Professional Documents
Culture Documents
Consequence Modelling: Risk Assessment Data Directory
Consequence Modelling: Risk Assessment Data Directory
Consequence
modelling
International Association of Oil & Gas Producers
ublications
Global experience
The International Association of Oil & Gas Producers has access to a wealth of technical
knowledge and experience with its members operating around the world in many different
terrains. We collate and distil this valuable knowledge for the industry to use as guidelines
for good practice by individual members.
Disclaimer
Whilst every effort has been made to ensure the accuracy of the information contained in this publication,
neither the OGP nor any of its members past present or future warrants its accuracy or will, regardless
of its or their negligence, assume liability for any foreseeable or unforeseeable use made thereof, which
liability is hereby excluded. Consequently, such use is at the recipients own risk on the basis that any use
by the recipient constitutes agreement to the terms of this disclaimer. The recipient is obliged to inform
any subsequent recipient of such terms.
This document may provide guidance supplemental to the requirements of local legislation. Nothing
herein, however, is intended to replace, amend, supersede or otherwise depart from such requirements. In
the event of any conflict or contradiction between the provisions of this document and local legislation,
applicable laws shall prevail.
Copyright notice
The contents of these pages are The International Association of Oil and Gas Producers. Permission
is given to reproduce this report in whole or in part provided (i) that the copyright of OGP and (ii)
the source are acknowledged. All other rights are reserved. Any other use requires the prior written
permission of the OGP.
These Terms and Conditions shall be governed by and construed in accordance with the laws of England and Wales. Disputes arising here from shall be exclusively subject to the jurisdiction of the courts of
England and Wales.
contents
1.0
2.0
2.1
2.1.1
2.1.2
2.1.3
2.2
2.2.1
2.2.2
2.2.3
2.3
2.3.1
2.3.2
2.3.3
2.4
Explosion modelling..................................................................................... 22
2.4.1
2.4.2
2.4.3
2.5
2.5.1
2.5.2
2.5.3
2.6
2.6.1
2.6.2
2.6.3
3.0
3.1
3.2
3.3
3.4
4.0
5.0
6.0
6.1
6.2
References .......................................................................... 32
References for Sections 2.0 to 4.0 .............................................................. 32
References for other data sources.............................................................. 34
OGP
Abbreviations:
BLEVE
CFD
CHRIS
CSTR
CV
DAL
DNV
EU
FV
HSE
HVAC
IDLH
JIP
LDx
LFL
LPG
MSDS
PDR
QRA
SLOD
SLOT
SVP
TNO
Onderzoek
TR
UVCE
VCE
OGP
1.0
2.0
OGP
These model most of the consequences set out above apart from smoke. However, they
are designed for onshore studies and not all of the models included will be appropriate
for offshore use, in particular in enclosed modules. The sections below give guidance
on the appropriate use of these models.
In addition, there are freeware packages that can be downloaded for the internet but
these do not come with any training or support, or with any guarantee of code quality;
2
OGP
the commercial packages listed above do include these and come from reputable
organizations with quality management systems.
In addition, freeware calculators may be found for specific consequences (e.g.
BLEVE) but these suffer the same disadvantages listed above for general consequence
modelling.
Com putational Fluid Dynam ics
Computational Fluid Dynamics (CFD) can be used to obtain numerical solutions for
ventilation, dispersion and explosion problems for both offshore platforms and onshore
plants. CFD simulations are becoming increasingly common as the computing power of
standard desktop computers grows. The NORSOK standard Z-013 [21] specifies use of
CFD in its probabilistic approach to explosion risk assessment. The objective of the
probabilistic assessment is to generate realistic (representative) overpressures for an
area based on probabilistic arguments. Ventilation, gas leaks, dispersion as well as gas
explosions are considered by establishing probable explosion scenarios, performing
explosion simulations and establishing probability of exceedance curves.
The application of CFD for gas explosion studies is common for offshore platforms and
is increasingly used onshore in cases where the explosion risk is significant and a
better description of the physics is required in order to give a more robust estimate of
the risk.
CFD simulations essentially solve the conservation equations for mass, momentum and
enthalpy in addition to the equations for concentration and flammable gas effects. The
equations are generally closed using the turbulence model. Most of the
commercially available CFD packages (see below) are based on the Finite Volume (FV)
method which uses an integral form of the conservation equations. Essentially, the
solution domain is subdivided into a number of control volumes (CV) at the centroid of
which lies a computational node where the variable values are calculated. The
conservation equations are applied to each CV and interpolation is used to express
variable values at the CV surface in terms of the centre values.
The most widely used commercially available CFD packages are:
2.1
Release modelling
Release modelling also called discharge or source term modelling is mainly used to
determine the rate at which a fluid is released to the environment in a loss of
containment incident, together with the associated physical properties (e.g.
temperature, momentum).
A simple approach is to calculate the initial rate and to assume that this is constant over
time. This is often used for studies of onshore facilities, especially where the offsite
risk is the motivation for the study.
OGP
A more sophisticated approach is to model the time dependence of the release rate.
This is often used for studies of offshore facilities, where the time dependence has a
significant impact on the likelihood, in particular, of the initial event escalating. The
modelling required is more complex but avoids certain issues that arise when initial rate
modelling is used:
Initial rate modelling can lead to over-prediction of the size of a jet fire over time but
under-predict its duration or the time for which it exceeds a critical length (e.g. to
other equipment)
Initial rate modelling can lead to over-prediction of the impact of toxic gas or smoke
effects
Where gas or non-flashing liquid would be released from an orifice, simple formulae
exist to calculate the initial rate, in particular Bernoullis equation for liquids (strictly,
incompressible fluids).
