Noname manuscript No.

(will be inserted by the editor)

Large eddy simulations of CH4 fire plumes

G. Maragkos · B. Merci

Received: date / Accepted: date

Abstract Large eddy simulations of large-scale CH4 fire plumes (1.59-2.61 MW)
with two different CFD packages, FireFOAM and FDS, are presented. It is in-
vestigated how the vorticity generation mechanism and puffing behavior of large-
scale fire plumes differs from previously studied iso-thermal buoyant plumes of the
same scale. In addition, the predictive capabilities of the turbulence and combus-
tion models, currently used by the two CFD codes, to accurately capture the fire
dynamics and the buoyancy-generated turbulence associated with large-scale fire
plumes are evaluated. Results obtained with the two CFD codes, typically used for
numerical simulations of fire safety applications, are also compared with respect
to the average and rms velocities and temperatures, puffing frequencies, average
flame heights and entrainment rates using experimental data and well-known cor-
relations in literature. Furthermore, the importance of the applied reaction time
scale model in combination with the Eddy Dissipation Model is examined. In
particular, the influence of the considered mixing time scales in the predicted cen-
terline temperatures is illustrated and used to explain the discrepancies between
the two codes.
Keywords LES; fire plume; FireFOAM; FDS

1 Introduction

The study of fire plumes has received a lot of attention due to the great environ-
mental impact and dangers often associated with them. More specifically, large-
scale fire plumes, often encountered after natural fires and industrial accidents, can
lead not only to severe air pollution due to the release of unburned hydrocarbons
and soot into the atmosphere but can also trigger more severe cascading effects

G. Maragkos and B. Merci

Department of Flow, Heat and Combustion Mechanics, Ghent University,
St. Pietersnieuwstraat 41, B-9000 Ghent, Belgium
Tel.: +32 9264 32 90, Fax: +32 9264 35 75
E-mail: Georgios.Maragkos@UGent.be
2 G. Maragkos, B. Merci

involving explosions, causing huge losses of human lives and material properties.
The use of Computational Fire Dynamics (CFD) has, nowadays, become an inte-
gral part of fire protection design of real life applications. Detailed knowledge of
the behavior and the dynamics of large-scale fire plumes is, therefore, essential for
fire safety engineers working on everyday life applications involving; fire detection,
fire suppression and venting, design of smoke control systems, fire heating of struc-
tural elements of buildings or thermal radiation hazards [1, 2]. CFD models can
not only be used to evaluate the effectiveness of current or future fire protection
systems but also to answer ‘what if’ questions and be used towards a cost-effective
design without compromising fire safety in buildings. A deeper understanding of
the capabilities and limitations of the available CFD codes available for performing
numerical simulations of fire related applications will provide the fire safety engi-
neers with the appropriate guidelines to carefully design and perform their CFD
analysis when evaluating fire safety systems in buildings and industrial facilities.
The unsteady behavior and flow evolution of fire plumes of different scales,
even though extensively studied in the past [3–6], still remains challenging due
to the coupled physical processes involved, e.g., buoyancy-generated turbulence,
combustion, thermal radiation and soot generation. Three regions are typically
distinguished in fire plumes [7]: the flame region, close to the fire source, where
continuous flames exist; the intermittent region where the flames strongly fluc-
tuate exhibiting a characteristic puffing behavior; and the plume region, where
combustion products are convected downstream and mixed with fresh air. What
is typically encountered in large-scale fire plumes is a rapid transition of the flow,
from laminar to fully turbulent, usually within a few inlet diameters from the fire
source. Turbulence is greatly enhanced by strong buoyancy forces which result in
a periodic shedding of large toroidal vortices (puffing behavior) near the base of
the fire source. This puffing behavior is a key element in the fire dynamics of such
plumes since it controls the air entrainment towards the reaction zone that will in
turn influence the heat release rate, the composition of the combustion products
and the soot generation.
The specific motivation of the present numerical study is three-fold: first, to
study the dynamics and oscillatory behavior of large-scale fire plumes and examine
how the vorticity generation mechanism and puffing frequency of such plumes differ
from previously studied iso-thermal buoyant plumes [8, 9] of the same size; second,
to evaluate the predictive capabilities of the turbulence and combustion models,
currently used in the fire community, of accurately capturing the fire dynamics and
the buoyancy-generated turbulence associated with large-scale fire plumes; third,
to apply two well-known CFD codes, typically used for numerical simulations of fire
safety applications, and evaluate the simulation results of transient and mean flow
dynamics of large-scale fire plumes against experimental data. The comparison of
the two CFD packages is expected to be useful for fire safety engineers working on
real-life applications of fire scenarios. As such, Large Eddy Simulations (LES) of 1
m in diameter CH4 fire plumes are conducted with FireFOAM (version 2.2.x) and
the Fire Dynamics Simulator (FDS) (version 6.1.2) and the numerical results are
compared against experimental data and well-known correlations in literature.
It is worth noting, however, that it is not the authors’ intention to directly
compare the two CFD codes at hand, e.g., use identical numerical set-ups and
modeling choices in the simulations, since such an attempt would not be possible
due to limitations and differences of available models in each software. Given the
Large eddy simulations of CH4 fire plumes 3

small differences in the modeling options employed in the two codes, this study
aims at illustrating the potential of each software separately by comparing the nu-
merical results with experimental data and correlations available in literature using
the best available model options in each code. The work presented in the paper
aims to focus on the strengths of each code as they stand but also to report their
weaknesses so as to improve their models and, hence, their predictive capabilities.
The present study is a significant extension of the previously conducted numerical
study by Maragkos et al. [10], incorporating improvements in the selected physical
models and a more detailed analysis. Within this mindset, the present paper also
aims at expanding the work on small-scale CH4 fire plumes, previously performed
by Wang et al. [11] with FireFOAM (version 1.6) to large-scale CH4 fire plumes.

2 Experimental case

Comparisons are made to the experiments performed in the Fire Laboratory for
Accreditation of Modeling by Experiment (FLAME) facility at Sandia National
Laboratories in Albuquerque, New Mexico, reported by Tieszen et al. [12, 13]. The
facility is a 6.1 m cubical enclosure with a 2.4 chimney located on the top of the
chamber. The fire source, 1 m in diameter, is placed in the middle of the facil-
ity, positioned 2.45 m above the floor and is surrounded by a 0.51 m wide steel
floor (ground plane). Particle Image Velocimetry (PIV) was used for velocity field
measurements of CH4 fire plumes with heat release rates ranging from 1.59 to
2.61 MW. However, no temperature measurements were performed. The experi-
mental uncertainty for the measured quantities was in the order of 20% and 30%
for the mean and turbulent statistics, respectively. It is worth noting that the re-
ported experimental uncertainty is relatively high in this case. Fire experiments, as
opposed to combustion experiments, typically involve much larger domains (mak-
ing e.g. ambient conditions to be less well-controlled), are of larger scale (e.g.
larger diameters of fuel inlet), often have less well-defined fuel source conditions
(e.g. turbulence intensity), the presence of radiation and soot generation can be
significant (e.g. particularly for hydrocarbon fuels) and difficulties in the exact
measuring of certain quantities are present (e.g. surface or in-depth temperatures
in case of flame spread, etc). All these factors combined contribute in typically
having bigger experimental uncertainties when it comes to fire experiments as op-
posed to combustion experiments. One of the the main reasons for considering
this test case in our work is the fact that it is part of the MaCFP Working Group
(http://www.iafss.org/macfp/) which emphasizes on making systematic progress
in fire modelling based on fundamental understanding of fire phenomena [14, 15].
In the absence of a better large-scale test case with smaller experimental uncer-
tainties these experiments were deemed sufficient for consideration by the authors.
An overview of the initial and boundary conditions of the different experimental
tests considered in this study is presented in Table 1.
This set of large-scale fire plume tests performed by Tieszen et al. [12, 13] has
provided an experimental database useful for validation of CFD codes. Several
numerical studies using different modeling approaches have appeared in literature
in the past focusing on these particular experiments. Ferraris et al. [16] performed
large eddy simulations in order to investigate the possibility of employing the Con-
ditional Source Estimation (CSE) as a model of low computational cost for large
4 G. Maragkos, B. Merci

Table 1 Initial and boundary conditions of the experiments [12, 13].

00
Test Fuel D ṁfuel Q̇c Tfuel T∞ P∞
(m) (kg/s) (MW) (K) (K) (kPa)
#14 CH4 1.0 0.040 1.59 284 285 80.6
#24 CH4 1.0 0.053 2.07 286 290 81.0
#17 CH4 1.0 0.066 2.61 274 278 81.1

fire simulations. Black et al. [17] performed RANS simulations with two differ-
ent turbulence treatments, a steady RANS solution with a model for buoyancy-
generated turbulence, and an unsteady solution with closure models based on a
temporal filter width. DesJardin et al. [18] used LES to apply flamelet modeling
combined with an alternative closure for the conditional dissipation rate, based on
a transport equation for the mixture fraction filtered probability density function.
Xin et al. [19] performed LES in order to validate an earlier version of the Fire
Dynamics Simulator (FDS) while Pasdarshahri et al. [20] applied LES with Open-
FOAM using the one-equation turbulence model. Hu et al. [21], also within the
LES context, used detailed chemistry based on the the laminar flamelet approach
and investigated the influence of chemical kinetics on the vortical structures of
large-scale fires. Finally, these large-scale fire plume experiments have also been
considered by the FDS developers for validation purposes of their code with nu-
merical results reported in the software’s validation guide [22].

