International Journal of Computer Applications (0975 – 8887)

Volume 70– No.11, May 2013

Experimental Study and CFD Simulation of Pool Fires

Alireza Alizadeh Attar Mojtaba Pourmahdian Bagher Anvaripour

Corresponding Author MSc student of Petroleum Petroleum University of
Petroleum University of University of Technology Technology
Technology

ABSTRACT
Liquid fuels(hydrocarbon and peroxide)are flammableand
there is a risk of pool fires during their storage and
transportation. In order to measure and develop effective
methods for protection and considering safety distances,
experimental studies and CFD simulations of pool fire of two
hydrocarbon fuels (i.e. gasoline and kerosene) and peroxide
(TBPB (tert-butyl peroxybenzoate)) were performed.
Experiments revealed that the pool fires of liquid organic
peroxide show fundamentally very different characteristics
e.g. generally much higher mass burning rate, largerflame
length as well as higher temperature and subsequently higher
irradiance in comparison to liquid hydrocarbon pool fires. The
three well-known flame zones of pool fires is well captured by
CFD simulations and the predicted axial flame temperature
profiles.The safety distances accurately predicted by CFD
simulations when predicted time averaged maximum flame
temperature is used instead to the experimentally measured
values in calculations performed in this work. Fig 1: Existence of three zones in a turbulent buoyant
diffusion flame
Keywords
CFD simulation,pool fire,temperature,irradiance A zone of un-burnt fuel vapor above the liquid fuel, which is
usually close to a constant conical shape.
1. INTRODUCTION
Pool fire is the most common consequence of any accidentally A combustion region above this zone but here there is
released flammable liquid. The occurrence of such scenario intermittency and obvious turbulence in the reactive flow.
may appear in a storage facility or during the transportation of The non-reacting buoyant plume which is generally fully
dangerous goods. A number of measures have been developed turbulent in nature and is characterized by decreasing
for specifying the safety distances from such fires for the velocity and temperature with height and lateral position.
people and places. The characteristics of large pool fire e.g.
burning rate, flame length and radiation are the most Physical characteristics associated with the pool fire have a
common parameters of interest. There have a lot experimental direct impact on the different zones and this impact is
work [2] been done on the measurement of these generally defined by means of measurable quantities.
characteristics. Recently some computational work has also
been reported by various groups [4]. The prediction of flame 3. EXPERIMENTAL PROCEDURE
temperature,velocity, irradiance and surface emissive power The arrangement of tests is shown in “Figure 2” where the
were carried out by using CFD simulation with various sub- fuel pan was placed on top of a weighing instrument. The
models for turbulence and chemistry. The decision about the thermocouples were located at increased axial distance above
distance for their safe storage and transportation is made on the liquid pool surface. The video camera is located at a
the basis of standard fire tests. Burning rate, flame defined location (depending on the intensity of the flame)
temperature and irradiance are the important parameters of from the flame. The fire experiments from pool diameter of
concern. 0.06 m to 0.18 m were carried out in the laboratory. The size
and the materials of the pans are given in “Table 1”.
2. DYNAMICS OF POOL FIRES Flame lengths were measured by using the movie sequence
The structure of most pool fires may be split into a number of recorded by a video camera. The instantaneous images of the
fairly well-defined zones [1,3,5-7]. These zones are shown in flame obtained during the main burning period were then
“Figure 1”[7] and the physical processes along with their converted into time averaged value of the flame length.
technical importance is described below.
The liquid fuel itself : In deep pools there may be a significant
convective flow within the fuel which can affect the fuel
vaporization rate.

