Energy and Buildings 147 (2017) 155165

Contents lists available at ScienceDirect

Energy and Buildings

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

Effect of furniture and partitions on the surface temperatures in an

ofce under forced ow
F.A. Dominguez Espinosa , Leon R. Glicksman
Department of Mechanical Engineering, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA

a r t i c l e i n f o a b s t r a c t

Article history: Computational Fluid Dynamics (CFD) simulations were performed to study the effects of surfaces typi-
Received 4 November 2016 cally found in an ofce, such as partitions and desks, on the temperatures of the surfaces surrounding
Received in revised form 10 March 2017 the occupants. An accurate estimation of these temperatures is needed to assess the thermal comfort
Accepted 21 April 2017
conditions in a space, thus ensuring that the occupants are satised with their thermal environment. A
Available online 3 May 2017
parameter quantifying the area available to convect heat to the air was proposed to account for the effect
of surfaces of different sizes and positions. A higher amount of available convective area was found to
Keywords:
decrease the temperatures of the surfaces in the room, and vice versa. Although the air temperature also
Room surface temperatures
CFD
increased with additional available convective area, this change was found to be small (typically below
Mean radiant temperature 15%). Moreover, the shape of the air temperature prole far from the surfaces remained unchanged when
Operative temperature changing the amount of available convective area. Therefore, the operative temperature that the occu-
Thermal comfort pants experience is lower in spaces with a large amount of available convective area. Expressions to
estimate the temperatures of the surfaces, generally within 10% of the air temperature rise in the space,
were proposed. These expressions can be useful during the early stages of the design process, when CFD
simulations are impractical and the exact location and type of furniture in a space are not well known.
Together with a model to predict the air temperature prole, these correlations can improve the accuracy
of thermal comfort assessment.
2017 Elsevier B.V. All rights reserved.

1. Introduction pute the operative temperature experienced by the occupants.

The thermal environment conditions that ensure that occupants
are comfortable in a space refer not only to the air temperature
or velocity near them, but also to the temperatures of the sur-
faces that surround the occupants. These surfaces exchange heat
with the occupants via radiation, so surfaces that are too hot or
cold can result in unacceptable thermal conditions in the space,
even if the air temperature is by itself acceptable [1]. Asymmetry
in the temperatures of these surfaces can also result in thermal
dissatisfaction [1]. The Graphical Comfort Zone Method for Typi-
cal Indoor Environments and the Optional Method for Determining
Acceptable Thermal Conditions in Naturally Ventilated Spaces the
adaptive comfort method of the ASHRAE Standard 55-2010 take
into account the radiation exchange between people and surfaces
by means of the operative temperature [1].
Models commonly used to assess the thermal conditions in
ofces, including CFD and multi-zone models, can be used to com-
However, the geometries of these models rarely include surfaces
found in a real ofce such as partitions and desks (e.g. [2]). These
surfaces can have an important effect on the temperature of all the
surfaces in the room, including the ceiling, oor and walls, because
they facilitate the transfer of heat from the sources (e.g. lighting,
equipment and occupants) to the air.
Ultimately, the temperatures of the surfaces in an ofce depend
on several factors, including convective heat transfer coefcients
associated with the surfaces and the view factors between them,
among others. Taking into account all these parameters is time-
consuming and might not be appropriate during the early stages
of the design process, when the exact position of the furniture and
partitions in an ofce is not well dened. Hence, this work presents
a simple method to estimate the temperatures of the surfaces in a
room that does not require exact knowledge of the position of the
furniture or the partitions.
1.1. Heat transfer paths in an ofce

Corresponding author. In a general sense the heat transfer processes in a space transport
E-mail address: alonso@mit.edu (F.A. Dominguez Espinosa). thermal energy from the source(s) to the heat sink(s). Heat is trans-

http://dx.doi.org/10.1016/j.enbuild.2017.04.067
0378-7788/ 2017 Elsevier B.V. All rights reserved.
156 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165

Nomenclature
volumetric thermal expansion coefcient of air
Tr air temperature rise across room
 dimensionless temperature

Roman symbols
A tting constant
Aconv unheated area available for convection heat transfer
Aconv dimensionless unheated convective area
Arhw Archimedes number based on inlet height
Ad dimensionless desk area
Ap dimensionless partition area
Aw dimensionless inlet area Fig. 1. Thermal circuit representation of a space with heat gains generated at ceiling
B tting constant level (a) without partition and (b) with a partition.