3 Governing equations

FireFOAM, developed by FM Global, is a C++, object-oriented, LES code based

on OpenFOAM. It is a second order accurate, finite volume code with implicit time
integration and unstructured mesh capabilities [23]. FDS, developed by NIST, is a
low-Mach number, LES code written in Fortran. It is a second order accurate, finite
difference code with explicit time marching that uses structured grids. Given the
fact that both FireFOAM and FDS are very comprehensive codes, all the technical
details related to the numerical methods and solution algorithms employed in each
CFD package cannot be presented here. The reader is directed to [24, 25] for more
details on FireFOAM and in [26] for FDS. Both codes solve conservation equations
for mass, momentum, chemical species and sensible enthalpy using Favre-filtered
quantities (e.g. φe = ρφ/ρ):
∂ρ
+ ∇ · (ρe
u) = 0 (1)
∂t
  
∂(ρeu) 2
+ ∇ · (ρeuue) = −∇p + ∇ · µef f ∇e u + (∇eu)T − (∇ · u e)I + ρg (2)
∂t 3
   
∂(ρYek ) νsgs 000
+ ∇ · (ρeuYek ) = ∇ · ρ Dk + ∇Yek + ω̇k , (k = 1, ..., Ns − 1) (3)
∂t Sct
   
∂(ρehs ) Dp νsgs
+ ∇ · (ρeuehs ) = + ∇ · ρ Dth + ∇e hs − ∇ · q̇r00 + q̇c000 (4)
∂t Dt P rt
where ρ is the density, u
e is the velocity, p is the pressure, µef f = µ + µsgs is the
effective dynamic viscosity, µ is the laminar viscosity, µsgs = ρνsgs is the sub-grid
Large eddy simulations of CH4 fire plumes 5

scale viscosity, I is the identity tensor, g is the gravitational acceleration, Yek is

the species mass fraction, Dk is the species mass diffusivity, Sct is the turbulent
000
Schmidt number, ω̇k is the species reaction rate, e hs is the sensible energy, Dth is
the thermal diffusivity, P rt is the turbulent Prandtl number, q̇r00 is the radiative
000
heat flux, q̇c000 = ∆Hc ω̇F is the heat release rate per unit volume due to combustion
(assuming complete combustion, i.e. χ = 1) and ∆Hc is the heat of combustion of
the fuel.
Both codes have a variety of turbulence and combustion models available for
use. In the current study, the well-known constant coefficient Smagorinsky model
[27] is used to model turbulence in FireFOAM (since there is no dynamic formu-
lation of the model available yet) while its extension, the dynamic Smagorinsky,
proposed by Lilly [28] with the model constant calculated by a dynamic procedure,
is used in FDS. The sub-grid scale dynamic viscosity is then calculated as:

µsgs = ρ(cs ∆)2 |S|

e (5)

where cs is a model parameter, ∆ is the LES filter size (e.g. cubic root of the
cell volume) and Se = 21 ∇e uT is the strain rate tensor. In FireFOAM, the


u + ∇e
model parameter cs is assigned a constant positive value while in FDS it can vary
locally in both time and space. For stability reasons in FDS, negative cs values are
clipped to zero (cs ≥ 0), hence, back-scattering effects are not considered (energy
transfer from the small to the large scales). For the dynamic implementation of
the Smagorinsky model, a test filter of twice the grid size is employed in FDS. As
part of a sensitivity study, the modified Deardorff model [29] is also applied in the
numerical simulations with FDS since it is the standard turbulence model of the
code. Within this model, the sub-grid scale dynamic viscosity is calculated as:
p
µsgs = ρcv ∆ ksgs (6)

where cv = 0.1 [30] is a model constant and ksgs is the sub-grid scale kinetic
energy. The sub-grid scale kinetic energy is calculated by the same model (the
only available option in the software) in both the dynamic Smagorinsky and the
modified Deardorff turbulence models in FDS (see Eq. (10) presented below).
Unless stated otherwise, results with the dynamic Smagorinsky are presented by
default in the paper. Some results with the modified Deardorff model are also
presented in the paper for comparative purposes only.
The combustion model applied in both codes employs a one-step, infinitely
fast, irreversible chemical reaction for CH4 :

CH4 + 2(O2 + 3.76N2 ) → CO2 + 2H2 O + 7.52N2 (7)

combined with a modified Eddy Dissipation Model (EDM) model [31] to deal with
turbulence-chemistry interactions. Within the EDM model the fuel mass reaction
rate is calculated as:
000 min(YeF , YeO2 /s)
ω̇F = ρ (8)
τmix
where YeF and YeO2 are the fuel and oxygen mass fractions, respectively and s is
000
the stoichiometric oxygen-to-fuel mass ratio. The ω̇k for the others species are
obtained through simple stoichiometric relations. The simplified consideration of
6 G. Maragkos, B. Merci

infinitely fast chemistry is a typical assumption made in CFD codes for fire simu-
lations given the fact that, in most cases, it is the prediction of temperature and
main species concentrations that are of relevance in fire safety and not so much the
intermediate species, which would require the use of detailed chemistry. Within
this framework, no chemical kinetics effects are considered within the combustion
model, rather, chemistry is simplified into the calculation of mixing time scales
under different conditions between the fuel and oxidizer. The main difference be-
tween the combustion model in the two codes lies in the calculation of the mixing
time scale, τmix . In FireFOAM, the turbulent (ksgs /sgs ) and molecular (∆2 /α)
diffusion time scales are considered when calculating the fuel mass reaction rate,
proving this way an estimate of the fuel-air mixing, under turbulent and laminar
flow conditions, respectively. On the other hand, a more detailed reaction time
scale model is used in FDS in which the chemical time scale is compared with the
mixing times for diffusion p (∆/DF , where DF is the fuel mass p diffusivity), sub-
grid scale advection (∆/ 2ksgs ) and buoyant acceleration ( (2∆/g) [32]. This
reaction time scale model proposes a scaling regime for coarse mesh resolution
based on buoyant acceleration. As explained in [32], for fires which are generally
buoyancy-driven flows, buoyant acceleration is expected to control the mixing at
relatively coarse scales. Hence, a time scale based on a constant acceleration that
scales with the square root of the filter width is proposed. Both τchem and τf lame
are effectively not used in the current simulations but pose the extreme limits that
τmix can take in the reaction time scale model.
The estimation of the sub-grid scale kinetic energy, ksgs , needed for calculating
the mixing scale scale, τmix , in the EDM combustion model is done differently in
the two codes. In FireFOAM, different ways of calculating ksgs exist depending
on the applied turbulence model. Within the Smagorinsky model, the assumption
of local equilibrium is used and ksgs is obtained from a balance equation as:
3/2
ce ρksgs
ρ(Se : B) + =0 (9)
∆

where ‘:’ denotes the double inner product, B = 32 ksgs I − 2ck ksgs ∆ SeD is the
p

sub-grid stress tensor, SeD is the deviatoric component of the strain rate tensor, and
ck = 0.05, ce = 1.048 [33] are two model constants. The sub-grid scale dissipation
3/2
rate, sgs , is then approximated as sgs = ce ksgs ∆−1 . On the other hand, in FDS
ksgs is approximated based on scale similarity arguments as:

1
(u − û)2 + (v − v̂)2 + (w − ŵ)2

ksgs = (10)
2
where u, v, w are the average values of u, v, w in the cell centers and û, v̂, ŵ are
the weighted averages of u, v, w in the adjacent cells.
A commonly used radiation model is adopted in this study in which the ra-
diative intensity is treated as a function of both spatial location and angular
direction and is obtained by solving the radiative transfer equation (RTE) by the
finite volume discrete ordinates model (fvDOM) accounting for attenuation and
augmentation of radiation by absorption and emission, respectively:

∂ Ie
eIeb − κ
=κ eIe (11)
∂s
Large eddy simulations of CH4 fire plumes 7

where I is the radiation intensity, s is the solid angle, Ib is the black body radia-
tion intensity and κ is the absorption coefficient. Under the assumption of a grey
gas (κP = κ) and considering an optically thin flame, neglecting scattering, the
radiative heat flux is then calculated as:
 Z 
∇ · q̇r00 = κ
eP 4π Ieb − IdΩ
e =κep (4σ Te4 − G)
e (12)
4π

where κP is the mean Planck absorption coefficient, σ is the Stefan-Boltzmann

constant, T is the (filtered) temperature and G is the total irradiance. The mean
Planck absorption coefficients for the chemical species are temperature-dependent
and calculated by the RADCAL model [34]. It is worth mentioning that contri-
butions from turbulence-chemistry interaction (TRI) are neglected in Eq. (12).
However, in FDS the source term of the RTE in the flame region is calculated as:
ep (4σC Te4 − G)
∇ · q̇r00 = κ e (13)
where C is a constant, computed at each time step, which corrects the emission
term in order to compensate for both errors in emissivity calculations and in TRI.
A radiative fraction of χr = 0.2 [35] is used in the present study, which represents
the global fraction (lower limit) of energy that is released from combustion as
thermal radiation.
No soot model was employed in either code since soot generation during CH4
combustion is not significant [36]. The selected models for the numerical simulation
with FireFOAM and FDS are outlined in Table 2 and Table 3, respectively.

Table 2 Models used in the FireFOAM simulations.

FireFOAM 2.2.x
Turbulence model Constant Smagorinsky
µsgs = ρ(cs ∆)2 |S|
e
cs = 0.1 [8]
Combustion model Eddy Dissipation Model (EDM)
000 min(Y
eF ,Y
2
eO /s)
ω̇F = ρ τmix
∆2
 
ksgs
τmix = min ,
CEDM sgs Cdif f α
| {z } | {z }
τturb τlam
CEDM = 4, Cdif f = 2 [25]
Radiation model Radiative Transfer Equation (RTE)
Finite Volume Discrete Ordinates Method (FVDOM)
48 solid angles
q̇r00 = κp 4σT 4 − G

Soot model -

4 Numerical set-up

The computational domain employed in the numerical simulations is 4 m × 6 m of

cylindrical shape for FireFOAM (Figure 1(a)) and 4 m × 6 m × 4 of rectangular
8 G. Maragkos, B. Merci

Table 3 Models used in the FDS simulations.