9
 International Journal of Computer Applications (0975 – 8887)
Volume 70– No.11, May 2013

Fig 3: A scheme of the 3-D meshes applied in simulations

Fig 2:Typical test set up
Altogether more than 400,000 control volumes are used for
the solution of transport equations described before. Since the
Table 1. Table captions should be placed above the table mesh-dependent studies are important for the numerical
solution to be independent on mesh size, so it was made sure
Pool that the simulations are mesh independent. The different grid
Pan material diameter Pan Height(m) Pan wall(m) size and the number of the control volumes that had been used
(m) in this work is shown in “Table 2”. The results of the
steel 0.06 0.05 0.002 prediction of axial time averaged flame temperature
steel 0.11 0.05 0.002 distribution using different grids for the pool fire of the
kerosene in a pool diameter of 0.06 m is shown in “Figure 4”.
steel 0.18 0.05 0.002 As can be seen the further refinement of the grids from grid-2
to the grid-3 have minor effect on the axial flame temperature
distribution of the kerosene thus the sizes corresponds in grid-
4. MODELING AND SIMULATION 3 is applied for the simulations carried out in this work.
4.1 Procedure of CFD simulation Table 2
In present work, pool fires of different hydrocarbons are
simulated. The CFD simulation is done by using different sub- Max x Max y Max z Number
case
models contained in softwares ANSYS FLUENT. The size(m) size(m) size(m) of Grids
different pool diameters are used to show that the Grid-1 0.02 0.02 0.03 182909
experimentally found dependence of various parameters such
Grid-2 0.02 0.02 0.015 210195
as temperature and thermal radiation on pool diameter in the
calculations realistically reproduced the experimental data. Grid-3 0.01 0.01 0.015 315490
The main purpose of CFD simulation was to determine the
temperature T, Surface Emissive Power (SEP) and irradiance
E. In the present CFD simulations, turbulence modeled using
RSM turbulence model with a buoyancy correction term and
Large Eddy Simulation (LES). Due to the absence of detail of
the combustion reaction of different fuels studied in this work,
a single step reaction and eddy-dissipation model used for
combustion modeling. Soot formation is modeled using one-
step Khan and Greeves model and the Discrete Ordinate (DO)
radiation model is applied for thermal radiation modeling
which is the most detailed radiation model applied in Ansys-
FLUENT.

4.2 Geometry,mesh and boundary Fig 4

condition The fire domain initially contains air under ambient
In this work three-dimensional (3-D) simulation of pool fires conditions: T = 298 K and P = 1.013 bar. The definition of
has been carried out. The domain is modeled as a cylindrical simplified geometry leads to a significant reduction of the
shape surrounded the pool of liquid fuel. The cylindrical time. As the fluid in domain an air is assumed to consist of
domain discretized to the hexahedral non uniform structured nitrogen, oxygen and carbon dioxide. The existence of the
control volumes. The details of the mesh and boundary types species carbon dioxide and water at the beginning of the
are given in “Figure 3”. The mesh is very refined at the pool reaction when eddy dissipation model is used is essential. The
surface and in the inner part of domain with increasing cell mass fractions of individual species in temperature in domain
dimension as moving to the side boundaries. The number of is T = 298 K and the pressure is defined relative to the
the meshes varied according to the size of the pool. The reference pressure P = 1.013 bar. In the inlet boundary the fuel
meshes and geometries which are used in the present is assumed to be already evaporated and the fuel vapor
simulations are created using Ansys-ICEM software coming from the inlet boundary surface with defined constant
temperature and a constant mass flow rate of fuel equal to the
mass burning rate, experimentally determined for different

10
 International Journal of Computer Applications (0975 – 8887)
Volume 70– No.11, May 2013

fuels. The inlet is surrounded with a low rim (“Figure 3”) and 5.2 Flame height
adiabatic ground area. The remaining areas in the The experimental visible flame height of different flammable
computational domain are open boundaries with defined
materials as a function of pool flame diameter is depicted in
ambient conditions. At the time t = 0 s, the flame is ignited in
the experiment, or begins in the computational modeling. “Figure 6”. As shown the visible flame height of each
From that time initially grows in the size until sometime after flammable material is increased with an increase in pool
its full size is reached. In the first test calculations have shown diameter which is due to the higher mass burning rates that
that the flames reach their full size after about t = 5 s. In extends flame visibility to a larger length. Also as can be seen
simulations the constant mass burning rate is used. For the the flame of TBPB extends about 2 to 4 times of the flame of
pool fires simulations time steps in a range of 0.0001 s ≤ Δt ≤ kerosene and gasoline which is due to the 10 to 30 times
0.001 s are chosen to reach achieved convergence level of
larger burning rates of TBPB in comparison to the kerosene
minimum 103 independently on pool diameter but depending
and gasoline as stated before. A schematic of the flame of
on the sub-models. Number of iterations per time step varied
from 10 to 40 depending on chosen sub-models, where at the each material is shown in “Figure 7”.
beginning of simulation a large number of 20 – 40 iterations
per time step is used and after the convergence is reached the
number of iterations decreased to minimum 10-15.

