Marine Structures 24 (2011) 73–96

Contents lists available at ScienceDirect

Marine Structures
journal homepage: www.elsevier.com/locate/
marstruc

Quantitative assessment of hydrocarbon explosion and fire

risks in offshore installations
Jeom Kee Paik a, *, Jerzy Czujko b, Bong Ju Kim a, Jung Kwan Seo a,
Han Seong Ryu a, Yeon Chul Ha a, Piotr Janiszewski b, Beata Musial b
a
Research Institute of Ship and Offshore Structural Design Innovation, Pusan National University, 30 Jangjeon-Dong, Geumjeong-Gu,
Busan 609-735, Republic of Korea
b
Nowatec AS, Stroket 8, 1383 Asker, Norway

a r t i c l e i n f o a b s t r a c t

Article history: A risk-based design framework should involve both risk assessment
Received 19 January 2011 and risk management. This article introduces and describes a number
Accepted 3 February 2011 of procedures for the quantitative assessment and management of
fire and gas explosion risks in offshore installations. These proce-
Keywords: dures were developed in a joint industry project on the explosion and
Risk-based design
fire engineering of floating, production, storage and off-loading units
Hydrocarbon explosions
(the EFEF JIP), which was led by the authors. The present article
Fires
Quantitative risk assessment reports partial results, focussing on defining the frequency of fires
Frequency analysis and explosions in offshore installations. Examples of the aforemen-
tioned procedures’ application to a hypothetical floating, production,
storage, and off-loading unit (FPSO) are presented. A framework for
the quantitative risk assessment of fires and explosions requires the
definition of both the frequency and consequences of such events.
These procedures can be efficiently applied in offshore development
projects, and the application includes the assessment of design
explosion and fire loads as well as the quantification of effects of risk
control options (RCO) such as platform layout, location and number
of gas detectors, isolation of ignition sources etc.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction

More than 70% of the accidents that occur in offshore installations stem from hydrocarbon explo-
sions and fires, which are extremely hazardous, involving blasts and heat [1], and have serious
consequences for human health, structural safety and the surrounding environment.

* Corresponding author.
E-mail address: [email protected] (J.K. Paik).

0951-8339/$ – see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.marstruc.2011.02.002
74 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

Recent decades have seen a number of explosion and fire accidents in offshore installations, well
documented in statistical data [2]. Fig. 1 presents a photo of the Piper Alpha accident, which occurred
on 6 July 1988, and Fig. 2 a photo of the more recent Deepwater Horizon accident, which occurred on
20 April 2010 in the Gulf of Mexico.
Since the Piper Alpha accident, a substantial amount of effort has been directed towards the
management of explosions and fire in offshore installations. Risk-based approaches have begun to
replace traditional prescriptive approaches in offshore design, and 26 joint industry projects have been
undertaken since 1990 [3]. In spite of these efforts, however, accidents continue to occur, as evidenced
by the recent Deepwater Horizon incident.
The authors are leaders of the ongoing 27th Joint Industry Project on the Explosion and Fire
Engineering of Floating, Production, Storage and Off-loading Units (FPSOs) (EFEF JIP hereafter) [3–5].
The aim of the EFEF JIP is to develop state-of-the-art technologies for the quantitative assessment and
management of the risk of hydrocarbon explosions and fires in offshore installations, a review of which
is presented in Refs. [6,7].
A framework for the quantitative assessment and management of the risks associated with fires and
gas explosions, Refs. [8–11], requires the identification of both the frequency and consequences of these
incidents. This paper describes methods of defining the frequency of fires and gas explosions, and
presents examples of these methods’ application to a hypothetical FPSO. The EFEF JIP also involves
definition of fire and explosion design loads and the damaging structural consequences of the inci-
dents, and the results of these investigations will be presented in future, separate papers.

2. EFEF JIP procedure for fire risk assessment and management

Although hydrocarbon explosions and fires often accompany each other in offshore installations, they
are by nature different phenomena [12,13]. Fire is a combustible vapour or gas that combines with an
oxidiser in a combustion process that is manifested by the evolution of light, heat and flame. The impact
of overpressure from explosions and that of elevated temperature from fire are the primary concerns in
terms of the actions that result from hazards within the risk assessment and management framework.
Fig. 3 presents the EFEF JIP procedure for the fire risk assessment and management of offshore
installations [4]. A set of credible fire scenarios is established via sampling techniques on the basis of
the following eight random variables.

- Wind direction (X1)

- Wind speed (X2)
- Leak rate (X3)

Fig. 1. Piper Alpha accident (6 July 1988).

 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 75

Fig. 2. Deepwater Horizon accident (20 April 2010).

- Leak duration (X4)

- Leak direction (X5)
- Leak position in the x direction (X6)
- Leak position in the y direction (X7)
- Leak position in the z direction (X8)

Fig. 3. EFEF JIP procedure for the fire risk assessment and management of offshore installations [4].
76 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

Computational fluid dynamics (CFD) simulations are performed for each of the scenarios estab-
lished to characterise the time- and space-dependent fire load profiles in terms of temperature and
heat amount. The accuracy of these CFD simulations is significantly dependent on the CFD modelling
techniques employed, and thus prior validation of these techniques through comparison with exper-
imental results is essential.
Risk is defined as the product of fire frequency and consequences, with the former calculated as the
product of leak frequency and ignition probability. The design fire loads can be defined in terms of
temperature and heat amount (flux or dose). The structural consequences of fire loads are identified via
nonlinear structural analysis. If the calculated risk level is greater than the acceptable risk level, then
the system must be redesigned, and/or such risk control options as a fire wall, passive fire protection or
deluge/water spray must be adopted. Acceptable risk level is normally defined in terms of probability of
exceedance of damage of main safety functions or probability of accident escalation.
Similar to the case of the aforementioned CFD simulations, it is also very important that the
modelling techniques employed for the nonlinear structural consequence analysis of fires be validated.
It is strongly recommended that an experimental database of fire loads and structural failure modes
obtained in full-scale test structures be compiled if possible.

