind of the test chamber
calculated when this mechanism is neglected. Other errors could compared to the vortices calculated by the other two models.
be due to the possibility of air infiltration through small Especially the ‘‘realizable’’ model leads to the smallest vortices at
experimental building cracks and also due to the existence of this particular area of the flow. However, since there were no
small obstacles such as packets used for the equipment storage. experimental data at this area, these differences serve just for air
Finally, it may be concluded that all models used for the movement evaluation purposes. Furthermore, the area of impor-
numerical simulation are satisfactory, giving qualitatively similar tance is mainly the internal flow domain, as indoor air quality
performance, for both experimental arrangements, while giving issues are investigated in the present study.

Fig. 14. Experimental case A: (a) velocity magnitude (m/s), (b) temperature distribution (K), (c) iso-relative humidity (%), (d) iso-PMVNV and (e) iso-PPD (%).
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1679

Fig. 15. Experimental case B: (a) velocity magnitude (m/s), (b) temperature distribution (K), (c) iso-relative humidity (%), (d) iso-PMVNV, (e) iso-PPD (%) and (f) iso-PD (%).

3.2. Thermal comfort study The distributions of local velocity, relative humidity, tempera-
ture, PMVNV and PPD, for both experimental cases, are presented in
Thermal comfort was examined under the experimental Figs. 14 and 15 at the traverse plane of the chamber. Referring to the
conditions which correspond to summer days and consequently experimental case A (Fig. 14), it can be observed that the tem-
to high temperatures. Space restrictions dictate that only results perature difference at all vertical sites that represent occupied zones
obtained by the RNG k–e model are presented, as this model does not exceed 3 8C (threshold value for thermal comfort [24]).
presents the best agreement with experiments especially for
temperature for both cases studied. For example, in case A, the
‘‘realizable’’ model led to the highest temperature differences at all
experimental locations (see Fig. 11b) and thus it leads to under-
prediction of occupant’s thermal perception. The standard k–e
could also be chosen due to similar discrepancies with those of the
RNG. However, because of the better formulation validity of the
latter for flows that include strong streamline curvature and
vortices [37], like in the present one, it has been selected for
presentation. It should be emphasized that thermal comfort could
equally well be studied using the other two models, because of the
flexibility of the computer program developed for the present
work, which calculates thermal comfort parameters, and is
compatible with any available flow distribution. Fig. 16. Temperature variation at the middle of the chamber.
1680 G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681

In Fig. 16, for example, the maximum temperature at the middle domain, while it can obtain relatively high values near the
vertical direction of the chamber is 307.1 K, while its minimum partition, due to high velocity gradients that occur there.
value is 305.8 K, even though the external temperature was 7 8C
above the initial wall temperatures. 4. Conclusions
As far as relative humidity is concerned, it is observed that the
internal values are higher than the external, due to the An experimental method has been developed to determine the
temperature decrease inside the chamber and the small pressure airflow pattern and indoor thermal environment in case of natural
differences between the calculated local pressure and the external cross-ventilation. Two experimental arrangements were examined
atmospheric pressure. The PMVNV values reveal an unacceptable for noon and afternoon hours under hot summer- and moderate
internal thermal environment due to high external temperatures. wind-conditions. Furthermore, a mathematical model was devel-
Specifically, this factor ranges from a mean value of 1.9 in the bulk oped and applied to study the indoor environment computation-
flow to 2.91 near the inlet-door area, representing warm and hot ally, using finite-volume techniques and three high-Reynolds
sensations in the occupied zone, respectively. This is also obvious RANS models. Steady-state simulations were performed incorpor-
using the PPD factor which received high values especially near the ating – as stable as possible – measurements for some of the
partition (65%) and in front of the inlet door (87%). Consequently, in boundary conditions applied. A sensitivity study was performed
case of both doors being open, the reflective insulation applied for concerning the impact of the terrain’s roughness and also of the
the walls is not enough for the establishment of the desired fluctuating recorded wind speeds and incidence angles on the
thermal comfort conditions, due to the strong thermal load numerical results. It was found that no significant differences
entering the room by the external hot air masses. In case of occurred for flat- and typical-rural’s terrain roughness. A similar
moderate external temperatures, PPD could be minimized due to observation is true referring to a 10% fluctuating wind speed and a
sufficient mixing, provided by the non-symmetrical locations of 15% fluctuating incidence angle. The numerical predictions