FDS 6.1.2
Turbulence model Dynamic Smagorinsky
µsgs = ρ(cs ∆)2 |S|
e
cs : determined dynamically
Combustion model Eddy Dissipation Model (EDM)
000 min(Y
eF ,Y
2
eO /s)
ω̇F = ρ τmix
τmix = max[τchem , min(τd , τu , τg ,q
τf lame )]
∆2 2
τd = D , τu = √ ∆ , τg = 2∆
g
F (2ksgs )

Radiation model Radiative Transfer Equation (RTE)

Finite Volume Discrete Ordinates Method (FVDOM)
100 solid angles
q̇r00 = κp 4σCT 4 − G : Inside flame zone


00
q̇r = κp 4σT 4 − G : Outside flame zone


C: correction factor
Soot model -

shape for FDS (Figure 1(b)). The 1 m in diameter fire source is modeled by
specifying the corresponding mass flow rate reported in the experiments. The
surrounding 0.51 m wide steel floor (ground plane) was modeled as adiabatic. The
bottom plane (y = 0 m), outside the steel floor, along with the sides and the outlet
of the computational domain were set as open, allowing air to be entrained into
the domain. The mesh was refined around the centerline (half an inlet diameter
outside the fire source and up to the outlet) which resulted in grid sizes of ≈ 1.5
cm in the areas of interest and 66 cells across the fire source. The grid spacing
used outside the refined region was 3 cm for FDS and varied from 3 cm (just
outside the refined region) to 6 cm (sides of the domain) in FireFOAM. The total
number of cells was then ≈2.38 million for FireFOAM and ≈8.65 million for FDS.
It is exactly this different meshing technique employed in the numerical simulations
that results in the difference in the total number of cells used with the two software.
Nevertheless, this is of secondary importance since approximately the same cell
size (≈ 1.5 cm around the centerline) is used in the main area of interest (above
the fuel inlet and on the centerline). All numerical simulations were set to run
for 50 sec with a varying time step, limited by a maximum Courant number of
0.8. Averaging was employed during the last 45 sec of the numerical simulations.
In FireFOAM, the equations are advanced in time using a second order implicit
Euler scheme. All quantities are assigned to the cell centers (collocated grid) with
velocities linearly interpolated to the cell faces. The convective terms are second
order centrally differenced. For scalar transport, a second order TVD scheme using
a Sweby limiter is applied while the diffusive terms are centrally differenced and
corrected for the non-orthogonality of the mesh. A PISO algorithm is used for
the pressure-velocity coupling with a Rhie-Chow interpolation to avoid odd-even
decoupling. The pressure equation is solved by a linear GAMG solver. In FDS,
the governing equations are advanced in time by using a second order explicit
Runge-Kutta scheme. Spatial derivatives are estimated with second-order finite
differences on a rectangular grid, with scalar quantities assigned to the cell center
and velocities assigned to the cell faces. Convective terms are upwind biased,
Large eddy simulations of CH4 fire plumes 9

based on a Courant-Friedrichs-Lewy condition, in the predictor step and downwind

biased in the corrector step. If the Courant number is near unity, this corresponds
to a nearly fully upwind scheme, while for a Courant number much less than unity,
the scheme is more centralized. This second-order finite differencing scheme of FDS
cannot, however, fully resolve sharp gradients on a relatively coarse grid. Steep
gradients can cause local over-shoots and under-shoots in quantities like density
and species mass fraction. For this reason, a flux correction scheme is performed in
both the predictor and corrector steps to ensure that the quantities stay bounded.
The second-order TVD scheme, SuperBee, is used for scalar transport and the
diffusive terms are central differenced. A fast Fourier transform based solver solves
the Poisson equation. The turbulent Schmidt and Prandtl numbers were set to 0.7.
The initial and boundary conditions used in the numerical simulations were set
according to the experiments, previously reported in Table 1. For the numerical
simulations with both codes, the corresponding mass flow rate and a fixed value for
temperature was applied at the inlet. In FireFOAM, a mixed boundary condition
is assigned for velocity at the sides of the domain, setting zero gradient for any
outward flow and calculating the inlet velocity from pressure. A Dirichlet boundary
condition is assigned for pressure at the sides of the domain, which fixes total
pressure and when velocity changes, the pressure is adjusted accordingly. In FDS,
all boundaries of the computational domains were considered as ‘open’ (natural
ventilation) allowing for flow into or out of the computational domain depending
on the local pressure gradient. An overview of the numerical set-up used for the
numerical simulations with the two codes is presented in Table 4.

(a) (b)

Fig. 1 Computational mesh used in the simulations with (a) FireFOAM and (b) FDS.
10 G. Maragkos, B. Merci

Table 4 Numerical set-up used in the simulations.

FireFOAM 2.2.x FDS 6.1.2

Domain 4m×6m 4m×6m×4m
(cylindrical) (rectangular)
Number of cells ≈ 2.38 million ≈ 8.65 million
Grid type Non-uniform Non-uniform
Grid size (centerline) ≈ 1.5 cm 1.5 cm
Cells across inlet 66 66
Run time 50 sec 50 sec
(45 sec averaging) (45 sec averaging)
Time step Varying Varying
(max Co = 0.8) (max Co = 0.8)

5 Results

5.1 Instability generation

A key element for accurately predicting the instability generation at the base of
the source and the resulting puffing frequencies of buoyant helium and fire plumes
is having good mesh resolution in the near-field region of the plumes. In both
cases there are vastly different scales involved: the small scales, related to mix-
ing, and the large scales, related to the characteristic puffing motion. For all the
test cases considered in this study, the conditions at the source were laminar (in-
let velocities of only a few cm/s), but very quickly buoyancy-generated turbulence
makes the flow turbulent (within an inlet diameter distance). The triggering mech-
anism [8, 9, 18] is the generation of instabilities at the edges of the source due to
baroclinic and gravitational torques. If coarse grids are used in the numerical simu-
lations then small-scale mixing is not well captured and temperatures are typically
under-predicted, resulting in smaller density gradients. As a consequence, such in-
stabilities are significantly smoothened out at the base of the source, resulting in
no clear puffing frequency. It was found that grid sizes of 1.5 cm are sufficiently
small to accurately predict the instability generation at the base of the source
and to obtain a correct puffing frequency. This grid size can be expressed in a
dimensionless way if we define a Plume Resolution Index (PRI) [22], expressing
the number of grid cells of length ∆x (m) that span the characteristic diameter of
the fire, D∗ (m), defined as:
D∗
P RI = (14)
∆x
with the characteristic fire diameter expressed as [37]:
 2/5
Q̇
D∗ = √ (15)
ρ∞ cp T∞ g

where Q̇ (kW) is the heat release rate, ρ∞ (kg/m3 ) is the ambient density, cp
(kJ/kg·K) is the heat capacity, T∞ (K) is the ambient temperature, g (m/s2 ) is
the gravitational acceleration while ∆x (m) is the width of the grid cells at the
Large eddy simulations of CH4 fire plumes 11

fire source. The higher the P RI values the more ‘resolved’ is the fire dynamics in
the numerical simulations. Nevertheless, P RI should only be used as an indicator
of the quality of the mesh of a given numerical study and it is not intended as a
replacement of a grid sensitivity study. Such a grid sensitivity study is presented at
the end of the paper for completeness. In literature, P RI values ranging between
5-15 have proven to give satisfactory accuracy with an acceptable computational
time for many fire scenarios [22], values up to 16 were used when simulating fire
plume scenarios [38] while values up to 40 were reported to be sufficient for a 1-m
in diameter turbulent buoyant helium plume [39]. In the present study, the P RI
values were approximately between 84 (for Test #14)-102 (Test #17) which are
sufficiently large enough to adequately capture the flow evolution and unsteady
behavior of these fire plumes. Similar P RI values have also been used by the FDS
developers in their validation study of the same test case [22].
The puffing cycle of a fire plume, shown in Figure 2, is similar to the one of
a buoyant plume and can be broken down into four distinct stages: the genera-
tion of the instability near the edge of the source due to baroclinic and gravita-
tional torques (Figure 2(a)), the growth of the instability towards the center of
the source (Figure 2(b)), the triggering of the Rayleigh-Taylor instability which
generates large toroidal vortices (Figure 2(c)) and the convection of these vortices
downstream until they break down due to generated secondary instabilities and
Kelvin-Helmholtz instabilities (Figure 2(d)). These stages are then repeated in
every puffing cycle [8, 9, 18]. The low inlet velocity and the strong deflection of the
plume from the axial direction at the source edges, indicate that vorticity from the
source is not responsible for the formation of turbulent structures at the plume-air
interface. Rather, they are formed by buoyancy-driven vorticity generation [8].
The vorticity generation mechanism in fire plumes differs from previously stud-
ied iso-thermal buoyant plumes [8, 9]. In the latter, there is a monotonic decrease
of density with increasing radius from the center of the source and the generated
vorticity is limited to the base of the source (near the two-fluid interface) where
the density gradients are maximum. In the former, density initially decreases with
increasing radius from the centerline until the reaction zone where it starts in-
creasing again moving towards the air side, resulting in a more localized vorticity
distribution. This effect generates a set of counter-rotating vortices on either side
of the flame zone, causing fuel and oxidizer to be brought into the reaction zone
[18]. This is illustrated in Figure 3, where the baroclinic and gravitational torques
with FireFOAM for Test #17 are presented at different heights. The baroclinic
torque is generated due to misaligned density and pressure gradients while the
gravitational torque due to misaligned density gradients and gravity. Both the
baroclinic and gravitational torques are responsible for the instability generation
at the base of the source (Figure 3(a)), however, only the influence of the baroclinic
torque remains significant with increasing height from the fire source (Figure 3(c)).
The same observations were also made in the study of a turbulent, buoyant helium
plume of the same source size [8]. At height y = 0.5 m (0.5D) the magnitude of the
inner baroclinic torque (at locations r = ±0.05 m) exhibits about half the magni-
tude of the outer baroclinic torque (at locations r = ±0.15 m) since the density
gradients from the center of the fire source to the reaction zone are significantly
smaller than the density gradients from the reaction zone to the surrounding air
(Figure 3(b)). However, on the centerline and up to the location of the maximum
flame temperature, density is decreasing with increasing height so that the den-
12 G. Maragkos, B. Merci

(a) (b)

(c) (d)

Fig. 2 Puffing cycle with FireFOAM for Test #17 at time (a) 30.0 s, (b) 30.2 s, (c) 30.4 s
and (d) 30.6 s.

sity gradients also become smaller, thus decreasing the magnitude of the inner
baroclinic torque ((Figure 3(c))).

