Journal of Loss Prevention in the Process Industries 13 (2000) 419–431

www.elsevier.com/locate/jlp

Prediction of vapour cloud explosions using the SCOPE model

J.S. Puttock *, M.R. Yardley, T.M. Cresswell
HSE Consultancy, Shell Global Solutions, Shell Research and Technology Centre, Thornton, P.O. Box 1, Chester, CH1 3SH, UK

Abstract

The SCOPE 3 model (Shell Code for Overpressure Prediction in gas Explosions) has been developed to predict the overpressures
which could be generated by gas explosions in vented enclosures, such as offshore modules. SCOPE 3 attempts, wherever possible,
to model the underlying physical processes in an explosion. This phenomenological approach gives greater confidence in predictions
for full-scale events than methods based simply on correlations of experimental data.  2000 Shell International Oil Products.
Published by Elsevier Science Ltd. All rights reserved.

Keywords: Explosion; Model; Vapour cloud; Overpressure

1. Introduction ments. We now regularly use the EXSIM CFD model

(Hjertager, Enggrav, Forrisdahl & Solberg, 1991) for
As concern has increased about the possible conse- detailed explosion assessments. Other similar models are
quences of a gas explosion in industrial plant, so has the FLACS (van Wingerden et al., 1993) and AutoReagas
need for accurate predictions of such explosions (van den Berg, The, Mercx, Mouilleau & Hayhurst,
increased. The predictions are improved by carrying out 1995).
experiments to improve our understanding, and by The improvements in CFD modelling do not, how-
developing mathematical models. We use a hierarchy of ever, negate the usefulness of the middle way —
predictive tools to assess explosion hazards, each having phenomenological modelling. Among the advantages
uses in appropriate circumstances. The simplest are:
approach is the provision of simple guidelines based on
correlations of experimental data, for example the Con- 1. less geometrical detail is needed in order to run the
gestion Assessment Method (Puttock, 1995). This allows model than for CFD, so that screening runs can be
rapid estimates to be made of the possible explosion done more easily or before detailed design has been
overpressure in a particular plant. van den Berg (1985) completed;
and Baker, Tang, Scheier and Silva (1994) have 2. computer run time is short (typically less than one
developed similar methods. second), so it is easy to perform many ‘what if’ runs,
At the other extreme of complexity are the Compu- testing the effect of design modifications, or sensi-
tational Fluid Dynamics (CFD) models. These do not tivity runs.
solve the full equations of fluid mechanics, as has some-
times been claimed, and they have needed extensive tun- Despite the relative simplicity of this approach, it is sig-
ing to allow good approximation of the effects of the nificantly different from correlation-based methods; the
many significant obstacles which are smaller than the aim is to describe the physics of the various physical
computational grid; nevertheless, major advances have processes involved in the explosion, rather than blindly
been made recently, and very good predictions are now to fit to overall results of explosion experiments.
being obtained in blind prediction of large-scale experi- SCOPE has its origins in the methodology published
by Cates and Samuels (1991). This was a simple spread-
sheet-based approach for an idealised geometry, taking
* Corresponding author. Tel.: +44-151-373-5594; fax: +44-151- the most obvious assumptions for each physical process
373-5845. in an explosion. The method was found to predict well
E-mail address: [email protected] (J.S. Puttock). for the series of 67 explosion experiments performed by

0950-4230/00/$ - see front matter  2000 Shell International Oil Products. Published by Elsevier Science Ltd. All rights reserved.
PII: S 0 9 5 0 - 4 2 3 0 ( 9 9 ) 0 0 0 4 5 - 5
420 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431

Det norske Veritas in a 35 m3 vented box1. It was resented as a series of grids positioned normal to the
expected that the method would be better able to predict main flow direction from the ignition point to the main
outside the range of existing experiments than corre- vent. Additional venting can be added as either side
lation-based methods. Improvements to many of the sub- vents of total area As or a rear vent of area Ar.
models in the Cates and Samuels formulation led to the Immediately following ignition at the back wall, the
release of the SCOPE 2, (Shell Code for Overpressure shape of the flame is an expanding hemisphere. How-
Prediction in gas Explosions) in 1993. Since then, the ever, once the flame reaches the sides of the box, it stops
understanding of explosion mechanisms has developed increasing in size, but retains its roughly hemispherical
further and the current model, SCOPE 3, has been in use shape as it advances down the box. The distance of lead-
since early in 1997. This paper gives an overview of the ing edge of the flame from the back of the box is denoted
formulation of SCOPE 3, and the results of validation. by X.
The basic principles of the model are unchanged from
the Cates and Samuels approach: a one-dimensional
2. Model overview model based on a series of obstacle grids, with the flow
through each grid used to determine turbulence and
It remains an important principle in the modelling that hence turbulent combustion downstream of the grid.
the formulation, although simplified, is based on an Fig. 2 summarises the overall structure of the model.
understanding of the physical processes involved in the
development of the explosion. The robustness of such 3. Differential equations
an approach derives from the fact that, wherever poss-
ible, the parameters used are derived from specific 3.1. Basic differential formulation
experiments to study the individual physical processes,
rather than ‘calibrated’ by adjusting the model predic- As the gas air mixture within the geometry under
tions to fit explosion data. investigation is burned, additional volume is generated
The environment to be modelled is idealised as a by the expansion of the gas. Of this ‘extra volume’, some
vented enclosure that is assumed to be filled with a flows out of the vents, and the remainder contributes to
homogeneous gas–air mixture, as shown in Fig. 1. At an increase of pressure. To take the balance between
one end of this box is a vent (the ‘main’ vent) of area vent flow and pressure generation properly into account,
Av. Ignition is assumed, on a worst-case basis, to take SCOPE 3 keeps track of the development of two vari-
place in the centre of the opposite face2. The length of ables as a function of time; these variables represent the
the box, along the direction from main vent to opposite amounts of unburned and burned gas in the box.
wall, is L, and the cross-sectional area is A. Obstacles, Consider the stage where the flame has already
such as vessels or pipework, within the box are rep- reached the sides of the box, and assume that there are
no side vents. The mass of unburned gas in the box, Mu,
is decreasing as some is consumed by the flame and
some is pushed out of the vent. If the flame area is sA,
the turbulent burning velocity Ut, and the unburned gas
density ru, then the rate of consumption by the flame is
sAUtru. The mass flow through the main vent is
UvcDAvru, where cD is the discharge coefficient and Uv
depends only on the pressure P. Thus
dMu
⫽⫺sAUtru⫺UvcDAvru (1)
dT
where T is time. (Some of the burned and unburned gas
Fig. 1. The basic geometry. may be expelled through side vents; this aspect is
covered later.) We now consider the burned gas. In this
case, it is not sufficient to keep track of the mass, since
1
Det norske Veritas. ‘Gas explosion research programme’, 1982– this does not provide information on the thermodynamic
83. These reports are unpublished; however, basic results are listed by state of the gas. It turns out that it is convenient to use
Cates and Samuels (1991).
2
Ignition at other locations could be accommodated, at the cost of the quantity B=P1/gVb, where g is the ratio of specific
making the model more complex. The initial aim in developing the heats3. Then the equation for the rate of increase of
model was to represent the worst-case explosion; this is given by burned gas is
ignition as far a possible from the vent. Ignition at the centre of a
symmetrical enclosure (vents at both ends) can be represented in the
3
current model, taking half the enclosure and using the symmetry, as We treat both unburned and burned as ideal gases. The ratio of
there would be no flow across the central plane. specific heats is not the same for unburned and burned gas. However,
 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431 421