C tting constant
cp air heat capacity at constant pressure air temperature prole, instead they rely on the assumption of per-
D tting constant fect mixing of the air, so the temperature in a zone is uniform and
E tting constant equal to the temperature of the outlet. This so-called well-mixed
F tting constant temperature, Tair , can be obtained using the steady-state energy
G tting constant conservation equation:
H room height
Q
hc convective heat transfer coefcient Tair = Toutlet = Tinlet + (1)
mcp
hr radiative heat transfer coefcient
h* dimensionless height coordinate where Toutlet is the outlet temperature, Tinlet is the inlet tempera-
hw inlet height ture, Q are the total heat gains in the room, m is the mass ow rate
hw dimensionless inlet height of air through the space and cp is the heat capacity of air at constant
I tting constant pressure. The well-mixed temperature serves as a convenient ref-
J tting constant erence for the present work because its value is independent of
K tting constant the amount of furniture or the distribution of heat gains in a room
L room length under forced ow (wind-driven natural ventilation or mechanical
l* dimensionless length coordinate ventilation). Moreover, this temperature can be calculated using
m airow rate multi-zone models according to interior and, in the case of natural
Q total heat gains ventilation, exterior conditions.
Rfc oor convective resistance A thermal resistor network is a useful tool to represent the heat
Rpc partition convective resistance transfer paths in a room. As a simple example, consider a very long
r
Rc,f radiative resistance from the ceiling to the oor and wide space, where the view factors from the ceiling to the walls
r
Rc,p radiative resistance from the ceiling to a partition are small enough to be neglected. Assume also that the tempera-
r
Rp,f radiative resistance from the partition to the oor ture of the surfaces are uniform and that the only heat source is
due to lighting (or solar heat gains), which is located at the ceiling
Ta mean air temperature
level, while the other surfaces are adiabatic. In a real room, except
Tair air well-mixed temperature
for other heat sources, the ceiling is generally the warmest surface
Tceiling ceiling temperature
in the space because hot air that rises due to buoyancy stagnates
Toor oor temperature
under it. As a consequence, the convective heat transfer coefcient
Tinlet inlet/supply temperature
associated with the ceiling is signicantly lower than that of the
To operative temperature
other surfaces in the room. The only effective mechanism available
Toutlet outlet temperature
to transfer heat from the ceiling is via radiation. The same holds
Tpartition partition temperature
for the simple example considered here, therefore it is possible to
Trad mean radiant
assume that the convective heat transfer from the ceiling to the
u inlet velocity
air is negligibly small. This is a common assumption in ventila-
x x coordinate
tion studies (e.g. [4]). Fig. 1a shows the thermal resistor network of
y y coordinate r is the radiation thermal resistance
this example. In this gure, Rc,f
z z coordinate
between the ceiling and the oor, and Rfc is the convective thermal
resistance from the oor to the air. Tceiling , Toor and Tair are the
temperatures of the ceiling, oor and the well-mixed temperature
ferred from active surfaces such as lights, equipment and people of the air, respectively. Fig. 1b shows the same space but with a
r and Rr are the radia-
thin partition in the room. In this gure, Rc,p
by convection to the air and by radiation to the walls, oor, parti- p,f
tions and furniture, which is then transferred to the surrounding tion thermal resistances between the ceiling and the partition, and
air. In this work, it is assumed that lights, equipment and occu- between the partition and the oor, respectively. Rfc is the convec-
pants are the only heat sources in the room, while the air in the tive thermal resistance from the partition to the air and Tpartition
space is the only heat sink in steady state. The temperature of the is the temperature of the partition. The well-mixed temperature of
air in the space is not uniform, instead it is higher near the ceiling the air does not change when adding the partition. Now some of the
in spaces with either displacement or mixing ventilation systems radiation from the ceiling is directed to the partition, reducing the
[3]. However, multi-zone models cannot be used to compute the direct radiation from the ceiling to the oor. The total surface area
 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
Fig. 2. A naturally ventilated ofce with partitions in between rows of occupants.
The area of the partition is lp hp , which in the example shown in this gure is equal
to 60% of the ofces side walls area of L H.
in the room below the ceiling is increased. This area absorbs radia-
tion from the ceiling and transfers the heat by convection to the air.
Since the total area is increased, the average surface temperatures
are reduced.
Similar consequences are expected when adding horizontal sur-
faces, for example desks. The desks reduce the view factor between
the ceiling and the oor, but allow an indirect connection through
them. A fraction of the heat gains radiated onto the top surfaces of
the desks will be conducted to their bottom surfaces and then radi-
ated to the oor. The net result is an increase in the area available
to convect heat to the air. Increasing this area reduces the effective
thermal resistance between the heat sources and the heat sink.
It is possible to reduce the effective thermal resistance by means
other than adding partitions or furniture. For a given room height,
length and heat gain intensity, a thinner room has surfaces with
temperatures closer to the well-mixed temperature than wider
rooms, because the walls have a greater participation in the radia-
tion exchange with the heat sources.
Adding convective area inuences the local air temperature dis-
tribution within the room. This, in turn, could alter the operative
temperature experienced by the occupants, an effect that is not con-
sidered in zonal models. The following sections of this work present
a method to estimate the effect of additional convective area on the
temperature of the air and the surfaces of a space, thereby allowing
a more accurate assessment of thermal comfort when CFD is not
practical.
2. CFD model
Different CFD domains were used to assess the effect of available
convective area on the temperature of the surfaces surrounding the
occupants to test the hypothesis proposed in the previous section.
Fig. 2 shows a naturally ventilated ofce with partitions between
rows of occupants, while Fig. 3 shows the same ofce with desks
or tables instead of partitions. The geometry of the CFD simulations
is based on an actual ofce building in Tokyo served by a hybrid
naturalmechanical ventilation system, but it is intended to be rep-
resentative of open plan ofces with few obstacles to the airow
between the inlets and the outlet. The domain is 15 m long and 3
m height, L and H, respectively, in these gures. The width of the
ofce, W, is large compared to its length and height, so Figs. 2 and 3
show only a fraction of the width of the ofce. The space has inlets
along one of the walls, each measuring 0.15 m 0.5 m and with
their centers located at 1 m from the oor. The distance between
157
Fig. 3. A naturally ventilated ofce with desks. The area of the desks, including the
top and bottom surfaces, is measured with respect to the oor area, W L. In this
case, the tables cover 80% of this area.
the centers of two adjacent inlets is 2 m. The outlet is a 0.1 m height
opening near the ceiling that extends the entire width of the room.
The center of the outlet is located at 2.8 m from the oor.
Each occupant in the space is represented by a rounded-top
cylinder, a simple shape that was found to replicate the behavior of
the convective ow generated by a sitting human and the radiation
exchange between a sitting person and the surfaces of a room like
the one considered here [3,5]. Each occupant is 0.31 m radius and
1.13 m high. The separation between occupants is 2 m (centerline to
centerline); the rst occupant is located at 2 m from the inlet wall.
There are ve occupants per row, located at 1 m from the partitions
or, equivalently, exactly in the middle between inlets.
Each desk between occupants is 75 cm long with a thickness