3. EFEF JIP procedure for explosion risk assessment and management

Hydrocarbons can explode through ignition when combined with an oxidiser (usually air). Thus,
when the temperature rises to the point at which hydrocarbon molecules react spontaneously to an
oxidiser, combustion takes place. This hydrocarbon explosion causes a blast and a rapid increase in
overpressure.
Fig. 4 presents the EFEF JIP procedure for the explosion risk assessment and management of
offshore installations [4]. Similar to that employed for fire risk assessment and management, a set of
credible explosion scenarios is established via sampling techniques based on eight random variables,
which are listed in Section 2.
In contrast to the fire-related procedure, however, gas dispersion analysis is required to identify the
gas cloud characteristics of each of the explosion scenarios before the explosion load profiles can be
characterised. CFD simulations are employed for such analysis, and – again – the accuracy of their
results is greatly dependent on the modelling techniques employed. It is thus essential that these
techniques be validated through comparison with experimental gas dispersion results prior to their use
in the CFD simulations for gas dispersion analysis.
These simulations result in the following eight variables of gas cloud characteristics.

- Gas volume (Y1)

- Gas concentration (Y2)
- Gas cloud position in the x direction (Y3)
- Gas cloud position in the y direction (Y4)
- Gas cloud position in the z direction (Y5)
- Gas cloud size in the x direction (Y6)
- Gas cloud size in the y direction (Y7)
- Gas cloud size in the z direction (Y8)

Based on the results of the gas dispersion analysis, it is then necessary to establish a set of credible
explosion scenarios that will be applied in CFD simulations to identify the explosion load profiles with
time and space. Again, for the reasons outlined above, it is very important that the modelling tech-
niques used for these simulations be validated in the aforementioned manner.
Calculation of the explosion risk requires definition of the explosion frequency and related
consequences, the former of which is defined as the product of leak frequency and ignition probability.
The probabilistic exceedance curve approach is applied to obtain the design explosion loads in terms of
overpressure, drag force and pressure impulse. The consequences can be defined via nonlinear
structural analysis for explosion design loads. Again, the modelling techniques employed in this
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 77

Fig. 4. EFEF JIP procedure for the explosion risk assessment and management of offshore installations [4].

analysis must be validated, and it is strongly recommended that an experimental database of explosion
loads and structural failure modes obtained in full-scale test structures be compiled if possible. If the
calculated risk level is greater than the acceptable risk level, then the system must be redesigned and/
or such risk control options as blast walls or blast partitions (decks) must be adopted.

4. Frequency analysis of explosions or fires

Fire and gas explosions cannot occur without ignition even in the event of a flammable gas or oil
leak. Fire or gas explosion frequency is expressed as the frequency per offshore platform per year, and
can be calculated as follows.

Explosion=fire frequency ¼ leak frequency  ignition probability: (1)

In current maritime industry practice, the following approaches are usually applied to analyse the
frequency of fires and explosions [8].

- Historical accident frequency data

- Fault tree analysis
- Theoretical modelling
- Event tree analysis
- Human reliability analysis
- Judgemental evaluation
- Bayesian analysis
78 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

Fig. 5 represents a sample calculation of leak frequency using event tree analysis. However, the
above-mentioned approaches are often unable to identify gas leak frequency and ignition probability
properly, and more refined methods for their calculation are thus required on the basis of simulations.
The EFEF JIP has developed more refined methods to calculate the frequency of fires and explosions.
Applied examples of the methods are presented in the next section.

5. Applied examples

5.1. Topside of hypothetical FPSO

The topside of a hypothetical very large crude oil carrier class FPSO is now considered as an applied
example. Fig. 6 depicts the layout of this topside [14]. The accommodation is located in the bow or stern
area, as it is necessary to maximise the separation between the accommodation (including the prin-
cipal evacuation systems) and the major hydrocarbon hazards. The topside modules are divided into
a process area and a utility area. The former includes space for hydrocarbon-containing equipment, the
flare tower, and compression and separation equipment, whereas the latter includes space for utility
and power-generation equipment.
When designing the layout of topside facilities, the safety of the personnel on board should be
maximised by minimising the potential hazards arising from the process equipment. For example,
a minimal amount of piping should be used, and the piping must be adequately protected from such
hazards as dropped objects and the dynamic flexing of the vessel hull. The process area must also be
located as far as possible from the accommodation, whereas the utility area may be located between
the two.

5.2. Selection of credible fire/explosion scenarios

In the present study, the Latin Hypercube Sampling (LHS) technique was employed to select the
scenarios. The LHS technique is based on the idea of Latin square – an n  n table filled with n different
symbols in such a way that each symbol occurs exactly once in each row and exactly once in each
column. Latin Hypercube is the generalization of this concept to an arbitrary number of dimensions,
whereby each symbol is the only one in each axis-aligned hyperplane containing it. More detailed
information on the basis and application of LHS technique can be found in Ref. [20].

Fig. 5. Explosion event tree analysis for a leak event.

 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 79

Fig. 6. (a) Layout of the FPSO topside considered. (b) Three-dimensional view of the FPSO topside considered.

When sampling a function of N variables, the range of each variable is divided into M equally
probable strata (intervals) as shown in Fig. 7. One sample is chosen from each stratum (e.g., assuming
uniform probability over the stratum). M-th column in the N-th dimension of the hypercube corre-
sponds to the value from M-th stratum of the N-th random variable. Sample points are then placed to
satisfy the Latin hypercube requirements – see Fig. 8. This forces the number of divisions, M, to be equal
for each variable. Also note that this sampling scheme does not require more samples for more
dimensions (variables), which is one of the main advantages of this sampling scheme.
Probability P of each of M samples generated by the LHS technique for N variables is obtained as
follows.