Some example release rates are shown in Figure 2.2, Figure 2.3 and Figure 2.4 for
selected representative materials. These were obtained using DNVs PHAST software.
Equations for modelling time-varying releases of gas, including blowdown, are given in
the CMPT Guide to quantitative risk assessment for offshore installations [1]. This also
includes a simple method for calculating the flash fraction of a liquid such as
unstabilized crude.
Modelling releases from ruptured pipelines is rather more complex as the pipeline
pressure decreases away from the release point over time and so the flow rate
decreases with time, especially for gases. It is therefore normal to use software tools
for discharge modelling.
OGP
Note: at 1 barg and 5 barg the releases are vapour; at higher pressures they are two-phase.
OGP
Figure 2.4 Release Rates for Kerosene-type Liquid at 20C (density = 714
kg/m 3 )
2.1.2
There is a range of software tools available that include release modelling. As with all
software, its range of validity and limitations need to be understood. For example, the
thermodynamics of mixtures may be modelled by an average equivalent pure
component. However, as computer power increases, this limitation is increasingly
being eliminated in favour of full multicomponent thermodynamics.
Software can model some or all of the following:
Releases from vessels containing liquid that flashes as the pressure decreases
These models are generally appropriate for use onshore and offshore.
When the fluid after release is two-phase, the modelling needs to predict the liquid
droplet size so that the amount of liquid that rains out (falls to the ground or water
surface) can be calculated as part of the dispersion modelling (Section 2.2).
SPT Groups OLGA software (http://www.sptgroup.com/products/olga) can be used to
model time dependent releases from pipeline networks and includes multiphase flow
capability.
It should be noted that a release from a high pressure reservoir will normally be quite
complex with sonic flow, expansion and compression shocks. In safety studies, this
OGP
complex outflow is often not calculated and the boundary conditions for the jet are
given at surrounding pressure. Both the specified momentum and the temperature
(density) of this jet may be important for the dispersion simulation and thereby the
resulting gas cloud size. Often this boundary condition is specified as pure gas at sonic
velocity at surrounding pressure or lower. This is not conserving momentum and
should not be used when momentum is important for dispersion.
2.1.3
Following a full bore rupture there will be flow from both sides of the break. The
consequences of a full bore rupture of a buried pipeline can be modelled as follows:
1. Initial high flow rate: consider immediate ignition as a fireball, using mass released
up to the time when this mass equals the fireball mass giving the same fireball
duration.
2. Ensuing lower flow rate(s): model dispersion and delayed ignition with low
momentum (velocity) as the flows from both sides of the break are likely to interact.
The following figure illustrates a possible simplification into quadrants of release
directions for a leak from a buried pipeline. The text beside suggests an approach to
modelling these for medium and large leaks, based on these having sufficient force to
throw out the overburden (and even concrete slabs, if placed on top).
1.
Vertical release.
Model as vertical release
(upwards) without modification of normal discharge
modelling output, i.e. full discharge velocity.
2, 3. Horizontal release.
Model at angle of 45
upwards with velocity of 70 m/s.
4.
Downward release. Model as vertical release
(upwards) with low (e.g. 5 m/s) velocity to reflect loss of
momentum on impact with ground beneath.
For small horizontal or downward leaks, the force exerted by the flow is unlikely to
throw out the overburden, hence the flow will only slowly percolate to the surface. The
following approach is suggested for all release directions:
Remodel release with a very low pipeline pressure (1 barg for operating pressure
>10 barg, 0.1 barg for operating pressure < 10 barg), to simulate diffusion through
the soil, with the hole size modified to obtain the same discharge rate as above.
2.2
Dispersion modelling is used to determine how the fluid released spreads in the
environment: usually air but also water1.
Onshore, dispersion is usually modelled for releases into the open air
Dispersion in soil is considered in environmental rather than safety risk studies and is outside
the scope of this datasheet.
OGP
When a release is in the open air, several mechanisms may cause it to disperse. These
are illustrated in Figure 2.5. Not all releases go through all phases. A gas release on an
offshore platform may go directly from turbulent jet to passvie dispersion. A release
from a stack may be passive from the stack tip. The vapour in a release of refrigerated
LPG will be dense from the start.
Figure 2.5 Mechanism s of Atm ospheric Dispersion of Vapour
OGP
2.2.1
Very little dispersion modelling can validly be done using simple formulae. That which
can is as follows:
1. Passive (Gaussian) dispersion
2. Gas build-up in enclosed volumes
To calculate the quantity of flammable gas, for explosion modelling (see Section
2.4)
OGP
rate. This diameter together with the gas flow rate can then be used as input to a
Gaussian plume model.
Some example dispersion modelling results (distances to LFL) are given in Figure 2.7
and Figure 2.8. These were obtained using DNVs PHAST software.
Figure 2.7 Dispersion Distances to LFL for Vapour Releases at 20C
Note: F1.5 refers to F stability, 1.5 m/s wind speed; D5 refers to D stability, 5 m/s
wind speed.
10
OGP
Note: F1.5 refers to F stability, 1.5 m/s wind speed; D5 refers to D stability, 5 m/s
wind speed.
2.2.2
Box models, which calculate vapour cloud dimensions and concentrations from
bulk properties.