the openings. The same conclusions are valid for the experimental obtained by all turbulence models were generally in acceptable
case B (Fig. 15). The maximum PPD is calculated 76.1% and is now agreement with the experimental measurements; and, thus, they
observed near the partition (Fig. 15e), rather than near the south provide an ‘‘image’’ of the airflow which may represent natural
wall as in case A (Fig. 14e). This is due to the change of wind ventilation as a result of prevailing wind effects. The RNG k–e model
direction that leads to hot air infiltration through the east door. The performed relatively better, especially for temperature predictions
mean value of PPD in the occupied zone is around 62% and 55% for and it was chosen and used further for thermal comfort estimation
cases A and B, respectively, corresponding to 38% and 45% purposes, under both measured experimental conditions. For this
acceptability of the indoor conditions. For both cases, internal reason, the extended PMV model was implemented in the CFD model
wall temperatures are similar. While the infiltrating air’s using the expectancy factor and the reduction of the metabolic rate
temperature is reduced by 3.4 8C from case A to case B, the according to a linear interpolation technique. The adaptive model for
corresponding PPD reduced by approximately 7% in the occupied natural ventilation was also used for additional thermal comfort
zone. Thus, indoor thermal environment was slightly improved estimation purposes. It is concluded that the indoor thermal
during the afternoon hours (case B) but still remains unacceptable. environment considered is unsatisfactory in terms of thermal
This is not surprising as the infiltrating outdoor air enters the perception using either model (for case B), as for both cases studied
chamber at a vertical incidence angle and thus provides maximum thermal acceptability was calculated below the recommended 80%,
thermal load. Since the recommended PPD values for an acceptable referring to both local and global assessment. However, due to the
indoor environment are below 20% (80% acceptability) [26], the non-symmetrical locations of the openings, natural cross-ventilation
current indoor environment is prohibitive for both cases studied. can provide well-mixed conditions, leading to low temperature
This is also true, using the adaptive model for case B concerning differences in the occupied zone and minimize local air draughts,
global thermal comfort assessment. The optimum indoor tem- even in regions close to internal obstacles which may represent
perature for thermal comfort is calculated 27.875 8C, using Eq. (20), furnishings or any building equipment. Finally, it is concluded that
by setting the external temperature as the mean value of the reliable predictions may be obtained using numerical simulations,
30 min measurements. Thus, the maximum indoor temperature within relatively modest computer resources.
for 80% acceptability should be 31.375 8C, while the average
temperature predicted by the CFD model is approximately 31.8 8C Acknowledgments
(Fig. 15b). This means that indoor environment is again predicted
out of the optimum temperature range. Referring to local thermal This research work was co-funded by the European Social Fund
comfort using the adaptive model, the same conclusion is true, (75%) and National resources (25%) through the Operational
especially near the partition. Due to high temperatures of the Program for Educational and Vocational Training II (EPEAEK II)
infiltrating air at this area, predicted temperature is at least 1 8C ‘‘Archimedes II’’.
higher than the optimum one for 80% acceptability. However,
temperature differences at all vertical directions are below the References
threshold value, as in case A (see Fig. 16). It is meaningful, for this [1] M. Orme, Estimates of the energy impact of ventilation and associated financial
case, to investigate also the local PD factor, standing for air expenditures, Energy and Buildings 33 (3) (2001) 199–205.
draughts occurring when temperatures are lower than the value of [2] C. Alloca, Q. Chen, L.R. Glicksman, Design analysis of single-sided natural ventila-
tion, Energy and Buildings 35 (8) (2003) 785–795.
human thermal neutrality (34 8C) under sedentary activity. It is [3] Energy consumption guide 19, Energy efficiency in offices, Energy efficiency
observed that, even though the indoor thermal conditions could be office/HMSO, London, 1993.
considered unacceptable in terms of thermal perception (PMVNV/ [4] P.F. Linden, The fluid mechanics of natural ventilation, Annual Review of Fluid
Mechanics 31 (1999) 201–238.
PPD, adaptive), the draught sensation of any occupant remained at
[5] E. Gratia, I. Bruyere, A. De Herde, How to use natural ventilation to cool narrow
low acceptable values (<20%) [24] in the bulk flow (see Fig. 15f). office buildings, Building and Environment 39 (10) (2004) 1157–1170.
This was due to the low indoor velocities, especially in the [6] Hazim B. Awbi, Chapter 7—Ventilation, Renewable and Sustainable Energy
Reviews 2 (1–2) (1998) 157–188.
recirculation zones, and also to low turbulence intensity. However,
[7] O.S. Asfour, M.B. Gadi, A comparison between CFD and network models for
PD can exceed recommended values in front of an inlet door predicting wind-driven ventilation in buildings, Building and Environment 42
because of the dominant maximum velocity of the internal (12) (2007) 4079–4085.