5. RESULT AND DISCUSSION

5.1 Flammable Material burning rate
The dependence of mass burning rate of flammable material
as a function of pool diameter for different materials is shown
in “Figure 5”. The mass burning rate of each of the materials
is increased with an increase in diameter. The heat flux from
the flame to the liquid surface increases with an increase in
liquid surface area which results in an increase in flammable
liquid evaporation rate in larger diameters.

Fig 6

Fig 5: Experimental measured mass burning rates for

gasoline, kerosene and TBPB
Also as can be seen organic peroxide TBPB burns almost 10
to 30 times of the common hydrocarbon fuels kerosene and
gasoline. This is due to the basis of chemical structure of the
fuel and the higher flame temperature of TBPB in comparison
to the kerosene and gasoline as shown in “Table 3”. which
results in larger radiation heat flux to the liquid surface based
on Stefan Boltzmann radiation heat law that describes the
radiation heat flux as a 4.th power of the temperature.
Fig 7
Table 3
Experimental measured maximum flame temperature in
5.3 Flame Temperature
The CFD predicted instantaneous iso-volumes of flame
different diameters.
temperature distribution for the simulations using RSM and
Diameter (m) 0.06 0.11 0.18 LES turbulence models for a pool fire of 0.18 m diameter of
Temperature (K) different flammable materials is shown in “Figure 8”. As can
Kerosene 1110 1331 1321 be seen in all of the simulations the three different zones of a
Gasoline 990 1181 1250 fire is well captured. But as can be seen the high turbulence
TBPB 1345 1569 1603 and intermittency in the plume zone is not well captured using
RSM model while the turbulence in this zone is well captured
using high physics LES turbulence model.

11
 International Journal of Computer Applications (0975 – 8887)
Volume 70– No.11, May 2013

Fig 9
Fig 8.1:gasoline The experimental measured and CFD predicted time averaged
axial flame temperature profiles for pool fire of each
flammable material in different pool diameters is shown in
“Figure 10”. As can be seen in all the cases the trend of the
experimental data is well captured using simulations although
there is some quantitative discrepancy between CFD and
experiments specially in the reaction zone which could be due
to the low physics chemistry kinetic model which models the
combustion reaction as a one-step reaction using a first order
eddy-dissipation reaction model. Also as shown in “Figure
11” the simulations using the LES turbulence model for the
0.18m pool diameter is in much better agreement than the
RSM turbulence model in “Figure 10.3” which is due to its
better performance in modeling turbulence field that has an
Fig8.2:kerosene important impact on the combustion model predictions and
flame flow field.

Fig 10.1:pool diameter 0.06m ,RSM

Fig 8.3:TBPB
The instantaneous flame temperature is monitored in some
representative point in the simulation domain for each of the
burning materials as indications of flow development and for
simulation using LES turbulence model of a 0.18m pool
diameter of gasoline is shown in “Figure 9”. As can be seen
generally it takes about 5-6 seconds for each simulation to
reach the quasi-steady state depending on the burning material
and pool diameter. Once the fully developed, quasi-steady
state is reached, the time-averaged quantities are calculated.
To ensure the convergence of the averaged quantities the
averaging processes are performed for 25 seconds.
Fig 10.2:pool diameter 0.11m, RSM

12
 International Journal of Computer Applications (0975 – 8887)
Volume 70– No.11, May 2013

Fig 12.3:TBPB

Fig 10.3:pool diameter 0.18m, RSM 5.4 Flame irradiation

The flame irradiance E can be calculated by surface emissive
power (SEP) of the flame using the following equations:

E  F ,R  E FSEP
(1)