CFD models, which divide the computational domain representing the space
through which the fluid disperses, into small volume elements where physical
properties are calculated explicitly.
In general, plume models do not allow for the influence of terrain, assuming a flat,
unobstructed surface. Plume models cannot model well the near field characteristics of
dispersion within a congested or confined area such as an offshore module or the
middle of a process unit. However, for far field (i.e. in open areas) dispersion and
when numerous release cases need to be run, plume models are ideal.
The software used needs to be selected with an understanding of the phenomena
(identified in Section 2.2) likely to occur for the cases being modelled, to ensure that the
software can adequately model them. For example:
A Gaussian plume model would not be appropriate for a gas release under pressure,
which will initially disperse as a turbulent jet (see Figure 2.5)
The results from dispersion modelling need to be examined to ensure they are sensible,
i.e. that they match expectations about their behaviour.
OGP
11
12
OGP
Although the NORSOK approach is for offshore installations, a similar approach can be
applied to explosion analysis for onshore installations. CFD modelling of ventilation
and dispersion is also useful for evaluating optimal geometry layout and location of gas
detectors [22,23]. CFD has also found some application in modelling dispersion in
complex topography (e.g. along a pipeline route), although it is not cost-effective to use
it routinely to model explicitly all scenarios typically represented in a QRA.
2.3
Fire modelling is typically used to calculate the flame dimensions for 2 purposes:
To determine whether a flame can reach a target for escalation (e.g. other
equipment)
Flash fire an ignited vapour plume, whose dimensions are typically determined
directly from the dispersion modelling as the distance to LFL
Jet fire an intense, highly directional fire resulting from ignition of a vapour or
two-phase release with significant momentum
Pool fire from an ignited liquid pool2 or sea surface gas pool resulting from a
subsea gas release (e.g. from a pipeline or wellhead)
Offshore installations often have grated decks, so a liquid spill will fall through the
grating onto the sea surface. If ignited, the resulting sea fire may engulf one or more
legs of the installation as well as risers and conductors.
Boilover when a full surface fire occurs in an oil storage tank, heat will slowly
conduct downwards to any layer of water in the bottom of the tank; this will then
vaporise and the resulting expansion will hurl boiling oil upwards out of the tank.
Fireball/BLEVE
Strictly, a BLEVE (Boiling Liquid Expanding Vapour Explosion) is simply explosively
expanding vapour or two-phase fluid. A BLEVE results from a hot rupture of a
vessel typically containing hydrocarbons such as LPG3, stored and maintained as a
liquid under pressure, due to an impinging or engulfing fire. A flammable material
will be ignited immediately upon rupture by the impinging/engulfing fire and will
burn as a fireball.
A fireball would also result from immediate ignition of a release resulting from cold
catastrophic rupture of a pressurised vessel.
The initial phase of a gas pipeline rupture should also be modelled as a fireball.
Crater Fire from ignition of a release from a buried pipeline. For vertical and
horizontal releases (see Section 2.1.3), the corresponding jet fire can be modelled.
For downward releases, the hole size corresponding to the low release velocity can
be taken as the diameter of a gas pool burning as a pool fire.
Note that it is not the liquid that burns but rather the vapour above it. The heat of the flame
vaporises the liquid beneath to provide the fuel supply.
3
BLEVEs of hydrocarbons up to butane or perhaps pentane are credible. A BLEVE of a vessel
containing a toxic material such as chlorine stored as a liquid under pressure is also credible
and should be considered if relevant. BLEVEs of heavier hydrocarbons such as crude oil or
petroleum do not occur.
OGP
13
An appropriate model for the type of fire that could result from ignition of the release
being considered can be selected. This will also depend on the time/location of ignition:
for example, for a high momentum vapour release, ignition close to the source will
result in a jet fire; ignition at a point away from the source will result in a flash fire or
explosion (see Section 2.4), which may also burn back to a jet fire.
Whatever model is selected, the following parameters of the flame have to be
calculated:
Flame dimensions
2.3.1
Some simple models for calculating flame dimensions are given in the sub-sections
below. Calculation of thermal radiation received by a target (e.g. a person) is not
straightforward, although an approximation can be used for a fireball due to its
spherical symmetry (see Section 0), and is best done using software. The simple flame
size models below are therefore best used either when only the flame dimensions are
required or to provide direct input to a flame radiation model.
2.3.1.1 Jet Fire
A simple correlation for the length L (m) of a jet flame due to Wertenbach [5]:
L = 18.5 Q0.41
Based on calculations using the Chamberlain model [7], the following rough
relationships for distance along the flame axis to various thermal radiation levels have
been calculated:
37.5 kW/m2:
13.37 Q0.447
12.5 kW/m2:
16.15 Q0.447
5.0 kW/m2:
19.50 Q0.447
Some example jet fire thermal radiation results for horizontal releases are presented in
Figure 2.9 and Figure 2.10. These were obtained using DNVs PHAST software, which
used the Chamberlain model [7].
14
OGP
Figure 2.9 Jet Fire Therm al Radiation Distances at Ground Level for
Propane Releases at 1 m Elevation
Figure 2.10 Jet Fire Therm al Radiation Distances at Ground Level for
Releases at 10 m Elevation
OGP
15
Mass Burning
Rate (kg/m 2 s)
0.05
0.06
0.05
0.08
0.08
0.14 on land [30]
0.24 on water [30]
0.11 on land
0.22 on water
Burning velocity
(m m /s)
0.07
0.07
0.07
0.11
0.13
0.242
0.422
0.21
0.42
Notes
16
OGP
Note: The shape of the curves for 12.5 kW/m is explained by the decreasing flame surface
emissive power with increasing pool diameter.