G.M. Stavrakakis et al. / Energy and Buildings 40 (2008) 1666–1681 1681

[8] L. Mora, A.J. Gadgil, E. Wurtz, C. Inard, Comparing zonal and CFD model predic- [28] N.C. Markatos, G. Cox, M.R. Malin, Mathematical modeling of buoyancy-induced
tions of indoor airflows under mixed convection conditions to experimental data, smoke flow in enclosures, International Journal of Heat and Mass Transfer 25 (1)
in: Proceedings of 3rd European Conference on Energy Performance and Indoor (1982) 63–75.
Climate in Buildings, Lyon, France, October 23–26, 2002. [29] R.J. de Dear, G.S. Brager, Developing an adaptive model of thermal comfort and
[9] L. Mora, A.J. Gadgil, E. Wurtz, Comparing zonal and CFD model predictions of preference, ASHRAE Transactions 104 (1a) (1998) 145–167.
isothermal indoor airflows to experimental data, Indoor Air 13 (2003) 77–85. [30] R.J. de Dear, G.S. Brager, D. Cooper, Developing an adaptive model of thermal
[10] Y. Jiang, D. Alexander, H. Jenkins, R. Arthur, Q. Chen, Natural ventilation in comfort and preference, ASHRAE RP-884 Final report, American Society of Heat-
buildings: measurement in a wind tunnel and numerical simulation with large ing, Refrigerating and Air-Conditioning Engineers, Atlanta, 1997.
eddy simulation, Journal of Wind Engineering and Industrial Aerodynamics 91 (3) [31] KIMO thermo-anemometer multi-probes VT 200F, technical data sheet.
(2003) 331–353. [32] Supco DLTH temperature/humidity data loggers, Instruction manual, USA, 2002.
[11] G. Evola, V. Popov, Computational analysis of wind driven natural ventilation in [33] G.M. Stavrakakis, M.K. Koukou, M.Gr. Vrachopoulos, N.C. Markatos, Study of
buildings, Energy and Buildings 38 (5) (2006) 491–501. airflow pattern and thermal environment in naturally ventilated buildings, in:
[12] S. Kato, S. Murakami, A. Mochida, S. Akabayashi, Y. Tominaga, Velocity-pressure CD Proceedings ENERTECH2007, Athens, Greece, October 18–21, 2007.
field of cross ventilation with open windows analysed by wind tunnel and [34] J. Franke, C. Hirsch, A.G. Jensen, H.W. Krüs, M. Schatzmann, P.S. Westbury, S.D.
numerical simulation, Journal of Wind Engineering and Industrial Aerodynamics Miles, J.A. Wisse, N.G. Wright, Recommendations on the use of CFD in wind
41–44 (1–3) (1992) 2575–2586. engineering, COST C14 International Conference on Urban Wind Engineering and
[13] M. Ohba, K. Irie, T. Kurabuchi, Study on airflow characteristics inside and outside a Buildings Aerodynamics, von Karman Institute, Rhode-Saint-Genèse, Belgium,
cross-ventilation model, and ventilation flow rates using wind tunnel experi- 05.-07.05, 2004.
ments, Journal of Wind Engineering and Industrial Aerodynamics 89 (14–15) [35] T. Bartzanas, T. Boulard, C. Kittas, Effect of vent arrangement on windward ventila-
(2001) 1513–1524. tion of a tunnel greenhouse, Biosystems Engineering 88 (4) (2004) 479–490.
[14] D.R. Ernest, F.S. Bauman, E.A. Arens, The effects of external wind pressure [36] N.C. Markatos, Computational fluid flow capabilities and software, Ironmaking
distributions on wind-induced air motion inside buildings, Journal of Wind and Steelmaking 16 (4) (1989) 266–273.
Engineering and Industrial Aerodynamics 41–44 (1–3) (1992) 2539–2550. [37] A. Yakhot, S. Orszag, Renormalisation group analysis of turbulence: I. Basic theory,
[15] Y. Jiang, Q. Chen, Study of natural ventilation in buildings by large eddy simula- Journal of Scientific Computing 1 (1) (1986) 1–51.
tion, Journal of Wind Engineering and Industrial Aerodynamics 89 (13) (2001) [38] T.H. Shih, W.W. Liou, A. Shabbir, Z. Yang, J. Zhu, A new k–e eddy viscosity model
1155–1178. for high Reynolds number turbulent flows, Computers Fluids 24 (3) (1995)
[16] E. Dascalaki, M. Santamouris, A. Argiriou, C. Helmis, D. Asimakopoulos, K. Papa- 227–238.
dopoulos, A. Soilemes, Predicting single sided natural ventilation rates in build- [39] B.E. Launder, D.B. Spalding, Mathematical Models of Turbulence, Academic press,
ings, Solar Energy 55 (5) (1995) 327–341. 1972.