of 7.5 cm, while the large desk at the end of the room is 3 m long
and of the same width. The desks of contiguous rows touch each
other, so their width is the same as the width of the ofce. The
number of desks were varied so their surface area (including top
and bottom surfaces) covers either 40% or 80% of the area of the
oor (L W). As an example, Fig. 3 shows an ofce where the
desks cover 80% of the oor area. Desks of three materials with
different thermal conductivities were used: wood (kt = 0.17 W/mK)
and metal (kt = 17.3 W/mK), as well as a case with adiabatic desks
(kt = 0 W/mK). Each simulation had desks of only one material.
The size of the partitions was also varied. The area of each one,
lp hp , was changed so that it would be equal to 0% (no partitions),
20%, 40%, 60% and 100% of the ofces side wall area, L H. The
length of the partitions was varied while their height was kept con-
stant at 2.25 m (except for the 0% and 100% cases). The location of
the center of the partitions was kept at 7.5 m from the inlet.
All these domains are representative of very wide spaces, so
the L H walls have no effect on the heat exchange of the room.
To study the effect of these walls two additional domains were
simulated. These domains were rooms with a width of 4 m (Fig. 4)
and 6 m. Together with the cases with partitions covering 0% and
100% of the side wall area, the simulations cover a wide range of
space widths, from very thin rooms to very wide open-plan ofces.
The boundary conditions of the CFD model were dened as fol-
lowing:
1 Inlets A uniform air velocity was specied at each one of the
inlets, the value of which was modied for different simula-
tions. The air temperature at the inlets was dened to be 295 K
(21.85 C).
2 Outlet The outlet was dened as a zero gauge pressure boundary.
158 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
Fig. 4. A naturally ventilated ofce with a width of 4 m.
Fig. 5. Mesh used in the study seen on a plane that extends along the length of
the room through the occupants. The inset shows the ne ination layer over the
surfaces in the room (in this case over an occupant and the oor). This is a 2D section
of a 3D mesh.
3 Ceiling and occupants These surfaces generated a uniform
amount of heat ux to represent the heat gains due to lighting
(in the ceiling) and due to the occupants and equipment. Approx-
imately 25% of the total heat gains were dened to be evenly
generated at the ceiling; the rest, by the occupants. The appro-
priateness of this division has been demonstrated before [6]. The
magnitude of the heat gains were also modied in different sim-
ulations.
4 Tables At the surfaces of the tables an energy balance was
enforced. Heat was allowed to be radiated to and from the sur-
faces of the tables, convected to the air or conducted through the
tables.
5 Other surfaces The rest of the surfaces, including the partitions,
were dened as adiabatic with no-slip. An adiabatic, no-slip sur-
face interacts with the ow of air, so heat can be transferred to
and from the surface by convection and radiation. There is also
non-zero shear stress at these surfaces.
Air was modeled using the Boussinesq approximation, with a
reference density of 1.14 kg/m3 . The viscosity and volumetric ther-
mal coefcient of expansion of the air were set at 1.8 kg/ms and
3.3 103 1/K, respectively.
All the cases were simulated using Fluent (part of the ANSYS 14.5
suite), solving the RANS equations for turbulent ow and the energy
conservation equation. The turbulence model used is the RNG k.
This model has been shown to accurately simulate room airow
dynamics [7]. The near-wall regions were meshed using very ne
elements (y+ 5), as required by the enhanced wall treatment
used in this study [8,9]. Fig. 5 shows an example of a mesh used in
these simulations.
Thermal radiation exchange was included using the surface-
to-surface model. This radiation model treats the air as perfectly
transparent to radiation, which is a valid approximation for air with
low humidity content [10,11]. The emissivity of all the surfaces was
dened to be 1. This simplication is justied given that typical
materials in an ofce, except for polished metals, have emissivities
above 0.9 [12]. The radiative view factors between all the surfaces
in the room were calculated using ray tracing.
The cases were run as transient simulations and stopped once
steady state was reached. Steady state was dened to be reached
when both the airow rate at the outlet and the air temperature
rise across the space stopped changing signicantly in value, and
when the value of the temperature rise across the space was within
2% of the value calculated using the energy conservation equation
with the selected airow rate.
2.1. Validation of the CFD model
The CFD code used here was validated by the following studies:
1 Grid independence Results with four different grid sizes
(107,364, 593,598, 726,882 and 2,332,858 elements) were com-
pared against each other in terms of the shape and magnitude
of the resulting vertical air thermal prole. The prole was not
found to change signicantly among the four grid sizes. In partic-
ular, the average difference in temperature between the largest
two meshes at every height was 3%, with a largest deviation of 6%.
Additionally, the mean temperature of the surfaces in the room,
the ceiling, oor and the walls, were compared. The difference in
the temperature of the surfaces between the four grid sizes was
less than 0.4 C, even though the grid size changes by more than
2000%. Finally, following the procedure in Ref. [13], the extrap-
olated relative error and the ne-grid convergence index for the
average temperature of the ceiling and the oor were computed
for the three largest meshes, nding them all below 0.1%. The
largest mesh size was used for the simulations presented in this
work and is shown in Fig. 5.
2 Turbulent air jet development The same CFD code of this study
was used to replicate experimental results of axisymmetric and
3D turbulent jets. Centerline velocity, volume ow rate and jet
width were compared against axisymmetric jet theory and, when
possible, experimental data [6]. For the specic case of a rectan-
gular opening, the shape of the inlet used in this work, the error
in the dimensionless centerline velocity of the CFD simulation
is less than 10% when compared to theoretical models. Similar
magnitude of errors (less than 10%) were found for the dimen-
sionless ow rate and dimensionless jet radius at a distance of
one diameter from the inlet [6].
3 Turbulent thermal plumes The CFD code was compared against
theory and experiments of thermal plumes in terms of their cen-
terline velocity, excess temperature, ow rate, and inertial and
thermal plume width [6]. The difference between CFD and the-
ory and experiments was also found to be below 10% in all the
variables. Additionally, CFD simulations were compared against
experimental measurements of centerline velocity of thermal
plumes generated by a standing person nding a largest differ-
ence of 7% between the simulations and the experiments [5].
4 Thermal stratication in water scale models and full-sized
rooms The CFD code used here was compared against the exper-
iments of Nansteel and Greif [14], Olson et al. [15] and Ray [16].
The CFD simulations replicated, both in shape and temperature
magnitude, the experiments by Nansteel and Greif with an R2
of 0.97 and by Olson et al. with an R2 of 0.84 [17]. Airow rates
under buoyancy-driven ventilation were compared to the scale
experiments by Ray [16], nding an average deviation of 13.5%.
In addition, CFD simulations using a similar geometry than in
this study were compared against experiments by Chen and
Glicksman [18] using a climate chamber. The maximum devi-
 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
ation between the simulation and the experiments far from the
oor was approximately 13% [6].
A detailed description and complete set of results of the val-
idation studies related to jet and plume development, as well as
thermal stratication previously mentioned, can be found in Refs.
[5,6,17].
3. Dimensional analysis
The Buckingham  theorem was used to determine a set of
dimensionless numbers that determine the temperature of the sur-
faces in the room according to the ow and heat load conditions in
the space. The dimensionless numbers related to the ow dynam-
ics in the room obtained by this method have also been obtained by