Fig. 2. An overview of the SCOPE model.

dB 1/g dB 1/g
⫽P sAUtE (2) ⫽P s(2pX2)UtE (5)
dT dT
where E is the expansion ratio. The expressions used to
calculate Ut, Uv and s are discussed below.
3.3. Dimensionless version
It is also useful to keep track of the mass of gas, Me,
emitted from the main vent, for use in the external The algebra is considerably simplified if the variables
explosion calculations: are converted to dimensionless form. Nondimensionalis-
dMe ing with A, L, P0 (the atmospheric pressure), r0u (the gas
⫽UvcDAvru (3) density at pressure P0) and C⬅√g(Po/r0u) (the speed of
dT
sound in the unburned gas at pressure P0), we write
X U TC P
x⫽ , u⫽ , t⫽ , p⫽ (6)
3.2. Early flame development L C L P0
Mu B A Av
In the early stage after ignition, until the flame reaches mu ⫽ , b⫽ , a⫽ 2, av⫽ 2
ALr0u ALP01/g L L
the sides of the box, the equations are different, since
the flame is expanding as a hemisphere, rather than trav- Then Eqs. (1) and (2) reduce to:
elling longitudinally down the box. At this stage, the
flame surface area is 2pX2, and the rate of consumption
of unburned gas 2pX2Ut; so the differential equations
dmu 1/g
dt 冉
⫽p ⫺sut⫺uvcD
av
a 冊 (7)

db
become ⫽sEutp1/g (8)
dt
dMu
⫽⫺s(2pX2)Utru⫺UvcDAvru (4) and for the hemispherical stage:

冉 冊
dT
dmu 1/g av
⫽p ⫺2px2ut⫺uvcD (9)
dt a
db
use of a single value for γ simplifies the formulation considerably, and ⫽2px2Eutp1/g (10)
dt
so one value is used, a mean of those for unburned and burned gas–
air mixtures.
422 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431

3.4. Derived quantities wrinkling of the flame front and an acceleration of the
flame. Reliable experimental measurements of this pro-
Other variables can be derived from b and mu; i.e. the cess are sparse, but the available data can be fitted by
pressure is determined by the total quantity of gas in the taking the burning velocity to increase linearly with dis-
box, and the flame position by the relative amounts of tance:
burned and unburned gas. In dimensionless terms, the
relations are simply Ut⫽U0⫹kR (13)
p1/g⫽mu⫹b (11) where R is the radius of the flame4.
Analysis of the data of Lind and Strehlow (1975) leads
and to a value of 0.25 s⫺1 for k in the case of propane (at
b b stoichiometry l=1.25) and 0.16 s⫺1 for methane
x⫽a⬘⫹ for ⬎2a⬘ (12)
mu⫹b mu⫹b (l=1.05). The experiments of Brossard et al. (1985)

冉 冊
show self-acceleration in ethene, and the plots from one
b 27a⬘2 1/3
b test allow a rough estimate to be made for k as 0.64 s⫺1.
x⫽ for ⬍2a⬘
mu⫹b 2 mu⫹b Degener and Forster’s (1990) turbulent burning velocity
where experiments also include some measurements without
turbulence. Resulting values of k are reasonably consist-

冪p
1 a ent with the above. These values are used in the model.
a⬘⫽
3
5.1. The limit of self-acceleration