E FSEP   F (T F4 T amb
4
) (2)

where the τ is the atmospheric transmissivity, F , R is the
view factor from flame (F) to receiver (R) and is calculated
using following equation [9]:

Fig 11:pool diameter 0.18m, LES 1  B  C  C  B 

F , R   arctan   arctan   
2  1  B 2
 1 B  1C
2 2
 1C
2
  (3)
The CFD predicted flame development of 0.18 m diameter
pool fire for different flammable materials which is studied in
this work is monitored using iso-volumes of flame where B and C are defined as the ratio of flame width to
temperature in different solution times using LES turbulence distance between flame and receiver (∆y) and flame height
model in “Figure 12”. As can be seen the flame development (H) to distance between flame and receiver (∆y) respectively
of a typical peroxide (TBPB) due to its different chemical for approximation of flame as a rectangular surface [8,9]. σ is
the Stefan Boltzmann constant ( 5.6710 W / (m K ) ).
properties is about two times faster than the common 8 2 4

hydrocarbon fuels gasoline and kerosene. This fact is

important specially in the case of accidents and requires faster
is the flame emissivity which is calculated in this work
operations to controlling the fire.
using following procedure.Babrauskas (1983) [10] applied
conservation of energy for the liquid surface as given below:

m f" hg  q r"  qc"  q rr"  q loss

"
(4)where is the
mass loss rate per unit area (assumed identical to the burning

rate), is the total heat of gasification, is the radiant

flux absorbed by the pool, is the heat received

convectively, is the re-radiant heat loss due to the

surface of the pool being at an elevated temperature and

Fig 12.1:gasoline loss is the lumped wall conduction losses and unsteady

terms. Lumped wall conduction losses ( ) and re-radiant

heat losses ( ) are usually small and hence neglected. Thus

the “(4)” can be simplified as :

m f" hg  q r"  qc" (5)

Fig 12.2:kerosene

13
 International Journal of Computer Applications (0975 – 8887)
Volume 70– No.11, May 2013

Fay [1] calculated the convective heat flux from flame to

liquid surface for circular pool fires as following relation:

  H c 
qc"  1.30 103  air gd    (6)
  1 f 
substitution of the radiation heat flux from Stefan-Boltzmann
and convective heat flux from “(6)” into “(5)” and some
mathematical manipulations results in following equation for
flame emissivity :

  H c  
m f" h  1.30 103  air gd  
  1  f   (7)
F 
T F4
Fig 13.2:kerosene
The flame irradiances E is computed using relation (1) with
both experimental and CFD predicted maximum time
averaged flame temperature in different horizontal distance ∆y
from pool rim. The results plotted as a function of non
dimensional (∆y/d) distance from fire for each flammable
material studied in this work and is shown in “Figure 13”. As
can be seen from the figures in all of the cases the CFD
predicted results are in good agreement with experimental
measured results. If one follows the criterion described in [11]
i.e. NFPA (also given in “Figure 13”) for designing the tanks
and process equipment’s is 5 kW/m^2 based on the
phenomenon of skin burn caused by radiant heat exposure
whereas EN 1473 recommends this limit to be 1.5 kW/m^2
[11,12].This safety distances is calculated for both
experimental measured and CFD predicted maximum time
averaged flame temperature of each flammable material
studied in this work in the case of pool fire of 0.18 m diameter
and the results is reported in “Table 4” and “Table 5”. Thus as Fig 13.3:TBPB
shown this safety distance can be predicted with CFD with
good accuracy. Table 4
Experimental and CFD predicted safety distance
corresponding to NFPA standard
gasoline kerosene TBPB

Experimental(m) 0.79 0.95 2.07

CFD (m) 0.6401 0.86 1.71
Error (%) 18.9793 9.4733 17.3913

Table 5
Experimental and CFD predicted safety distance
corresponding to EN 1473 standard.
gasoline kerosene TBPB