N
1
P ¼ (2)
M
In the present example, 50 scenarios were randomly selected using the Latin hypercube sampling
technique. The method employed was as follows. The probability density distribution of each of the
eight random variables is divided into 50 ranges, with the interval of each range determined to ensure
that the area below the curve between the probability density versus random variable is equal. The
representative value of the random variable for each segment is defined as the average value of that
segment. The centre of the area is taken as the representative value of the segment at the tail of the
probability density distribution. The selected scenarios for fires and explosions are indicated in Tables
A1 and A2, respectively.
80 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

Probability density

X1 X2 Xi X

Fig. 7. Method employed to select the representative value of the random variable for each range.

5.3. Calculation of leak frequency – by combined historical data and simulation

Leak frequency per FPSO per year can be identified on the basis of a historical database as a function
of the leak amount, where this amount ¼ leak rate  leak duration. The discrete model or continuous
probability density function can then be applied to identify the relationship between the leak
frequency and amount. The probability density distribution often follows the Weibull function, which
is given as
 
l  x l1  x l
f ¼ exp  ; (3)
a a a

Fig. 8. Illustration of Latin hypercube sampling technique – for case with two variables and eight samples.
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 81

where x ¼ a random variable, a ¼ a scale parameter and l ¼ a shape parameter. If both the rate and
duration of the leak are known, then its amount can be defined. The leak frequency can be determined
from Equation (3) as a function of the leak amount, which is a random variable.
An illustrative example is given here. The UK Health and Safety Executive (HSE) has compiled
a database of FSPO leak frequencies in association with the leak amount (amount of fuel released) and
number of FPSOs. Fig. 9 shows the HSE measurement database, for which measurements were taken
over the 15 years from 1994 to 2008 from a total of 135 FPSOs operating in UK waters. The database is
classified into three groups based on the amount of fuel released, namely, major, significant and minor
releases, as shown in Fig. 9(a). The definitions and criteria of the three groups are as follows [3].
Major: Potential for a quick and significant impact on the local area, e.g., affecting the temporary
refuge, blocking escape routes, escalating to other areas of the installation, and causing serious injuries
or fatalities. A major leak, if ignited, is likely to result in a major accident, i.e., it is of a size capable of
causing multiple casualties or rapidly escalating to the point of affecting the temporary refuge and
escape routes. The following criteria are considered for three types of major releases.

Total number of major, significant and minor releases

a 50

Major
45 43
Significant
40
Minor

35 33
31
29
30
2727

25 23 23
21
20
19
20 18
17 17 17
16
14 14 15 15
15 13

9
10 8 8 8
7
6
4
5 3 3 3
2 2 2 2 2
1 1
0
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Year

Fig. 9. (a) HSE leak database releases (in terms of amount) from FPSOs operating in UK waters between 1994 and 2008. (b) HSE leak
database (in terms of number of vessels reported and their releases) for FPSOs in UK waters from 1994 to 2008.
82 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

 Gas releases: either a leak amount > 300 kg or a leak rate > 1 kg/s and a leak duration > 5 min.
 Liquid releases: either a leak amount > 9000 kg or a leak rate > 10 kg/s and a leak
duration > 15 min.
 Two-phase releases: either an amount of released liquids > 300 kg or a liquid release rate > 1 kg/s
and a release duration > 5 min.

Significant: Potential to cause serious injuries or fatalities among personnel within the local area and
to escalate within that local area, e.g., by causing structural damage, secondary leaks or damage to safety
systems. A significant leak, if ignited, has the potential to cause an event severe enough to be viewed as
a major accident or to be of sufficient size to lead to significant escalation within the immediate area or
module. The criteria for significant releases fall between those for major and minor releases.

 Gas releases: capable of jet fires between 5 and 10 m lasting for between 2 and 5 min or a release
rate of between 0.1 and 1.0 kg/s lasting for 2–5 min and producing flammable clouds of between 10
and 3000 m3 in size.
 Liquid releases: pool leaks between 2 and 10 m in diameter lasting between 5 and 15 min.
 Two-phase releases: possible combinations of gas and liquids.

Minor: Potential to cause serious injury to personnel in the immediate vicinity, but very unlikely to
escalate or cause multiple fatalities. A minor leak, even if ignited, is not expected to result in a multiple
fatality or significant escalation, but it could cause serious injuries or fatalities local to the leak site or
within the particular module in which it occurs. The following criteria are considered for three types of
minor releases.

 Gas releases: either a leak amount < 1 kg or a leak rate > 0.1 kg/s and a leak duration < 2 min.
 Liquid releases: either a leak amount < 60 kg or a leak rate < 0.2 kg/s and a leak duration < 5 min.
 Two-phase releases: either an amount of released liquids < 1 kg or a liquid release rate < 0.1 kg/s
and a release duration < 2 min.

Table 1 presents details of the frequency of leaks per FPSO over a 15-year period, and Table 2 that of
those per FPSO per year as a function of leak amount. Fig. 10(a) and (b) plot the data in Tables 1 and 2,
respectively.
Within the gas explosion and fire risk assessment framework, the leak amount is often unknown,
although the gas cloud volume or size may be known. In this case, it is necessary to establish the
relationship between the leak amount and the gas cloud volume.
Fig. 11 and Table 3 show a sample relationship between a leak amount and the equivalent stoi-
chiometric gas cloud volume, which is obtained through CFD simulations of gas dispersion which is
described in Section 5.7. The equivalent stoichiometric gas cloud volume is defined such that the
flammable gas cloud volume obtained via these simulations can be converted into an equivalent gas
cloud with a purely stoichiometric gas concentration.
Fig. 12 depicts a sample relationship between a flammable gas cloud volume and the equivalent
stoichiometric gas cloud volume. It can be seen that, in this sample case, the latter is approximately 40%
of the former. Once the gas cloud volume is known, the leak amount can be approximated on the basis
of the relationship between it and the gas cloud volume. Therefore, the leak frequency can also be
defined once the gas cloud volume is known.