nondimensionalizing the set of equations that govern room airow
[19]. These dimensionless groups are the Archimedes, Reynolds and
Prandtl numbers, and the nondimensional temperature.
The Archimedes number, Arhw , is dened in this work according
to the following equation:
gT r h3w
Arhw = (2)
u2 Aw
where g is the acceleration of gravity, is the volumetric ther-
mal expansion coefcient of air, Tr is the air temperature change
across the room (Tr = Toutlet Tinlet ), hw is the height of the inlets,
u is the inlet velocity, Aw is the inlet area. The Buckingham  theo-
rem allows for the arbitrary selection of the normalizing factors (e.g.
characteristic lengths, areas, velocities, etc.) when constructing a
dimensionless number. In this work the form of the Archimedes
number presented in Eq. (2) is used, instead of an alternative ver-
sion presented before (e.g. [19]), because it has been shown that
the expressions that correlate the temperature of the surfaces in
the room, as well as its air thermal stratication prole, to its
Archimedes number in this form are simple [3,6].
Conceptually, the Archimedes number compares the relative
strengths of the buoyancy forces in the room and the momentum
of the supply air. For this reason, the Archimedes number is related
to the path that the supply air jet follows [3,20]. Moreover, it has
been shown that the air stratication prole of the room depends
on its Archimedes number. Rooms with high Archimedes number
tend to develop a temperature prole of displacement ventilation.
Low Archimedes numbers are related to more turbulent mixing in
the lower part of the room, but less mixing in the upper part of
the room than in displacement ventilation [3,6]. There is also more
variation of the air temperature along the length of the room in a
space with a low Archimedes number [3].
The Reynolds number, ReAw , compares the importance of iner-
tial and viscous effects, while the Prandtl number, Pr, compares
momentum and thermal diffusivities in a uid [11].
The dimensionless temperature, , is given by:
T Tinlet
= (3)
Toutlet Tinlet
where T is the temperature to be normalized, Tinlet is the tem-
perature of the air at the inlet, and Toutlet is the temperature of
the air at the outlet. The well-mixed temperature in dimensionless
form is  a,wm = 1. All the temperatures in this work are expressed
as dimensionless temperatures, including the air temperature and
the temperatures of the surfaces in the room (ceiling, oor, walls,
partitions, tables, etc.).
The boundary conditions of the governing equations result in
dimensionless groups related to the geometry and position of the
inlets and outlets, the strength of the heat sources, the geometry
and location of the tables and partitions and the geometry of the
159
room itself. Some simplications are needed to make the daunting
task of analyzing the effect of all these parameters manageable.
The aspect ratios of the inlet and outlet were found to have a
marginal effect on the air temperature prole, nevertheless their
vertical positions are important [6]. In this work, the vertical posi-
tion of the outlet is dened to be located near the ceiling, a common
assumption in ventilation studies, as this is generally the case in
real commercial buildings. The vertical position of the inlet was
kept constant at 1 m from the oor. Although this height is not
uncommon of windows in some naturally ventilated buildings, it
is not representative of the full range of inlet heights in natu-
rally ventilated buildings, nor of other ventilation systems, such
as mechanically driven mixing ventilation where the air supply is
located closer to the ceiling. The effect of the inlet height has only
been studied qualitatively, so additional research to properly estab-
lish the effect of this parameter on the temperature of the surfaces
in a room is needed. The effects of the length of the room, its height,
and the number, position and geometry of the heat sources have
been studied in exploratory work before [3,6]. Although the effects
of these parameters were found to be generally negligible, more
exhaustive work is necessary to conrm this nding [3].
The air temperature prole far from the heat sources has been
shown to be affected marginally by the exact division of the total
heat gains between lighting (and/or solar gains) at the ceiling and
occupancy [6]. The only exception is when the entire amount of
heat gains in the room are due to lighting or solar gains at the ceil-
ing [6]. In addition, further tests carried out for a different study
showed that the way in which the lighting heat gains are modeled
(as a uniform heat source in the ceiling, as a thin strip along the ceil-
ing or as a separate body that represents suspended ofce lighting
xtures) does not have a signicant impact on the temperature pro-
le of the air far from these heat sources. In particular, for the same
total amount of heat gains, the temperature prole of the air in the
occupant zone changed by less than 5% (and commonly below 2%)
of the room temperature change when changing either the way the
heat sources are modeled, or when changing the percentage of the
total heat gains that were generated by the occupants (with the rest
generated by the lighting sources).
Furthermore, it is not necessary to determine the explicit depen-
dence of the stratication prole on the Reynolds and Prandtl
numbers. Because the Prandtl number is a very weak function of
temperature under typical room conditions (for room air this num-
ber changes by less than 1% under a change in temperature of
25 C [11]), the only practical way to change the Prandtl number
to analyze its effects is by changing the uid that is being studied.
Therefore, in the present study that deals only with air in room ven-
tilation, the Prandtl number was considered to be constant and the
dependence of the stratication on it was thus not studied.
Nielsen [19] has shown experimentally the negligibly small
effect of the Reynolds number on the airow dynamics of rooms
under displacement ventilation, due to similar ow structures at
the very high Reynolds numbers expected in realistic spaces. Sev-
eral other experimental works have shown the small effect of the
Reynolds number on the ow dynamics in rooms, over a wide
range of Reynolds numbers, particularly far from the walls [21] and
when compared to the signicant effect of the Archimedes number
[22,23]. This negligible effect is expected as long as the Reynolds
number is large enough to be in the fully turbulent regime, regard-
less of the driving force of the ow. The Reynolds number based
on inlet conditions, ReAw , used in this study always exceeded 3900,
which is in the fully turbulent jet regime [24]. Additionally, the
independence of the thermal prole with respect to the Reynolds
number was tested and conrmed using the CFD model of this work
[3].
Finally, several dimensionless groups are related to the posi-
tion, orientation and size of the partitions and tables, which include
160 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
the radiative view factors and convective heat transfer coefcient
associated with these surfaces in their denition. However, the
objective of this work is to provide a simple parameter that can
be used to estimate the effect of these surfaces on the tempera-
ture of all the surfaces in the room. Moreover, the exact position
of the surfaces might not be known during the design process of
the space. For these reasons, a simple dimensionless number that
is intended to capture all these parameters is proposed in this work.
This dimensionless number measures the amount of area available
to convect heat to the air. For the simulations with tables, for exam-
ple, the convective area of the entire room includes the oor, the
walls and the surface area of the tables, including the top and bot-
tom ones. The dimensionless available convective area is dened
as:
Aconv
Aconv =
WL
where Aconv is the unheated area that is in contact with air, and W
and L are the width and the length of the room, respectively. In the
case of a very wide space, it is still possible to use this parameter
with a simple modication. In this case, W refers to the width of the
repeating geometry. For example, W refers to the distance between
partitions in Fig. 2, or between occupants in Fig. 3.
Then, under the previous simplications, the dimensionless
temperature can be expressed as:
 
