COMPUTATIONAL METHODS IN ENGINEERING AND SCIENCE
EPMESC X, Aug. 21-23, 2006, Sanya, Hainan, China
©2006 Tsinghua University Press & Springer
Numerical Simulation of Atrium Fire using Two CFD Tools
V. K. Sin1, L. M. Tam1,2, S. K. Lao1, H. F. Choi1*
1
Department of Electromechanical Engineering, Faculty of Science and Technology, University of Macau, Macau,
China
2
Institute for the Development and Quality, Macau, China
Email: ma46558@umac.mo
Abstract Building fires lead to casualties which are mainly caused by smoke. Therefore, smoke management system
is one of the most important factors to be considered in Fire Safety Engineering (FSE). Its major aim is to ensure that
lives are protected from heat and smoke [1]. Applying full scale experimental study on fire dynamics research is
relatively expensive and time-consuming. Thus, computational fluid dynamics (CFD) has been widely used in
assessments of fire hazards. The most popular models for fire simulation using CFD are zone model and field model.
According to [2], atria, covered shopping malls, convention centers, airport terminals, sport arenas, and warehouses
are examples of large spaces for which conventional zone-model approaches are not always effective. In this paper,
numerical simulation of fire-produced smoke movement in a large complex space proposed in [3] has been carried out
in field model with two CFD tools, STAR-CD and FDS. STAR-CD is a general-purpose, unstructured CFD code that
allows extremely complex scenarios often encountered in many situations to be simulated. Fire Dynamics Simulator
(FDS) is a fire model based on large-eddy simulation CFD code developed by the National Institute of Standards and
Technology (NIST). The data obtained from calculations based on these two field models are being compared. Results
of the transient development of temperature distribution and velocity distribution are presented and analyzed in this
study. It is found that the trend of the results obtained from both of these two CFD tools are in good agreement.
Key words: performance-based code, fire safety engineering, field model, STAR-CD, FDS
INTRODUCTION
Along with the rapid development of the society and architectural techniques, many high-rise and large space
buildings have appeared. The atrium building has become a commonplace. From the viewpoint of Fire Safety
Engineering (FSE), the application of traditional prescriptive regulations can hardly legitimately evaluate the fire
protection design of new complex structures. Gradually, a new approach, performance-based method has been
developed and being applied to meet these needs in recent years. In line with this world-wide trend, the U.S. and
Canada are planning to introduce performance-based codes in the near future. This performance-based approach
enables flexibility in design that can lead to lower construction costs without lowering the level of safety [4].
Fire simulation with performance-based code has been widely used in assessments of fire hazards due to its
cost-effectiveness and versatility compared to the traditional full-scale experimental study. Two different modeling
techniques are commonly used in fire simulation: zone modeling and field modeling, where the latter is also referred to
as Computational Fluid Dynamics. According to [2], atria, covered shopping malls, convention centres, airport
terminals, sport arenas, and warehouses are examples of large spaces for which conventional zone model approaches
are not always effective. The problem of shopping malls for fire simulation is that they are complex in geometry. They
are generally different in style, size and can be very non-standard in their design. In addition, the physics of fire is also
extremely complicated; it is a mixture of heat convection, radiation and conduction [5]. The paper presents numerical
simulations of fire-induced thermal and fluid flow in shopping mall. Results with different modeling parameters, such
as computational grids, turbulence models, are compared to quantify their relative effects.
__
207
__
DESCRIPTION OF STAR-CD AND FDS
Many CFD codes have been developed for fire science and fire safety engineering in the past decades. Increased use of
computer modeling is due to the advancement of computing technology and the availability of various improved and
faster computation method. Friedman [6] and Stephen [7] have conducted a comprehensive survey of computer models
for fire and smoke. In this article, two CFD tools are used: STAR-CD and FDS.