5.4. Calculation of ignition probability – by historical data

The ignition probability in a module can be approximated by considering the features of the ignition
type and source. There are two categories of ignition type, namely, continuous ignition and inter-
mittent ignition. Ignition sources can be divided into several categories, such as rotating equipment
(e.g., pumps, generators and compressors), electrical and other equipment, personnel, and hotwork.
Adjustments may be necessary, depending on the condition of the equipment in terms of age, main-
tenance, and manning or technology level [15].
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 83

Table 1
Details of leak frequency per FPSO over 15 years in UK waters.

Year (1994–2008) Number of leaks Number of FPSOs Leak frequency (per FPSO)

Minor Significant Major Minor Significant Major Total

1994 3 8 0 1 3.000 8.000 0.000 11.000
1995 4 6 0 2 2.000 3.000 0.000 5.000
1996 2 7 0 2 1.000 3.500 0.000 4.500
1997 8 14 0 6 1.333 2.333 0.000 3.667
1998 14 17 2 9 1.556 1.889 0.222 3.667
1999 15 21 3 9 1.667 2.333 0.333 4.333
2000 27 27 1 12 2.250 2.250 0.083 4.583
2001 16 17 0 13 1.231 1.308 0.000 2.538
2002 23 18 2 13 1.769 1.385 0.154 3.308
2003 31 13 0 16 1.938 0.813 0.000 2.750
2004 33 15 0 12 2.750 1.250 0.000 4.000
2005 29 19 3 12 2.417 1.583 0.250 4.250
2006 23 17 1 6 3.833 2.833 0.167 6.833
2007 43 20 2 12 3.583 1.667 0.167 5.417
2008 9 8 2 10 0.900 0.800 0.200 1.900
Sum 280 227 16 135 31.226 34.945 1.500 67.665

Leak frequency (per FPSO per year) 2.081 2.330 0.100 4.511

Table 4 presents a sample calculation of ignition probability on an FPSO topside in the case of
continuous ignition. It is assumed that no hotwork such as welding is undertaken on the module
during operation of the offshore installation. In terms of ignition probability in conjunction with
personnel workspace and electrical and other equipment, the exposed areas are approximated. It is
assumed that ignition probability so predicted can be applied to the entire target module. In reality,
however, ignition probability is not constant, but rather is time- and/or space-dependent. More
sophisticated assessments will thus require time- and/or space-dependent ignition probability models.

5.5. Calculation of fire frequency – by combined historical data and simulation

Fire frequency can be defined using Equation (1) with leak frequency and ignition probability. The
former is determined as the leak amount, which is calculated as a product of the leak rate and leak
duration. Ignition probability is calculated using the method described in Section 5.4. Table A1 presents
sample calculations of fire frequency for various leak scenarios with the leak amount known.

5.6. Calculation of explosion frequency – by combined historical data and simulation

Explosion frequency can also be defined using Equation (1) with leak frequency and ignition
probability. The former is determined as the leak amount, which is calculated as a product of the leak
rate and leak duration. Gas explosion CFD simulations are undertaken with gas cloud volume and size,
which are obtained via gas dispersion analysis as described in Section 5.7. In the probabilistic analysis
of explosion loads obtained through such simulations, the leak rate, duration and amount are
unspecified, although the gas cloud volume and size are given. In this case, the relationship between

Table 2
Leak frequency per FPSO per year as a function of leak amount.

Type Leak amount (kg) Density (per FPSO over 15 years) Frequency (per FPSO per year)
Minor 0w1 31.226 2.081
Significant 1 w 300 34.945 2.330
Major >300 1.500 0.100

Total 67.665 4.511

84 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

b 2.5
Leak frequency(per FPSO year)

2.0

1.5

1.0

0.5

0.0
0 200 400 600 800 1000
Leak amount(kg)

Fig. 10. (a) Variation in leak frequency per FPSO over 15 years in UK waters. (b) Leak frequency per FPSO per year as a function of
leak amount.

the gas cloud volume and the leak amount is employed to estimate the latter when the former is
known. Table A2 presents sample calculations of explosion frequency in various leak scenarios when
the equivalent stoichiometric gas cloud volume is known.

5.7. Simulation-based analysis of frequency

Total frequency of leak per year is treated as a deterministic value, but gas dispersion analysis is
necessary to identify the probability of ignition. The gas dispersion analysis is often undertaken based
on the following assumptions.

i
(1) Leak rate is constant and leak duration tleak is limited by amount of fuel in the leaking segment.
This is a very simplified approach, because in reality pressure in the leaking segment and thus the
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 85

12000

10000

Leak amount (kg)

8000

6000

4000

2000

0
1 10 100 1000 10000
Equivalent volume (m3)

Fig. 11. Sample relationship between leak amount and equivalent stoichiometric gas cloud volume.

leak rate will drop due to loss of containment and possibly due to action of emergency blow-down
system. In this case, leak duration is calculated from the formula:

i amount of fuel in the leaking segment ðkgÞ

tleak ¼ (4)
leak rate ðkg=sÞ

i
(2) Probability of ignition of gas cloud in i-th dispersion scenario Pign corresponds to the moment
when all the fuel has leaked. This means that presence of flammable cloud in the period between
leak end and the moment when the cloud has been vented away does not contribute to probability
of ignition.
(3) Each dispersion simulation is stopped when gas cloud reaches steady state in terms of constant
position and size (i.e., position and volume of gas cloud are constant) – see Fig. 13. Position and
volume of the gas cloud for any time point exceeding the simulated time are the same as at the end
of the simulation. Dispersion analysis is normally carried out using modern CFD codes such as
FLACS [21] or similar.

