International Journal on Architectural Science, Volume 7, Number 2, p.

35-46, 2006

APPLYING LARGE EDDY SIMULATION IN THE STUDY OF FIRE AND

SMOKE SPREAD AT UNDERGROUND CAR PARK

T.T. Chow, Zhang Lin and C.F. Tsang

Building Energy & Environmental Technology Research Unit
Division of Building Science & Technology, City University of Hong Kong
Tat Chee Avenue, Hong Kong, China

(Received 30 March 2006; Accepted 27 December 2006)

ABSTRACT

In this study, the large eddy simulation technique has been applied to predict the spread of smoke and fire in a
public underground car park. The simulation platform has been the fire dynamics simulator, which was
developed by the US National Institute of Standards and Technology (NIST). The results show that in such a
confined space the smoke from a severe fire can spread quickly and dangerously. With the use of the
conventional balanced ventilation system as the means of smoke extraction, a higher ventilation rate tends to
accelerate the spread of fire, smoke, carbon dioxide and carbon monoxide. At the same time, this leads to a
higher temperature environment which can endanger the human life. The probability of flashover was also
found to increase with higher ventilation rates. The findings suggest that the current fire safety design
recommendations are insufficient. Several suggestions on improving the public car park safety are raised, such
as to avoid the use of excessive ventilation rate, to increase the number of emergency exits, and to shorten the
vehicle exit routes.

1. INTRODUCTION extraction systems are to be provided for those

greater than 7,000 m3 in volume. The mechanical
The underground car park has been the subject of means of smoke extraction gives a better guarantee
study in recent years due to the increasing of the flow rate since it affects little by the stack
prevalence of this type of structure in modern cities. and wind interaction. It is also effective to prevent
The nature of the car park as a relatively confined smoke from spreading out into the escape routes
space and the high risk of vehicle fire make them and to other functional areas of the building. A
potentially dangerous. In these days, the fire risk simple design is to make direct use of the
potential can be assessed through fire modeling. ventilation system already installed for public
Fire modeling is generally divided into three health. As a general design practice, the ventilation
categories. The first is the empirical method, which rate is usually 6 to 8 air changes per hour (ACH).
involves costly experimentation and also physically
an element of risk. The second method is the zonal
method which assumes certain characteristics of 2. LITERATURE REVIEW
the fire, such as its stratification. The final method
is through the use of computational fluid dynamics Numerous types of CFD software are available in
(CFD) analysis, which can be based on finite the market and able to simulate combustion and
difference, finite element or finite volume methods. fire effects. The accuracy of various turbulence
models such as k-epsilon, renormalized k-epsilon
The poor performance of conventional ventilation and other combinations was assessed by Chow [3].
systems in clearing out smoke was demonstrated It was also found that similar results could be
through a series of fire tests by Colt International obtained from these various models. Kandola and
[1]. A new method involving impulse and Morris analyzed the results of simulation studies
induction was outlined. Safety systems in relation that were carried out using AEA CFD-FLOW3D
to underground car park were investigated by [4]. The hazards of smoke were highlighted, since
Chow [2]. The fire environment was studied by a the smoke from distant fires could affect people
fire zone model (CFAST), and based on this the and fire safety equipment. The attempts of AEA
recommendations for future car park design were Technology in validating CFD codes were
given. The activation time of the sprinkler was discussed by Sinai et al. [5]. Practical examples
found critical and suggested to be longer than the were also presented, such as the case of an
escape time of occupants. According to the fire explosion on an oil rig platform. The PHOENICS
regulation of Hong Kong, automatic sprinkler, fire CFD software was used by Mawhinney et al. to
hydrant and hose reel systems are to be provided at simulate domestic fires [6]. It was found that the
underground car parks. In addition, smoke simulated results agreed well with observed trends.