OGP
17
2.3.1.3 Boilover
Boilover can be modelled as a pool fire with:
Figure 2.13 shows a procedure to determine the model required for a gas or 2-phase
release. A similar approach can be taken for a liquid release.
Lees [9, pp16/286ff] suggests possible approaches and other models for compartment
fires. Although written as applying to fires inside buildings, the text can also be applied
offshore.
In the former case there is an adequate supply of air to ensure complete combustion of the fuel;
in the latter case the ventilation is limited and the fuel is not fully combusted.
18
OGP
Figure 2.13 Procedure for Fire Model Selection (Gas or 2-phase Release)
Note: in a highly confined volume with limited ventilation (e.g. a platform leg), even a small fire
may be ventilation controlled.
2.3.1.5 Fireball/BLEVE
Several models for fireball duration and diameter have been developed. Most are simple
correlations between these quantities and fireball mass5. One model is due to Prugh
[10]:
Diameter, D (m): D = 6.48 M0.325
Duration, td (s): td = 0.825 M
0.26
When the release is two-phase, the fireball may not consume all the liquid. One possible
assumption is that the fireball mass is calculated assuming 3 the adiabatic flash fraction at the
burst pressure, constraining this to be 1.0.
OGP
19
2.3.2
transmissivity:
The software packages listed in Section 2.0 model the fire types listed in Section 2.3,
apart from compartment fires. They will model the flame dimensions and orientation,
and thence the thermal radiation received.
For a compartment fire, if the fire inside the module is a diffusive fire smaller in volume
than the module, it can be modelled as a pool fire with the dimensions suggested in
Section 2.3.1.4; the surface emissive power can be taken to be the same as that of the
unimpinged jet fire.
2.3.3
CFD models can be used to determine the fire loading on critical areas on both offshore
structures and onshore plants. The Oil and Gas UK guidance [24] provides a state-ofthe-art review of CFD fire modelling. In particular, it is stated that although CFD models
provide a more realistic representation of the flow physics, there are uncertainties
associated with modelling turbulent flow and combustion as well as in definition of fire
source and ambient conditions. Commonly used software for fire modelling include
Kameleon FireEx and CFX. Kameleon FireEx is typically used for fire modelling on
offshore platforms and onshore plants; CFX is more commonly for low geometry
scenarios, e.g. fire and smoke modelling in tunnels.
For CFD fire modelling, it may be best to reduce the size of the problem by modelling
only a subset of the installation. Otherwise, the run times for the analyses would be very
long. The procedure for running the fire analyses can be summarised in the following
steps:
1. Define leak size and select realistic leak locations;
2. Select leak directions. Typically, the analyses are run for up to 6 leak directions;
3. Run the fire simulations for different leak rates for each leak location and direction
until steady state conditions are reached.
Huser [25] describes a probabilistic procedure for the design of process against fires
using CFD modelling. The probabilistic assessment provides a Dimensioning Accidental
Load (DAL) fire that is used for design of the structure and allows for the development
of a consistent methodology (similar to explosion approach) for calculating fire loads.
The methodology is illustrated in Figure 2.14.
20
OGP
[25] has shown that for CFD simulations of jet fires the following parameters are
important (i.e. resulting in more than 20% variation in the heat loads when all other
parameters are kept constant):
OGP
21
Jet direction
Deluge
The probabilistic approach can be used to generate a fire exceedance curve from which
the DAL fire can be obtained.
2.4
Explosion modelling
For QRA and associated studies, explosions are usually taken to mean vapour cloud
explosions (VCEs). However, other types of explosion are possible (see Figure 2.1):
Dust explosions
Runaway reactions
In addition, BLEVEs and vessel bursts generate overpressures that may be significant.
However, this section focuses on VCEs.
Huge advances in understanding and modelling of VCEs have been made in the last
decade since the Spadeadam tests. For offshore, the NORSOK standard Z-013 [11] has
established a comprehensive but computationally demanding approach to explosion
modelling, requiring use of an advanced CFD tool. Whilst originally developed
specifically for platforms in Norwegian waters, this approach is being adopted in other
areas of the North Sea. Although CFD models cannot yet be incorporated directly within
(offshore) QRAs, output from QRA is increasingly expected to be used in them.
Onshore, CFD is less well established in QRA whilst the application of simpler models
available in general purpose software is becoming more sophisticated and considered
fit for purpose. However, where design or layout decisions may critically depend on
explosion risks, use of CFD for specific scenarios would give additional robustness to,
and confidence in, the results. Another issue where CFD would assist is where terrain
effects are important, for example if a facility is built on a slope or at the foot of a hill: in
this case dispersion would be significantly modified compared with that which would
result over flat ground.
The recent advances in understand of explosions referred to above mean that the
previous classification of VCEs as unconfined, semi-confined or confined can now be
considered over-simplistic. It would be better to talk about degrees of confinement and
congestion6. TNOs Multi-Energy model [12], discussed further in Section 2.4.2, allows
for 10 levels of confinement/congestion, ranging from the equivalent of a UVCE
(Unconfined Vapour Cloud Explosion) through to highly confined/ congested volumes
such as can be found in a densely packed process area of an onshore plant. In this and
similar models, some assessment or assumption needs to be made outside of the
model as to the maximum overpressure. In CFD modelling, the distinction between
levels of confinement/ congestion disappears since the geometry is defined and the
software itself calculates the maximum overpressure.