Experimental(m) 1.73 1.9296 4.1706

CFD (m) 1.48 1.7694 3.5694
Error (%) 14.4522 8.3022 14.4152
Fig 13.1:gasoline

14
 International Journal of Computer Applications (0975 – 8887)
Volume 70– No.11, May 2013

6. CONCLUSION [2] Thomas PH, Baldwin R, Heselden AJM. “Buoyant

In order to characterize and measure safety distances to diffusion flames: some measurements of air entrainment,
protect personnel and equipment against the accidental heat transfer, and flame merging.” 10th Symposium on
encountered pool fires, a series of experiments and CFD Combustion, Cambridge, UK, 1965, pp. 983–996.
simulations using different materials and different pool [3] McCaffrey B. “Purely buoyant diffusion flames—some
diameters is performed. To approach this aim pool fire of two experimental results.” NBSIR 79-1910, National Bureau
hydrocarbon fuels (i.e. gasoline and kerosene) and a peroxide of Standards, Washington, 1979
(TBPB (tert-butyl peroxybenzoate)) in different pool
[4] Sinai YL, Owens MP. “Validation of CFD modeling of
diameters (i.e. 0.06, 0.11, 0.18 m pan diameter) is studied.
unconfined pool fires with crosswind: flame geometry.”
The results are as follows: Fire Safety J 1995;24:1–34.
1- Experiments revealed that the pool fires of liquid organic
peroxides show fundamentally very different characteristics [5] B. J. McCaffrey; C. L. Beyler; G. Heskestad: “SFPE
e.g. generally much higher mass burning rates, large flame Handbook of Fire Protection Engineering”, Third
lengths as well as high temperatures and subsequently higher Edition, 2002, ISBN: 087765-451-4.
irradiances in comparison to liquid hydrocarbon pool fires. [6] T. Steinhaus; S. Welch; R. O. Carvel; J. L. Torero:
The higher temperature flame of TBPB in comparison to the “Large-scale pool fires”, Journal of Thermal Science
gasoline and kerosene results in higher heat flux on to the (2007)101–116.
liquid surface which results in higher vaporization rate of
liquid pool and larger pool flames. [7] P. K. Raj: “Large LNG fire thermal radiation”, AIChE
2- The three well-known flame zones of pool fires is well Spring Meeting April 13(2005)219-232.
captured by CFD simulations and the predicted axial flame [8] P. Joulain: “Behaviour of pool fires: state of the Art and
temperature profiles are quantitatively close to the measured new insights, Proceedings”, 27th Int. Combustion
values. Symposium, (1999)2691–2706.
3- The CFD predictions using LES turbulence model in
[9] A. Schonbucher; B. Arnold; K. Banhardt; V. Bieller; H.
comparison to the RSM turbulence was in better agreement
Kasper; M. Kaufman; R. Lucas; N. Scheiß: , 21
with experiments specially in the plume zone which is due to
Symp.(Intl.) on Combust., The Combustion Institute
the higher physical accuracy of LES model and it’s capability
1986.
to model dispersion of large eddies in plume zone.
4- The CFD simulations showed that the pool flame of TBPB [10] Babrauskas, V., 1983. “Estimating large pool fireburning
about two times faster develops in comparison to common rates”. FireTechnology 19, 251–261.
hydrocarbon fuels gasoline and kerosene. [11] P. K. Raj: “A review of the criteria for people exposure
5- The safety distances accurately predicted by CFD to radiant heat fluxes fromfires”, Journal of
simulations when predicted time averaged maximum flame HazardousMaterials 159(2008)61-71.
temperature is used instead to the experimentally measured
values in calculations performed in this work. [12] EN 1473, European Standard on the Installation and
Equipment for Liquefied Natural GasDesign of On-shore
7. ACKNOWLEDGMENTS Installations, prEN 1473:2005(E), Standard Version
This project is supported by consultancy funding generated in 2.1c2, Prepared by the TechnicalCommittee CEN/TC
the research account under Areas of Strength in Fire Safety 282, AFNOR Secretariat, Brussels. See Installation and
Engineering of the Petroleum University of Technology. equipment for LNG, Table A.2, p 83, The Secretariat of
whichisheld by AFNOR, 1 September 2005.
8. REFERENCES
[1] J.A. Fay, “Model of large pool fires”, J. Hazard. Mater.
136 (2006) 219–232.

15