35
International Journal on Architectural Science

A review of CFD fire modeling and the important ∂ ui ∂ ui u j 1 ∂ p ∂ 2 ui

aspects of CFD applications such as mesh size was + = +υ (2)
∂t ∂ xj ρ ∂ xi ∂x j ∂x j
given by Wang et al [7]. Steckler et al. used
experimental fire data to validate several CFD
models [8]. Novozhilov reviewed the current trends where i and j stand for 1, 2 or 3 and correspond to x,
in compartment fire modeling [9]. His paper y and z co-ordinates. p is the static pressure, ρ is
discussed the importance of validation and the density and ν is the kinematic viscosity of the
experimental studies. Large eddy simulations were fluid. The problem with turbulence modeling is
discussed in detail as well. Also given was a largely due to the wide variation of length scales.
comparison between zone and field models, and the In order to describe the physical processes
continuing trend towards field models. adequately a high grid resolution is needed. This
requires large amounts of computing power and
Fires in tunnels where vehicle explosions are memory. In addition, the large range of time scales
frequent have been studied extensively through further complicates the problem. To overcome
both CFD and full scale tests. The vehicle tunnel is these problems, various solutions have been
similar to a basement car park in that both are proposed including the RANS (Reynolds Averaged
confined structure where smoke and fire can trap Navier Stokes) solution which gives the time
people. den Boer et al. carried out full scale fire averaged values. These solutions are based on the
tests on a new tunnel built in the Netherlands [10]. well known eddy viscosity models. Large eddy
The CFD PHOENICS software was used to simulation is another type of approximation where
simulate several of the tests. It was found that the the large scales are simulated directly, and the
accuracy of the software was reasonable, smaller scales are computed based on simple
considered that onsite conditions could not be turbulence model.
simulated completely and accurately. The smoke
was found over predicted in some cases. CFD fire The FDS has two options: large eddy simulation
simulation is economical in use and inherently (LES) and direct numerical simulation (DNS). The
safer than full scale fire tests. Tabarra et al. used solution for the small grade scales are based on the
scale model tests to validate CFD models applying Smagorinsky model [14]. Further, there are two
to a tunnel. The aerodynamic interaction of supply types of combustion models, of which the choice
vents and fire was studied, for the purpose of depends on whether the LES or DNS simulation
generating technical data for ventilation system has been chosen. In LES simulation, a mixture
design [11]. The use of CFD modeling of train fires fraction combustion is utilized. In this case the
in underground stations was studied by Deng et al. large scale convective and radiative heat transfer
[12]. The paper discussed grid selection, physical processes are calculated directly while the small
modeling, transient versus steady state flows. The scales are approximated. The soot yields are taken
problems associated with fire modeling such as from the FDS database which came from the SFPE
setting up of appropriate boundary conditions were Handbook of Fire Protection Engineering [15].
also addressed in detail. Numerous texts give illustrations and explanations
about the differences when comparing LES, DNS,
In the current study, the large eddy simulation and typical eddy viscosity models such as k-epsilon.
method was applied to study smoke and flame The FDS program uses finite differences for
propagation at the type of public underground car solution of the above equations (1) and (2) while
park commonly found in Hong Kong. This was finite volume is used to solve the thermal radiation
performed through the use of a public-domain transport equation. Lagrangian particles simulate
software – the Fire Dynamics Simulator. the smoke and air movement. The use of the finite
difference approach obviously limits the program
to rectilinear meshes and basic geometries, but
3. LARGE EDDY SIMULATION increases the computation speed.

The Fire Dynamics Simulator (FDS) software was

developed based on the solution of the Navier 4. VALIDATION EXAMPLES USING
Stokes equations for momentum and energy FDS
conservation in fluid flow [13]. The incompressible
Navier Stokes equations, written in tensorial Hadjisophocleous and McCartney presented the
notations, are: guidelines on the use of CFD to model fire [16]. In
particular the NIST fire dynamics simulator FDS
∂ ui was covered in detail. Several issues were
=0 (1) examined, including grid resolution, combustion
∂ xi
modeling, smoke modeling and radiative fraction.
Ryder et al. carried out a validation test of the FDS
and simulator. Four case examples were used and