OGP
2.4.1
Historically, simple TNT equivalence models have been used for modelling explosion
overpressures from unconfined VCEs (UVCE) onshore. However, these require the
explosive mass to be calculated: as this is an output from dispersion modelling, manual
calculation of explosion overpressures is not likely to be undertaken.
Another old approach for onshore QRA [13] calculates the distance to specified levels
of damage directly from the explosion energy by a simple correlation. Again, this
requires the explosive mass to be calculated.
2.4.2
As in the Multi Energy model, the overpressure vs. scaled distance is a set of curves (in
this case 11) that span the range of input selections.
These models are appropriate for use in studies of onshore facilities including marine
terminals.
2.4.2.2 Offshore explosions
For offshore installations, non-CFD software has been used to estimate maximum
overpressures in modules using relatively simplified methods that nevertheless take
account of the broad features of module geometry. For example, DNV have used their
programs COMEX and NVBANG in numerous studies, however these programs are not
available commercially and are not recommended for non-specialists in explosion
modelling.
However, in offshore applications the maximum overpressure itself is usually not used
directly in the risk calculations. Rather, it represents the worst case combination of
module fill, release location and ignition location. In a real situation, this combination is
OGP
23
The representative gas clouds from the CFD dispersion analysis (see Section 2.2.3) can
be ignited and explosion analysis carried out. The Oil and Gas UK guidance [24] reports
that it is not recommended to use dispersed non-homogeneous and turbulent gas
clouds in CFD explosion simulations due to the lack of testing/validation for this
application. Instead, an equivalent quiescent stoichiometric gas cloud, that gives similar
overpressures to the non-homogeneous and turbulent clouds, has to be calculated.
As an example of how this can be done, the FLACS software automatically calculates a
parameter (referred to as Q5) that converts the non-homogeneous cloud into an
equivalent quiescent gas cloud. It should be noted that the duration of the equivalent
gas cloud may be shorter than the non-homogeneous one resulting in a difference in
the structural response.
The explosion simulations should be carried out for various gas cloud sizes and
shapes, gas cloud locations and ignition locations. For each gas cloud size, the gas
cloud location and ignition location should be varied. In particular, it is important to
locate the clouds close to critical and congested areas of equipment and piping.
The ignition location will also have a strong impact on the explosion loads. Generally,
the CFD analyses are run with two different locations namely ignition location at centre
of cloud and at edge of cloud. Depending on the geometry and layout, edge ignition will
sometimes produce the higher (than central ignition) explosion overpressures due to
the increased flame distance.
Results in terms of explosion overpressures can be output at monitor points at predefined locations and drag forces can be obtained for design of critical equipment and
piping.
2.5
Modelling of smoke and gas ingress to the TR or living quarters usually forms part of an
offshore QRA and could also be used in onshore studies. More generally, modelling of
smoke generation and dispersion can be useful to determine the likelihood of escape
routes being impaired or of people out-of-doors being overcome by smoke.
Smoke and gas ingress modelling has up to 4 stages:
Source Term Dispersion Ingress Effects
24
OGP
The source term comes from the release rate modelling (Section 2.1): directly for gas
and from suitable ratios of (mass of smoke) / (mass of hydrocarbon released).
Dispersion can be modelled as suggested in Section 2.2. Since smokes largest
constituent is nitrogen (i.e. the unburnt part of the air involved in combustion), one
approach used has been to model the smoke as hot, dense nitrogen, giving it a
molecular weight and temperature equal to those estimated for the combustion gases.
However, the high temperature invariably results in a rapidly rising smoke plume that
doesnt match experience. For example, photographs of smoke from the Piper Alpha
disaster show the plume travelling almost horizontally. One possible reason is that the
soot particles in the smoke increase the plumes density. Hence this approach is not
recommended for 3D results. However, it may be used to determine the smoke
concentration at a given distance horizontally from the release point, assuming as a
worst case that this is the centreline concentration.
2.5.1
The CMPT Guide to quantitative risk assessment for offshore installations [1] provides data
and references on smoke generation, composition, dispersion, visibility reduction,
ingress to TR and impact.
A series of linked models has been used in offshore QRAs for BP and other operators:
Smoke generation:
Component
Fuel Type*
Light
Heavy
a) Fuel Controlled
Carbon Monoxide (ppm)
400
800
Carbon Dioxide (%)
10.9
11.8
Oxygen (%)
0
0
Smoke Temperature (C)
1,000
1,000
Particulates (dB/m)
15
47
b) Ventilation Controlled
Carbon Monoxide (ppm)
30,000
31,000
Carbon Dioxide (%)
8.2
9.2
Oxygen (%)
0
0
Smoke Temperature (C)
600
600
Particulates (dB/m)
29
70
* The light composition is used for gas jet fires. The heavy composition is used for
condensate fires.
Dispersion: based on a dilution factor, which is a function of fuel burn rate and of
distance from source (does not take into account wind speed or the presence of
barriers).
Figure 2.15 shows dilution factors, based on calculations using FLACS [17], for
different release rates.