4. Laminar burning velocity and expansion ratio The process of self-acceleration is unlikely to continue
indefinitely, since the extent to which surface insta-
The laminar burning velocity has two important roles bilities can increase the flame area is limited. The above
in explosion modelling. Firstly, it describes the behav- data give no clear indication of where this limit occurs.
iour of the flame just after ignition, before it encounters The SOLVEX experiments (Bimson et al., 1993),
any obstacles. Here, the flame is moving slowly and involved explosions in a 550 m3 vented box; the first
hence this period is relatively long when compared with two experiments, without obstacle grids, provided use-
the duration of the main pressure pulse. Secondly, the ful data.
laminar burning velocity is used in the determination of SCOPE predictions of flame travel have been com-
the turbulent burning velocity, as discussed later. pared with SOLVEX measurements. For methane, with-
The values for U0 in stoichiometric mixtures at normal out a limit to self-acceleration, the model overpredicts
pressure are derived from Warnatz (1981). The form of the flame speed in the later stages, but if the self-acceler-
the variation of laminar burning velocity with stoichi- ation is stopped when the flame radius reaches 3m, a
ometry is derived from the data of Taylor (1991), as good fit is found. A similar result is found for propane
reported by Tseng, Ismael and Faeth (1993). Since the if the self-acceleration is stopped after 2 m. For both
values at stoichiometric are not entirely consistent with these gases, these limits correspond to a cessation in the
other data, we have scaled Taylor’s curves to meet the acceleration when the burning velocity reaches two and
desired stoichiometric values. A polynomial fit was used a half times the laminar burning velocity U0; this limit
for each gas; in many cases a single polynomial does is used in the SCOPE model.
not provide an adequate fit over the whole range, and so
separate polynomials are specified in different regions.
Both the laminar burning velocity and the expansion 6. Turbulent burning velocity
ratio change as the pressure is increased (adiabatically).
P.H. Taylor (private communication) used the Sandia A suitably accurate and fully theoretical expression
computational model (Kee, Grcar, Smooke & Miller, for turbulent burning is not available. Experiments to
1985) which incorporates full chemistry and transport, measure turbulent burning velocity are also difficult,
to investigate the properties of laminar flames under mainly due to the problems of creating a large volume
adiabatic compression. Power laws were fitted to the containing homogeneous turbulence and allowing a
results and are used in the model. flame to propagate without boundary effects. Experi-
ments at the largest values of Reynolds number likely
5. Self-acceleration of the flame (quasi-laminar to be encountered in full-scale gas explosions have not
burning)

Even in the absence of turbulence, the expanding 4

Cates and Samuels (1991) introduced a delay distance, after which
laminar flame develops instabilities that cause some the acceleration starts, but this does not appear to be necessary.
 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431 423

been performed. Thus, in SCOPE, a semi-empirical for (ut/ul)H should be biased more to the higher turbu-
approach has been taken. The aim in choosing a turbu- lence data.
lent burning relation has been to derive an expression The fitted expression was of the form
that is believed to have roughly the right functional form, (ut/ul)H=c1(KLe)−c2(u⬘/ul); KLe was a parameter, because
but with constants that allow the expression to be fitted the Bradley et al. (1992) compilation of data is plotted
to experimental data. The choice of functional form is in terms of K Le. The best fit was obtained with c1=1.00,
important, for example, to achieve a correct dependence and c2=0.33.
of the burning rate on turbulent length scale. Following this, two changes were made to reach the
The turbulent burning velocity is given by a smoothed final expression: 0.33 was changed to the simple 1/3.
transition between two turbulent burning velocity Secondly, it is believed that the effect of turbulent stretch
relations, one that describes the behaviour at low turbu- on burning velocity is represented by a (stretch)
lence intensity, Ut|L, and the other at high turbulence Markstein number Ma, and so it is preferable to have
intensity, Ut|H. A relationship that provides a smooth the expression dependent on K Ma, rather than K Le. To
transition between the two regimes, in reasonable agree- make this change K Le was replaced by K Ma/4.1. This
ment with experimental data (see below) is: simplistic substitution was based initially on Eq. [5] of

再冉 冊 冉 冊 冎
−4 −4 −1/4
Searby and Quinard (1990), taking the reduced acti-
ut ut ut vation energy β to be 10; however, data on Markstein
⫽ ⫹ (14)
ul ul L ul H numbers confirm that this is a reasonable approximation
for stoichiometric and lean methane and propane, which
covers much of the data included in the Bradley et al.
The relation used for low turbulence intensities, fol-
(1992) compilation.
lowing Gulder (1990) is:
Thus the turbulent burning velocity used in SCOPE 3

冉冊
ut
ul L
⫽1⫹0.7
u⬘
ut 冉冊 1/2
R1/4
l
(Fig. 3) is given by Eqs. (14) and (15) and the follow-
ing expression:

where Rl=UL/ν, L is the integral scale of the turbulence,

and the constant 0.7 has been obtained by fitting data
冉冊ut
ul H
⫽1.60(KMa)−1/3
u⬘
ul
(16)
from the Leeds combustion bomb (Abdel-Gayed, Al-
Khishali & Bradley, 1984; Bradley, Lau & Lawes, It may be noted that this expression is not applicable for
1992). This can be written as: negative Markstein number. When Ma is zero or nega-

冉冊
ut
ul L
⫽1⫹0.28K −1/2
u⬘
ul 冉冊 3/2
(15)
tive, Eq. (15) is currently used for all u⬘.
There are no suitable direct experimental determi-
nations of the effect of length scale on turbulent burning
where K is the Karlowitz number calculated, following velocity; this aspect is important since the response of
Abdel-Gayed et al. (1984), as K=0.157(U⬘/Ul)R−1.2
l . the model to changes of scale is largely determined by
In considering the case of high turbulence intensity, the form of the turbulent burning velocity expression.
we note that Bray (1990) fitted the following expression
to data on turbulent burning velocities:
ut u⬘
⫽0.875K −0.392
ul ul
and Bradley et al. (1992) obtained a good fit for their
compilation of a wide range of data using:
ut u⬘
⫽0.88(KLe)−0.3
ul ul

We have performed a new fit for two reasons. Firstly,

we wished to extend to higher Reynolds numbers than
can be obtained in the laboratory by using the compu-
tations of Bradley, Gaskell and Gu (1994); very large
length scales and thus Reynolds numbers can apply in Fig. 3. The fitted turbulent burning velocity. The dashed curves are
from Eqs. (14)–(16), taking Ma=4.1Le. The solid lines show the trends
real industrial explosions. The second factor was that we of the data from experiments (Abdel-Gayed et al., 1984) and, at higher
were adjusting the (ut/ul)H formulation so that the com- u⬘, the computations of Bradley at al. (1992). All are shown for values
bined expression (14) fitted the data best. Thus the fit of K Le equal to 0.10 (uppermost), 0.14, 0.21, 0.30, 0.43, 0.63 and 1.0.
424 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431

Comparisons of SCOPE 3 predictions with data from therefore, the dip to 0.4 is ‘bridged’ in the formula used
the SOLVEX large-scale explosion experiments and the in SCOPE 3:

再 冎
corresponding small-box experiments are therefore very
encouraging; these show that the change of explosion 1 Log10(Re)⬍5.3
cs⫽ (18)
pressure with scale (for a factor of 220 increase in 0.7 5.5⬍Log10(Re)
volume) is reproduced very well by the model. (The
UD
experiments are described later in the Section 13.) where, Re= , U is flow velocity, D is cylinder diameter
v
and v is gas viscosity. cs is taken to decrease from 1 to
0.7 linearly in log10(Re) from 5.3 to 5.5.
7. Flame shape
The drag of a grid of such obstacles as a function of
the blockage ratio σ is derived from Roach (1987)
After the end of the hemispherical expansion, i.e.
when the flame reaches the walls of the box, a value
must be chosen for s, the ratio of the flame surface area
to the box cross-sectional area. As in SCOPE 2, this is
cg⫽0.52cs 冉冉 冊 冊
1
(1−σ)2
⫺1 (19)

taken to be the same as the surface area/base area ratio

for a hemisphere, i.e. 2. However, for sufficiently long For sharp-edged obstacles, e.g. square bars, there are
aspect ratios, it might be expected that this should be similar expressions. The drag of individual bars is
increased, since a flame tends to elongate as it travels dependent on incident turbulence intensity. The follow-
down a long tube. On the other hand, only the front part ing is based on data from Graf and Mansour (1975):
of the flame will have passed through the latest grid, and

冦 冧
so will be contributing at the fastest burning rate. 2 I⬍0.01
Based on data from the small-box experiments (I−0.01)
(Degener & Forster, 1990), the value of s is increased cs⫽ 2−0.5 0.01ⱕI⬍0.025 (20)
0.015
linearly with distance if the aspect ratio exceeds 1.3:
1.5 0.025ⱕI
If xrⱖ1.3 (17)
U⬘/U
s⫽2(1⫹0.7(xr⫺1.3)) I⫽
Log10(Re)
where xr⬅x/√a⬅X/√A. This was originally fitted using
UD
experiments with aspect ratio L/√A⬅1/√a up to 2; it was Re=
later found to provide good results for further experi- ν
ments with aspect ratio up to 4.9. For grids of sharp-edged obstacles, the dependence on
If the main vent is significantly smaller than the cross- the blockage s is given by Hoerner (1950):
section of the module, then the flow converges as it csσ
approaches the vent. The flame similarly narrows as it c g⫽ (21)
(1−σ)2
follows the flow. A grid near a small vent thus has a
smaller effect than would otherwise be expected, since
the part of the flame front which reaches the turbulent Naudascher and Farell (1970) have tabulated experi-
region downstream of the grid has a smaller area. This mental data on grid drag coefficients from ten different
effect is now taken into account in the model. sources, involving single-plane and biplanar grids of cir-
cular rods and sharp-edged bars. The values are plotted
in Fig. 4 together with curves from Eqs. (19) and (21).
8. Grid effects Fig. 4 shows good agreement between the formulae and
these data.
The drag coefficient cs of a cylinder is dependent on cg is used in determining the turbulence generated by
flow velocity, decreasing from 1 at low Reynolds num- a grid, which is taken to be proportional to √cgU, where
ber to about 0.4 at a Reynolds number of 5×105, and U is the flow velocity through the grid as the flame
thereafter increasing to 0.7 (Kwok, 1986). The existence reaches it.
of the minimum is a problem for application in the
model. Although the formal model assumption is a flow 8.1. Additional flame area downstream of a grid
uniform across a grid, the reality is that there would be
a variation of velocity across the width of the box; in As a flame passes through a row of obstacles the time-
the vicinity of a minimum in cs, the average drag would averaged, large-scale flame surface area increases as a
be higher than given by the drag relevant to the average result of the flame ‘fingering’. This mechanism of
flow velocity. To avoid underprediction in such cases, ‘macroscopic’ flame area generation is different from the
 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431 425

simple arrays of cylinders used in most idealised experi-

ments. In particular, there is a great range of length
scales present. The effect of this is to increase the
‘macroscopic’ flame area generation above that per-
taining to rows of uniform cylinders. We have performed
experiments to demonstrate these effects, using idealised
obstacles but with a range of obstacle sizes. Using the
results of these experiments and others, we account for
these effects by increasing the macroscopic flame area
generation in the model for the higher levels of
obstacle complexity.

9. Vents

9.1. Vent flow

Standard relations for compressible vent flow are used

in the model, allowing correct calculation of the depen-
dence of vent flow on pressure up to and including
choked flow. A ‘vent flow velocity’ Uv is defined as the
mass flow Ṁ through the vent divided by the vent area,
the discharge coefficient and the gas density; dividing
Fig. 4. (a) The drag coefficient cg of grids of round pipes [Eq. (19)]; by the speed of sound gives the dimensionless ‘vent
(b) The drag coefficient cg of grids of square pipes [Eq. (21)]; flow velocity’
Ṁ
u v⫽ (23)
stochastic, small-scale increase in flame area due to cDAvρC
wrinkling, which is taken into account in the turbulent
burning velocity formulation. Because of the lack of uv thus defined is the same for all vents where unburned
detailed experimental data in this area, computational gas is being vented, being a function only of the pressure
fluid dynamics (CFD) (Hjertager et al., 1991) was used ratio p. If p is less than the critical pressure ratio

冉 冊
γ
to give some indication of the behaviour of an initially γ+1 γ+1
planar flame as it passes through a regular grid and a pcrit⫽ (24)
2
simple model was derived from the computational data.
It was found that the flame area showed a near-linear then venting is subsonic, otherwise sonic flow occurs in
increase with flame position (measured from the the vent. The expressions for uv are:
obstacles) in a region close to the obstacles; the gradient
2 γ−1 2
was roughly proportional to the square root of the block- u2v⫽ (p γ ⫺1)p− γ 1ⱕpⱕpcrit (25)
age ratio σ. γ−1