36
 International Journal on Architectural Science

compared with experimental data; good agreement of the enclosing walls; both are 10 m wide. A
was found [17]. Devaud and Weckman tested the number of emergency exits are also at the boundary
ability of the FDS simulator in modeling walls. Exhaust outlets and supply inlets are
combustion, fire and soot, in the case of a fuel spill uniformly distributed at the ceiling, typically for
near an aircraft fuselage [18]. Reasonable mixing ventilation systems. In many cases, the
agreement between experiments and computations sources of ignition energy in vehicles are the same
was found. The NIST FDS was used by as those associated with structure fires. In our study,
Lautenberger et al. to predict soot formation for the modeled fire was started on top of a vehicle.
hydrocarbon flames [19]. A large outdoor fire was The fire was increased linearly up to 5 MW from
simulated by Vidmar and Petelin [20] using FDS. t = 0 to 100 s, as recommended by Morgan and
The results were compared with fire safety analysis Gardener [22]. Two case studies were performed
reports. More discussions and examples of the using different ventilating rates of supply air at
validation efforts can be found in the NIST FDS ambient temperature, i.e. 7 ACH for Case 1 and 14
manual [21]. ACH for Case 2. The supply velocities were
respectively 3 ms-1 and 6 ms-1.

5. PHYSICAL MODEL The following three meshes were investigated to

determine the most time effective solution: 60 × 30
The physical model of a basement car park, × 10, 100 × 50 × 20 and 140 × 80 × 30. The second
including the ventilation supply inlet and exhaust mesh was finally chosen based on the satisfactory
outlet positions, and the mesh grid, is outlined in accuracy and speed of computation. The program
Figs. 1 to 3. This 60 m × 30 m × 5 m underground files were then run on a Pentium 5 1.8 GHz PC for
car park is of size commonly found in basements of 12 hours. The program simulated three minutes real
commercial shopping malls in Hong Kong. Two time, starting from the fire outbreak.
vehicle exits are at the west side and the east side

Vehicle exit

Vehicle exit

Fig. 1: Isometric view of the physical model of the underground car park

Supply inlets

Exhaust outlets

Fig. 2: Locations of supply inlets and exhaust outlets at ceiling

37
International Journal on Architectural Science

Fig. 3: Selected mesh grid of the underground car park model

6. RESULTS AND DISCUSSION Case 2. The temperature then drops gradually due
to the decrease in combustion efficiency, which
The case studies indicate the sensitivity of the fire results from the build up of smoke and
and smoke spread to the ventilation system consequential reduction in oxygen supply.
performance. Comparison between Cases 1 and 2, Comparing the temperatures at t = 40 s and at t =
as in Figs. 4 and 5, shows that the smoke and fire is 100 s, it is found that for both cases, the areas of
significantly increased by the change in ventilation high temperature become smaller as the time
rate. Initially during the first 20 s, as shown in Figs. elapses. Generally, the temperature can become
4 and 5, the fire (light color) and smoke (dark color) unbearable after 30s. The doubling of the
propagations can be seen very similar. As time ventilation rate increases the overall temperature in
progresses, the fire in Case 2 is more dispersed and the occupied zone significantly.
intense than in Case 1. The smoke produced is also
slightly less for Case 2, the one with more fire Fig. 7 shows the levels of carbon dioxide at the 1.5
present, although this increases later on as the m level. The average level of carbon dioxide is
simulation progresses. Obviously the increased slightly higher for Case 2 with the higher air
ventilation rate supplies more oxygen to the fire change per hour. This is particularly evident in the
and accordingly, increased the intensity of the fire. first 20 s of the fire starting. At around t = 60 s and
The bulk of the fire in Case 2 also seems to drift t = 100 s similar levels of CO2 are found, though in
towards the left hand side of the car park as Case 2 the levels are more evenly dispersed
evidenced when t = 100 s. throughout the car park. In Case 1, a more
polarized distribution with higher levels at one end
The profile of the smoke in Fig. 5 illustrates further of the car park is observed. The carbon monoxide,
the smoke spread pattern. In both cases a shown in Fig. 8, follows a similar pattern and
mushroom cloud rises up to the ceiling and then distribution for both cases studied.
spreads to the rest of the car park. The time for the
smoke to reach almost all sections of the Fig. 9 shows the visibility in metres for the 1.5 m
underground car park is about 20 s. The speedy level of the car park. It can be seen clearly that
spread within this short time span gives alarm to around t = 20 s the visibility decreases significantly.
the danger of these indoor spaces. The visibility reduces to less than 1 m afterwards.
In Case 2 the visibility is worse than Case 1 for
The temperature profiles, given in Fig. 6, show the most parts of the underground car park especially
instantaneous temperature levels at 1.5 m above the within the first 20 s. The initial stage of the fire is
ground level. The average temperatures for Case 2 particularly crucial for evacuation; it is when the
are clearly higher than Case 1 at all time instants. occupants can assess the situation and give
The maximum fire temperature occurs at around t = responses.
40 s, reaching 965°C for Case 1, and 1000°C for