Fig. 14 illustrates the necessary inputs and outputs of the dispersion analysis. The input data
includes:

- Geometry of FPSO
- Segments characteristics (inventory volume, temperature pressure, composition etc)
- Probability distributions of random variables (see Section 3)

Table 3
Relationship between leak amount and equivalent gas cloud volume.

Equivalent stoichiometric gas cloud volume (m3) Leak amount (kg)

V < 10 467.13
10  V < 100 779.27
100  V < 300 1045.45
300  V < 1000 1736.44
1000  V 2107.71
86 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

10000

Equivalent volume (m3)

8000

6000

4000

2000

0
0 5000 10000 15000 20000
Flammable volume (m3)
Fig. 12. Sample relationship between flammable gas cloud volume and equivalent stoichiometric gas cloud volume.

The output data of the dispersion analysis includes:

- Volume of equivalent stoichiometric gas cloud

- Field output for gas concentration
- Lower and upper flammability limits

It is noted that in the procedure leak duration is used only in the post-processing phase. Each
dispersion is calculated until gas cloud reaches a steady state in terms of volume and position.
Probability of gas ignition is calculated using probabilistic model of gas ignition which takes gas
concentration in time and space as input and returns probability of ignition as a function of time. The
procedure developed by Scandpower [22] is useful for this purpose. The main advantages of this
procedure are:

- It accounts for position of the gas cloud and potential ignition sources associated with time-
varying probability of gas detection.
- It accounts for various ignition sources: continuous, discrete and hot surfaces.
- It accounts for effectiveness of isolation of ignition sources upon detection of flammable gas.

Table 4
Sample calculation of ignition probability on an FPSO topside.

Item Gas Age Maintenance Manning Technology Subtotal Number Total

of units
per module
Hotwork 1.14E  04 0.00 0.00E þ 00
Rotating Pump 9.60E  05 0.90 1.00 1.00 0.80 0.72 9 6.22E  04
machinery
Compressor 2.30E  03 2 3.31E  03
Generator 3.50E  03 2 5.04E  03
Electrical equipment 2.60E  06 0.90 1.00 1.00 0.80 0.72 1931 m2 3.61E  03
Other equipment (1) 2.60E  06 0.90 1.00 1.00 0.80 0.72 1931 m2 3.61E  03
Other equipment (2) 1.30E  06 1.00 1.00 1.00 1.00 1.00 896 m2 1.16E  03
Personnel 3.00E  06 1.00 1.00 0.60 1.00 0.60 1931 m2 3.48E  03

Sum 2.08E  02
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 87

Fig. 13. Schematic of simulation time and extrapolation of results for dispersion analysis.

Also, the parameters of the Scandpower procedure [22] are:

(1) Ignition sources in the area, Pif (default value is 0.075).

The Pif parameter scales the ignition probability by the exposure (both with respect to duration
and cloud size). A relatively large contribution to ignition from distributed ignition sources (Pif)
means that the gas cloud accumulation in the module is important for the ignition probability. Pif is
considered the main tuning factor in the model. A basic unadjusted Pif value of 0.075 has been
selected for the reference area of 1500m2  10 m exposed for 3 min without isolation of ignition
sources.

(2) Discrete versus continuous ignition sources, ia and ib (default values are 0.5 and 0.75 respectively).

The relative contributions to ignition probability for discrete ignition sources are described by ib
(before isolation) and ia (after isolation). For large leaks ib has little influence on the ignition probability
if the effectiveness of detection and isolation are adequate. A relatively high value of ib makes the
ignition probability more dependent on leak duration (relatively higher ignition probability will be
predicted for small leaks). A relatively high value of ia makes the ignition probability less dependent of
cloud size and more dependent on leak duration.

(3) Effect of ignition source isolation, Piso (default value is 0.75).

The effect of ignition source isolation on ignition probability is quantified using the parameter
Piso. Piso ¼ 0 means that ignition source isolation has no effect with respect to ignition probability.
Piso ¼ 1 means that ignition source isolation effectively stops all ignition sources in the area. Effective
ignition source isolation will make the ignition probability less dependent on the cloud size,
provided the gas detection system is adequate. If ignition source isolation effectively reduces the
ignition probability upon gas detection, gas cloud buildup after isolation will not contribute to the
ignition probability. Effective ignition source isolation will make continuous ignition sources
dominant (over discrete) for large leaks, and time predicted to ignition must be short. The effect on
continuous ignition sources (that may be hot surfaces) is not immediate. This is taken into account
by the parameter Phot.

(4) Ignition by hot surfaces after isolation, Phot.

For continuous sources there is an additional delay related to the cooling time of hot surfaces. The
probability that the continuous source is still a potential source of ignition has been modelled expo-
nentially decreasing. Phot is the fraction of the isolated ignition sources that can still ignite a flammable
gas cloud. Phot is thus a time-dependent factor.
 88
J.K. Paik et al. / Marine Structures 24 (2011) 73–96
Fig. 14. Input and output from an individual dispersion analysis.
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 89

Fig. 15. Example results from the probabilistic model of gas detection. The solid bars and lines show calculations using field output,
while the transparent bars and dotted lines show results extrapolated based on assumption that gas cloud reached a steady state.

(5) Probability of detection, Pdet.

The probability of gas detection and that isolation has been physically carried out in the module is
denoted by Pdet.
The probability of gas detection Pdet described in Equation (6) above is time dependent and
calculated using a probabilistic model of gas detection which takes gas concentration in time and space
as input and returns probability of gas detection as a function of time. As a simple model, the following
parameters may be used.

