THERMAL SCIENCE: Vol. 13 (2009), No. 1, pp.
69-78 69
COMPUTATIONAL STUDY OF SMOKE FLOW CONTROL IN GARAGE
FIRES AND OPTIMISATION OF THE VENTILATION SYSTEM
by
Miloš J. BANJAC and Barbara M. NIKOLI]
Original scientific paper
UDC: 692.842:532.57
BIBLID: 0354-9836, 13 (2009), 1, 69-78
DOI: 10.2298/TSCI0901069B
With the aim of evaluating capabilities of a ventilation system to control the spread
of smoke in the emergency operating mode, thereby providing conditions for safe
evacuation of people from a fire-struck area, computational fluid dynamics simula-
tion of a fire in a semi-bedded garage was conducted. Using the experimental re-
sults of combustion dynamics of a passenger car on fire, optimal positions of venti-
lation openings were determined. According to recommendations by DIN EN
12101 standard, the operating modes of a ventilation system were verified and opti-
mal start time of the smoke extraction system was defined.
Key words: fire, smoke control, ventilation system, computational fluid dynamics,
smoke extraction
Introduction
A proper design and construction of a smoke extraction system, either during con-
struction of a new building or reconstruction of old ones, represents an important measure that
has to be taken to reduce hazard that people may be exposed to in case of fire. An appropriate
arrangement of smoke extraction openings, as well as proper selection of an operating mode
of a ventilation system, enable prevention of further fire spreading by decreasing fresh air sup-
ply, i. e. provide conditions for safe evacuation of people from fire-struck area by efficient
smoke extraction. Therefore, for proper design of these systems, apart from knowing combus-
tion dynamics of particular materials, the amount of heat and smoke produced during their
combustion, it is necessary to have a reliable methodology for predicting what would, in dif-
ferent operating modes of a smoke extraction system, happen with fire, i. e. how the air tem-
perature and velocity fields will change in a fire-struck area and in which direction smoke will
spread, i. e. remain.
“Conventional” experimental methods, as methodologies for solving the above-men-
tioned questions, are very rarely applied due to their high price and long period for calculation,
especially in the phase of a building design. Other semi-empirical methods for predicting the
aforementioned fields that rely on the use of experimental data on smoke spreading in seemingly
immovable air or in the presence of air jets, i. e. on the behaviour of air near openings, are very
unreliable. Among others, significant disadvantage of this approach is scarcity or complete lack
of information about physical properties of substances in parts of area outside of the zones near
extraction openings. Also, the use of such expressions and diagrams in real conditions usually
leads to considerable differences between the estimated and achieved fields of the aforemen-
70 Banjac, M. J., Nikoli}, B. M.: Computational Study of Smoke Flow Control in ...
tioned properties. A similar, if not worse situation is with nowadays conventional methodolo-
gies based on empirical data on the necessary number of air replacements and other similar very
rough calculations. Therefore, solving the described problem before the appearance of compu-
tational fluid dynamics (CFD) and its use in the design of ventilation and smoke extraction sys-
tems, as well as the combustion process and fire spreading, was a very difficult task.
Unlike the conventional approach, the numerical CFD-approach enabled, regardless
of the complexity of geometrical area and boundary spatial and time conditions, relatively easy
and, at the same time, very precise prediction of even very complex fields of velocity, tempera-
ture, and concentrations formed in the air in case of fire. This approach, based on the space
discretization and mesh generation, thus forming a very large number of finitely small control
volumes (CV) and setting and simultaneous solving of balance equations defined for each CV,
made it possible to gather information on fluid velocity, pressure, temperature, density, turbu-
lence level, concentration of particular substances, etc. for a significantly large number of points
in volume. This particular advantage of the CFD-approach has made it an almost ideal method
for designing ventilation, i. e. smoke extraction systems [1, 2].
Problem description
As part of a Mechanical Design (MD) of a heating and ventilation system of a particu-
lar block of flats, it was necessary to design the ventilating system, i. e. smoke extraction system
of a semi-bedded garage beneath the same building. The rectangular garage, with inner dimen-
sions of 15.8 × 15.8 × 3.2 m, had all four sides embedded into the ground up to 50% of its height.