冉 冊
γ⫹1
2 2γ−1 γ−1
d(Af/A)
d(x/D)
⫽1.7 σ 冑 (22) u v⫽
γ⫹1
p 2γ pcritⱕp

Further downstream of the obstacle, the flame must The discharge coefficient for an orifice is dependent
‘relax’ and will tend towards a planar flame front if there on the pressure ratio, increasing from 0.6 for 1/p=1 to
are no more obstacles in its path, owing to the flame’s 0.9 for 1/p=0.
burning across the wakes.
Thus the model includes a flame area enhancement 9.2. Side vents and rear vents
factor, which increases rapidly downstream of a grid (at
a rate proportional to √σ), and then decays slowly. A further beneficial consequence of taking the differ-
ential equation approach described above is that side
8.2. Obstacle complexity vents can be taken into account. It is straightforward to
include the loss of burned and unburned gas through the
The obstacles, e.g. vessels and piping, in a real plant side vents in Eqs. (7)–(10). For example, if the area of
environment are typically much more complex than the side vents ahead of the flame is Asa, then an additional
426 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431

term ⫺cDuv(asa/a) is added to the right hand side of 10. External explosion
Eq. (7). The side vent areas ahead of and behind the
flame are determined by taking the boundary to be where As the flame burns through the enclosure, gas is
the edge of the hemispherical flame touches the side of pushed out through the vent owing to the expansion
the box. No allowance is taken for the distortion of the associated with the combustion. This gas forms a highly
flame shape by the presence of side vents. turbulent jet and a ‘starting vortex’ outside the vent.
The simple form of the side vent flux given above is When the flame emerges through the vent, this gas burns
correct for the venting of unburned gas, which has den- rapidly to form the ‘external explosion’ (Harrison &
sity ρu at pressure P0. If the gas density is lower, the Eyre, 1987). The external explosion is responsible for
mass flow through the vent decreases in proportion to overpressure effects away from the box. It also affects
√ρ, but the volume flux increases in proportion to √ρ. the internal pressure, which generally peaks after the
Thus for burned gas the dimensionless ‘vent flow velo- occurrence of the external explosion.
city’ is √E uv, where uv is defined in Eq. (23), since The dynamics of the turbulence and flame travel in the
E=ρu/ρb (This is only strictly correct if E is constant, but external explosion, and its interaction with the internal
it is taken to be a reasonable approximation as E varies pressure, are complex and not well understood. Minor
with pressure.) This formulation was introduced into the changes to the SCOPE 2 formulation with respect to the
model and was found to give good predictions of the peak internal pressure have taken place, but the
effects of various side vents without any adjustment. geometry formulation remains unchanged.
A rear vent, i.e. a vent in the ignition wall, is simply
10.1. External explosion geometry
handled in the same way as a side vent in that burned
and unburned gas is expelled through the vent. However,
An idealised shape is assumed for the external
the calculation of the relative sizes of the venting area
explosion, as shown in Fig. 5. The mushroom of
ahead of and behind the flame front depends upon
unburned gas, at the time of flame emergence, is mod-
whether the flame front is still expanding over the sur- elled as a hemisphere surmounting a cylinder. The cross-
face of the rear wall rather than travelling down the sectional area of the cylinder is taken to be the area of
length of the box. A pessimistic approach is taken in that the vena contracta outside the vent, i.e. cDAv. There are
the vent area is assumed to be positioned away from the two parameters to determine: the radius of the hemi-
ignition location; so no rear venting takes place until the sphere, Rh, and the length of the cylinder. The amount
flame ball has grown to a size where the base area of of gas expelled through the main vent up to this time,
the hemisphere is larger than the cross sectional area of Me, has been calculated using Eq. (3). A correlation
the module minus the rear vent area. between jet tip position and head diameter was derived
from an experiment in the Buxton 2.5 m3 box (see Sec-
tion 13) this completes a set of equations that can be
9.3. Vent opening solved for length of the cylinder and the radius of the
hemisphere.
The pressure in the external explosion, Pext, is taken
The effect of vent covers (on side, main or rear vents) to be proportional to the pressure reached in the enclos-
can be simulated using SCOPE 3. For each vent, the user ure when the flame emerges from the main vent, Pemerg.
must specify an opening pressure and time delay. The
vent area is taken to be zero until the internal pressure
in the box first reaches the opening pressure. The vent
area then increases linearly with time so that the vent is
fully open after a further time equal to the time delay.
Quite complex behaviour can be simulated. For
example, if there are two vents with similar properties,
the user is advised not to enter identical opening press-
ures since there will in practice be some differences
between the vents. In such cases, the predicted result is
often that the pressure rises rapidly at first and opens
one vent; as a consequence, the pressure falls again and
does not rise high enough to open the other vent until
late in the explosion. At that stage the second vent may
not have much effect. We have observed such behaviour
in experiments with vent covers.
Fig. 5. The idealised external explosion.
 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431 427

The correlation used has Pext/Pemerg dependent on the when this will occur, but the assumption that the external
ratio of vent area to box cross-sectional area, the aspect explosion always adds to the internal pressure is con-
ratio of the box, and Me. It is also found experimentally servative.
that the external explosion pressure is lower for a single External explosion effects are only considered for the
main vent than when there are several openings in that main vent, far from the ignition, not for side vents. The
wall; in the latter case, the vent blockage is similar to flame reaches a side vent earlier than the main vent; at
an extra obstacle grid. This is also taken into account in this time the maximum flow and turbulence due to the
the correlation. accelerating flame have not developed. Thus the burning
The correlation does not allow for the compression of rate outside the side vent is generally much lower than
the gas in the external explosion when the pressure is in the main external explosion. In addition, the volume
high. At high pressures the expansion ratio from the of gas is smaller.
ambient initial condition is low; hence the ratio of flame
speed to turbulent burning velocity is lower. The over-
pressure depends on the flame speed and so this needs 11. Stoichiometry changes
to be taken into account. Assume that the overpressure is
SCOPE 3 handles stoichiometry variations by altering
p⫺1⫽αU2f with Uf⫽Ut(E⫺1)
the laminar burning velocity of the fuel, and the
where α is a constant, Uf and Ut are flame speed and Markstein number, using multiple polynomials. For pro-
turbulent burning velocity, and E is the expansion ratio. pane, the Markstein number decreases for an increase in
An iterative calculation in the model effectively replaces stoichiometry, which, for a constant laminar burning rate
the above with would lead to an increase in the overpressure. Combined
with the decrease in laminar burning velocity between
Uf⫽Ut(Ep−1/γ⫺1)
stoichiometry of 1.1 and 1.3, this results in the overpres-
where γ is the ratio of specific heats. sure remaining roughly constant over this range. By con-
Since geometries without any obstacles do not easily trast, for methane, the Markstein number increases with
fit the same trends as cases with obstacles, they are increasing stoichiometry, hence the overpressure
treated as a special case, although empty boxes are not decreases rapidly with increasing stoichiometry on the
often encountered in real-life geometries. rich side.