 = f Arhw , hw , Aw , Aconv (5)
where hw is the dimensionless inlet height (normalized by the
height of the room) and Aw is the dimensionless inlet area (nor-
malized by the area of the wall on which the inlets are located).
This dimensionless temperature is given at a certain location in the
room, described by two dimensionless coordinates:
z results for a given set of dimensionless numbers is valid for all com-
h =
H
y
l =
L
where z and y are the z (along the height) and y (along the length)
coordinates at which the temperature is being measured. The x
coordinate (across the room width) was not considered, as the air
temperature was not found to vary signicantly in this direction
far from surfaces (by less than approximately 4%) [3].
4. Methodology
Simulations of the domains presented in Section 2 were per-
formed. To explore the effect of a given dimensionless group, the
value of this group was varied, while the rest of the dimensionless
numbers was kept the same. The Archimedes number was varied
by changing the inlet velocity and/or the internal heat gains. The
dimensionless free convective area, Aconv , was varied by changing
the number of tables or the size of the partitions as explained before.
The dimensionless inlet area was also changed.
The parameters of the simulations were chosen to be repre-
sentative of naturally ventilated spaces. The Archimedes number
considered ranged from a low value of 0.07, up to 7.91. Additional
simulations were run with Archimedes numbers of 63 and 100.
The dimensionless inlet areas used are 6.25 103 (0.0375 m2 each
inlet), 1.25 102 , 5 102 and 6.25 102 , although for some
cases additional area values of 2.5 102 and 3.75 102 were also
simulated. These dimensionless areas are typical of openings in nat-
urally ventilated buildings, although in natural ventilation there
is more variability in size [25]. The heat gains were kept below
125 W/m2 , four times the heat gains in a typical ofce in the United
States (approximately 30 W/m2 ) [12]. The heat gains were divided
between the occupants and the ceiling, to represent the heat load
(4)
Fig. 6. A section of an ofce (includes only one row of occupants), showing the
locations of the measurement planes. In all the different domains used in this work,
the measurement planes extend along the entire width of the space.
due to occupancy and equipment, and due to lighting, respectively.
The temperature rise across the room, Tr , was kept between 2.5 C
and 13 C. The inlet velocity was kept between 0.1 m/s and 6 m/s,
with approximately 75% of the velocities used below the value of
2 m/s. The upper limit for the inlet velocity was chosen to be 6 m/s,
because at this velocity the local air velocity near the occupants was
at most 0.7 m/s, thus staying under the limit value of 0.8 m/s for
sedentary activities of the Graphical Elevated Air Speed Method to
determine comfort conditions of the ASHRAE 55-2010 standard [1].
Nevertheless, by working with dimensionless numbers, the exact
numerical values used in the simulations are less important, as the
(6) binations of parameters that result in the same dimensionless set,
as long as there is not a signicant change in the ow and ther-
(7) mal dynamics in the room (e.g. as long as the ow in the room is
turbulent, et cetera) [26].
Air and surfaces temperatures are reported. Air temperatures
were measured at the planes labeled Measurement planes in
Fig. 6, which are located at a distance from the inlet of 5 m, 9 m,
and 12 m, or equivalently at l* = 1/3, 3/5 and 4/5. Although Fig. 6
shows only one row of occupants with no desks or partitions, the
air temperature was measured at these planes in simulations with
partitions and tables, and in the wider ofces. These planes were
chosen to be located between occupants near the inlet, near the
middle of the length of the room, and near the outlet. Additionally,
average temperatures were obtained for the ceiling, the oor, the
walls and the partitions.
For all the cases, the resulting operative temperature expe-
rienced by the occupants was also computed. The operative
temperature, To , is dened as:
hc Ta + hr Trad
To = (8)
hc + hc
where hc is the convective heat transfer coefcient associated with
the occupants, hr is the radiative heat transfer coefcient, Ta is the
mean air temperature near the occupants and Trad is the mean radi-
ant temperature [1]. Mathematically, the operative temperature is
the weighted average of the mean radiant temperature and the
air temperature around the occupants with the radiative and con-
vective heat transfer coefcients as the weights. The mean radiant
temperature is a function of the view factors between the occupant
and the surfaces surrounding said occupant [1]. Results are pre-
sented as plots of dimensionless operative temperature (calculated
using Eqs. (3) and (8)) against dimensionless available convective
area.
 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
Fig. 7. Dimensionless air temperature,  a , as a function of dimensionless height from the oor, h* , for spaces (a) with different widths, (b) with partitions, and (c) with desks,
at the measurement plane located at 9 m (l* = 3/5) from the inlet.
In this work, the air temperature surrounding the occupants was
estimated using the measurement planes at 5 m and 9 m from the
inlet because these planes are located between the occupants. The
air temperature surrounding the occupants was approximated as
the average air temperature at these measurement planes, from the
oor up to a height of 1.35 m (l* = 0.45), i.e. approximately 20 cm
above the head level of the occupants. The mean radiant tem-
perature was calculated using the average temperature of all the
surfaces (including partitions and tables, if applicable) and the view
factors were computed using ray tracing software. The accuracy of
the ray tracing software has been validated before [3]. The convec-
tive and radiative heat transfer coefcients for the occupants where
found to be similar with a value of approximately 5 W/m2 K. Thus,
they were assumed to be equal.
5. Results and discussion
5.1. Effect of convective area on the temperature of the air
Fig. 7 shows the vertical temperature stratication prole of
the air at the measurement plane at 9 m from the inlet, for the
spaces with different widths (Fig. 7a), with partitions (Fig. 7b) and
with desks (Fig. 7c). The spaces with partitions and desks are very
wide. Except for the change in the width, the addition of parti-
tions or desks, the spaces are identical, including equal inlet areas
and velocities, and heat gains. The vertical axis of the plots shows
the nondimensional temperature of the air,  a , as a function of the
dimensionless vertical height from the oor, h* . Each plot has sev-
eral curves identied by different colors. In Fig. 7a the different