STAR-CD [8] is a general purpose commercial CFD package which consists of two components (see Fig. 1): the
numerical analysis module STAR and the pre- and post-processing module pro-STAR. It provides rich models for
turbulence, combustion, radiation and multiphase physics. Those models can be used for a wide range of applications,
such as combustion engine design and aerospace industry. The flexibility of STAR-CD is that it can accept almost any
mesh structure or cell shape. It can also allow the user to write FORTRAN routines to define an unusual boundary
condition. However, the use of STAR-CD code to validate studies of fire application is still rare. In order to verify the
accuracy of modeling of fire related heat transfer and fluid flow by using STAR-CD, the authors have put effort to
study on the numerical simulations of fire in enclosure [9]. It was found that the temperature predictions agree well
with experimental data. It provides confidence that STAR-CD is capable of modeling fire generated heat transfer in an
enclosure. However, existing experimental data about fire testing for large spaces are rather insufficient and not easily
accessible. It is hard to evaluate the validation of STAR-CD in fire simulation with large spaces through comparisons
with experimental data. The objective of this study is to investigate the effectiveness of simulation in a large complex
structure by using STAR-CD and the results are compared with those obtained from FDS, another CFD code which
will be discussed in the next paragraph.
Figure 1: Overall STAR-CD system structure (Source: Computational Dynamics Ltd.)
Fire Dynamics Simulator (FDS) is a fire model based on large-eddy simulation CFD code developed by the National
Institute of Standards and Technology (NIST). It consists of two components (see Fig. 2), FDS and Smokeview. FDS
is a CFD code written in FORTRAN 90, and Smokeview is an OpenGL graphics program for visualization of results.
Fig. 2 illustrates an overview of how data files used by both FDS and Smokeview are related. FDS is well documented
Figure 2: An overview of FDS structure (Source: NIST)
__
208
__
and can be obtained from NIST at no cost. It is widely used by fire protection engineers to predict smoke generation,
hot gas plume behaviour, concentration of toxic species, thermal exposure and visibility. The limitation of FDS is that
the grid cells must be rectangular with the aspect ratios which are not very large. It means that only relatively small
cells can be used to form surfaces and steps are used to approximate the curved shapes. This limitation is due to the fact
that the Fast Fourier Transform (FFT) is used in FDS that can reduce calculation time to a fraction of those used with
conventional models [10].
One of the main differences between STAR-CD and FDS is the structure of the mesh. In FDS, mesh must be structured
being based on a rectangular block whereas STAR-CD uses an unstructured approach. The latter enables the modeling
of extremely complex geometries more easily and efficiently.
GEOMETRY OF SIMULATED BUILDING
Shopping malls are a typical case for the need to undertake fire risk assessment. A shopping mall with geometry
proposed in [3] was selected due to its complex shape of building. The shopping mall is an integration of an L-shape
and a ball-shape structure, which is located in Taipei. The sphere has a diameter of 58 meters. The mall has 12 stories
high and 7 underground levels. The height of the mall in simulation is about 74 meters. Because of the shape of the
building, the grid generation capability of FDS is challenged. Besides FDS, an additional program known as
DXF2FDS, which is also developed by NIST, is needed. DXF2FDS [11] is a program that converts 3D DXF cad files
to FDS input files. The geometry created by AutoCAD (see Fig. 3) is generated into a DXF file with 3DFACE
variables, which is further converted into FDS obstruction. On the contrary, the capability of grid generation of
STAR-CD is more powerful than that of FDS. It can save significant amounts of time to build up the complex
geometry. Fig. 4 illustrates the computational domain created by STAR-CD and FDS.
Figure 3: The geometry of the shopping mall created by AutoCAD
Figure 4: The computational domain created by STAR-CD (left) and FDS (right)
__
209
__
MATHEMATICAL MODEL
STAR-CD provides a wide variety of turbulence modeling capabilities. The k-ε turbulence model was selected to use
in this study. Although the k-ε turbulence model has its limitations, it is extensively implemented in fire application. Its
chief privileges are its low computational cost and its comparatively numerical stability. Its main limitation in the
context of smoke movement is that it assumes an isotropic eddy viscosity, which does not account for the non-isotropic
effects of buoyancy on turbulent mixing [12]. Reynolds-averaged Navier-Stokes equations and the k-ε turbulence
model are solved in STAR-CD. Further details for the governing equations can be found in [9].