5.2 Mean and rms quantities

According to Pope’s criterion [40], a turbulence resolution of more than 80% is

needed for accurate LES. Nevertheless, this criterion should be used cautiously
since depending on how the sub-grid scale kinetic energy, ksgs , is calculated, nu-
merical dissipation introduced by the numerical schemes used for discretization is
Large eddy simulations of CH4 fire plumes 13

1000 1,0 1000 1,0

800 0,9 800 0,9

600 600
0,8 0,8
)

)
400 400

Density (kg/m )

Density (kg/m )
-2

-2
3

3
0,7 0,7
Vorticity term (s

Vorticity term (s
200 200
0,6 0,6

0 0

0,5 0,5
-200 -200

0,4 0,4
-400 -400

0,3 0,3
-600 -600

-800 0,2 -800 Baroclinic torque Gravitational torque 0,2

Baroclinic torque Gravitational torque
Density Net Density
Net
-1000 0,1 -1000 0,1

-0,5 -0,4 -0,3 -0,2 -0,1 0,0 0,1 0,2 0,3 0,4 0,5 -0,5 -0,4 -0,3 -0,2 -0,1 0,0 0,1 0,2 0,3 0,4 0,5

Radial position (m) Radial position (m)

(a) (b)
1000 1,0

800 0,9

600
0,8
)

400
-2

0,7
Vorticity term (s

Density (kg/m )
3

200
0,6

0,5
-200

0,4
-400

0,3
-600

-800 Baroclinic torque Gravitational torque 0,2

Net Density
-1000 0,1

-0,5 -0,4 -0,3 -0,2 -0,1 0,0 0,1 0,2 0,3 0,4 0,5

Radial position (m)

(c)

Fig. 3 Baroclinic and gravitational torques with FireFOAM for Test #17 at height (a) y =
0.05 m, (b) y = 0.5 m and (c) y = 1.5 m.

not explicitly accounted for in the calculation of the turbulence resolution. This
type of criterion tends to lead to an over-estimation of the simulation quality if the
amount of numerical dissipation is increased, i.e. when the quality of the simulation
is actually lowered by using too dissipative schemes [41]. The turbulent resolution
in all numerical simulations (not shown here) remained above Pope’s suggested
value, however, the real value is likely lower due to the reasons mentioned above
and due to the tendency of the sub-grid scale models to being ”optimistic” in the
calculation of ksgs .
Simulation results of the mean and rms quantities as function of radial positions
at different heights, y, and on the centerline are presented in this section. The
symbols depict experimental data and predictions of experimental correlations
while the grey areas represent the experimental uncertainty of the measurements.
Focusing on the plume region (Table 5) of fire plumes, Heskestad’s experimental
14 G. Maragkos, B. Merci

correlations [42] give an estimate of the centerline axial velocities:

 1/3
g −1/3
v = 3.4 Q̇1/3
c (y − y0 ) (16)
c p T∞ ρ ∞
and centerline temperatures:
 1/3
T∞ −5/3
T = 9.1 Q̇2/3
c (y − y0 ) + T∞ (17)
gc2p ρ2∞

where T∞ , cp and ρ∞ are the temperature, heat capacity and density of ambient
air, respectively, and Q̇c = χr Q̇ is the convective heat release rate, with a radiative
fraction of χr = 0.2, as reported from the experimental study by Hamins et al.[35].
The virtual origin of the plume, y0 , depends on the diameter of the fire source, D,
and the total heat release rate, Q̇, and according to Heskestad’s correlation [42] it
can be estimated as:
y0 = 0.083Q̇2/5 − 1.02D (18)

Table 5 Summary of the different regions in the fire plumes [3].

Test Flame region Intermittent region Plume region

y y y
2/5 < 0.08
Q̇
0.08 < 2/5
Q̇
< 0.2
Q̇ 2/5 > 0.2

#14 y < 1.52 m 1.52 < y < 3.81 m y > 3.81 m

#24 y < 1.69 m 1.69 < y < 4.24 m y > 4.24 m
#17 y < 1.86 m 1.86 < y < 4.65 m y > 4.65 m

McCaffrey’s experimental correlations [3] can also be used to get an estimate

of the centerline axial velocities and temperatures in the plume region:
 − 1
y 3
v = 1.1 Q̇1/5 (19)
Q̇ 2/5

 2  − 5
1.1 y 3
T = T∞ + T∞ (20)
0.9(2g)1/2 Q̇2/5

5.2.1 Velocities

The predicted axial and radial velocities with FireFOAM and FDS at different
heights, y, are presented in Figures 4-9 for the test cases considered. It is observed
that axial velocities increase with higher elevations due to the flow acceleration
induced by strong buoyancy forces on the plume axis. With increasing distance
from the fire source, the velocity profiles also become wider due to the fire plume
spreading. On the contrary, the radial velocities obtain their maximum values close
to the fire source due to strong air entrainment towards the plume axis generated
by the vorticity generation mechanisms. The trends of the mean axial and radial
velocity profiles in the near-field region (y < 1.0 m) are well captured by both codes
for all cases examined and in most cases remain within experimental uncertainty.
Large eddy simulations of CH4 fire plumes 15

No significant differences are evident in the predicted results, revealing that both
CFD codes can predict both qualitatively and quantitatively the mean flow field
associated with large-scale fire plumes. Nevertheless, FDS maintains the ‘double
peak’ in the velocity profile longer than FireFOAM. This can be explained by the
higher temperatures obtained with FireFOAM (causing stronger acceleration) and
by the less turbulence observed in FDS (leading to less mixing) on the centerline,
presented later in the paper. Some discrepancies between the simulations and
the experiments are present, e.g., in the axial velocities at height y = 0.9 m for
Test #14 and in the radial velocities for the cases with higher heat release rate.
However, the possibility of errors in the experimental measurements should not
be excluded since, e.g., the reported velocities for Test #14 at height y = 0.9 m
are lower than at height y = 0.5 m (Figure 4) which is not consistent with the
velocity measurements of the rest of the tests.
The mean and rms centerline axial velocities are reported in Figure 10. The
rapid transition of the flow from laminar to turbulent, created by strong buoyancy
forces, is evident. Even though the inlet conditions in the experiments were laminar
(inlet velocities of only a few cm/s), the axial velocities reach up to 5-6 m/s within 1
diameter. It is observed that the laminar to turbulent transition is well predicted
by both codes and no significant differences are observed in the flame region.
In general, the predicted axial velocities by both FireFOAM and FDS compare
favorably with both the experimental results and the experimental correlations in
the plume region for all the test cases. Some over-prediction of the axial velocities
by FireFOAM in the plume region is, however, observed. Both codes predict the
same level of axial velocity fluctuations in all regions of the fire plumes with no
significant differences in the profiles.
By analyzing the axial evolution of the Smagorinsky constant in the FDS
simulations (Figure 11) it is observed that the constant gradually increases with
axial distance from the fire source until it reaches an almost constant value of
cs ≈ 0.13 at locations y/D > 3. This constitutes a 30% increase of the constant
in the FDS simulations with the dynamic Smagorinsky when compared to the
constant value of cs = 0.1 that was used in FireFOAM and can partially explain
the small differences in the axial velocities observed between the two codes in the
plume region. It is expected that if the model constant cs in the Smagorinsky model
in FireFOAM was determined dynamically, accounting for the local properties of
the flow, it would predict more accurately the velocities in the intermittent and
plume regions of these large-scale fire plumes.

5.2.2 Mixing time scales and reaction rates

Figure 12 presents the evolution of the centerline mixing time scales (τmix ) and
000
fuel reaction rates (ω̇F ) for Test #17 (the results for the other two cases were
in the same order of magnitude and are omitted to avoid repetition). In general,
significant differences are observed between the two codes. Higher mixing times
scales than FDS are evident with FireFOAM, very close to the inlet, at locations
y < 0.5 m. At heights y > 2.0 m the predicted mixing time scales with FireFOAM
(τmix = 0.12 s) remain approximately two times lower than FDS (τmix = 0.056 s)
at all axial locations examined. These lower τmix values in FireFOAM imply that
mixing between fuel and oxidizer requires less time (i.e. is faster) in FireFOAM
than in FDS, thus enhancing combustion and producing higher temperatures. It is
16 G. Maragkos, B. Merci

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)

Axial velocity (m/s)

Radial position (m) Radial position (m)

(a) (a)

Experiment Experiment
FireFOAM FireFOAM
FDS - Dyn. Smag.
FDS - Dyn. Smag.
FDS - Mod. Deardorff
FDS - Mod. Deardorff
Radial velocity (m/s)
Axial velocity (m/s)

Radial position (m) Radial position (m)

(b) (b)

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)
Axial velocity (m/s)

Radial position (m) Radial position (m)

(c) (c)

Fig. 4 Axial velocities for Test #14 at Fig. 5 Radial velocities for Test #14 at
height (a) y = 0.9 m, (b) y = 0.5 m and height (a) y = 0.9 m, (b) y = 0.5 m and
(c) y = 0.3 m. (c) y = 0.3 m.
Large eddy simulations of CH4 fire plumes 17

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)

Axial velocity (m/s)

Radial position (m) Radial position (m)