In the garage, there were 4 cylindrical concrete supporting pillars with the outer diameter of 0.7
m. According to the Civil Engineering Design, there is a movable door on the approach side,
planned for entrance and exit of vehicles. The door dimensions are 3.2 × 2.5 m. The garage is
supplied with three windows, with dimensions 1.0 × 0.6 m on one side and one opposite wall, at
the height of 2.2 m measured from the garage floor. The door for entrance and exit of people, i. e.
for their evacuation, stands on the opposite wall of the car door. The garage capacity is 12 pas-
senger vehicles in two rows.
In normal operating conditions, garage space is ventilated by natural ventilation or by
forced ventilation, in case of exceeding particular exhaust gas concentration in the air. In case of
forced ventilation, efficiency of two different constructive solutions were analysed. The first
technical solution included the ventilation channel routed to the centre of the garage, where the
suction opening was placed at the bottom channel side and it was used for extracting the air out
of the garage (optionally on lateral sides). The second technical solution of the design avoided
routing channels inside the garage, and one of the existing window openings had been trans-
formed into a suction opening of the ventilation system.
According to the MD, the use of a ventilation system was envisiged for the smoke ex-
traction in the case of fire start in the garage space i. e. fire on one of the parked cars. Further-
more, in case it works as a ventilation system, in compliance with the existing legal regulations
in this area [3], a fan should operate in a way that provides an air change rate of 6 changes per
hour; in case it works as a smoke extraction system, the air change rate should be 12 changes per
hour.
According to the terms of reference of the MD, in the part that elaborates on the prob-
lem of a smoke extraction system, the following items should be analyzed:
– determination of the optimal position of an extraction ventilation opening, i. e. extraction
opening for the smoke extraction system in which, in the designed operating mode of 12 air
changes per hour, smoke produced by the car on fire will not fill the whole garage space, i. e.:
THERMAL SCIENCE: Vol. 13 (2009), No. 1, pp. 69-78 71
– the smoke concentration should be reduced to minimum in the selected reference point -
near the door for entrance/exit of people, and
– temperature in the selected reference point should be as low as possible,
– verification of the capability of operating modes of a smoke extraction system in terms of the
ability to extract smoke out of the garage space for minimal time. In the absence of the
quantitative values of produced smoke quantity, it was considered that the smoke
concentration on boundary surfaces of the car on fire was equal to 100% (C = 1.00), and in
“fresh” air 0% (C = 0.00). According to the above-mentioned, it was defined that a smoke
extraction system should provide:
– smoke concentration near the door for entrance/exit of people lower than 30%, and
– temperature near the door for entrance/exit of people not higher than 40 °C.
Numerical calculation of velocity, temperature and
smoke field in the garage
Numerical model
The first step for the calculation of flow and temperature fields, i. e. smoke concentra-
tion fields formed in case of fire within the garage space, was generating a 3-D garage space
model by commercial CFD software package PHOENICS 3.4. According to the assumed physi-
cal situation, the facility layout was generated – 12 cars with dimensions 3.9 ´ 1.4 ´ 1.4 m, 4
concrete pillars and a ventilation
channel, i. e. position of appropriate
smoke extraction openings, windows
and doors.
Inner, virtual garage space was di-
vided into CVs in such a way that all
boundary surfaces of CVs were
aligned with the contours of appropri-
ate solid bodies and barriers within
the space. Thus generated mesh was
additionally balanced, i. e. made
thicker, by adding control volumes in
zones near windows and openings.
Total number of formed control vol-
umes was 39 ´ 43 ´ 26 = 43.602 (fig.