Smoke Ingress:
OGP
25
CO and CO2 build-up in the module are calculated using a CSTR model, taking as
input the smoke concentration immediately outside the TR and the TRs
ventilation rate
The CO2 concentration calculation also includes exhaled CO2 from personnel
inside
Smoke effects/toxicity
Considers toxicity of CO; effects of CO2, lack of oxygen and high air temperature;
visibility reduction
For gas ingress a set of dilution factors is used, equivalent to but different from those
used for smoke. A CSTR model is used for gas ingress, and fatalities in the TR are
assumed to occur if the gas concentration exceeds 60% of LFL.
2.5.2
For smoke dispersion in the open, general purpose consequence modelling software
such as the packages listed in Section 2.0 is sometimes used. However, the validity of
this approach and its superiority to the simple approach described in Section 2.5.1 are
uncertain.
For smoke and gas build-up within modules, multizone models such as COMIS can be
used. Multizone modelling involves solving mass balance equations for the flow
between different zones, thus allowing for partitioning due to smoke barriers, walls
between rooms, etc. Multizone models were developed primarily to predict airflow in
buildings, but they are also capable of predicting the transient transport of
26
OGP
CFD modelling can be used to provide a detailed prediction of the smoke distribution in
TR or living quarters. The effect of heat sources due to people and computing
equipment can be included in the analysis. However, smoke modelling using CFD can
be quite difficult due to the variability and uncertainty in the boundary conditions [26]. A
recent article by ODonnell et. al. [27] provides a comparison of different approaches to
smoke modelling namely the CSTR model, a multizone model and a CFD model. CFX
and Kameleon FireEx can be used to carry out detailed CFD smoke modelling.
The smoke and gas dilution factors used in the models described in Section 2.5.1 were
determined using FLACS, a CFD package. This or another CFD package could be used
directly to model smoke dispersion from a source in the same way as described in
Section 2.2.3 for gas dispersion modelling in general. However, the approach described
in Section 2.5.1 has generally been accepted as fit for purpose in QRAs.
CFD is more likely to be useful in design, for example in locating HVAC air intakes to
minimise the likelihood of smoke ingress. Although best practice is to place them on
the TR face away from potential smoke sources (i.e. fires), flow around bluff bodies
results in zones of recirculation and hence of enhanced smoke concentration.
2.6
Toxicity modelling
The toxic effects of a material may be acute (resulting from accidental exposure to a
high concentration over a short period of time) or chronic (resulting from continuous
exposure to a lower concentration over a long period of time, as a result of routine
emissions or a small, undetected leak). Different toxic materials have different
physiological effects: they may inhibit respiration (causing asphyxiation) through
inhalation, they may affect the central nervous system, they may be ingested or
absorbed through the skin. For the purposes of this datasheet, the discussion is limited
to acute effects and it is not necessary to consider the nature of the physiological
effects. The discussion addresses toxicity on the basis of dose-response relationships
(see below).
Offshore, besides smoke (as discussed in Section 2.5), toxic modelling is usually limited
to the effects of sour gas, i.e. H2S.
Onshore, besides H2S (in onshore hydrocarbon production, transport and processing),
other toxic materials are potentially of concern. Toxic consequences are invariably
bound up with toxic effects: that is, a model for toxicity is a model for lethality or lesser
effects.
Toxicity data is typically encountered in two forms when required for QRA: specified
concentrations such as the IDLH (Immediate Danger to Life and Health), or
concentration-lethality levels for different species such as rats. Such data can be found
in Material Safety Data Sheets (MSDS) or online reference sources such as CHRIS
www.chrismanual.com.
OGP
27
For QRA, a dose-response relationship is often used that relates the lethality to the dose
received at a point. At its simplest, the dose is given by (concentration time),
assuming the concentration remains constant over time. However, for many materials,
the effect of concentration is magnified and, for concentration C and exposure time t,
the relevant dose A is given by:
SLOT (Specified Level Of Toxicity) Dangerous Toxic Load: the dose that results in
highly susceptible people being killed and a substantial portion of the exposed
population requiring medical attention and severe distress to the remainder
exposed. It represents the dose that will result in the onset of fatality for an
exposed population (commonly referred to as LD1 or LD1-5)
Values of the SLOT and SLOD for selected materials are given in Table 2.3. As can be
seen in the final column, values of n for these materials range from 1 to 4.
Table 2.3 SLOT & SLOD Values for Selected Materials
Substance
Ammonia
SLOT
8
3.78 10
Carbon monoxide
40125
Chlorine
Hydrogen sulphide
1.08 10
12
2.0 10
Sulphur dioxide
4.66 10
Hydrogen fluoride
Oxides of nitrogen
SLOD
9
1.09 10
n
2
57000
4.84 10
13
1.5 10
2
4
7.45 10
41000
5
6.24 10
1
2
12000
96000
As stated above, the LD50 is often used in risk calculations. The HSEs approach allows
for calculation of the LD50 for any exposure duration.
The most sophisticated approach to determining toxicity adopts the same approach to
calculating the dose but allows the lethality to be calculated for any given concentration
and duration of exposure. This is the probit. A probit value Pr is calculated (for a
constant release rate and hence concentration7) as:
where a, b and n are all material specific constants (n is the same as above).
These constants have been published for many commonly encountered materials in a
number of sources [e.g. 9,20]. A table relating lethalities to probits can be found in
many places e.g. [9].
For a time varying release rate and hence concentration, the (C t) can be replaced by an integral
over time.
28
OGP
2.6.1
The software listed in Section 2.0 will calculate probits for toxic materials and thence
the lethality level as a function of distance from the release point or as contours of
different lethalities overlaid on a plan or map. In this way the lethality at any point can
be determined for a given wind direction.