38
 International Journal on Architectural Science

t=3s

t = 10 s

t = 20 s

t = 60 s

t = 100 s
7 ACH 14 ACH

Fig. 4: Isometric view of smoke and flame propagation with time after the fire out break

39
International Journal on Architectural Science

t=3s

t = 10 s

t = 20 s

t = 60 s

t = 100 s
7 ACH 14 ACH

Fig. 5: Elevation view of smoke and flame propagation with time

Fig. 10 shows the heat released rate per unit the fire gases, and reducing the fire size (and hence
volume (HRRPUV) from t = 3 to 100 s. The the smoke), the sprinklers are viewed as an
comparison between the two cases shows that the effective means of smoke control. For a fire size of
increased ventilation rate increases the 5 MW, its actuation time can be from 31 s to 66 s,
combustibility of the surrounding air. The fire depending on its response time index [2]. Since an
initially starts at the location of the vehicle. As the early discharge of water may lead to a quick
temperature increases, the surrounding gaseous production of hot steam, which subsequently may
mixture ignites as the fuel gasifying. The increased hurt the occupants, a suitable response time of the
ventilation rate distributes the gasified fuel more sprinklers could be approaching 60 s. This is triple
effectively hence the higher level of heat released the 20 s for the smoke and fire to fill up the entire
rate. The flashover point can be observed at around car park space. Another issue is that the smoke will
t = 40 s when the entire car park is undergoing further lower down the visibility and slow down
complete combustion in Case 2, and the the escape speed. For people who are not familiar
combustion in Case 1 is nearly (if not yet) complete. with the building, a visibility of 15 m to 20 m is
However, because of the lower ventilation rate in required for safe evacuation. If this is the case, the
Case 1 the car park is never in full combustion, car park design should facilitate the occupants and
probably due to inadequate distribution of fuel the vehicles to evacuate within 20 s or so. Hence
vapor. Hence from the above observations, there are needs to increase the number of
increasing ventilation rate increases the speed and emergency exits for the occupants, and to shorten
likelihood of the fire reaching the flashover point. the vehicle exit routes.
Also it can be seen that in the case of lower
ventilation rate the fire subsides slightly after t =
40 s. This could be because of the increased 7. CONCLUSIONS
proportion of smoke which reduces the
combustibility of the air in the underground car The FDS simulator has been made use of to predict
park. And the lower ventilation rate results in more the dynamics of a reasonably large fire at 5 MW in
soot produced owing to inefficient combustion. an underground car park. The CFD technique
allows an effective control of the boundary
Taking the evacuation speed of the occupants (who conditions and so an examination of the effect of
are not familiar with the environment) as 0.8 ms-1 various parameters. Our results show that in such a
and the diagonal length (67 m) of the car park as confined space the smoke and the flame can spread
the longest traveling distance, the maximum quickly and may fill up the entire volume space in
required escape time is therefore 84 s. For those less than one minute. This can be before the
occupants inside the vehicles, the escape time by actuation of the automatic sprinkler protection
means of the vehicles could be around 60 s, taking system, and shorter than the evacuation time of the
a vehicle traveling speed of 1 ms-1. On the other occupants. In this sense, the threat is high. For this
hand, the sprinkler heads in a car park are type of car park, it will be desirable to increase the
commonly started to operate at 68oC. By cooling

40
 International Journal on Architectural Science

number of exits, to shorten the vehicle exit routes, the smoke more and to reduce visibility, especially
and to make the exit signs clear to the occupants. during the initial phase of the fire outbreak. The
possibility of flashover was also found higher for
A higher ventilation rate tends to accelerate the fire increased air ventilation rates. Our findings suggest
spread. This also leads to a higher temperature that the current fire safety design recommendations
environment which can endanger the human life. In can be insufficient to help the practicing engineers.
terms of safety it appears questionable whether the More work to explore the potential hazards and the
use of a higher ventilating rate (for the sake of effective means of protecting human life should be
indoor air quality) is a good design solution. In proceeded in this area.
addition, increasing ventilation tends to disperse