- a ¼ distance between gas detectors, [m]

- c ¼ minimal detectable concentration of gas
- f ¼ probability of gas detection when a single detector is exposed to gas with concentration above c
for time tf
- r ¼ probability that a single gas detector is operative

In addition to the above-mentioned variables one has to define a monitoring region where the gas
detectors are located. During i-th time step Dti, there is:
Vi ¼ volume of gas with concentration above c and within the monitoring region
90 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

Fig. 16. Example probability of ignition from discrete ignition sources. The volume of flammable gas was obtained by CFD
simulations.

ni ¼ aV3i ¼ number of gas detectors engulfed by gas with concentration above c

pi ¼ 1  ð1  f ÞDti =tf ¼ probability of gas detection by a single detector
Probability of gas detection over the i-th is then calculated as follows.

 n
pi ¼ 1  1  pi i (5)

Time history of probability of gas detection (probability that gas will be detected before time ti) is
calculated as follows.

P0det ¼ 0 (5a)

 
Pidet ¼ Pi1
det det
þ 1  Pi1 $pi for i ¼ 1; 2; 3; . (5b)

Fig. 15 shows example results obtained from the probabilistic models of gas detection.
During i-th time step probability of ignition from discrete sources Pign_disc and continuous sources
Pign_cont is calculated from the following formulae.

VLFL UFL $Dt

Pign disc ¼ $P $½P $ia $ð1  Piso Þ þ ð1  Pdet Þ$ib  (6)
15000$180 if det
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 91

Fig. 17. Example probability of ignition from continuous ignition sources. The volume first time exposed to flammable gas was
obtained by CFD simulations.

Vnew
P ¼ $P $½ð1  P Þ$ð1  i Þ þ P ð1  ia Þ$ðP $P þ ð1  P ÞÞ (7)
where, ign cont 15000 if det b det iso hot iso

VLFL_UFL ¼ size of ignitable gas cloud at the i-th time step,

Vnew ¼ new volume exposed to flammable gas during i-th time step,
Dt ¼ length of the i-th time step.
The expression for calculating the ignition probability up to and including the i-th time step is given
as follows.

 
Pign i ¼ Pign i1 þ 1  Pign i1 $ Pign disc þ Pign cont  Pign disc $Pign cont (8)

Example results from the model (probabilities of ignition from discrete sources, continuous sources
and total as functions of time) are shown in Figs. 16,17 and 18.
As a simpler approach, representative gas cloud sizes and ignition probabilities are often used. Each
CFD dispersion simulation generates time history of gas cloud size up to steady state. The NORSOK Z-
013 procedure [23] does not specify at which time point the gas cloud size should be taken as
representative for given dispersion scenario. In the present procedure it is proposed to take gas cloud
i
size at the end of the leak (tleak ) and assign to it ignition probability corresponding to the same time –
see Fig. 19. If end of leak is greater than length of CFD simulation, then the last gas cloud size recorded
in the CFD simulation should be taken (based on the assumption that cloud is in steady state till the end
92 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

Fig. 18. Example total probability of ignition (from both discrete and continuous ignition sources). These graphs show the sum of
probabilities presented in Figs. 16 and 17.

of the leak). This approach is rather conservative, as every dispersion will be associated with the
highest possible probability of ignition and in most cases also with the greatest volume of the gas
cloud. On the other hand, such approach is consistent and simple to apply.
total is calculated based on the total frequency of leaks and
The total frequency of ignitions Fign
probabilities of ignition as follows.
P i i
Pdisp $Pign
total total total total i
Fign ¼ Fleak $Pign ¼ Fleak $ P i
(9)
Pdisp
i
where,
total ¼ total frequency of leak events (year1),
Fleak
total ¼ total frequency of ignition events (year1),
Fign
total ¼ total probability of ignition,
Pign
i
tleak ¼ leak duration in i-th dispersion scenario (sec),
i
Pdisp ¼ probability of occurrence of i-th dispersion scenario,
i
Pign ¼ probability of ignition of gas cloud in i-th dispersion scenario,
ij
Pexpl ¼ probability of occurrence of j-th explosion scenario if the cloud from i-th dispersion scenario
is ignited,
Fexpl ¼ frequency of occurrence of j-th explosion scenario (year1).
ij
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 93

Fig. 19. Representative volume of gas cloud and probability of ignition.

Total frequency of ignition represents also the number of explosion/fire events per year (or other
time interval). Figs. 16 and 17 give examples of calculation of probability of ignition from discrete and
continuous ignition sources, respectively, while Fig. 18 illustrates combined effects of both ignition
sources accounted in the probability calculation.

6. Concluding remarks

This paper presents the procedures for the quantitative assessment and management of fire and
hydrocarbon explosion risks in offshore installations developed during the EFEF led by the authors. As
discussed herein, the use of simulation-based methods, such as CFD simulations and nonlinear finite
element methods, is essential in such quantitative risk assessment and management [16,17]. This paper
focuses on the frequency analysis of fires and explosions, whilst other of the EFEF JIP’s results have been
published in separate papers [18,19,24] and more are planned in future publications.
Simulation-based methods incur substantial computational costs and require in-depth engineering
knowledge. Finding a way to reduce the computational burden without sacrificing accuracy constitutes
a significant challenge. The EFEF JIP aims to achieve the following.

- Develop the best procedure for the quantitative assessment and management of fire and explosion
risks.
- Establish the smallest set of credible explosion and fire scenarios possible.
- Develop appropriate modelling techniques for CFD simulations and nonlinear structural conse-
quence analysis.
- Develop the best procedure to define explosion and fire design loads using the probabilistic
approach.
- Develop useful risk acceptance criteria.