curves show the air temperature prole in spaces with different
dimensionless widths, W* W/L. The plot in Fig. 7b shows the
results for different partition areas, Ap , in a wide ofce. The par-
tition areas are normalized with respect to the rooms side area, i.e.
Ap (lp hp )/(L H). Finally, the different curves in Fig. 7c repre-
sent different surface areas (including the top and bottom surfaces)
and thermal conductivities of the desks in wide spaces. The dimen-
sionless desk area, Ad , is normalized with respect to the room oor
area, W L.
Although as expected the air temperature increases with
increasing convective area (higher air temperature in thinner
spaces and with more partition and desk area), the temperature
prole of all the spaces is similar in magnitude and shape. Similar
results were found for the other measurement planes. The devia-
tion in air temperature between the rooms with different widths is
generally below 15%, corresponding to less than 1 C for the param-
eters used to create Fig. 7, for the region between 0.1 h* 0.9 in
161
all the measurement planes. However, the deviation near the oor
is as high as 40%, while near the ceiling is between 26% and 30%.
The deviation in the air temperature between the cases with
different partition areas is substantial (up to 40%) near the oor
and the ceiling. However, the deviation due to the partition size in
the temperature of the air between the dimensionless heights of
0.1 h* 0.9 is generally below 14%, corresponding to 0.7 C. The
air temperature prole in the space with desks show an impor-
tant difference near the tables (h* 0.2). The temperature of the
air surrounding the tables is signicantly higher than at other loca-
tions near them, however the prole seems to be insensitive to
the thermal conductivity of the desks. The high temperature of the
desks is due to the radiation heat transfer they receive from the
active sources, particularly from the ceiling. This heat is then trans-
ferred to the surrounding air by convection, thereby resulting in the
increased air temperatures seen. Nevertheless, the air temperature
prole is similar in shape and the deviation of the dimensionless
temperature of the air is generally below 15% far from the tables
and far from the ceiling and the oor.
This means that it is possible to use thermal stratication models
for rooms with no additional convective area to estimate the air
temperature prole (e.g. [3,6]) in rooms of different widths, with
partitions and desks, as long as the results are used for the air region
away from the ceiling and the oor (0.1 h* 0.9) as well as the
desks. The error incurred by using these stratication models would
be approximately 15%.
5.2. Surface temperatures
Determining the shape of the temperature of the surfaces as a
function of the dimensionless length, l* , or dimensionless height,
h* , is of dubious usefulness due to the dependence of these tem-
peratures on the exact location of the occupants, the stagnation
point of the cool air jet from the inlet, and/or the furniture and
partitions [3]. For this reason, the following discussion is limited
to the average temperatures of the surfaces in the room. Average
temperatures can still provide valuable information for the opera-
tive temperature to the building designer to assess thermal comfort
conditions.
Fig. 8 shows the average dimensionless temperature of the
ceiling, the oor and the walls as a function of the dimension-
less convective area in the room, Aconv . All these cases share the
same Reynolds number, ReAw = 35, 379, dimensionless inlet area,
Aw = 1.25 102 , and inlet height, hw = 1/3. Different curves show
the results for different values of the Archimedes number, as indi-
cated in the gure.
162 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
Fig. 8. Average dimensionless temperature of the (a) ceiling, (b) oor, and (c) walls as a function of the convective area, Aconv . Different Archimedes numbers, Arhw , results
in different curves. All simulations had an inlet area of Aw = 1.25 102 .
Fig. 8 shows that the temperatures of all the surfaces decrease
with increasing area available to convect heat to the air. For
example, for an Archimedes number of Arhw = 0.6 the temper-
ature of the ceiling decreases from being twice the well-mixed
temperature for the innitely wide room, to being 1.5 times the
well-mixed temperature for the 2 m wide space, a difference of
2.5 C and 5 C for typical room air temperature rises of 5 C and
10 C, respectively. The temperatures of the walls and the oor
also decrease with increasing convective area. For Arhw = 0.6, the
reduction corresponds to 2 C for a room temperature rise of 5 C.
When the dimensionless convective area in the room is Aconv = 4.3,
the temperatures of the walls and the ceiling are just 1 C above
the well-mixed temperature each, for a room temperature rise
of 5 C. This gure also shows that both horizontal and verti-
cal surfaces (partitions and tables) approximately lie on a single
temperatureconvective area curve. The only substantial exception
is seen in the ceiling temperature for the table that covers 40% of
the oor area. The temperature of the ceiling for this case is above
the trend of the rest of the cases, signaling that this horizontal sur-
face is less efcient at convecting heat from the sources to the air.
Vertical surfaces are more efcient at convecting heat to the air
because they do not block the direct radiation from the ceiling to
the oor. Moreover, the convective heat transfer coefcient associ-
ated with the surface of the desk facing downwards is signicantly
lower than that of the vertical surfaces. This lower convective coef-
cient is due to the lower air velocities under a warm surface facing
downwards, as the buoyant forces keep a layer of near stagnant
warm air in contact with the surface.
Fig. 8 also shows that the dimensionless temperatures of all the
surfaces decrease with increasing Archimedes number. It was seen
in the results of the CFD simulations that with a low Archimedes
number, the supply air jet throw is longer, so the jet touches the
oor farther from the inlet and continues moving along the oor
towards the outlet wall [3,27]. The high supply momentum causes
the air that reaches the outlet wall to rise before returning towards
the inlet at a higher height [3,27]. This air is entrained by the sup-
ply jet and, because its temperature is higher than the supply jet,
it increases the temperature of the jet [3,27]. For example, the
centerline dimensionless temperature of the jet when it reached
the oor is jet,h =0 1.3 for an Archimedes number of Arhw = 0.6.
With a higher Archimedes number, the cold air does not mix sig-
nicantly with the warm air in the upper part of the room, as