(a) (a)

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)
Axial velocity (m/s)

Radial position (m) Radial position (m)

(b) (b)

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)
Axial velocity (m/s)

Radial position (m) Radial position (m)

(c) (c)

Fig. 6 Axial velocities for Test #24 at Fig. 7 Radial velocities for Test #24 at
height (a) y = 0.9 m, (b) y = 0.5 m and height (a) y = 0.9 m, (b) y = 0.5 m and
(c) y = 0.3 m. (c) y = 0.3 m.
18 G. Maragkos, B. Merci

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)

Axial velocity (m/s)

Radial position (m) Radial position (m)

(a) (a)

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)
Axial velocity (m/s)

Radial position (m) Radial position (m)

(b) (b)

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

FDS - Mod. Deardorff FDS - Mod. Deardorff

Radial velocity (m/s)
Axial velocity (m/s)

Radial position (m) Radial position (m)

(c) (c)

Fig. 8 Axial velocities for Test #17 at Fig. 9 Radial velocities for Test #17 at
height (a) y = 0.9 m, (b) y = 0.5 m and height (a) y = 0.9 m, (b) y = 0.5 m and
(c) y = 0.3 m. (c) y = 0.3 m.
Large eddy simulations of CH4 fire plumes 19

14 14
Experiment Experiment

12 Heskestad
12 Heskestad

McCaffrey McCaffrey

FireFOAM FireFOAM

10 FDS - Dyn. Smag. 10 FDS - Dyn. Smag.

Axial velocity (m/s)

Axial velocity (m/s)

FDS - Mod. Deardorff FDS - Mod. Deardorff Mean
Mean
8 8

6 6

4 4
Rms Rms

2 2

0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Axial position (m) Axial position (m)

(a) (b)
14
Experiment

12 Heskestad

McCaffrey

FireFOAM

10
Axial velocity (m/s)

FDS - Dyn. Smag.

Mean
FDS - Mod. Deardorff

4
Rms

0
0 1 2 3 4 5 6
Axial position (m)

(c)

Fig. 10 Centerline axial velocities for (a) Test #14, (b) Test #24 and (c) Test #17.

worth noting that the predicted sub-grid scale kinetic energies, ksgs , with the two
codes at different heights and on the centerline (not shown here) were on the same
order of magnitude, nevertheless, differences in the radial profiles up to a factor
of two were evident. The differences that are observed in the mixing time scales
are partially attributed to the fact that only turbulent and molecular mixing is
accounted for in the version of the EDM combustion model used in FireFOAM,
whereas, for the determination of τmix in FDS, the chemical mixing time scale is
compared with the mixing times for diffusion, advection and buoyant acceleration.
More specific, the different definition of τturb in FireFOAM and τu in FDS (or the
different choice of constants in their derivation) leads to significant differences
between these two time scales. The particular choice of constants here, CEDM = 4
and ce = 1.048, leads to τu ≈ 2−3τturb (depending also on the calculated values of
ksgs ), which is clearly portrayed in Figure 12. More specifically, the mixing time
scale due to molecular mixing in FireFOAM, τlam is, in this case, consistently
higher than τturb and is not taken into account at all in the EDM model, rather
20 G. Maragkos, B. Merci

0.20
FireFOAM: Constant Smagorinsky

0.18 FDS: Dynamic Smagorinsky - Test #14

FDS: Dynamic Smagorinsky - Test #24

0.16
FDS: Dynamic Smagorinsky - Test #17

Smagorinsky constant
0.14

0.12

0.10

0.08

0.06

0.04

0.02

0.00

0 1 2 3 4 5 6

Axial position (m)

Fig. 11 Centerline evolution of the Smagorinsky constant, cs .

only τturb is used. On the other hand in FDS, τg is important close to the fire
source and τd becomes dominant further downstream. These differences in the
mixing time scales between the two codes also contribute to the resulting fuel
000
mass reaction rates where the peak ω̇F values with FireFOAM are more than
double than those of FDS. These high fuel reaction rates with FireFOAM in the
flame region enhance combustion and produce hotter combustion products than
FDS that are convected in the plume region, resulting in higher temperatures
(shown below).

5.2.3 Temperatures

The predicted centerline temperatures with both codes are presented in Figure 13.
In general, the predictions of FDS agree well with the experimental correlations
in the plume region where over-predictions of up to 50 K are observed. This was
expected since the axial velocity profiles were also well predicted. On the other
hand, the centerline temperatures predicted by FireFOAM are over-predicted by
up to 100 K in the plume region. This is a direct consequence of the previously
reported slightly higher axial velocities compared to FDS (Figure 10) and higher
reaction rates (Figure 12) convecting hotter combustion products into the plume
region. Additionally, small differences are evident in the profiles of the tempera-
ture fluctuations at almost all axial locations examined, with FDS having higher
fluctuations than FireFOAM.

5.2.4 Turbulent kinetic energies

Results of the turbulent kinetic energies (sum of resolved and modeled) are pre-
sented in Figure 14 for Test #24. The simulation results with FireFOAM agree well
with the experimental values and remain close to the experimental uncertainty,
Large eddy simulations of CH4 fire plumes 21

0.25 1.0

FireFOAM - Test 17
0.9
FDS - Test 17

0.20 0.8

Reaction rate (kg/m s)

0.7

3
0.15 0.6
(s)

turb
mix

0.5

0.10 0.4

0.3
g
0.05 0.2

u
0.1
turb
0.00 0.0

0 1 2 3 4 5 6

Axial position (m)

000
Fig. 12 Centerline evolution of the mixing time scales (τmix ) and reaction rates (ω̇F ) for
Test #17.

revealing that the turbulent fluctuations in the near-field region of the fire plumes
are well captured by the code. The turbulent kinetic energies predicted by FDS
are less than FireFOAM, particularly around the centerline of the fire plumes, but
remain in reasonable agreement with the experiments.

5.3 Puffing frequency

From experiments performed on pool fires of different source sizes and fuels it has
been observed that the puffing frequency is strongly dependent on the fire source
diameter and almost independent on the fuel type. Correlations for predicting the
puffing frequencies of pool fires have been derived based on experiments performed
by, e.g., Zukoski [43]:
 1/2
g
f = 0.5 Hz (21)
D
and Cetegen [44]: √
f = 1.5 D Hz (22)
The puffing frequency corresponds to the number of puffing cycles encountered,
counted here as a maximum peak in the axial velocity followed by a minimum, in
the given timeline examined. This puffing behavior is seen in Figure 15(a), pre-
senting the time trace (6 sec) of the centerline axial velocity at height y = 0.5 m
for Test #17. The time signals taken from the numerical simulations have been
shifted along the time axis to make the comparison with the experiments more
clear (i.e. inspection by eye to overall match the locations of the maxima observed
in the experiments). A total of 10 cycles, corresponding to the passage of large-
scale structures, are distinguished in the experimental data. In general, both codes
are able to reproduce the puffing behavior of the CH4 fire plume, also predicting
22 G. Maragkos, B. Merci

1600 1600
Heskestad Heskestad

McCaffrey McCaffrey
1400 FireFOAM
1400 FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

1200 FDS - Mod. Deardorff 1200 Mean FDS - Mod. Deardorff

Temperature (K)

Temperature (K)
Mean
1000 1000

800 800

600 600
Rms Rms

400 400

200 200

0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Axial position (m) Axial position (m)

(a) (b)
1600
Heskestad

1400 McCaffrey

FireFOAM

1200 Mean
FDS - Dyn. Smag.

FDS - Mod. Deardorff

Temperature (K)

1000

800

600
Rms

400

200

0
0 1 2 3 4 5 6
Axial position (m)

(c)

Fig. 13 Centerline temperatures for (a) Test #14, (b) Test #24 and (c) Test #17.

the same number of cycles in the same time period examined. Nevertheless, the
amplitude of the axial velocity with FDS is smaller when compared to FireFOAM
and the experimental data. However, if the time trace of temperature is consid-
ered at the same location (y = 0.5 m) then a totally different behavior is observed
between the two codes Figure 15(b). In this case, a puffing behavior is clearly ob-
served only with FireFOAM while the temperature in the FDS simulations remains
almost constant. This lack of temperature fluctuations in FDS was previously also
seen in Figure 13.
Fast Fourier transformations of the time signals (taken between 5-50 sec in the
numerical simulations) of the axial velocities and temperatures at height y = 0.5 m
above the fire source were performed. The predicted amplitude spectrums are
presented in Figures 16-17, respectively, and the resulting puffing frequencies pre-
sented in Tables 6-7, respectively. The smaller amplitude in the predicted centerline
axial velocities with FDS, previously seen in Figure 15, is obviously confirmed. It
is also observed that, for the same fuel, increasing the HRR from 1.59 (Test #14)
Large eddy simulations of CH4 fire plumes 23

Experiment Experiment

FireFOAM FireFOAM

FDS - Dyn. Smag. FDS - Dyn. Smag.

Turbulent kinetic energy (m /s )

Turbulent kinetic energy (m /s )

2

FDS - Mod. Deardorff

2
FDS - Mod. Deardorff
2

2
Radial position (m) Radial position (m)

(a) (b)

Experiment

FireFOAM

FDS - Dyn. Smag.