The assessment procedures for risk of hydrocarbon explosions and fires developed within EFEF JIP
can be efficiently applied in offshore development projects. The possible applications include the
94 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

assessment of design explosion and fire loads, as well as the quantification of effects of RCO (risk
control options) such as platform layout, location and number of gas detectors, isolation of ignition
sources etc.
As pointed out in Ref. [7], the results of CFD simulations and nonlinear structural consequence
analysis greatly depend on the modelling techniques employed, and it is thus essential that these
techniques be validated through comparison with experimental results before CFD or structural failure
computations begin. In this regard, it is strongly recommended that experimental results be obtained
for use as a reference in such validation. Furthermore, the scale effect is significant in association with
explosions and fires in both CFD simulations and structural failure analysis, and it is thus desirable that
the aforementioned experimental results be obtained using full-scale test structures rather than small-
scale test models.

Acknowledgements

The present study is part of the EFEF JIP (Phase I). The authors are pleased to acknowledge the
support of their partners in this project: the American Bureau of Shipping, ComputIT AS, Daewoo
Shipbuilding and Marine Engineering, Gexcon AS, Hyundai Heavy Industries, and the UK Health and
Safety Executive. This work was also supported for two years by Pusan National University Research
Grant.

Appendix

Table A1
Sample calculations of fire frequency for various leak scenarios.

Scenario Leak Leak frequency Ignition Fire frequency

no. amount (kg) (per FPSO per year) probability (per FPSO per year)
F01 906.773 2.5627E  03 2.08E  02 5.3326E  05
F02 1662.103 2.5627E  03 2.08E  02 5.3326E  05
F03 4622.968 2.5627E  03 2.08E  02 5.3326E  05
F04 3627.480 2.5627E  03 2.08E  02 5.3326E  05
F05 4478.329 2.5627E  03 2.08E  02 5.3326E  05
F06 97.772 2.5884E  01 2.08E  02 5.3861E  03
F07 309.319 2.5627E  03 2.08E  02 5.3326E  05
F08 1102.698 2.5627E  03 2.08E  02 5.3326E  05
F09 151.480 2.5884E  01 2.08E  02 5.3861E  03
F10 142.260 2.5884E  01 2.08E  02 5.3861E  03
F11 285.955 2.5884E  01 2.08E  02 5.3861E  03
F12 1896.404 2.5627E  03 2.08E  02 5.3326E  05
F13 1205.579 2.5627E  03 2.08E  02 5.3326E  05
F14 3354.590 2.5627E  03 2.08E  02 5.3326E  05
F15 955.355 2.5627E  03 2.08E  02 5.3326E  05
F16 2074.801 2.5627E  03 2.08E  02 5.3326E  05
F17 1280.702 2.5627E  03 2.08E  02 5.3326E  05
F18 1232.832 2.5627E  03 2.08E  02 5.3326E  05
F19 1691.120 2.5627E  03 2.08E  02 5.3326E  05
F20 1542.128 2.5627E  03 2.08E  02 5.3326E  05
F21 1389.622 2.5627E  03 2.08E  02 5.3326E  05
F22 842.398 2.5627E  03 2.08E  02 5.3326E  05
F23 1375.690 2.5627E  03 2.08E  02 5.3326E  05
F24 438.009 2.5627E  03 2.08E  02 5.3326E  05
F25 1405.137 2.5627E  03 2.08E  02 5.3326E  05
F26 134.913 2.5884E  01 2.08E  02 5.3861E  03
F27 8797.933 2.5627E  03 2.08E  02 5.3326E  05
F28 298.587 2.5884E  01 2.08E  02 5.3861E  03
F29 959.817 2.5627E  03 2.08E  02 5.3326E  05
F30 2079.683 2.5627E  03 2.08E  02 5.3326E  05
F31 227.127 2.5884E  01 2.08E  02 5.3861E  03
F32 2171.604 2.5627E  03 2.08E  02 5.3326E  05
 J.K. Paik et al. / Marine Structures 24 (2011) 73–96 95

Table A1 (continued )

Scenario Leak Leak frequency Ignition Fire frequency

no. amount (kg) (per FPSO per year) probability (per FPSO per year)
F33 9368.119 2.5627E  03 2.08E  02 5.3326E  05
F34 1730.442 2.5627E  03 2.08E  02 5.3326E  05
F35 303.250 2.5627E  03 2.08E  02 5.3326E  05
F36 3543.961 2.5627E  03 2.08E  02 5.3326E  05
F37 141.140 2.5884E  01 2.08E  02 5.3861E  03
F38 436.565 2.5627E  03 2.08E  02 5.3326E  05
F39 2028.569 2.5627E  03 2.08E  02 5.3326E  05
F40 2479.109 2.5627E  03 2.08E  02 5.3326E  05
F41 3800.438 2.5627E  03 2.08E  02 5.3326E  05
F42 2763.912 2.5627E  03 2.08E  02 5.3326E  05
F43 1090.084 2.5627E  03 2.08E  02 5.3326E  05
F44 581.881 2.5627E  03 2.08E  02 5.3326E  05
F45 1516.910 2.5627E  03 2.08E  02 5.3326E  05
F46 2348.702 2.5627E  03 2.08E  02 5.3326E  05
F47 927.221 2.5627E  03 2.08E  02 5.3326E  05
F48 1534.487 2.5627E  03 2.08E  02 5.3326E  05
F49 1544.305 2.5627E  03 2.08E  02 5.3326E  05
F50 205.259 2.5884E  01 2.08E  02 5.3861E  03

Table A2
Sample calculations of explosion frequency in various scenarios.