[17] Y. Jiang, Q. Chen, Buoyancy driven single-sided natural ventilation in buildings [40] G.M. Stavrakakis, N.C. Markatos, Computational prediction of buoyancy-driven
with large openings, International Journal of Heat and Mass Transfer 46 (6) (2003) airflow in air-conditioned enclosures, in: Proceedings of 2nd IC-SCCE, Athens, July
973–988. 5–8, 2006.
[18] K.A. Papakonstantinou, C.T. Kiranoudis, N.C. Markatos, Numerical simulation of [41] E. Galea, N.C. Markatos, The mathematical modeling and computer simulation of
air flow field in single-sided ventilated buildings, Energy and Buildings 33 (1) fire development in aircraft, International Journal of Heat and Mass Transfer 34
(2000) 41–48. (1) (1991) 181–197.
[19] N.C. Markatos, Computer analysis of building ventilation and heating problems, [42] S.V. Patankar, D.B. Spalding, A calculation procedure for heat, mass and momen-
in: Proc. Int. Conf. on Passive and Low Energy Architecture, Pergamon Press, Crete, tum transfer in three-dimensional parabolic flows, International Journal of Heat
1983, pp. 667–675. and Mass Transfer 15 (10) (1972) 1787–1806.
[20] Q. Chen, Using computational tools to factor wind into architectural environment [43] J.H. Ferziger, M. Peric, Computational Methods for Fluid Dynamics, third ed.,
design, Energy and Buildings 36 (12) (2004) 1197–1209. Springer-Verlag, Berlin, Heidelberg, New York, 2002.
[21] A. Mochida, H. Yoshino, T. Takeda, T. Kakegawa, S. Miyauchi, Methods for [44] K.B. Shah, J.H. Ferziger, A fluid mechanicians view of wind engineering: large eddy
controlling airflow in and around a building under cross-ventilation to improve simulation of flow past a cubic obstacle, Journal of Wind Engineering and
indoor thermal comfort, Journal of Wind Engineering and Industrial Aerody- Industrial Aerodynamics 67&68 (1997) 211–224.
namics 93 (6) (2005) 437–449. [45] J. Franke, Introduction to the prediction of wind effects on buildings with
[22] ISO Standard 7730-94, Moderate thermal environments—determination of PMV computational wind engineering (CWE), Lecture III: Boundary conditions, com-
and PPD indices and specification of the conditions for thermal comfort, Inter- putational domains and grids, Advanced schools 2006 at International Centre for
national Standards Organization, Geneva, 1994. Mechanical Sciences (CISM), Udine, Italy, September 18–22, 2006.
[23] P.O. Fanger, A.K. Melikov, H. Hanzawa, J. Ring, Air turbulence and sensation of [46] N.C. Markatos, The mathematical modeling of turbulent flows, Applied Mathe-
draught, Energy and Buildings 12 (1) (1988) 21–39. matical Modeling 10 (3) (1986) 190–220.
[24] Thermal environmental conditions for human occupancy, ANSI/ASHRAE 55, 1992. [47] Fluent Inc. Fluent User’s Guide, Version 6.2, Fluent Inc., Lebanon, NH, USA, 2005.
[25] P.O. Fanger, J. Toftum, Extension of the PMV model to non-air-conditioned [48] Huser Asmund, Nilsen Pa1 Jahre, Skatun Helge, Application of k–e model to the
buildings in warm climates, Energy and Buildings 34 (6) (2002) 533–536. stable ABL: pollution in complex terrain, Journal of Wind Engineering and
[26] Thermal environmental conditions for human occupancy, ANSI/ASHRAE Standard Industrial Aerodynamics 67 and 68 (1997) 425–436.
55, 2004. [49] N.J. Cook, The Deavis and Harris ABL model applied to heterogeneous
[27] E. Bacharoudis, M.Gr. Vrachopoulos, M.K. Koukou, D. Margaris, A.E. Filios, S.A. terrain, Journal of Wind Engineering and Industrial Aerodynamics 66 (3)
Mavromatis, Study of the natural convection phenomena inside a wall solar (1997) 197–214.
chimney with one adiabatic and one wall under a heat flux, Applied Thermal [50] R.H. Perry, D.W. Green, Perry’s Chemical Engineers’ Handbook, McGraw-Hill,
Engineering 27 (13) (2007) 2266–2275. 1999.

Vous aimerez peut-être aussi