www.elsevier.com/locate/jlp
Evaluation of a CFD porous model for calculating ventilation in
explosion hazard assessments
C.E. Fothergill
ab
, S. Chynoweth
a
, P. Roberts
a,
, A. Packwood
b
a
Shell Global Solutions, P.O. Box 1, Chester, CH1 3SH, UK
b
Department of Mechanical and Materials Engineering, University of Surrey, GU2 7XH, UK
Abstract
In the past, gas explosion assessment relied on worst case scenarios. A more realistic approach is to look at the probability of
explosions and their likely severity. The most exible way of investigating many different scenarios is to estimate a ventilation
ow, feed this into a ammable volume calculation and then calculate the explosion severity. The procedure allows many parameters
to be varied efciently. A Computational Fluid Dynamics porous model is evaluated for modelling the ventilation ow through
congested regions, including a new method that has been developed to derive the resistance. Comparison with velocity measurements
from a large scale model of an offshore module showed that overall the CFD model performs very well, especially considering
that the homogenous porosity block does not model any of the internal obstructions and therefore would not predict any local ow
effects. This gives condence that the overall ow pattern is sufciently close to the local ow patterns, to be used in explosion
assessments. The porous approximation in CFX is found to underpredict the turbulence intensity in the obstacle array compared to
the explosion model EXSIM. Improving the turbulence prediction in the porous model would be valuable, so a relatively simple
method of increasing the turbulence in porous regions is proposed. The CFD model will provide the non-uniform natural ventilation
owelds of complex regions for future explosion assessments at a hierarchy of levels.
2003 Shell Research Ltd. Published by Elsevier Science Ltd. All rights reserved.
Keywords: CFD; Porous model; Explosion assessment
1. Introduction
In the past, gas explosion assessment relied on worst
case scenarios. A more realistic approach is to look at
the probability of explosions and their likely severity.
For instance, a gas release may not result in a uniform
ammable mixture throughout the module; and the
pressures created from a worst case explosion arising
from the ignition of a stoichiometric gas cloud extending
through the entire module have been proven to be higher
than those created by a realistic gas cloud (Johnson,
Cleaver, Puttock, and Van Wingerden, 2002).
The most exible way of investigating many different
scenarios is to estimate a ventilation ow, feed this into
a ammable volume calculation and then calculate the
t
(brU
i
)
x
j
(b
i
rU
j
U
i
)
x
j
(b
i
s
ij
) b
p
x
i
R
i
where U
i
is the mean velocity vector in the i-direction,
b is the volume porosity, b
i
is the area porosity in the i
direction, and
R
i
b(R
C
i
R
F
i
|U
i
|)U
i
R
Cx
(kgs
-1
/m
-3
) is the resistance constant in the x-direc-
tion and R
Fx
(kg/m
4
) is the speed resistance factor in the
x-direction. The linear resistance term R
Cx
can be set to
zero due to the highly turbulent nature of the ow.
The porous approximation has been used to model
two-dimensional porous fences (Takahashi, Du, Wu,
Maki & Kawashima, 1998). Packwood (2000) used a
resistance based on empirical data from Hoerner (1965)
and tested the performance of a modied k- and the
Reynolds stress model for ow downstream of fences.
Overall the much simpler, modied k- model worked
surprisingly well. The three-dimensional problem has
been largely neglected, although Speirs (1997) experi-
mental study showed that the main ow features were
common to both the two and three-dimensional cases.
Another study of 3D volumes compared CFD PDR
results to experiments (Hoang, Verboven, De Baerde-
maeker & Nicola , 2000) and found 26% error in the
CFD simulations, but this was thought to be mostly due
to mesh quality.
Despite these studies, methods for representing com-
plicated geometries as porous regions in CFD are not
well established. The value for resistance depends on
both the shapes of the blockages within a module and
the distribution of large volume and small volume
obstacles. The drag on the ow through a porous region
can be estimated by calculating the drag due to each
individual obstacle and then summing these values to
arrive at the overall resistance.
The speed resistance factor R
Fi
(kg/m
4
) equates to
R
F
i
D
Vu
2
where V is the total volume, u is the averaged ow speed
and D is the drag. SI units are used throughout. The drag
on an individual object can be written
D
1
2
ru
2
C
D
A
where r is density, C
D
is the drag coefcient, and A is
the frontal area. The effective drag of the entire module
and all of its internal congestion was then taken to be
the sum of the drag of each obstacle:
D
i
1
2
ru
2
C
(i)
D
A
(i)
where C
D
(i)
and A
(i)
are values for an individual
obstacle.Therefore,
D
1
2
ru
2
i
C
(i)
D
A
(i)
and
R
F
i
1
2
r
C
(i)
D
A
(i)
V
The formulation above uses CAD data to calculate the
actual drag for four categories of obstruction, and each
of the shapes have different drag coefcients, available
in the literature (Munson, Young & Okiishi, 1998).
Implicit in the above formulation is the assumption
that obstacles are sufciently well spaced. This issue has
been addressed in another study using a resistance para-
meter, (TNO, 1989) where corrections were made for
obstacles lying in the wake of another obstacle. In Pack-
woods (2000) simulations the main deciency found is
that it overestimates the total resistance in cases where
grids are closely spaced, due to the partial shielding
effect of an upstream grid on a downstream one. A tech-
nique which sums the drag of all of the obstacles within
the module may similarly overpredict the overall resist-
ance.
The Engineering Sciences Data Unit Item 74040
(1974) provides a method for estimating the pressure
drop across tube banks which takes into account the
spacing of the obstacles. This method was used to test
the effect of spacing. It assumes that all of the obstruc-
tions are circular cylinders, and that the ow is con-
strained to go through the tube matrix rather than around
it. The resistance of an array of typical dimensions and
1 m spacing were calculated using both the ESDU
method and the method that sums the drag. A difference
of 7.5% was found in the ventilation. This is negligible
when considering the role of the ventilation gures in
the calculation of gas cloud size and explosion assess-
ments, but indicates that the method could be
improved nonetheless.
4. Validation
The ability of the porous block approximation in the
CFX code to represent ow through congested modules
was evaluated using measurements from a full scale
module and the results from another code, EXSIM
(http://www.exsim.com). EXSIM is a CFD based
software tool specically designed for explosion model-
ling, which models more of the complex geometry
within a congested region.
Simulations were carried out of the ow through a
module of dimensions 28 m 12 m 8 m with one
long side, the adjacent short side and the top blocked.
Five different wind speeds were modelled, ranging from
2.77.8 m/s, and the same wind direction was used in
344 C.E. Fothergill et al. / Journal of Loss Prevention in the Process Industries 16 (2003) 341347
every case, with the wind impinging on the short blocked
side of the module. Velocities from CFX were compared
at the locations where measurements were taken in the
module. The velocities for the different windspeeds were
non-dimensionalised by the ambient windspeed in each
case. The non-dimensionalised values from each differ-
ent wind speed simulation were then averaged to give a
single value at each location for the measurements and
the model.
The experimental programme was carried out in a
Joint Industry Project with 11 participating companies
(BG Technology, 1999). Velocities were measured
inside an experimental rig representative of an offshore
module, in different wind speeds. The rig was on open
land with trees and buildings in the vicinity. Ten anem-
ometers were placed in two planes, one at eight m along
the length of the rig, and the other at 20 m along the
length of the rig.
The domain in the CFX model extended ve module
heights around the module and 15 heights downwind, in
accordance with guidance found in Casey and Win-
tergerste, (2000). The domain was divided into cells on
a 0.5 m grid. The inlet velocity was tted with a log
law prole
U
ln
z
z
o