Turbulent kinetic energy (m /s )
2

FDS - Mod. Deardorff

2

Radial position (m)

(c)

Fig. 14 Turbulent kinetic energies for Test #24 at height (a) y = 0.9 m, (b) y = 0.5 m and
(c) y = 0.3 m.

to 2.61 MW (Test #17), even though it increases the air entrainment near the
base of the fire plumes (see later), does not significantly alter the predicted fre-
quencies in the numerical simulations. This is in line with what has been reported
experimentally by Zukoski [43] and Cetegen [44]. Overall, the predicted puffing
frequencies based on the axial velocity with both codes are in very good agree-
ment with the (limited) experimental puffing frequencies reported by Tieszen et al.
[12, 13] but also with well-known correlations reported in literature. The predicted
frequencies are also in good agreement with the experimental data by Pagni [45]
(Figure 17) for a wide range of fire plumes, revealing an almost inversely linear re-
lationship between the puffing frequency and the fire source diameter. The puffing
frequencies based on temperature compare well with the experimental data and
the experimental correlations for both the FireFOAM and the FDS simulations.
24 G. Maragkos, B. Merci

8
Experiment

FireFOAM
7
FDS

6
Axial velocity (m/s)

1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0

Time (s)

(a)
1600
FireFOAM

FDS

1400

1200
Temperature (K)

1000

800

600

400

200
30 31 32 33 34 35 36 37 38 39 40
Time (s)

(b)

Fig. 15 Time trace at height y = 0.5 m for Test #17 of (a) axial velocity and (b) temperature.

Table 6 Predicted puffing frequencies based on axial velocity.

Zukoski: Cetegen:
Test Experiment Eq. (21) Eq. (22) FireFOAM FDS
(Hz) (Hz) (Hz) (Hz) (Hz)
#14 - 1.57 1.50 1.58 1.69
#24 - 1.57 1.50 1.65 1.75
#17 1.65 1.57 1.50 1.66 1.64
Large eddy simulations of CH4 fire plumes 25

1.0 250

0.9 FireFOAM

f=1.58 Hz FDS
Axial velocity amplitude (m/s)

Temperature amplitude (K)

0.8 200
FDS f=1.67 Hz
f=1.69 Hz
0.7

FireFOAM
0.6 150
f=1.58 Hz
0.5

0.4 100

0.3

0.2 50

0.1

0.0 0

0.1 1 10 100 0.1 1 10 100

Frequency (Hz) Frequency (Hz)

(a) (a)
1.0 250

0.9
Axial velocity amplitude (m/s)

FireFOAM
Temperature amplitude (K)

0.8 200
f=1.65 Hz

0.7 FDS

f=1.75 Hz
0.6 150

0.5 FireFOAM

f=1.60 Hz
0.4 100 FDS

1.71 Hz
0.3

0.2 50

0.1

0.0 0

0.1 1 10 100 0.1 1 10 100

Frequency (Hz) Frequency (Hz)

(b) (b)
1.0 250

0.9
Axial velocity amplitude (m/s)

Temperature amplitude (K)

0.8 200

0.7 FireFOAM

f=1.66 Hz
0.6 FDS 150

f=1.64 Hz
0.5 FireFOAM FDS

f=1.69 Hz 1.66 Hz
0.4 100

0.3

0.2 50

0.1

0.0 0

0.1 1 10 100 0.1 1 10 100

Frequency (Hz) Frequency (Hz)

(c) (c)

Fig. 16 Predicted amplitude spectrum of Fig. 17 Predicted amplitude spectrum of

axial velocity at height y = 0.5 m for (a) temperature at height y = 0.5 m for (a)
Test #14, (b) Test #24 and (c) Test #17. Test #14, (b) Test #24 and (c) Test #17.
26 G. Maragkos, B. Merci

Table 7 Predicted puffing frequencies based on temperature.

Zukoski: Cetegen:
Test Experiment Eq. (21) Eq. (22) FireFOAM FDS
(Hz) (Hz) (Hz) (Hz) (Hz)
#14 - 1.57 1.50 1.58 1.67
#24 - 1.57 1.50 1.60 1.71
#17 1.65 1.57 1.50 1.69 1.66

Fitted line

Pagni (1990)
10
FireFOAM

FDS
f (Hz)

0.1

0.01 0.1 1 10 100

D (m)

Fig. 18 Predicted puffing frequencies against experimental data from literature.

5.4 Flame height

Accurate prediction of flame height is in an important aspect for CFD codes since
it can be important in cases related to flame spread such as, e.g., in external façades
of buildings and areas involving combustible materials. The incident heat flux on
a material surface is directly related to the predicted flame temperature by the
CFD codes and the correct rate of pyrolysis of the material will strongly depend
on the accurate prediction of the flame height in the numerical simulations.
An estimate of the flame height, L, for buoyant diffusion flames and pool fires
has been given experimentally by Heskestad [46]:
L
= 15.6N 1/2 − 1.02 (23)
D
where N is a non-dimensional number given by a modified Froude number [47]:

cp T Q̇2
N= (24)
gρ2 (∆Hc/s)3 D5
The predicted flame heights are presented in Table 8, along with the flame
heights given by Heskestad’s correlation (Eq. (23)). A common criterion to calcu-
late the average flame height in numerical simulations is to determine the elevation
Large eddy simulations of CH4 fire plumes 27

where there is a 500-600 K difference between the average flame temperature and
the surrounding air temperature [36]. In the present study a temperature differ-
ence of 550 K was considered. The predicted flame heights by both codes, even
though strongly dependent on the temperature criterion used to determine them,
are generally in good agreement with Heskestad’s correlation (±15%). The pre-
dicted flame heights also lie within the wide range of experimentally measured
flame heights of different fuels and fire source sizes, presented in Figure 19, given
as a function of the dimensionless heat release rate, Q∗ , proposed by Zukoski [48]:
Q̇
Q∗ = (25)
ρ∞ cp T∞ (gD)1/2 D2

Table 8 Predicted flame heights based on the position on the centerline with ∆T = 550 K.

Test Q∗ Heskestad: Eq. (23) FireFOAM FDS

(m) (m) (m)
#14 1.81 3.89 4.21 3.23
#24 2.34 4.48 4.86 4.08
#17 2.95 4.86 4.98 4.56

10
9 Fitted line

CH - D=50 cm: Cetegen (1984)

8 4

CH - D=19 cm: Cetegen (1984)

7
4

1.2<D<2.4 m: Kung & Stavrianidis (1982)

6 Gasoline - 0.03<D<23.6 m: Blinov & Khudiakov

CH - D=31 cm: Alvarez (1985)

4

5 C H O - D=91 cm: Alvarez (1985)

3 6

CH COCH - D=152 cm: Wood et al. (1971)

3 3

4
L/D

CH - Heskestad (1984)
4

CH - D=1 m: FireFOAM
4

CH - D=1 m: FDS
4
1

0.1 1 10

*
Q

Fig. 19 Predicted flame heights, based on the position on the centerline with ∆T = 550 K,
compared to experimental data from literature.
28 G. Maragkos, B. Merci

5.5 Entrainment

Accurate prediction of the air entrainment at different heights is another key el-
ement of capturing the dynamics and unsteady behavior of fire plumes in the
numerical simulations. Air entrainment not only brings fresh air towards the reac-
tion zone for the combustion to take place but also cools the combustion products
while they are convected downstream from the fire source. Estimates of the en-
trainment rates in the flame, intermittent and plume region of fire plumes can be
estimated based on Heskestad’s experimental correlations [49]:

0.0056Q̇c (y/L)
 (y < L)
ṁ = (26)
0.071Q̇1/3 (y − y )5/3 + 1.92 × 10−3 Q̇

(y > L)
c 0 c

The predicted entrainment rates at different planes given by Heskestad’s cor-

relation (Eq. (26)) and the two codes are summarized in Table 9. In general, good
agreement of the entrainment rates is evident with FireFOAM at almost all loca-
tions examined, except very high in the domain (close to the outlet plane of the
domain). The entrainment rates with FDS are consistently lower, particularly in
the intermittent and plume regions of the fire plumes.

Table 9 Predicted entrainments at different heights.

Test #14 Test #24

y Hesk.: Eq. (26) FireFOAM FDS Hesk.: Eq. (26) FireFOAM FDS
(m) (kg/s) (kg/s) (kg/s) (kg/s) (kg/s) (kg/s)
0.5 0.92 1.28 1.02 1.04 1.31 1.08
1.0 1.83 2.34 1.76 2.07 2.39 1.84
2.0 3.66 4.28 3.12 4.14 4.40 3.27
3.0 5.49 6.09 4.54 6.21 6.31 4.73
4.0 8.46 7.91 6.14 8.28 8.26 6.35
5.0 11.66 9.91 7.84 12.58 10.40 8.03
6.0 15.37 12.09 8.80 16.54 12.64 9.04

Test #17
y Hesk.: Eq. (26) FireFOAM FDS
(m) (kg/s) (kg/s) (kg/s)
0.5 1.20 1.42 1.05
1.0 2.41 2.61 1.79
2.0 4.81 4.82 3.23
3.0 7.22 6.94 4.78
4.0 9.62 9.07 6.54
5.0 13.50 11.35 8.39
6.0 17.67 13.75 9.46

It is interesting to relate the entrainment rates with the predicted flame heights
in the numerical simulations with the entrainment number, n, expressing the ratio
of entrained air up to the flame tip with the stoichiometric air needed for com-
plete combustion. Different values of the constant have been reported in literature;
Tamanini [50] produced values of n = 9 based on numerical studies, Quintiere and
Large eddy simulations of CH4 fire plumes 29

Grove [51] reported n = 9.6 for axi-symmetric and linear plumes, Delichatsios and
Orloff [52] n = 10 for turbulent buoyant fires while Heskestad [53] and McCaffrey
& Cox [54] values of n = 12. It is generally suggested that the entrainment num-
ber will have values of n = 10 ± 5 at the flame tip [36]. Figure 20 presents the
mass flow rates of Table 8 (excluding the mass flow rate of fuel at the source),
normalized by the stoichiometric air required for complete combustion, and plot-
ted against the axial location, non-dimensionalized by the predicted flame heights
(previously presented in Table 7). Heskestad’s correlation, with the entrainment
number n = 12 at the flame tip, is also presented for comparative purposes. The
results reveal that the entrainment rates predicted with both codes are linear.
FireFOAM’s entrainment number at the flame tip (y/L = 1) is n = 12 − 15 while
for FDS is n = 7 − 9, both close to the reported values in literature.