2.6.3
CFD will provide as output the concentration at any point. This could be used together
with a SLOT/SLOD value or probit to calculate lethality at that point. Contour plots of
toxic lethality are not available from CFD software but could probably be generated from
tabular output.
3.0
3.1
General validity
The approaches described in Section 2.0 are based on published sources that are
widely known and accepted.
All modelling of physical phenomena is imperfect. Any use of software must be within
the limitations set out for the software, and even then the analyst must carry out a
reality check on the results. For example: a jet fire model applied to a large, high
pressure gas release will predict a jet flame several hundreds of metres long; the
analyst must consider whether this is credible, or whether the flame will impinge on an
obstruction within this distance.
Depending on the application, a simple model may be fit for purpose, or detailed
modelling (e.g. using CFD) may be required. Whilst it may be considered desirable to
use CFD as much as possible, the resources (time, trained personnel, and budget)
required to use it effectively are rarely available; hence it is usually used to address
specific problems or to provide results for a limited set of scenarios that can be applied
or extrapolated to all the scenarios being modelled in a QRA.
In the early stages of design, the detailed design information required for CFD to give
accurate predictions of overpressures is not available and hence decisions based on
CFD results may result in under-design for the potential overpressures.
OGP
29
3.2
Uncertainties
All modelling suffers from uncertainties. For a given set of input (initial) conditions, it is
unlikely exactly to match the physical outcome that would result in reality from the
same initial conditions. Indeed, numerous physical realisations of the same release
would give different results, whereas consequence modelling software gives the same
result each time8. Sources of uncertainty in consequence modelling for QRA include
the following:
A QRA only models a limited range of cases, so the conditions of an actual release
are unlikely to match exactly any of the cases modelled in a QRA
Ambient conditions (wind speed, wind direction) do not stay constant over the
duration of a release as is modelled
Box models for dispersion, and models of equivalent complexity for other
phenomena, cannot deal with solid or porous barriers (buildings, process units,
bund walls, etc.)
3.3
As set out in Section 2.0, whilst simple models are available for some consequences,
and a range of numerical results for some consequences are given there, some
consequence modelling requires the use of either general purpose or CFD software. To
decide which is the best approach it is necessary to decide:
Will the analysis need to be updated in the future, or the results interrogated? If so,
who will do this?
If the scope is a full, detailed QRA, then most or all of the 6 steps described in Section
2.0 will need to be undertaken. This means that the output from one step of the analysis
will become the input to the next step, and it is important to make the links between the
steps as straightforward and robust as possible. This in turn suggests that general
purpose consequence modelling software where the transitions from one model to the
next are automated is preferable to using a mixture of models from different sources
(perhaps with some implemented in spreadsheets, others coded). However, in this case
the automated transitions may be black box-like and so the analyst needs to
understand fully how these work to ensure that the results represent physical reality.
(For example, that a modelled jet fire is a credible outcome.)
If a coarse QRA of a simple installation is to be undertaken, a simpler approach may be
acceptable. This could use the correlations given or referred to in Section 2.0, or the
consequence results presented in that section.
Monte Carlo modelling could be used to vary slightly the input parameters but this does not
appear to be done routinely. Another type of dispersion modelling, random walk modelling,
likewise does not appear to be used for QRA.
30
OGP
For a QRA of an offshore installation with enclosed modules, use of CFD for explosion
modelling is now routinely used. For a new installation, it will almost certainly have to
be used in order to design for explosions. For an existing installation, explosion
modelling predating the Blast and Fire Engineering for Topside Structures JIP will probably
have been revised using CFD. Thus it is likely that the necessary CFD modelling will
have been done, or at least that the geometry model has been built and it will be
relatively straightforward to obtain any additional results required.
For QRAs of onshore installations, use of the TNO Multi Energy Model or the Baker
Strehlow Tang model (see Section 2.4.2.1) is strongly recommended over use of earlier
VCE models.
For problems of a more limited nature, in particular decisions about significant
investment in relation to fire or explosion and especially in relation to offshore
structures, it is advisable to use CFD in order to maximise the robustness of the
analysis and the confidence in the results.
CFD modelling requires considerable experience and expertise to use effectively. It is
rare for a risk analyst skilled in all aspects of QRA to possess the required degree of
specialist expertise. CFD analysis should therefore be assigned or contracted to
personnel with the required expertise.
3.4
Generally, the numerical grid in CFD models is not fine enough to resolve the smaller
items of equipment and pipe work which are responsible for a large part of the
turbulence generated during an explosion. Most of the software (FLACS, EXSIM,
AutoReaGas) uses a so-called distributed porosity concept (Porosity, Distributed
Resistance (PDR) model) to account for the objects which cannot be represented by the
grid. The porosity model is used to calculate the turbulence source terms due to those
small items and the flame speed enhancement arising from flame folding in the sub grid
wake.
Explosion relief panels and yielding walls can also be represented by modifying the
porosity in the region where they occur.
It is important that all the geometric details are properly represented in a CFD model due
to their importance in pressure build-up. The particular areas where gas explosion
analyses are carried out must be modelled with a high degree of accuracy. In the early
design stages, no detailed description of the geometry exists and this may pose a
problem with regard overpressure prediction. There are two ways in which this problem
can be circumvented namely by applying a factor for equipment growth to account and
by adding anticipated congestion to obtain final expected object density and
distribution.