Figure 1. Mesh of control volumes in virtual garage space
1). (color image see on our web site)
Mathematical model
For the calculation of flow and temperature fields of air formed within the garage, a
two-equation k-e turbulent model was used [4]. This universal turbulent model was chosen due
to its confirmed reliability in predicting the flow fields during flows with the Mach number con-
siderably lower than 1 [5]. Apart from three, i. e. four basic balance equations describing
non-stationary incompressible fluid flow for each previously defined control volume:
– continuity equation
¶r ¶
+ ( rU i ) = 0 (1)
¶t ¶xi
72 Banjac, M. J., Nikoli}, B. M.: Computational Study of Smoke Flow Control in ...
– modelled Reynolds equation
¶ ¶ ¶P ¶ é 2 ù
( r Ui ) + ( rU i U j ) = - + ê( m f + m t ) Si j - kdi j úû + rFi (2)
¶t ¶xj ¶ xi ¶ xj ë 3
– and energy balance equation
¶ ¶ ¶P ¶ éæ m t ö ¶H ù
( rH ) + (rUj H ) =U i + 2m t S i j S i j + êçç a f + ÷ ú (3)
¶t ¶x j ¶xi ¶ x i ëè Pr h ÷ø ¶ x i û
with additional
– balance equation of smoke “concentration” (mass fraction of smoke in the air)
¶ ¶ ¶P ¶ éæ mt ö ¶C ù
( rC ) + ( r U j C ) = Ui + 2m t S i j S i j + êçç DC + ÷ ú (4)
¶t ¶x j ¶xi ¶xi ëè Sh t ÷ø ¶ x i û
this turbulence model was defined with
– transport equation for the turbulence kinetic energy
¶ ( rk ) ¶ ¶ éæ m t ö ¶k ù m t gi ¶ r
+ ( rU i k ) = P k + êçç m f + ÷÷ ú-r - re (5)
¶t ¶xi ¶ x i ëè Pr k ø ¶ x i û Pr h ¶ x i
– and transport equation for the dissipation rate
¶ ( re) ¶ e e2 m g ¶r ¶ éæ m t ö ¶e ù
+ ( r U i e) = C e1 P k - C e2 r + C e3 r t i + êçç m f + ÷ ú (6)
¶t ¶xi k k Pr h ¶ x i ¶ x i ëè Pr e ÷ø ¶ x i û
In the aforementioned equations, according to a standard procedure, Sij was defined as
the main strain-rate tensor:
1 æ ¶U i ¶U j ö
Si j = ç + ÷ (7)
2 çè ¶ x j ¶ x i ÷ø
and Pk, the volumetric production rate of k by shear forces is
æ ¶U ö
P k = m t çç ¶U i + j ÷÷ ¶ Ui (8)
è ¶ xj ¶xi ø ¶ x j
Modelling of the Reynolds stresses tensor was based on the Boussinesq hypothesis:
æ ¶U i ¶U j ö 2
ti j = m t ç + ÷ - kdi j = m t S ij - 2 kdi j (9)
ç ¶x j ¶ x i ÷ø 3 3
è
where the eddy viscosity – mt was defined by the equation
k2
m t = (C DC m ) r (10)
e
Since the value of molecular diffusivity of smoke into the air was negligible compared
to the turbulent (molar) diffusivity, it was neglected during the calculation.
The values of the empirical constants of this model, as well as the values of the Prandtl
(enthalpy), i. e. Schmidt turbulent number, are given in tab. 1.
Table 1. Empirical constants of k-e model
Prk Pre CDCm Ce1 Ce2 Ce3 k Prh Sht
1.0 1.314 0.09 1.44 1.92 1.0 0.41 0.41 0.81
THERMAL SCIENCE: Vol. 13 (2009), No. 1, pp. 69-78 73
Apart from the k-e turbulent model, and as a standard procedure for two-equation tur-
bulent models, the Reynolds enthalpy flux, i. e. the Reynolds flux of smoke concentration was
modelled in accordance with the principles of the simple gradient-diffusion hypothesis:
m t ¶H
- rhui = (11)
Pr h ¶x i
i. e.
m ¶C
- rcui = t (12)
Sh t ¶x i
Since both fluids – air and smoke – can be considered as ideal gases, i. e. their mixture,
regardless of fractions of particular components, can be treated as ideal gas, for determining
flow and temperature fields and smoke concentration fields, the so-called scalar variable mark-
ing method was used.