Recently, large eddy simulation has attracted the engineers’ attention for application in fire modeling. It has the
potential to model the turbulent flow phenomena more realistically. FDS is built upon this concept. In FDS, the
Smagorinsky form of Large Eddy Simulation (LES) is applied. The concept of LES is that the large-scale eddies are
computed directly and the sub-grid scale dissipative processes are modeled. The LES is inherently time-dependent
because the Navier-Stokes equations are not time-averaged. FDS solves a form of the Navier-Stokes equations
appropriate for low Mach number, thermally expandable, multi-component mixture of ideal gases and the governing
equations are listed below [13]:
Conservation of Mass:
∂ρ
v
+ ∇ ⋅ ρu = 0
∂t
(1)
Conservation of Momentum:
v
v
v⎞
v v
⎛ ∂u
ρ ⎜ + (u ⋅ ∇)u ⎟ + ∇p = ρg + f + ∇ ⋅ τ
⎝ ∂t
⎠
(2)
Conservation of Energy:
∂
v Dp
v
( ρh) + ∇ ⋅ ρhu =
− ∇ ⋅ q r + ∇ ⋅ k∇T + ∑ ∇ ⋅ hl ρDl ∇Yl
Dt
∂t
l
(3)
Conservation of Species:
∂
v
( ρYl ) + ∇ ⋅ ρYl u = ∇ ⋅ ρDl ∇Yl + m& l′′′
(4)
∂t
v
v
v
where ρ is density, u is velocity vector; p is pressure; g is gravity vector; τ is viscous stress tensor; f is external
v
force vector (excluding gravity); h is enthalpy; q r is radiative heat flux vector; T is temperature; k is thermal
& l′′′ is mass production rate of lth species
conductivity; D is diffusion coefficient; Yl is mass fraction of lth species. m
per unit volume.
The viscous stress tensor, τ , in the momentum equation is defined as
v 2
v v⎞
⎛
3
⎝
⎠
v
where I is the identity matrix and the deformation tensor is given by
τ = μ ⎜ 2def u − (∇ ⋅ u ) I ⎟
[
v 1 v
v
def u ≡ ∇u + (∇u ) t
2
(5)
]
(6)
The dynamic viscosity, μ , can be defined as
μ ijk = ρ ijk (C s Δ) 2 S
(7)
and
2
2
2
2
2
2
⎛ ∂v ⎞
2
v
⎛ ∂u ⎞
⎛ ∂w ⎞ ⎛ ∂u ∂v ⎞ ⎛ ∂u ∂w ⎞ ⎛ ∂v ∂w ⎞
⎟⎟ − (∇ ⋅ u ) 2
S = 2⎜ ⎟ + 2⎜⎜ ⎟⎟ + 2⎜ ⎟ + ⎜⎜ + ⎟⎟ + ⎜ +
⎟ + ⎜⎜ +
3
⎝ ∂x ⎠
⎝ ∂z ⎠ ⎝ ∂y ∂x ⎠ ⎝ ∂z ∂x ⎠ ⎝ ∂z ∂y ⎠
⎝ ∂y ⎠
(8)
where Cs is an empirical constant; Δ = (δxδyδz )1 / 3 , is length on the order of a grid cell; S is magnitude of the stress tensor.
__
210
__
In an LES calculation, the thermal conductivity and material diffusivity are related to the turbulent viscosity by
k ijk =
c p , 0 μ ijk
Pr
; ( ρD) ijk =
μ ijk
(9)
Sc
where c p , 0 is the specific heat of the dominant species of the mixture; Pr is the Prandtl number and S c is the Schmidt
number. Based on simulation of smoke plumes, Pr and S c are 0.5.
MODEL OF FIRE SOURCE
For shopping malls, the combustible material was assumed to be the furniture. A realistic plan area for fire was set the
same dimensions as those of the furniture, i.e. 2m x 1.5m. According to the study of Morgan [14], the size of a fire was
recognized as 5MW, which is a realistic assumption. In STAR-CD, a prescribed heat source model was used. A fire
was represented by a fixed volume in which heat was released. The fire was assumed to grow according to a fast
t-squared curve. In FDS, the heat release rate can be either prescribed or predicted. In the simulation, prescribed heat
release rate was chosen rather than the predicted one. FDS uses a mixture fraction combustion model. The mixture
fraction is a conserved scalar quantity that is defined as the fraction of gas at a given point in the flow field that
originated as fuel [13]. The model, based on the reaction of fuel and oxygen, is infinitely fast. The advantage of the
mixture fraction approach is that all of the species transport equations are combined into one, reducing the
computational cost [15]. Fig. 5 gives an isometric view of the shopping mall on which the fire is located.