The Oil and Gas UK guidance [24] reports on a detailed investigation of a typical North
Sea integrated deck platform which showed that, for good prediction of overpressures,
definition of all major equipment, boundaries (decks, TR), all piping with diameters >
0.2 m, and primary/ secondary structures with cross-section dimensions > 0.13 m is
required.
In addition, it is important to define the CFD grid to extend quite a large distance from
the area of interest to avoid too strong influence from open boundaries.
OGP
31
4.0
Key general sources for suitable consequence modelling methods are the Guide to
quantitative risk assessment for offshore installations [1] and Lees Loss Prevention in the
Process Industries [9]. These have been supplemented by more specific published
papers and books as listed in Section 6.1: all of these are believed to have found wide
acceptance in the QRA community including with regulatory authorities.
The general purpose software packages listed in Section 2.0 are all commercially
available. Validation data for them, if required, should be sought from the software
providers. The EU SMEDIS project [28] in particular has compared the leading
dispersion models with results from experimental measurements.
The basis of the suggested approach to modelling releases from buried pipelines
(Section 2.1.3) is confidential work carried out by DNV on behalf of clients (personal
communication). No published methodology has been found.
The basis of the suggested approach to modelling boilover (Section 2.3.1.3) is the Dyfed
Fire Brigade video of the Amoco Milford Haven refinery tank fire. The flames from the
boilover reached a height of 3000 feet, or about 10 times the tank diameter; however,
they were not continuous or constant to this height over a typical period of interest, and
were partly obscured by smoke. Hence a height of 5 times tank diameter appears
reasonable.
For explosion modelling, FLACS and AutoReaGas have been extensively validated
against experimental data, in particular from the Phase 2 and Phase 3 JIP Blast and Fire
Engineering for Topside Structures experiments carried out at Spadeadam and elsewhere.
FLACS is also currently being validated for hydrogen as part of the EU HySafe
programme. Details of FLACS and AutoReaGas validation are available on their
respective websites (see Section 2.0).
5.0
For further information, the data sources referenced in Sections 2.0 to 4.0 should be
consulted. Some additional references are given in Section 6.2.
On the subject of subsea releases, two major reports 32], [33] were published in 2007
and 2008 and should be consulted if detailed information is required (i.e. if subsea
releases appear to pose a significant risk).
6.0
References
6.1
1. Spouge, J, 1999. A guide to quantitative risk assessment for offshore installations, CMPT
publication no. 99/100, ISBN 978-1-870553-36-0 / 1 870553 36 5. Now available from
the Energy Institute www.energyinst.org.uk.
2. Czujko, J (ed.), 2001. Design of Offshore Facilities to Resist Gas Explosion Hazard
Engineering Handbook, Sandvika: CorrOcean ASA.
3. BP Amoco, CERC and BG Technology, 2000. Workbook on Gas Accumulation in a
Confined and Congested Area, Joint Industry Project Gas Build Up from High Pressure
Natural Gas Releases in Naturally Ventilated Offshore Modules. [Believed to be
available only to sponsors but summarised in the following reference.]
4. Cleaver, R P and Britter, R E, 2001. A Workbook Approach to Estimating the Flammable
Volume Produced by a Gas Cloud, Paper R416, FABIG Newsletter: Issue 30, 5-7.
32
OGP
OGP
33
24. Oil and Gas UK & HSE, 2007. Fire and Explosion Guidance, publication no. EHS24.
Available from Oil and Gas UK http://www.ukooa.co.uk/ukooa/.
25. Huser, A, 2006. Probabilistic procedure for design of process areas against fires,
FABIG Newsletter 44. Available from FABIG www.fabig.com.
26. Talberg, O, Hansen, O R, Bakke, J R, and Wingerden, K. Application of a CFD-based
probabilistic explosion risk assessment to a gas-handling plant, conference paper
available from CMR-Gexcon http://www.gexcon.com/download/ERA_00-Paper.pdf.
27. ODonnell, K, Deevy, M, and Garrard, A, 2007. Assessment of mathematical models
for prediction of smoke ingress and movement in offshore installations, FABIG
Newsletter 48. Available from FABIG www.fabig.com.
28. Daish, N C, Britter, R E, Linden, P F, Jagger, S F, and Carissimo, B, 1999. Scientific
Model Evaluation techniques applied to dense gas dispersion models in complex
situations, Intl Conf. on Modelling the Consequences of Accidental Releases of
Hazardous Materials, CCPS, San Francisco, California, September 28 October 1.
29. Mudan, K S, and Croce, P A, 1988. Fire Hazard Calculations for Large Open
Hydrocarbon Fires, Fire Protection Engineering, Section 2 Chapter 4, Society of Fire
Protection Engineers, National Fire Protection Association.
30. Cleaver P, & Johnson, M, 2004. LNG Behaviour, Fire and Blast Issues related to LNG,
FABIG Technical Review Meeting, London & Aberdeen, October 6 7.
6.2
31. CCPS 1994. Guidelines for Evaluating the Characteristics of Vapour Cloud Explosions,
Flash Fires and BLEVES, New York: American Institute of Chemical Engineers.
32. Fannelp, T K, and Bettelini, M, 2007. Very Large Deep-Set Bubble Plumes From
Broken Gas Pipelines, Report No. 6201, Project No. 99B43, Petroleum Safety Authority
Norway.
33. Tveit, O J, and Huser, A, 2008. Risiko knyttet til gassutslipp under vann. Viderefring
2007, Spredning over havet, Petroleum Safety Authority Norway.
34
OGP
www.ogp.org.uk