Regarding (molecular) viscosity, it was assumed that there was square thermodynamic
temperature dependency:
mf = –4.9468·10–6 + 4.5839·10–8T + 8.0974·10–11T 2 [m2s–1]
whereas, for a specific thermal capacity of air at constant pressure, i. e. its thermal conductivity,
it was assumed that they had constant values, cp = 1004 Jkg–1K–1, lf = 2.63·10–2 Wm–1K–1.
Due to relatively high combustion product temperatures generated during car burning,
in order to gain higher precision, heat transfer by radiation was also covered by numerical calcu-
lation. The so-called immersed-solid (Immersol) radiation model [6, 7] was used. Within the
space between solids, the distribution of radiosity, i. e. sT34 , can be represented as the following
equation:
¶ æ ¶T3 ö
çç G rad ÷ = ( a + s)( E12 - sT34 ) (13)
¶x i è ¶x i ÷ø
where: E12 stands for the phase-surface-average of sT 4, a = 0.1 stands for the absorptivity of the
fluid medium, and s = 0.1 stands for the scattering coefficient of that medium. Grad is defined as
the reciprocal of Grad = 0.75[(a + s + 1/wgap)], and wgap stands for the distance between adjacent
walls.
Boundary and initial conditions
According to the real physical situation, i. e. designed layout of the garage and smoke
extraction system, it was necessary to specify boundary spatial and initial conditions for veloc-
ity, temperature (energy), and smoke concentration fields.
– Boundary conditions for contact between air and solid surfaces
The “wall” function model was used for specifying boundary conditions near the solid
surfaces within the garage, related to the velocity field. Since the used turbulent model belongs
to a class of the high Reynolds turbulent model, for determining values of variables, i. e. their
flows next to the plate, wall functions of the logarithmic area of the turbulent boundary layer
were used [5, 8].
At the same time, the following was assumed as boundary conditions for the tempera-
ture field:
– garage walls were adiabatic surfaces with thermal emissivity of 0.9,
– cars which were not on fire were made of steel sheet metal with thermal emissivity of 0.9,
and
74 Banjac, M. J., Nikoli}, B. M.: Computational Study of Smoke Flow Control in ...
– contour of the automobile on fire was
treated as a thermal source contour, i. e.
smoke source.
Thermal emission dynamics of this ther-
mal source had a shape of a “saw” function.
This shape of the function was generated by
the approximation of experimental data [9] in
case of one automobile burning in an under-
ground garage (fig. 2).
Smoke production dynamics was defined
by assuming proportionality between smoke
quantity produced during combustion and
Figure 2. Heat release rate of one car amount of heat released during that process
[10]. During this, in the absence of quantita-
tive values of produced smoke quantity, it was considered that the smoke concentration on
boundary surfaces on the automobile on fire equalled 100% (C = 1.00), and in “fresh” air 0%
(C = 0.00).
– Boundary conditions on window panes and doors
Since the variable values in the domain outside sections (windows) are completely un-
known, except in the case of garage smoke extraction openings, the so-called condition of "con-
stant" pressure [11] as a boundary condition was used. This condition consists of specifying zero
derivatives in the direction normal to the outgoing plane, i. e. specifying second boundary con-
ditions for all values of dependent variables, except for velocities normal to the outgoing plane,
i. e. ¶F/¶x2ïout = 0 and ¶F/¶x3ïout = 0 (F = k, e, H, and C), and specifying a pressure value equal
to the ambient pressure Pïout = Pamb. Values of missing velocities were defined indirectly, by us-
ing pressure values.
The door for entrance of vehicles was shut, as well as the door for the evacuation of
people. Thermal emissivity of these surfaces was specified to be 0.9.
The assumed outside air temperature was 17 °C.
– Boundary conditions at smoke extraction openings
Velocity field at the smoke extraction opening was specified by using appropriate val-
ues of the volumetric air flow.
– Initial conditions
The assumed temperature of the air and all objects within the garage space equalled at
the initial moment. Smoke concentration in the air of the whole garage space equalled zero, and
the air was assumed to be absolutely still.
The analysis was done on only one operating mode with 12 air replacements per hour
and it was specified that the smoke extraction systems would be activated 1 minute and 2 min-
utes after the outbreak of fire.