20

Heskestad (1986)
*
18
FireFOAM - Q =1.81
*
FireFOAM - Q =2.34 FireFOAM
16 *
FireFOAM - Q =2.95 n=12-15
*
14 FDS - Q =1.81
*
FDS - Q =2.34
12
/(sm )

*
FDS - Q =2.95
f

10
air
m

FDS
4
n=7-9

0.0 0.2 0.4 0.6 0.8 1.0 1.2

y/L

Fig. 20 Predicted air entrainment numbers at flame tip.

6 Grid sensitivity

In this section, a selection of results from a grid sensitivity analysis with both CFD
codes is presented in Figures 21-23 for completeness of the current study. Results
for the mean velocities and temperatures at height 0.9 m above the fire source and
on the centerline with three different grid sizes (i.e. 1.5, 3.0 and 6.0 cm) for the
three tests (i.e. #14, #24 and #17) are presented. It is very interesting to observe
that the FDS results are not strongly grid dependent and comparable predictions
are obtained with 1.5, 3.0 and 6.0 cm for both the velocities and temperatures at
all the different locations examined. Similar behavior was observed for the FDS
30 G. Maragkos, B. Merci

results with the modified Deardorff on the same grid sizes (not shown here to
avoid repetition). On the other hand, the FireFOAM results show a very strong
grid dependency and reveal that relatively fine meshes ought to be used in the
numerical simulations in order to have accurate predictions with the code. As
expected, in both cases, the results with the finest grid employed (i.e. 1.5 cm)
compare better with the experimental data.
The reasons for the observed differences can be attributed to a number of
factors related to the modelling choices in the numerical simulations with Fire-
FOAM. First, the Smagorinsky turbulence model constant, cs , remains constant
to 0.1 regardless the cell size employed in the simulations effectively making the
fire plume more laminar as the cell size increases from 1.5 cm to 6 cm. On the
other hand, the variation of the Smagorinsky constant on the coarser grids (i.e. 3
cm and 6 cm) with FDS exhibited similar behavior as the 1.5 cm cell size case,
mostly remaining below 0.1 at heights up to y = 1.0 m and then increasing up
to about 0.13. The variation of cs with grid size in the FDS simulations is not
presented here to avoid repetition, nevertheless, its evolution with axial position
is similar to the one presented in Figure 11. This higher Smagorinsky coefficient in
the FireFOAM simulations, results in more laminar-like structures at the base of
the fire plumes which increases the predicted velocities and the flame temperatures
on the centerline and results in the fire plume to break at distances much further
downstream when compared to FDS. This is also confirmed by examining the rms
values of temperature on the centerline on the coarse grids (i.e. 3 cm and 6 cm),
presented in Figures 21(c), 22(c) and 23(c), which remain significantly lower with
FireFOAM than with FDS and imply more laminar-like flame structures. Second,
the resulting mixing time scales in FireFOAM remained consistently lower than
FDS with increasing grid size (i.e. 3 cm and 6cm) very close to the fuel inlet
similarly to what was reported in Figure 12 for a grid size of 1.5 cm. It is due
to the different definition of τturb in FireFOAM and τu in FDS (or the different
choice of constants in their derivation) that leads to significant differences between
these two time scales and the resulting reaction rates and flame temperatures as
well. Finally, a small contributing factor is attributed to the radiation modeling
approach employed in FireFOAM which is grid-sensitive due to the dependence
of the emission term of the RTI on the resolved temperature (i.e. T 4 ). The FDS
results are less sensitive on this aspect due to the constant C employed in Eq. (13)
which corrects the emission term based on a prescribed, by the user, global ra-
diative fraction [22]. The resulting radiative fractions in the numerical simulations
with the 1.5, 3 and 6 cm cell sizes were 24.8%, 28.3% and 25% for FireFOAM
while 25.3%, 22.2% and 19% for FDS.

7 Conclusions

Large eddy simulations of large-scale CH4 fire plumes with two different CFD
packages, FireFOAM (version 2.2.x) and FDS (version 6.1.2), have been presented.
It was verified that the predominant mechanism for vorticity generation at the
base of the fire source is due to the baroclinic and gravitational torques. However,
with increasing axial distance from the fire source, only the influence of the baro-
clinic torque remains significant. In comparison to previously studied iso-thermal
helium plumes [8, 9], the non-monotonic behavior of density in the reaction zone of
Large eddy simulations of CH4 fire plumes
Axial velocity (m/s)
31

Axial velocity (m/s)

Experiment Experiment

FireFOAM - 1.5 cm FireFOAM - 1.5 cm

FireFOAM - 3.0 cm FireFOAM - 3.0 cm

FireFOAM - 6.0 cm FireFOAM - 6.0 cm

FDS - 1.5 cm FDS - 1.5 cm

FDS - 3.0 cm FDS - 3.0 cm

FDS - 6.0 cm FDS - 6.0 cm

Radial position (m) Radial position (m)

(a) (a)

Experiment Experiment

FireFOAM - 1.5 cm FireFOAM - 1.5 cm

FireFOAM - 3.0 cm FireFOAM - 3.0 cm

FireFOAM - 6.0 cm FireFOAM - 6.0 cm

FDS - 1.5 cm FDS - 1.5 cm

FDS - 3.0 cm FDS - 3.0 cm

Radial velocity (m/s)

Radial velocity (m/s)

FDS - 6.0 cm FDS - 6.0 cm

Radial position (m) Radial position (m)

(b) (b)
1800 1800
FDS - 1.5 cm Heskestad FDS - 1.5 cm
Heskestad
McCaffrey FDS - 3.0 cm
1600 McCaffrey FDS - 3.0 cm
1600 FireFOAM - 1.5 cm FDS - 6.0 cm
FireFOAM - 1.5 cm FDS - 6.0 cm
FireFOAM - 3.0 cm
FireFOAM - 3.0 cm
1400 FireFOAM - 6.0 cm
1400 FireFOAM - 6.0 cm
Temperature (K)

Temperature (K)

1200 1200
Mean Mean
1000 1000

800 800

600 600
Rms Rms
400 400

200 200

0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Axial position (m) Axial position (m)

(c) (c)

Fig. 21 Grid sensitivity for Test #14 on Fig. 22 Grid sensitivity for Test #24 on
(a)axial velocities at height y = 0.9 m, (b) (a)axial velocities at height y = 0.9 m, (b)
radial velocities at height y = 0.9 m and radial velocities at height y = 0.9 m and
(c) centerline temperatures. (c) centerline temperatures.
32 G. Maragkos, B. Merci

Experiment

FireFOAM - 1.5 cm

FireFOAM - 3.0 cm

FireFOAM - 6.0 cm

FDS - 1.5 cm

FDS - 3.0 cm

Radial velocity (m/s)

Axial velocity (m/s)

FDS - 6.0 cm

Experiment

FireFOAM - 1.5 cm

FireFOAM - 3.0 cm

FireFOAM - 6.0 cm

FDS - 1.5 cm

FDS - 3.0 cm

FDS - 6.0 cm

Radial position (m) Radial position (m)

(a) (b)
1800
Heskestad FDS - 6.0 cm

1600 McCaffrey FDS - 1.5 cm

FireFOAM - 1.5 cm FDS - 1.5 cm
FireFOAM - 3.0 cm
1400 FireFOAM - 6.0 cm
Temperature (K)

1200
Mean
1000

800

600
Rms
400

200

0
0 1 2 3 4 5 6
Axial position (m)

(c)

Fig. 23 Grid sensitivity for Test #17 on (a)axial velocities at height y = 0.9 m, (b) radial
velocities at height y = 0.9 m and (c) centerline temperatures.

these large-scale fire plumes is responsible for the creation of two counter-rotating
vortices, enhancing the transport of fuel into the reaction zone [18].
There was, overall, good qualitative and quantitative agreement between the
simulation results obtained with the two codes, but also with the experiments per-
formed by Tieszen et al. [12, 13] and well-known correlations reported in literature.
The predicted mean and rms velocities and temperatures in the near-field region of
the fire plumes, the average flame heights, entrainment rates, the puffing behavior,
and resulting frequencies were well predicted and comparable between both codes.
The main differences observed in the results obtained with the two codes were fo-
cused in the centerline mean and rms temperatures and entrainment rates. More
specifically, FireFOAM over-predicted the temperatures in the plume region by
approximately 100 K while almost no temperature fluctuation were evident with
FDS. The differences observed in the centerline temperatures were attributed to
the different reaction time scale models employed in the two codes which resulted
in mixing time scales with FDS (τmix = 0.12 s) that were approximately double
Large eddy simulations of CH4 fire plumes 33

those of FireFOAM (τmix = 0.056 s). This led to significantly higher reaction
rates with FireFOAM in the flame and intermittent regions which resulted in hot-
ter combustion products to be convected in the plume region. FDS consistently
under-predicted the entrainment rates when compared to FireFOAM, particularly
in the plume region where discrepancies of up to 45% were evident. The mean and
rms results with FDS showed to be much less grid dependent than FireFOAM for
mesh sizes of 1.5, 3.0 and 6.0 cm. The relatively strong grid dependency observed
with FireFOAM poses a limitation when performing numerical simulations with
the code since fine grid sizes need to be employed.
Updating some of the physical models of FireFOAM (i.e. related to turbulence,
combustion and radiation) to the state-of-the-art modeling choices available in lit-
erature would improve the predictive capabilities of the code and enhance its range
of applicability. The use of dynamic models for turbulence and combustion, where
there is no possibility of pre-determination or calibration of the model constants,
would be ideal and desirable.