buoyant forces keep it at oor level. The centerline temperature of
the jet at the stagnation point, for Arhw = 100, was found to be only
jet,h =0 0.6. In addition, air velocities were higher in the lower
part of the room with higher Archimedes number, because the cold
jet experiences a larger acceleration due to buoyant forces [3,27].
The centerline velocity of the supply jet at 5 mm from the stag-
nation point over the oor was 2.3 times higher in the case with
Arhw = 100 than with Arhw = 0.6, when the velocities were nor-
malized with respect to their value at the inlet. The same higher
air velocities in the lower part of the room with high Archimedes
number were found when comparing two simulations with the
same inlet velocity. Similar ndings have been reported in exper-
imental work that measured the air velocity distribution from air
diffusers and its dependence on Archimedes number [28]. Thus,
lower jet temperatures and higher air velocities cause a decrease in
the temperature of the oor when the Archimedes number is high.
The ceiling and walls are also cold because they are connected via
radiation to the cold oor.
Fig. 8 demonstrates that as the Archimedes number increases,
the effect of the convective area decreases and the dimensionless
temperatures of all the surfaces tend to get closer to a value of 1,
the well-mixed temperature.
The effect of the inlet area can be seen by comparing Figs. 8 and 9.
The former shows the results for a dimensionless inlet area of
Aw = 1.25 102 , while the latter is for Aw = 5.00 102 . Sim-
ulations for a dimensionless partition area of Ap = 0.4 were not
performed for the dimensionless area of Aw = 5.00 102 because,
as seen in these gures, it is not needed to establish a trend. At a
xed Archimedes number, the dimensionless temperatures of all
the surfaces increase with increasing inlet area. The dimensionless
temperature of the ceiling increased the most when increasing the
inlet area, with an increase between 13% and 70%. The oor had the
lowest temperature change, between 0.5% and 32%, with respect to
a change in the inlet area.
The trend seen in Fig. 8a and 9a for the dimensionless tem-
perature of the ceiling, for example, can be approximated as the
 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
Fig. 9. Average dimensionless temperature of the (a) ceiling, (b) oor, and (c) walls as a function of the convective area, Aconv . Different Archimedes numbers results in
different curves. All simulations had an inlet area of Aw = 5.00 102 .
Table 1
Values of the tting constants in Eq. (9).
A B C D E F
0.33 0.61 0.39 1.63 0.52
Table 2
Values of the tting constants in Eq. (10).
Parameter Expression
G 9.73Aw + 1.04
I 13.51Aw + 0.61
multiplication of two exponential decays, one with respect to the
Archimedes number, Arhw , and one for the convective area, Aconv :
       Archimedes number, convective area and inlet area, as a function
 c = A exp BArhw +C D exp EAconv +F
where A, B, C, D, E, F are tting parameters that, in general, depend
on the inlet area, Aw . This equation captures the fact that the dimen-
sionless temperature decreases both with increasing Archimedes
number, Arhw , and increasing convective area, Aconv , and that these
changes decay approximately exponentially. Table 1 shows the
values for the tting constants for the case with an inlet area of
Aw = 1.25 102 . This t was obtained using the Curve Fitting
Toolbox for MATLAB and has a correlation factor of r2 = 0.97.
To account for different inlet areas, a trend that relates the result
in Eq. (9) with the dimensionless ceiling temperature of rooms with
different inlet areas is given next:
 c (Aw ) = G c (Aw = 1.25 102 ) + I
where G and I are tting parameters obtained using the expressions
in Table 2, and  c (Aw = 1.25 102 ) is given by Eq. (9) and Table 1.
A similar expression as that in Eq. (9) can be constructed to esti-
mate the dimensionless temperature of the oor and the walls.
However, by taking advantage of the fact that the dimensionless
temperatures of the oor and the walls are approximately equal,
and that they were found to be linearly related to the ceiling tem-
perature, regardless of surface type and/or Archimedes number,
(Fig. 10a), a simpler expression that depends on a lower number of
tting constants can be used instead:
 f
=  w = J  c + K
where J and, K are tting parameters that, in general, depend on
the inlet area, Aw . This line is compared to the data in Fig. 10a. The
163
Table 3
Values of the tting constants in Eq. (11).
Parameter Expression r2
2.25 J 2.43Aw + 0.83 0.97
K 24.14Aw 2 1.01Aw + 0.07 0.92
dashed lines in this gure encompass a 10% deviation from the t.
r2 The largest deviation from this line is seen in the simulation with
the lowest Archimedes number. Table 3 shows the functional forms
0.99
0.94 to obtain J and K as a function of the inlet area.
A simplied correlation for the temperature of the oor and
the walls that also ignores the effect of the inlet area is shown
schematically in Fig. 10b. In this gure, the data points show the
dimensionless temperature of the oor and walls, regardless of
(9) of the ceiling dimensionless temperature. The dashed lines in this
gure encompass an area with a deviation of 10% from the tting
line. Because the effect of the inlet area has been ignored, more data
points lie outside of the 10% region, and the correlation factor of
the t, at r2 = 0.95, is lower than when using Eq. (11). Neverthe-
less, most of the data points are still reasonably represented by the
simplied t. This simplied correlation is given by:
 f
=  w = 0.67 c + 0.21 (12)
5.3. Operative temperature
Fig. 11a shows a plot of dimensionless mean radiant tempera-
(10) ture against dimensionless available convective area, while Fig. 11b
shows a plot of dimensionless air temperature. The mean air tem-
perature was estimated using the average air temperature from the
oor to a height of 1.35 m, 25 cm above the head of the occupants,
at the 5 m and 9 m planes. These two quantities were then used to
compute the operative temperature that the occupants experience.
The results are shown in dimensionless form in Fig. 11c. These g-
ures are all for an inlet area of Aw = 1.25 102 , however similar
trends were seen for other values of the inlet area. Fig. 11a rein-
forces the results of the previous sections by demonstrating that the
mean radiant temperature of the occupants decreases with increas-
ing convective area. Fig. 11b shows that the air temperature does
not vary signicantly with respect to the convective area, how-
(11)
ever it is dependent on the Archimedes number. It is important to
mention that the air temperature around the occupants is lower
than the well-mixed temperature of the space at  wm = 1. For the
164 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165