10.2. Maximum internal pressure

12. Gas mixtures
The interaction between the external explosion and the
further development of the internal pressure is complex All properties used in SCOPE 3 for a given fuel are
and not well understood. In the majority of experiments specified in a ‘gas file’. A linked model ‘gasmix’ can be
in our rectangular boxes, the external explosion has the used to generate a new gas file based upon a mixture of
effect of increasing the internal pressure further. It is not pure fuels, or a single fuel at given ambient conditions.
clear whether this is best viewed as the effect of a press- This gas file can then be used by SCOPE in a similar
ure wave travelling into the box, or as the result of the way to the pure fuels supplied with the program. The
external pressure restricting the vent flow while burning laminar burning velocity of the mixture is estimated by
still continues into the corners of the box. For this reason performing a full calculation of the flame temperature;
the treatment of this phenomenon in SCOPE 3 is empiri- then the adiabatic flame temperature is used in conjunc-
cal. The experiments also show that the effect of the tion with a series of parameters for each pure fuel, which
external explosion on the internal pressure is larger are derived from the SANDIA flame code (Kee et al.,
where no grids are present in the module, consequently 1985), or other data, to determine the laminar burning
this is modelled differently. rate of the mixture. Some other parameters required by
The effect of the external explosion on the internal the SCOPE model, e.g. self-acceleration rates etc. are
pressure is modelled by adding a proportion of the exter- calculated by interpolation between the values used for
nal pressure Pext to the pressure computed when the the pure fuels.
flame reached the vent (Pemerg) to obtain the maximum
internal pressure. Surprisingly, this very simple approach
works well for many cases. The main exception is that 13. Validation
in some explosions, the internal pressure has decayed
markedly before the influence of the external explosion Significantly more data are now available for vali-
is felt; then the external explosion produces a peak that dation of the SCOPE 3 model than were available for
is not as high as the ‘emerging’ peak, i.e. the maximum SCOPE 2 validation. The model has been validated
pressure is Pemerg. We do not know enough to predict against the following data sets:
428 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431

1. Results from the Det norske Veritas experiments in a mental results is shown in Fig. 6. The central line here
35 m3 box, mentioned above. Data from 67 experi- shows perfect agreement, with the factor-of-two bands
ments are available; they were all performed using shown.
rich propane gas, with a variety of vent areas, using A histogram of the ratio of measured to predicted
up to four rows of obstacles, round and sharp. overpressure is shown as the bars in Fig. 7. Fig. 7 is
2. The Shell SOLVEX experiments in a 550 m3 enclos- centred on a value of one. In fitting a distribution to such
ure, using methane and propane (Bimson et al., 1993). ratios, it is not appropriate to use a normal distribution
3. The first phase of explosion experiments for the Joint when the ratios can diverge so far from one. It is better
Industry Project ‘Blast and Fire Engineering for Top- to use a log-normal distribution. This has the advantage
side Structures’ (BFETS) Phase 2 performed in a full- that unphysical negative values are not obtained, and a
scale rig at Spadeadam (Selby & Burgan, 1998) factor of two (say) overprediction is as far away from
(see below). one as a factor of two underprediction. If a normal distri-
4. The ‘1/6th scale’ vented explosion rig at our Buxton bution were used, quite different results would be
test site: obtained if the ratio of predicted to observed were
앫 The one-sixth scale versions of the SOLVEX assessed rather than the ratio of observed to predicted.
experiments (Bimson et al., 1993), using methane, The log-normal distribution is a normal distribution of
propane and ethene. the logarithms of the values (in this case, ratios). The
앫 The ‘S-series’ in the same rig included a statisti- width of the distribution will be expressed here as the
cally-designed set of variations of height, width, standard deviation of the natural logarithm of the values;
vent area (50 and 100%), side venting (with or this will be denoted the logarithmic standard deviation
without), number of grids (one or two), stoichi- (l.s.d.). It can be converted to a pressure ratio by apply-
ometry (approx. 1.1 or rich), and fuel (three fuels). ing the exponential function.
In addition, extra experiments were performed to It is clear in Fig. 7 that the mode of the distribution
cover the effects of a smaller vent (25%), and is near one, but that the distribution is skewed. This
further variations in stoichiometry. Eighty-seven again reflects the fact that the model gives a good rep-
experiments were performed in all. resentation for many cases, but there are some circum-
앫 Several other series, including use of sharp-edged stances where conservative assumptions have been
grids as well as round, with up to six grids, with taken. There are several ways in which a distribution
and without side venting, and variations of vent could be fitted to these data. It would be possible to use
location and shape. a single log-normal distribution, but this would not
5. The ‘1/6th scale’ rig at Buxton was replaced by a reflect the skew in the data, and would significantly over-
larger rig, which at its maximum is one quarter of the estimate the upper tail of the distribution. Thus we have
size of the SOLVEX box. This has greater flexibility chosen to fit the upper half of a (log-)normal distribution
for variations in geometry. The further series of to the values above one, and the lower half of a wider
experiments include the following: (log-)normal distribution to the values below one. The
앫 1/4 scale SOLVEX. results are shown as the points in Fig. 7. The tails of the
앫 Small and large aspect ratios, with length/width distribution are represented well. Numerically, the
ratio down to 0.74 and up to 4.0, and length/height results are that 38% of the points are in the upper distri-
from 1.1 to 6.0. This series included vents at both bution, with an l.s.d. of 0.427; the lower distribution has
ends as well as single vent, and central and end an l.s.d. of 0.666.
ignition. The scatter in the ratio of prediction to observation
앫 Variation of spacing between grids, including rows
of pipes close together, staggered and in line.