Optimization of the position an extraction
opening in the garage space
Optimal position of an extraction opening of a smoke extraction system, i. e. ventila-
tion system was selected by performing numerical simulations. All simulations were performed
THERMAL SCIENCE: Vol. 13 (2009), No. 1, pp. 69-78 75
for unsteady – transient conditions. According to the experimental data on the duration of fire,
time domain for each simulation was 40 minutes. Good convergence of the solution was
achieved for the time step of 5 seconds, with 200 iterations within each time step.
Numerical simulations were performed both for the case of routing the ventilation
channel to the centre of the garage and for the case of smoke extraction through one of the win-
dow openings. Several simulations were made for the case of fire spreading from one car to an-
other, but since such an event is unlikely to occur, in all other simulations, we considered the
case where only one car was on fire. Also, the position of the car on fire in several initial simula-
tions was changed. Afterwards, it was decided to use the second car from the entrance and exit
door as the representative position of the vehicle (the worst scenario).
By viewing all achieved results, it was determined that the temperature near the area of
the car on fire was between 200 and 370 °C, which corresponded to experimentally determined
values [12]. According to the criteria set by the terms of reference (ToR) of the project, it was
Figure 3. Garage air velocity field in plane z = 3 m, i. e. on the surface with smoke concentration of 30%
and air temperature of 40 °C, 5 minutes after the fire start, in case of two locations, (a) and (b), for smoke
extraction (color image see on our web site)
76 Banjac, M. J., Nikoli}, B. M.: Computational Study of Smoke Flow Control in ...
noted that the location of a smoke extraction opening should be as close as possible to the place
of fire, i. e. on the wall “behind” the car on fire. Since it was not possible to fulfil this condition –
only one smoke extraction location was requested by the project ToR, it was chosen, as an opti-
mal solution, that the smoke extraction location should be placed on the ceiling of the garage
centre, i. e. that the ventilation channel should be placed into the garage space. By this technical
solution, the smoke extraction opening must not be placed on the bottom, but exclusively on the
lateral channel sides. This is the only way in which it will be possible to extract smoke accumu-
lated in the ceiling area.
Since smoke spreads quickly within the garage space (fig. 3) in terms of smoke extrac-
tion, it was concluded that it would be more suitable if the smoke extraction system was acti-
vated at least within one minute following the fire outbreak.
Reviewing the achieved results, visually at first, and then by the verification of numer-
ical values, it was concluded that, under the foreseen values of the volumetric flow of 12 air
changes per hour, it was only possible to achieve specified parameters if the smoke extraction
channel was placed in the central zone of the garage space (fig. 4). In that manner, in the area
around the entrance and exit door for people, smoke concentration was not over 30% and the air
temperature was not over 40 °C.
Figure 4. Smoke spreading – surfaces with smoke concentration of 30%
After (a) 30 seconds; (b) 1 minute; (c) 2 minutes; and (d) 15 minutes following the outbreak of fire
(color image see on our web site)
THERMAL SCIENCE: Vol. 13 (2009), No. 1, pp. 69-78 77
Conclusions
Analysing and reviewing the results of the numerical simulation of a fire spread in a
semi-bedded garage, with dimensions 15.8 ´ 15.8 ´ 3.2 m, caused by the fire outbreak on one
car, the following could be concluded:
– smoke extraction system should be as close as possible to the place of fire, i. e. on the wall
“behind” the automobile on fire or, if there is only one location for smoke extraction, on the
ceiling in the central zone of the garage; in that case, smoke extraction openings must not be
placed on the bottom, but exclusively on the lateral channel sides, because only in this way
will it be possible to extract smoke accumulated in the ceiling area.
– the operating mode of a smoke extraction system with the rate of 12 air changes per hour
specified by DIN EN 12101 (Smoke and heat control systems) standard will make sure that,
in the space around the entrance and exit door for people, smoke concentration will not
exceed 30%, i. e. that the air temperature will not exceed 40 °C, and only will be provided if
smoke extraction system is located in the central part of the garage space and smoke
extraction systems are activated within 1 minute following the fire outbreak.