Fig. 10. Average dimensionless temperature of the oor,  f , or walls,  w , as a function of the average dimensionless temperature of the ceiling,  c . The dashed lines encompass
a 10% range from the line of Eq. (11). (a) Plot for spaces with different surfaces (marker shape and border color) and different Archimedes numbers (marker ll color) and
with dimensionless inlet area of Aw = 1.25 103 . (b) Plot that includes the results of all the inlet areas considered in this work.

Fig. 11. Average dimensionless (a) mean radiant, (b) mean air, and (c) operative temperature against dimensionless convective area, Aconv . The inlet area of the simulations
in this plot was Aw = 1.25 102 .

range of Archimedes numbers used in this study, the mean tem-
perature decreased with increasing Archimedes number, being as
low as 64% of the well-mixed temperature. Thus, approximating
the air temperature as the well-mixed temperature might result
in an inaccurate estimation of the operative temperature. Previ-
ous research has called attention to this problem of the well-mixed
assumption (e.g. [3,6,29]). Finally, Fig. 11c demonstrates that the
decrease in mean radiant temperature is important, as it causes a
reduction in the operative temperature, even if the air temperature
does not follow a simple trend, as hypothesized in this work.
6. Conclusions
The addition of partitions and horizontal surfaces substantially
reduces the average radiant temperature while only moderately
changing the air temperature near the occupants. This results in
a decrease in the operative temperature between 10% and 21%
of the overall temperature difference between the inlet and the
outlet. This corresponds to a temperature difference of up to 2 C
between the spaces with few and several additional surfaces in
typical naturally ventilated spaces. Due to the small effect of the
convective surfaces on the shape of the air temperature prole, it is
still possible to use thermal stratication models for rooms without
additional surfaces to estimate this prole away from the surfaces
in the room with an error below approximately 15%.
The free convective area, a parameter that effectively captures
the effect of the radiation view factors and convective heat trans-
fer coefcients of the surfaces typically found in an ofce, such as
partitions and desks, was proposed. Correlations to determine the
temperatures of the surfaces surrounding an occupant in a room
(ceiling, oor, walls) as a function of the Archimedes number, the
inlet area and the free convective area were developed. It was found
that regardless of the orientation or position of these surfaces, the
average temperatures of the surfaces in the room depend on this
parameter.
The temperatures of the room surfaces were found to decrease
with increasing Archimedes number. A high Archimedes number
was related to higher air velocities and lower supply jet temper-
ature at the stagnation point. The high velocity is a consequence
of the larger acceleration experienced by the supply air due to the
higher buoyant forces. Cold oors and walls are related to cold ceil-
ings and vice versa, as these surfaces are all connected via radiation
exchange. Indeed, a simple linear relation was found between the
oor and walls temperatures, and the ceiling temperature. In addi-
tion, it was found that for a given Archimedes number, a larger
supply area meant that the temperature of all the surfaces were
higher than with smaller inlet areas. However, it is also possible to
change the inlet height while keeping the same Archimedes num-
ber. This option was not explored in this work. Future work can
help expand this study by analyzing the effect of the inlet height
on the temperatures of the surfaces in a space.
The correlations proposed in this work, together with air strati-
cation proles, can help designers to assess comfort conditions in
buildings more accurately than is currently possible using models
 F.A. Dominguez Espinosa, L.R. Glicksman / Energy and Buildings 147 (2017) 155165
that rely exclusively on the well-mixed temperature. The correla-
tions presented here are simple enough to be used in multi-zone
models, given that its inputs are all available in multi-zone models.
Acknowledgments
This research was supported by the Mexican Council for Science
and Technology (CONACYT), the U.S. China Clean Energy Research
Center (CERC) and the Martin Family Society of Fellows for Sustain-
ability
References
[1] Thermal Environmental Conditions for Human Occupancy, American Society
of Heating, Refrigerating and Air-Conditioning Engineers, 2010.
[2] T. Catalina, J. Virgone, F. Kuznik, Evaluation of thermal comfort using
combined cfd and experimentation study in a test room equipped with a
cooling ceiling, Build. Environ. 44 (2009) 17401750.
[3] F.A. Dominguez Espinosa, Determining thermal stratication in rooms under
mixing and displacement ventilation, Massachusetts Institute of Technology,
2016 (Ph.D. thesis).
[4] E. Mundt, Displacement Ventilation Systems convection ows and
temperature gradients, Build. Environ. 30 (1995) 129133.
[5] M.-A. Menchaca Brandan, L.R. Glicksman, How to replicate computationally a
human plume with a simple geometry, and how not to, in: Proceedings of
Roomvent 2011 The 12th International conference on air distribution in
rooms, SCANVAC (Scandinavian Federation of Heating, Ventilating and
Sanitary Engineering Association), 2011.
[6] M.A. Menchaca Brandan, Study of Airow and Thermal Stratication in
Naturally Ventilated Rooms, Massachusetts Institute of Technology, 2012
(Ph.D. thesis).
[7] Q. Chen, Comparison of different k- models for indoor air ow computations,
Numer. Heat Transf. B Fundam. 28 (1995) 353369.
[8] ANSYS, FLUENT Theory Guide, ANSYS, 2011.
[9] H.B. Awbi, Calculation of convective heat transfer coefcients of room
surfaces for natural convection, Energy Build. 28 (1998) 219227.
[10] J.H. Lienhard V, J.H. Lienhard IV, A Heat Transfer Textbook, Dover Publications
Inc., 2011.
[11] A.F. Mills, Basic Heat and Mass Transfer, 2 ed., Prentice Hall, 1999.
[12] 2009 ASHRAE Handbook Fundamentals, American Society of Heating,
Refrigerating and Air-Conditioning Engineers, 2009.
165
[13] I.B. Celik, U. Ghia, P.J. Roache, C.J. Freitas, H. Coleman, P. Raad, Procedure for
estimation and reporting of uncertainty due to discretization in CFD
applications, J. Fluids Eng. Trans. ASME 130 (2008).
[14] M.W. Nansteel, R. Greif, Natural convection in undivided and partially divided
rectangular enclosures, J. Heat Transf. 103 (1981) 623629.
[15] D.A. Olson, L.R. Glicksman, H.M. Ferm, Steady-state natural convection in
empty and partitioned enclosures at high Rayleigh numbers, J. Heat Transf.
112 (1990) 640647.
[16] S.D. Ray, Modeling Buoyancy-Driven Airow in Ventilation Shafts,
Massachusetts Institute of Technology, 2012 (Ph.D. thesis).
[17] M.-A. Menchaca-Brandan, L.R. Glicksman, The importance of accounting for
radiative heat transfer in room airow simulations, in: Proceedings of
Roomvent 2011 The 12th International Conference on Air Distribution in
Rooms, SCANVAC (Scandinavian Federation of Heating, Ventilating and
Sanitary Engineering Association), 2011.
[18] Q. Chen, L. Glicksman, System Performance Evaluation and Design Guidelines
for Displacement Ventilation, 2003.
[19] P.V. Nielsen, Displacement ventilation in a room with low-level diffusers
Indoor Environmental Technology, vol. R8836, Department of Building
Technology and Structural Engineering, Aalborg University, 1988.
[20] T.-G. Malmstrm, Archimedes number and jet similarity, in: Proceedings of
the 5th International Conference on Air Distribution in Rooms ROOMVENT,
1996, pp. 415422.
[21] H.B. Awbi, M.M. Nemri, Scale effect in room airow studies, Energy Build. 14
(1990) 207210.
[22] D. Etheridge, M. Sandberg, Building Ventilation: Theory and Measurement,
John Wiley & Sons, 1996.
[23] C. Walker, G. Tan, L. Glicksman, Reduced-scale building model and numerical
investigations to buoyancy-driven natural ventilation, Energy Build. 43
(2011) 24042413.
[24] J.H. Lee, V.H. Chu, Turbulent jets and plumes, in: A Lagrangian Approach,
Springer Science & Business Media, 2012.
[25] A. Roetzel, A. Tsangrassoulis, U. Dietrich, S. Busching, A review of occupant
control on natural ventilation, Renew. Sustain. Energy Rev. 14 (2010)
10011013.
[26] A.A. Sonin, The Physical Basis of Dimensional Analysis, Technical Report,
Massachusetts Institute of Technology, 2001.
[27] F. Dominguez Espinosa, L.R. Glicksman, Determining thermal stratication in
rooms with high supply momentum, Build. Environ. 112 (2017) 99114.
[28] P.V. Nielsen, Velocity distribution in a room ventilated by displacement
ventilation and wall-mounted air terminal devices, Energy Build. 31 (2000)
179187.
[29] P.F. Linden, G.F. Lane-Serff, D.A. Smeed, Emptying lling boxes: the uid
mechanics of natural ventilation, J. Fluid Mech. 212 (1990) 309335.