The total data set comprises 308 experiments.

For comparison with the SCOPE predictions of
internal overpressure, it is appropriate to take the highest
(peak) pressure seen by each sensor inside the enclosure,
and either average over all such sensors or, better, take
the median value. The latter has been done for some of
these experiments. However, for many of the experi-
ments where there was little variation of peak pressure
with location within the enclosure, the highest value has
been taken. This gives a slightly conservative assess-
ment.
A plot of SCOPE 3.12 predictions against experi- Fig. 6. SCOPE 3.12 validation — peak internal pressure.
 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431 429

Fig. 7. The distribution of the observed/predicted ratio for peak internal pressure.

comes from two sources: the inability of the model to

give perfect predictions, and experimental variability.
We have analysed the repeatability of the experiments.
From these results we have taken the long-term experi-
mental variability (logarithmic standard deviation) to be
0.15. The corresponding variance should therefore be
subtracted from that of the measured/predicted ratio to
estimate the true scatter of the model predictions. As a
result, the l.s.d. for the upper half distribution becomes
0.400, and for the lower half distribution 0.649. This
implies that 3.2% of cases would be underpredicted by
more than a factor of two.
There is, of course, particular interest in the cases with
the largest underprediction. In Fig. 6 there are several
Fig. 8. SCOPE 3.12 validation — peak pressure in the external
points just outside the factor of two, but only five where explosion.
the underprediction ratio was above 2.24. Four out of
these five were experiments with a single grid, placed
close to a 100% end vent. Fortunately, this arrangement outs, most used simple, regular grids. In assessing the
is rarely encountered in practice. The worst underpredic- practical use of the model, it should be noted that some
tion relates to a DnV experiment which was repeated; further variability is associated with translating a com-
the repeat gave a lower overpressure. plex layout into the series of grids used by SCOPE.
The predicted and measured pressure in the external
explosion can also be compared. The comparison plot 13.1. Validation against ‘BFETS Phase 2’ data
for SCOPE 3 is shown in Fig. 8. There are fewer points
in this plot because in a number of experimental series, Phase 2 of the Joint Industry Project on Blast and Fire
including DnV, pressure in the external explosion was Engineering for Topside Structures has been co-ordi-
not measured. nated by the Steel Construction Institute (Selby & Bur-
This analysis has allowed us to estimate the ability of gan, 1998). The experiments, performed by BG Tech-
the model to predict well-defined experiments. Although nology, consisted of two series. The first series used an
some of the experiments involved complex obstacle lay- enclosure 8×8×25.6 m long, open only at the two ends,
430 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431

Table 1
SCOPE predictions for “BFETS Phase 2” Joint Industry Project

Test number Congestion density Ignition location Median peak overpressure (mbar)
Observed Predicted SCOPE 2 Predicted SCOPE 3

1 “Low” Central 410 690 460

6 “Low” End 280 780 420
7 “High” End 2000 1070 1540
12 “High” Central 1770 1690 2290

which was filled with a near-stoichiometric mix of natu- SCOPE is designed for use with confined, vented
ral gas and air. The first two experiments in this con- modules, and the current version cannot be applied in
figuration had a ‘low density’ obstacle congestion, very open conditions. The configuration of the second
designed as a scale-up of some medium-scale experi- series of experiments included too much venting for
ments performed in Norway some years ago. Then more SCOPE to run, so SCOPE predictions were not provided
congestion was added, including some pipework, and for this series.
representations of cable trays and pipe racks, to achieve
the ‘high density’ congestion.
Predictions using the unmodified SCOPE 2 model (as 14. Further developments
used since 1993) were submitted to the project. These
are summarised in Table 1; the ratios of predicted to A similar modelling approach can also be used for
observed median peak overpressure are plotted in the simulation of explosions in unconfined but congested
Fig. 9. It may be noted that in no case was there an plant. In that case, a ‘one-dimensional’ approach results
underprediction by more than a factor of two. Since we in a hemispherical flame expanding in an assumed
were using a factor of two as a guideline to the accuracy roughly symmetrical region of congestion; ignition is
of SCOPE (expecting 90% of cases to be within this taken to be at the centre.
factor), this could be regarded as satisfactory. However, Another enhancement that has recently been com-
it was clear that the increase in overpressure between the pleted is the modelling of the effect of water deluge on
‘low density’ and ‘high density’ cases was significantly explosion development. Provided the water sprays gen-
underestimated. These results prompted the experimental erate fairly large droplets, and there is sufficient flow
work on obstacle complexity and the related changes to acceleration in the development of the explosion, water
the model included in SCOPE 3. The Table shows that deluge can have a large mitigating effect. Thus quantifi-
the predictions of SCOPE 3 are an improvement. The cation of this effect is of considerable interest.
reductions of overprediction for test number 6 are princi-
pally due to the introduction of explicit modelling of rear
vents in the model. 15. Conclusions