References

1. B.J. Meacham, The evolution of performance based codes and fire safety design methods,
National Institute of Standards and Technology, NIST-GCR-98-761 (1998)
2. G. Heskestad, Dynamics of the fire plume, Phil. Trans. R. Soc. Lond. A, 356, 2815-2833
(1998)
3. B.J. McCaffrey, Purely Buoyant Diffusion Flames: Some Experimental Results, NBSIR
79-1910, National Bureau Of Standards (1979)
4. E.E. Zukoski, T. Kubota, B. Cetegan, Entrainment in Fire Plumes, Fire Safety J., 3,
107-121 (1980)
5. B.M. Cetegen, E.E. Zukoski, T. Kubota, Entrainment and flame geometry of fire plumes,
NBS-GCR 80-402, National Bureau of Standards, Gaithersburg, MD (1980)
6. P.H. Thomas, P.L. Hinkley, C.R. Theobald, D.L. Simms, Investigations into the flow of
Hot Gases in Roof Venting, Fire Research Technical Paper no. 7, HMSO, London (1995)
7. A. Hamins, J.C. Yang, T. Kashiwagi, An Experimental Investigation of the Pulsation
Frequency of Flames, Proc. Combust. Inst., 24, 1695-1702 (1992)
8. G. Maragkos, P. Rauwoens, Y. Wang, B. Merci, Large eddy simulations of the flow in the
near-field region of a turbulent buoyant helium plume, Flow Turb. Combust., 90, 511-543
(2013)
9. P.E. DesJardin, T.J. O’Hern, S.R. Tieszen, Large eddy simulation and experimental mea-
surements of the near-field of a large turbulent helium plume, Phys. Fluids, 16, 1866-1883
(2004)
10. G. Maragkos, B. Merci, Large Eddy Simulations of Large-scale CH4 Fire Plumes, Pro-
ceedings of the 2nd IAFSS European Symposium of Fire Safety Science (2015)
11. Y. Wang, P. Chatterjee, J.L. de Ris, Large eddy simulation of fire plumes, Proc. Comb.
Inst., 33, 2473-2480 (2011)
12. S.R. Tieszen, T.J. O’Hern, E.J. Weckman, T.K. Blanchat, Experimental study of the flow
field in and around a one meter diameter methane fire, Combust. Flame, 129, 378-391
(2002)
13. S.R. Tieszen, T.J. O’Hern, E.J. Weckman, R.W. Schefer, Experimental study of the effect
of fuel mass flux on a 1-m diameter methane fire and comparison with a hydrogen fire,
Combust. Flame, 139, 126-141 (2004)
14. B. Merci, J.L. Torero, A. Trouve, IAFSS Working Group on Measurement and Computa-
tion of Fire Phenomena, Fire Technol., 52, 607-610 (2016).
15. B. Merci, J.L. Torero, A. Trouve, Call for participation in the first workshop organized by
the IAFSS Working Group on Measurement and Computation of Fire Phenomena, Fire
Safety J., 82, 146-147 (2016).
16. S. Ferraris, J.X. Wen, S. Dembele, Large-eddy Simulation of a Large-scale Methane Pool
Fire, Fire Safety Science, 8, 963-974 (2005)
34 G. Maragkos, B. Merci

17. A.R. Black, Numerical Predictions and Experiment Results for a 1m Diameter Methane
Fire, ASME International Mechanical Engineering Congress and Exposition, 429-435
(2005)
18. P.E. DesJardin, Modeling of conditional dissipation rate for flamelet models with appli-
cation to large eddy simulation of fire plumes, Combust. Sci. Technol., 177, 1883-1916
(2005)
19. Y. Xin, S.A. Filatyev, K. Biswas, J.P. Gore, R.G. Rehm, H.R. Baum, Fire dynamics
simulations of a one-meter diameter methane fire, Combust. Flame, 153, 499-509 (2008)
20. H. Pasdarshahri, G. Heidarinejad, K. Mazaheri, Large eddy simulation on one-meter
methane pool fire using one-equation sub-grid scale model, MCS 7, Chia Laguna, Cagliari,
Sardinia, Italy, September 11-15 (2011)
21. M. Hu, A.C.Y. Yuen, S.C.P. Cheung, P. Lappas, W.K. Chow, G.H. Yeoh, Modelling of
temporal combustion behaviour in a large-scale buoyant pool fire with detailed chemistry
consideration, International Congress on Modelling and Simulation, Adelaide, Australia
(2013)
22. K. McGrattan, S. Hostikka, R. McDermott, J. Floyd, C. Weinschenk, K. Overholt, Fire
Dynamics Simulator Technical Reference Guide Volume 3: Validation, NIST Special Pub-
lication 1018-3, Sixth Edition (2015).
23. H. Jasak, A. Jemcov, Z̆. Tuković, OpenFOAM: A C++ Library for Complex Physics
Simulations, International Workshop on Coupled Methods in Numerical Dynamics IUC,
Dubrovnik, Croatia, September 19th -21st (2007)
24. Z. Chen, J. Wen, B. Xu, S. Dembele, Large Eddy Simulation of Fire Dynamics with the
Improved Eddy Dissipation Concept, Fire Safety Science, 10, 795-808 (2011)
25. N. Ren, Y. Wang, S. Vilfayeau, A. Truvé, Large eddy simulation of turbulent wall fires,
8th U.S. National Combustion Meeting, Paper 070FR-0056 (2013)
26. K. McGrattan, S. Hostikka, R. McDermott, J. Floyd, C. Weinschenk, K. Overholt, Fire
Dynamics Simulator Technical Reference Guide Volume 1: Mathematical model, NIST
Special Publication 1018, Sixth Edition (2014)
27. J. Smagorinsky, General circulation experiments with the primitive equations: I. The basic
experiment, Mon Weather Rev. 91, 99-164 (1963)
28. D.K. Lilly, A proposed modification of the Germano subgrid scale closure method, Phys.
Fluids A, 4, 633-635 (1992)
29. J.W. Deardorff, Numerical Investigation of Neutral and Unstable Planetary Boundary
Layers, Journal of Atmospheric Sciences, 29, 91-115 (1972)
30. S.B. Pope, Turbulent Flows, Cambridge University Press (2000)
31. B.F. Magnussen, B.H. Hjertager, On Mathematical Modeling of Turbulent Combustion
with Special Emphasis on Soot Formation and Combustion, Proc. Comb. Inst., 16, 719-
729 (1976)
32. R. McDermott, K. McGrattan, J. Floyd, A Simple Reaction Time Scale for Under-Resolved
Fire Dynamics, Fire Safety Science, 10, 809-820 (2011).
33. C. Fureby, G. Tabor, Mathematical and Physical Constrains on Large-Eddy Simulations,
Theoret. Comput. Fluid Dynamics, 9, 85-102 (1997)
34. W.L. Grosshandler, RADCAL: A Narrow-Band Model for Radiation Calculations in a
Combustion Environment, NIST technical note 1402 (1993)
35. A. Hamins, T. Kashiwagi, R. Buch, Characteristics of pool fire burning, Fire resistance of
industrial fluids, ASTM STP 1284, George E. Totten and Jurgen Reichel, Eds. American
society for testing and materials, Philadelphia (1996)
36. D. Drysdale, An Introduction to Fire Dynamics, 3rd Edition, John Wiley and Sons, Eng-
land (2011)
37. K. McGrattan, J. Floyd, G. Forney, H. Baum, Improved radiation and combustion routines
for a large eddy simulation fire model, Fire Safety Science, 7, 827-838 (2003)
38. NRC, Verification and Validation of Selected Fire Models for Nuclear Power Plant Appli-
cations, NUREG-1824, U.S. Nuclear Regulatory Commission, Washington D.C (2007)
39. W. Chung, C.B. Devaud, Buoyancy-corrected k- models and large eddy simulation applied
to a large axisymmetric helium plume, International Journal of Numerical Methods in
Fluids, 58, 57-89 (2008)
40. S.B. Pope, Ten questions concerning the large-eddy simulation of turbulent flows, New
Journal of Physics, 6, 1-24 (2004)
41. I. Celik, M. Klein, J. Janicka, Assessment Measures for Engineering LES Applications, J.
Fluids Eng., 131, 031102 (2009)
42. G. Heskestad, Engineering Relations for Fire Plumes, Fire Safety J., 7, 25-32 (1984)
Large eddy simulations of CH4 fire plumes 35

43. E.E. Zukoski, Properties of fire plumes, Combustion Fundamentals of Fire (ed. G. Cox),
101-219, Academic Press, London (1983)
44. B.M. Cetegen, T.A. Ahmed, Experiments on the periodic instability of buoyant plumes
and pool fires, Combust. Flame, 93, 157-184 (1993)
45. P.J. Pagni, Some unanswered questions in fluid mechanics, App. Mech. Rev., 43, 153-170
(1990)
46. G. Heskestad, Luminous heights of turbulent diffusion flames, Fire Safety J., 5, 109-114
(1983)
47. G. Heskestad, Peak gas velocities and flame heights of buoyancy-controlled turbulent dif-
fusion flames, Proc. Comb. Inst., 18, 951-960 (1981)
48. E.E Zukoski, Convective Flows Associated with Room Fires, Semi Annual Progress Report,
National Science Foundation Grant No. GI 31892 X1, Institute of Technology, Pasadena,
CA (1975)
49. B. Karlsson, J.G. Quintiere, Enclosure Fire Dynamics, CRC Press (2000)
50. F. Tamanini, Reaction rates, air entrainment and radiation in turbulent fire plumes, Com-
bust. Flame, 30, 85-101 (1977)
51. J.G. Quintiere, B.S. Grove, A unified analysis for fire plumes, Proc. Comb. Inst., 27,
2757-2766 (1998)
52. M.A. Delichatsios, L. Orloff, Entrainment measurements in turbulent buoyant jet flames
and implications for modeling, Proc. Comb. Inst., 20, 267-375 (1984)
53. G. Heskestad, Fire plume air entrainment according to two competing assumptions, Proc.
Comb. Inst., 21, 111-120 (1986)
54. B. McCaffrey, G. Cox, Entrainment and heat flux of buoyant diffusion flames, Report
NBSIR 82-2473, National Bureau of Standards (1982)