Figure 5: Fire location. (Left – STAR-CD, right – FDS)
BOUNDARY CONDITIONS
As an initial phase of research, the shopping mall was assumed to be well sealed. Hence, no inlet or outlet were
defined. So, the ventilation strategy and the fire suppression are not investigated in this work. In addition, a no-slip
condition was applied at the wall. Adiabatic walls were imposed at the shopping mall.
COMPARISON OF DIFFERENT MODELING APPROACHES
Even though modern workstations are powerful, CFD calculations are still expensive. Factors influencing the
computation time are mesh sizes, physical and numerical model. In order to obtain reasonable run-times, it is necessary
to study the sensitivity of these modeling parameters. Table 1 outlines the important modeling parameters chosen in
this study for the two CFD tools.
Table 1 Summary of the modeling parameters
Modeling
parameters
Fire model
STAR-CD
FDS
Volume heat source model
Mixture fracture combustion model
Turbulence model
k-ε model
LES model
Radiation model
Discrete Ordinate (S4) method
Finite volume method
__
211
__
RESULT AND DISCUSSION
Grid size and total number of grid cells used for coarse and fine grid in STAR-CD and FDS are shown in Table 2.
Temperature distributions on a plane through the fire at 30s, 120s, 180s, 240s, 360s and 420s are shown respectively in
Fig. 6(a) to 6(f) for coarse grid. Except in the initial stages of the fire, the overall transports of hot gas layer obtained by
these two CFD tools present the similar gross trends.
(a) 30s
(b) 120s
(c) 180s
(d) 240s
(e) 360s
(f) 420s
Figure 6: Temperature distribution (K) of coarse grid at various times (Left – STAR-CD, right – FDS)
Table 2 Summary of the different simulation
Simulation
STAR-CD
FDS
Grid size
Number of grid cells
CPU time (hour)
Coarse
1 x 1 x 1m
42,840 cells
13.16
Fine
0.5 x 0.5 x 0.5m
342,720 cells
61.14
Coarse
1 x 1 x 1m
300,960 cells
3.39
Fine
0.5 x 0.5 x 0.5m
2,592,000 cells
61.32
__
212
__
In order to test the grid sensitivity, a fine grid is used to perform the simulation again and the results are shown in
Figure 7a to 7f. The results of hot gas layer behavior between the two CFD tools are also similar to each other,
especially in development and temperature distribution of the hot gas layer. The hot gas layer rises to 30 m at
approximately 120 seconds after ignition. The temperature decreases as the hot layer rises. As the fire progresses, the
hot gas layer reaches the ceiling of the shopping mall and air is continuously entrained into the ascending plume. The
plume rises vertically until it is impeded by the ceiling of the mall after 240 seconds. At this moment, the plume is
constrained to flow in a predominantly horizontal direction, resulting in a ceiling jet (see Figure 7e). However, the flow
under the ceiling is turbulent; thus, the flow is not strictly horizontal due to the existing vortices. This phenomenon can
be observed in Figure 8 which shows the velocity field near the ceiling of the mall obtained from simulation by both
STAR-CD and FDS. In Figure 7f, it is found that the base of this hot gas layer may continue to descend after 420
seconds, and the temperature near the fire region predicted by STAR-CD is higher than that of FDS by 3.7%.
Furthermore, from the simulation of STAR-CD in Figure 7f, it is noted that there is an explicit stratification on both
sides of the source of the fire. This phenomenon may be caused by the turbulent model being used. Since the standard
k-ε turbulence model is used and which does not account for the non-isotropic effects of buoyancy on turbulent
mixing. It is possible that the grid is too coarse compared to the length scales of the key flow phenomena. In the
absence of experimental data, it is not easy to examine this uncertainty. From the viewpoint of fire safety, the closer the
hot layer remains to the ground, the higher the potential risk is for the public.