A phenomenological modelling approach can provide

a useful tool in the range of methods applied for the
prediction of explosion overpressures. With less need for
details of the small objects in the geometry, screening
runs can be done earlier or more easily than when using
CFD. Furthermore, computer run time is short; so large
numbers of sensitivity runs can easily be performed.
Despite the relative simplicity of the approach, its
basis in describing the physics of the various processes
involved allows much more confident extrapolation than
can be done with simple correlations.
SCOPE 2 validated reasonably well against the data
from the BFETS Phase 2, in no case underpredicting by
more than the claimed error bounds. However, it was
clear that insufficient account was being taken of the
effects of obstacle complexity. This has been corrected
Fig. 9. Ratio of predicted and observed median of measured peak in SCOPE 3. Other significant changes include a revised
overpressures in BFETS Phase 2, for SCOPE 2 and SCOPE 3. turbulent burning velocity formulation (including the
 J.S. Puttock et al. / Journal of Loss Prevention in the Process Industries 13 (2000) 419–431 431

effect of stoichiometry changes on laminar burning velo- Harrison, A. J., & Eyre, J. A. (1987). External explosions as a result
city and Markstein number), allowance for vents in the of explosion venting. Combustion Science and Technology, 52, 92.
Hjertager, B.H., Enggrav, S., Forrisdahl, J.E., & Solberg, T. (1991)
rear wall, treatment of gas mixtures and the effect of A case study of gas explosions in a process plant using a three-
varying ambient temperature and humidity. dimensional computer code. Guidelines for evaluating conse-
SCOPE 3 has been validated against data from over quences of fires and explosions from vapor clouds and BLEVE’s.
three hundred explosion experiments. AIChE Center for Chemical Process Safety.
Hoerner, S. F. (1950). Fluid dynamic drag. Midland Park, NJ: The
author.
References Kee, R.J., Grcar, J.F., Smooke, M.D., & Miller, J.A. (1985). A Fortran
program for modeling steady laminar one-dimensional premixed
Abdel-Gayed, R. G., Al-Khishali, K. J., & Bradley, D. (1984). Turbu- flames. Sandia report SAND85-8240.
lent burning velocity and flame straining in explosions. Proceed- Kwok, K. C. S. (1986). Turbulence effect on flow around circular cyl-
ings of the Royal Society of London, A391, 393. inder. Journal of Engineering Mechanics, 112, 1181.
Baker, Q.A., Tang, M.J., Scheier, E.A., & Silva, G.J. (1994). Vapour Lind, C.D., & Strehlow, R.A. (1975). Unconfined vapor cloud
cloud explosion analysis. Proceedings of the 28th Annual AIChE explosion study. Loss Symposium, Florida.
Loss Prevention Symposium, Atlanta, GA, USA. Naudascher, E., & Farell, C. (1970). Unified analysis of grid turbu-
Bimson, S.J., Bull, D.C., Cresswell, T.M., Marks, P.R., Masters, A.P., lence. Journal of Engineering Mechanics, 96, 121.
Prothero, A., Puttock, J.S., & Samuels, B. (1993). An experimental Puttock, J.S. (1995). Fuel gas explosion guidelines—the congestion
study of the physics of gaseous deflagration in a very large vented assessment method. Second European Conference on Major Haz-
enclosure. 14th International Colloquium on the Dynamics of ards Onshore and Offshore, IChemE, Rugby, UK (p. 267).
Explosions and Reactive Systems, Coimbra, Portugal. Roach, P. E. (1987). The generation of nearly isotropic turbulence by
Bradley, D., Lau, K. C., & Lawes, M. (1992). Flame stretch rate as a means of grids. Heat and Fluid Flow, 8, 82.
determinant of turbulent burning velocity. Proceedings of the Royal Searby, G., & Quinard, J. (1990). Direct and indirect measurements
Society of London, A338, 359. of Markstein numbers of premixed flames. Combustion and Flame,
Bradley, D., Gaskell, P. H., & Gu, X. J. (1994). Application of a Reyn- 82, 298.
olds stress, stretched flamelet, mathematical model to computations Selby, C.A., & Burgan, B.A. (1998) Blast and fire engineering for
of turbulent burning velocities and comparison with experiment. topside structures—phase 2: final summary report. Steel Construc-
Combustion and Flame, 96, 221. tion Institute publication 253, Ascot, UK.
Bray, K. N. C. (1990). Studies of the turbulent burning velocity. Pro- Taylor, S.C. (1991). Ph.D. Thesis, University of Leeds, UK.
ceedings of the Royal Society of London, A431, 315. Tseng, L.-K., Ismail, M. A., & Faeth, G. M. (1993). Laminar burning
Brossard, J., Desbordes, D., Leyer, J.C., Saint-Cloud, J.P., Di Fabio, velocities and Markstein numbers of hydrocarbon/air flames. Com-
N., Garnier, J.L., Lannoy, A., & Perrot, J. (1985). Truly unconfined bustion and Flame, 95, 410.
deflagrations of ethylene–air mixtures. 10th International Collo- van den Berg, A. C. (1985). The multi-energy method: a framework
quium on the Dynamics of Explosions and Reactive Systems. for vapour cloud explosion blast prediction. Journal of Hazardous
Cates, A. T., & Samuels, B. (1991). A simple assessment methodology Materials, 12, 1.
for vented explosions. Journal of Loss Prevention in the Process van den Berg, A.C., The, H.G., Mercx, W.P.M., Mouilleau, Y., &
Industries, 4, 287. Hayhurst. C.J. (1995). AutoReaGas-a CFD tool for gas explosion
Degener, M., & Forster, H. (1990). Investigation of flame propagation: hazard analysis. In: Loss Prevention and Safety Promotion in the
influence of turbulence on flame propagation. EC Research Area: Process Industries, 1, p. 349. Eds. J.J. Mewis, M.J. Pasman, E.F.
Major Technological Hazards Project no. EV4T0011-D(B), final de Rademaker, UK.
report (December 1990). van Wingerden, K., Storvik, I., Arntzen, B.J., Teigland, R., Bakke,
Graf, W. H., & Mansour, F. F. (1975). Turbulent drag coefficients of J.R., Sand, I.O., & Sorheim, H.R. (1993). FLACS-93, a new
sharp-edged obstacles. Journal of Hydraulic Research, 13, 127. explosion simulator. 2nd International Conference on Offshore
Gulder, O.L. (1990). Turbulent premixed combustion modelling using Structural Design against Extreme Loads, London.
fractal geometry. 23rd International Symposium on Combustion Warnatz, J. (1981) The structure of laminar alkane, alkene and acety-
(p. 835). lene flames. 18th International Symposium on Combustion (p. 369).