Nomenclature
a – absorption coefficient of the air, [m–1] Ui – mean velocity vector in tensor notation,
af – thermal diffusivity, [m2s–1] – [ms–1]
C – mass fraction of smoke in the air, [–] ui – fluctuating velocity vector in tensor
DC – mass diffusivity, [m2s–1] – notation, [ms–1]
Fi – body force per unit mass, [Nkg–1] wgap – distance between adjacent walls, [m]
gi – gravity force per unit mass, [Nkg–1] xi – position vector in tensor notation, [m]
H – mean enthalpy per unit mass, [Jkg–1]
h – fluctuating enthalpy per unit mass, [Jkg–1] Greek letters
k – kinetic energy of turbulent fluctuation per dij – Kronecker delta, [–]
– unit mass, [m2s–2] e – dissipation of turbulent kinetic energy,
P – mean static pressure, [Pa] – [m2s–3]
Prh – Prandtl number, [–]
mf – molecular viscosity, [Pa·s]
Prk – Prandtl number for k, [–]
Sij – mean strain-rate tensor, [s–1] mt – eddy viscosity, [Pa·s]
Sht – Schmidt turbulent number, [–] r – mass density, [kgm–3]
s – scattering coefficient of air, [m–1] s – Stefan-Boltzmann constant, (= 5.67·10–8),
T – temperature, [K] – [Wm–2K–1]
t – time, [s] tij – Reynolds stresses tensor, [Nm–2]
References
[1] Vidmar, P., Petelin, S., Methodology of Using CFD-Based Risk Assessment in Road Tunnels, Thermal
Science, 11 (2007), 2, pp. 223-250
[2] Stevanovi}, Ž., Markovi}, Z., Turanjanin, V., Numerical Simulation of Fire Spread in Terminal 2 of Bel-
grade Airport, Thermal Science, 11 (2007), 2, pp. 251-258
[3] Recknagel, H., et al., Heating and Air Conditioning: Including Hot Water and Cooling Technique (in Ser-
bian), Interklima, Vrnja~ka Banja, Serbia, 2004
[4] Launder, B. E., Spalding, D. B., The Numerical Computation of Turbulent Flows, Computer Methods in
Applied Mechanics and Engineering, 3 (1974), 2, pp. 269-275
[5] Wilcox, D. C., Turbulence Modelling for CFD, DCW Industries, La Canada, Cal., USA, 1993
[6] Spalding, D. B., Proposal for a Diffusional Radiation Model, Unpublished technical memorandum,
CHAM, London, 1994
[7] Ozisik, M. N., Radiative Heat Transfer, John Wiley and Sons, New York, USA, 1973
78 Banjac, M. J., Nikoli}, B. M.: Computational Study of Smoke Flow Control in ...
[8] Stevanovi}, Ž., Numerical Aspects of Turbulent Momentum and Heat Transfer (in Serbian), Faculty of
Mechanical Engineering, University of Niš, Ni{, Serbia, 2008
[9] Zhang, X.G., et al., Numerical Simulations on Fire Spread and Smoke Movement in an Underground Car
Park, Building and Environment, 42 (2007), 10, pp. 3466-3475.
[10] Vidakovi}, M., Fire and Architectural Engineering (in Serbian), Handbook, Fahrenheit, Belgrade, 1995
[11] Versteeg, H. K., Malalasekera, W., An Introduction to Computational Fluid Dynamics: The Finite Volume
Method, Prentice Hall, Upper Sadle River, N. J., USA, 1995
[12] Vidakovi}, M., Thermodynamics of Uncontrolled Combustion, Part I (in Serbian), Prometej, Novi Sad,
Serbia, 1994
Author's affiliation:
M. J. Banjac (corresponding author)
Faculty of Mechanical Engineering
University of Belgrade
16, Kraljice Marije Str., 11120 Belgrade, Serbia
E-mail: [email protected]
B. M. Nikoli}
Tehnikum Taurunum – Technical School
Department for Fire Safety
Belgrade, Serbia
Paper submitted: December 1, 2008
Paper revised: December 24, 2008
Paper accepted: January 1, 2009