<100 <100 400

°C
400
100 100

t = 10 s

<100 <100
200 900
800 700

900 200
900
200
400 400
t = 20 s

100 900
100 400 600
100 990
100
600 960 400 600
700 800

400 500 990

100 500 100

t = 40 s

600 900
100
100
950

300 200
900
200
300
600
600 950
400 400

t = 100 s

7 ACH 14 ACH

Fig. 6: Plan view of temperature distribution at 1.5 m level (°C)

41
International Journal on Architectural Science

0.03 0.03
0.1

0.1

<0.01 <0.01

t = 10 s

0.01 0.01
0.05
0.1
>0.1
0.05

0.03
<0.01 <0.01 0.03

t = 20 s

0.01 >0.1 0.01 >0.1

<0.01 <0.01

0.02 0.08

0.03
0.09 0.03
0.07
0.03 0.07 0.09
0.01

t = 40 s

>0.1 0.06 0.06

0.09 >0.1 0.09
<0.01 <0.01
0.07
0.05 0.01 0.02
0.09

0.07 0.09 0.1

0.07 0.09

t = 100 s

7 ACH 14 ACH

Fig. 7: Plan level of carbon dioxide at 1.5 m (mol/mol)

42
 International Journal on Architectural Science

60
100

300 300

<10 <10

t = 10 s

90
240 300 210
30
90
150
270
270

<10 120 <10 120

t = 20 s

30 300 300
30 300
240 210
<30 <30
60
150 300
240
60 270 270 300

240 100
90 210 30

t = 40 s

210 90 150 270

270
90 300
<30 <30 210
300 30
180 60
270 120
240
180
270 210
240 300 270 270

t = 100 s

7 ACH 14 ACH

Fig. 8: Plan level of carbon monoxide at 1.5 m (ppm)

43
International Journal on Architectural Science

4
7 4
1 7 1

10
16 4
7
19
>31 19 >31

t = 10 s

4 >31 10
31
13 4 4
22 19
16 1
1 10
22
4 22
>31 >31 7 13

t = 20 s

10 31 10 22
<1 <1

7 31
16
4 25
7 4
4
4 10 19
4
4

t = 40 s

<1 <1
7

4 7

10 4

t = 60 s

<1 <1
7 4

7
4

t = 100 s

7 ACH 14 ACH

Fig. 9: Plan view of level of visibility at 1.5 m (m)

44
 International Journal on Architectural Science

400 400

t=3s

400

400

t = 10 s
400
160
400
160
400 400

t = 20 s
400
400
400
360
400 400

400 400
400
320

t = 40 s
400
400 320 400

360
320 240

400 400
360
360 400

t = 60 s 400

280
400 360 400

400 400
400
400
400 280
320

t = 100 s
7 ACH 14 ACH

Fig. 10: HRRPUV (heat released rate per unit volume, kWm-3) at 1.5 m level

45
International Journal on Architectural Science

REFERENCES 13. K.B. McGrattan, H.R. Baum, R.G. Rehm, A.