(a) 30s
(b) 120s
(c) 180s
(d) 240s
(e) 360s
(f) 420s
Figure 7: Temperature distribution (K) of fine grid at various times (Left – STAR-CD, right – FDS)
__
__
213
(a) STAR-CD
(b) FDS
Figure 8: Velocity field (m/s) near the ceiling of the shopping mall, 420 seconds after ignition
By comparing the result of simulations with coarse and fine grid, a notable difference very close to the fire source is
observed, the hot gas layer is propagating more rapidly near the fire region for coarse grid. This occurrence may be
brought about by the influence of a relatively coarse grid resolution in the fire source. Though the computational time
is high for the use of a fine grid, more flow details can spontaneously be obtained by it. For instance, eddies are
obviously found near the ceiling. In the meantime, the coarse grid tends to omit the flow details, and this may lead to an
inaccurate result. Definitely, a finer grid offers more confidence for predicting the main feature of the flow. Also
shown in Table 2 are the CPU time for different simulations which were carried out on a personal computer with Intel
Pentium 4 of 2.8 GHz, 2 GB RAM. Even though the grid number used in FDS is much larger than those used in
STAR-CD (i.e., 7 times for coarse grid and 8 times for fine grid), the CPU time used in FDS is only 0.25 times of that
used in STAR-CD for coarse grid simulation, and the CPU time is almost the same for both STAR-CD and FDS for
fine grid.
CONCLUSION
Simulations have been undertaken for coarse and fine grid with two different CFD tools. It is found that the results of
temperature distribution, thickness and rate of propagation of hot gas layer obtained by STAR-CD and FDS based on
different modeling parameters are in good agreement. All results appear to be physically reasonable, but there are
uncertainties due to the difference in the fire model, physical models, and assumed boundary conditions. The results
could only be compared qualitatively because no experimental data were available. It is also noted that the CPU time
used for both two CFD tools are comparable in this study. In summary, CFD can be successfully used in support of a
fire safety assessment in unique geometry like the shopping mall investigated in this study.
REFERENCES
1. NFPA 92B. Guide for Smoke Management Systems in Malls, Atria and Large Areas. National Fire Protection
Association, Quincy, MA, 2000.
2. Chang CH, Banks D, Meroney RN. Computational fluid dynamics simulation of the progress of fire smoke in
large space, building atria. Tamkang Journal of Science and Engineering, 2003; 6(3): 151-157.
3. Yang CS. Design Analysis and Experimental Investigation of Smoke Management and Egress System of a Large
Shopping Mall. PhD dissertation, National Sun Yat - Sen University, Taiwan, China, 2003.
4. Yung DT, Benichou N. How design fires can be used in fire hazard analysis. Fire Technology, 2002; 38:
231-242.
5. Robert F. Fire protection in shopping malls. Fire Safety Engineering, 2002; 9(3): 18-20.
6. Friedman. An international survey of computer models for fire and smoke. Journal of Fire Protection
Engineering, 1992; 4(3): 81-92.
7. Olenick SM. An updated international survey of computer models for fire and smoke. Journal of Fire Protection
Engineering, 2003; 13(2): 87-110.
8. STAR-CD V3.22 User Guide. Computational Dynamics Ltd., http://www.cd-adapco.com.
__
214
__
9. Tam LM, Sin VK, Lao SK, Choi HF. CFD analysis of fire in a forced ventilated enclosure. Proceeding of
EPMESC X, Aug 21-23, 2006, Sanya, Hainan, China.
10. Klote JH, Milke JA. Principles of Smoke Management. American Society of Heating, Refrigeration and
Air-Conditioning Engineers, 2002.
11. http://fire.nist.gov/fds/
12. Gobeau N, Zhou XX. Evaluation of CFD to Predict Smoke Movement in Complex Enclosed Spaces. Research
report 255, Health and Safety Laboratory for HSE, 2004.
13. McGrattan KB, Baum HR, Rehm RG, Hamins A, Forney GP. Fire Dynamics Simulator, Technical Reference
Guide. Technical Report NISTIR 6467, NIST, Gaithersburg, Maryland, USA, January 2000.
14. Morgan HP. Smoke Control methods in enclosed shopping complexes of one or more stores: A design summary.
Building Research Establishment report, 1979.
15. McGrattan KB, Baum HR, Forney GP, Floyd JE. Improved radiation and combustion routines for a large eddy
simulation fire model. Fire Safety Science. Proc. 7th International Symposium International Association for Fire
Safety Science. June, 2003, pp. 827-838.
__
215
__