Hamins, G.P. Forney, J.E. Floyd and S. Hostikka,
1. Colt International, “Smoke ventilation for Fire Dynamics Simulator - Technical Reference
underground car parks”, Building Engineer, Vol. Guide, NISTIR 1081, National Institute of
80, No. 1, pp. 41-42 (2005). Standards and Technology, Gaithersburg, MD,
USA (2004).
2. W.K. Chow, “On safety systems for underground
car parks”, Tunnelling and Underground Space 14. J. Smagorinsky, “General circulation experiments
Technology, Vol. 13, No. 3, pp. 281-287 (1998). with the primitive equations. Part I: The basic
experiment”, Monthly Weather Review, Vol. 91,
3. W.K. Chow, “Numerical studies of airflows pp. 99-152 (1963).
induced by mechanical ventilation and air-
conditioning (MVAC) systems”, Applied Energy, 15. Society of Fire Protection Engineers, SFPE
Vol. 68, No. 2, pp. 135-159 (2001). Handbook of Fire Protection engineering, National
Fire Protection Association, Quincy, Boston,
4. B.S. Kandola and M. Morris, “Smoke hazard Massachusetts, USA (1988).
assessment using computational fluid dynamics
(CFD) modeling”, Institution of Chemical 16. G.V. Hadjisophocleous and C.J. McCartney,
Engineers Symposium Series, No. 134, pp. 171- “Guidelines for the use of CFD simulations for fire
194 (1994). and smoke modeling”, ASHRAE Transactions,
Vol. 111, Part 2, pp. 583-594 (2005).
5. Y.L. Sinai, M.P. Owens and P. Smith, “Advances
in CFD assessment of fire and smoke movement”, 17. N.L. Ryder, J.A. Sutula, C.F. Schemel, A.J. Hamer
Proceedings of the International Conference on and V. van Brunt, “Consequence modeling using
Offshore Mechanics and Arctic Engineering - the fire dynamics simulator”, Journal of Hazardous
OMAE, Vol. 2, pp. 389-395 (1995). Materials, Vol. 115, No. 1-3, pp. 149-154 (2004).

6. R.N. Mawhinney, E.R. Galea, N. Hoffmann and 18. C.B. Devaud and E.J. Weckman, “Numerical
M.K. Patel, “Critical comparison of a PHOENICS simulation of a pool fire and large object in a
based fire field model with experimental cross-wind”, Proceedings of the ASME Heat
compartment fire data”, Journal of Fire Protection Transfer/Fluids Engineering Summer Conference
Engineering, Vol. 6, No. 4, pp. 137-152 (1994). 2004, Vol. 4, pp. 217-220 (2004).

7. J. Wang, J.S. Hua, K. Kumar and S. Kurichi, 19. C.W. Lautenberger, J.L. de Ris, N.A. Dembsey,
“Modeling and validation of fire induced airflow J.R. Barnett and H.R. Baum, “A simplified model
in a room”, Proceedings of the 2004 International for soot formation and oxidation in CFD
Symposium on Safety Science and Technology, simulation of non-premixed hydrocarbon flames”,
PART A, Progress in Safety Science and Fire Safety Journal, Vol. 40, No. 2, pp. 141-176
Technology, Vol. 4, pp. 1164-1170 (2004) (2005).

8. K.D. Steckler, J.G. Quintiere and W.J. Rinkinen, 20. P. Vidmar and S. Petelin, “Large outdoor fire
“Flow induced by fire in a compartment”, model analysis”, Proceedings of the 2003
Symposium (International) on Combustion, pp. International Joint Power Generation Conference,
913-920 (1982). pp. 817-822 (2003).

9. V. Novozhilov, “Computational fluid dynamics 21. K.B. McGrattan, H.R. Baum, R.G. Rehm, A.
modeling of compartment fires”, Progress in Hamins and G.P. Forney, Fire Dynamics Simulator
Energy and Combustion Science, Vol. 27, No. 6, - Technical Reference Manual, NISTIR 6467,
pp. 611-666 (2001). National Institute of Standards and Technology,
Gaithersburg, MD, USA (2004).
10. D.J. den Boer, J.J. Mertens, N.J. van Oerle and J.F.
Wijnia, “Validation of CFD calculations of full 22. H.P. Morgan and J.P. Gardner, “Design principles
scale medium sized fires in a two lane road tunnel”, for smoke ventilation in enclosed shopping
11th International Symposium on Aerodynamics centres”, Building Research Establishment Report,
and Ventilation of Vehicle Tunnels, Vol. 2, pp. Fire Research Station, Borehamwood, UK (1990).
657-667 (2003).
11. M. Tabarra, B. Kenrick and R.D. Matthews, “CFD
validation of natural smoke movement in a model
tunnel”, American Society of Mechanical
Engineers, Fluids Engineering Division
(Publication) FED, Vol. 238, No. 3, pp. 543-546
(1996).
12. M.P. Deng, P.C. Miclea and D. McKinney, “CFD
modeling considerations for train fires in
underground subway stations”, American Society
of Mechanical Engineers, Fluids Engineering
Division (Publication) FED, Vol. 238, No. 3, pp.
547-555 (1996).

46