Scenario Equivalent Leak Leak frequency Ignition Explosion frequency

no. volume (m3) amount (kg) (per FPSO per year) probability (per FPSO per year)
E01 774.71 1736.44 2.10E  03 2.08E  02 4.37E  05
E02 405.21 1736.44 2.10E  03 2.08E  02 4.37E  05
E03 445.04 1736.44 2.10E  03 2.08E  02 4.37E  05
E04 9001.00 2107.71 2.10E  03 2.08E  02 4.37E  05
E05 1569.39 2107.71 2.10E  03 2.08E  02 4.37E  05
E06 731.68 1736.44 2.10E  03 2.08E  02 4.37E  05
E07 2536.12 2107.71 2.10E  03 2.08E  02 4.37E  05
E08 863.09 1736.44 2.10E  03 2.08E  02 4.37E  05
E09 4390.03 2107.71 2.10E  03 2.08E  02 4.37E  05
E10 2161.05 2107.71 2.10E  03 2.08E  02 4.37E  05
E11 647.58 1736.44 2.10E  03 2.08E  02 4.37E  05
E12 202.63 1045.44 2.10E  03 2.08E  02 4.37E  05
E13 1101.74 2107.71 2.10E  03 2.08E  02 4.37E  05
E14 325.39 1736.44 2.10E  03 2.08E  02 4.37E  05
E15 689.33 1736.44 2.10E  03 2.08E  02 4.37E  05
E16 1438.59 2107.71 2.10E  03 2.08E  02 4.37E  05
E17 1377.27 2107.71 2.10E  03 2.08E  02 4.37E  05
E18 1318.30 2107.71 2.10E  03 2.08E  02 4.37E  05
E19 2058.09 2107.71 2.10E  03 2.08E  02 4.37E  05
E20 525.12 1736.44 2.10E  03 2.08E  02 4.37E  05
E21 818.48 1736.44 2.10E  03 2.08E  02 4.37E  05
E22 65.45 779.27 2.10E  03 2.08E  02 4.37E  05
E23 285.09 1045.44 2.10E  03 2.08E  02 4.37E  05
E24 1153.34 2107.71 2.10E  03 2.08E  02 4.37E  05
E25 484.98 1736.44 2.10E  03 2.08E  02 4.37E  05
E26 365.37 1736.44 2.10E  03 2.08E  02 4.37E  05
E27 1261.45 2107.71 2.10E  03 2.08E  02 4.37E  05
E28 1502.53 2107.71 2.10E  03 2.08E  02 4.37E  05
E29 1713.40 2107.71 2.10E  03 2.08E  02 4.37E  05
E30 1051.59 2107.71 2.10E  03 2.08E  02 4.37E  05
E31 2693.84 2107.71 2.10E  03 2.08E  02 4.37E  05
E32 1639.54 2107.71 2.10E  03 2.08E  02 4.37E  05
(continued on next page)
96 J.K. Paik et al. / Marine Structures 24 (2011) 73–96

Table A2 (continued )

Scenario Equivalent Leak Leak frequency Ignition Explosion frequency

no. volume (m3) amount (kg) (per FPSO per year) probability (per FPSO per year)
E33 2877.41 2107.71 2.10E  03 2.08E  02 4.37E  05
E34 3098.09 2107.71 2.10E  03 2.08E  02 4.37E  05
E35 244.28 1045.44 2.10E  03 2.08E  02 4.37E  05
E36 1874.41 2107.71 2.10E  03 2.08E  02 4.37E  05
E37 3759.52 2107.71 2.10E  03 2.08E  02 4.37E  05
E38 606.33 1736.44 2.10E  03 2.08E  02 4.37E  05
E39 1206.52 2107.71 2.10E  03 2.08E  02 4.37E  05
E40 1962.96 2107.71 2.10E  03 2.08E  02 4.37E  05
E41 159.68 1045.45 2.10E  03 2.08E  02 4.37E  05
E42 2397.45 2107.71 2.10E  03 2.08E  02 4.37E  05
E43 3376.75 2107.71 2.10E  03 2.08E  02 4.37E  05
E44 2273.44 2107.71 2.10E  03 2.08E  02 4.37E  05
E45 908.61 1736.44 2.10E  03 2.08E  02 4.37E  05
E46 565.54 1736.44 2.10E  03 2.08E  02 4.37E  05
E47 1002.76 2107.71 2.10E  03 2.08E  02 4.37E  05
E48 1791.48 2107.71 2.10E  03 2.08E  02 4.37E  05
E49 955.13 1736.44 2.10E  03 2.08E  02 4.37E  05
E50 114.58 1045.44 2.10E  03 2.08E  02 4.37E  05

References

[1] HSE. Accident statistics for floating offshore units on the UK continental shelf (1980–2003). HMSO RR 353. London: Health
and Safety Executive; 2005.
[2] Vinnem JE. Offshore risk assessment: principles, modelling and applications of QRA studies. 2nd ed. London: Springer;
2007.
[3] Paik JK, Czujko J, Explosion and fire engineering and gas explosion of FPSOs (phase I): hydrocarbon releases on FPSOs –
review of HSE’s accident database. Final Report No. EFEF JIP-02. Busan, Korea: Research Institute of Ship and Offshore
Structural Design Innovation, Pusan National University; 2009.
[4] Paik JK, Czujko J, Explosion and fire engineering of FPSOs (phase II): definition of fire and gas explosion design loads. Final
Report No. EFEF-03–R2. Busan, Korea: Research Institute of Ship and Offshore Structural Design Innovation, Pusan National
University; 2010.
[5] Czujko J, Paik JK, Explosion and fire engineering of FPSOs (phase II): definition of gas explosion design loads. Final Report
No. EFEF JIP-04–R1. Busan, Korea: Research Institute of Ship and Offshore Structural Design Innovation, Pusan National
University; 2010.
[6] HSE, Review of analysis of explosion response. Report No. OTO 98174. London: Health and Safety Executive; 1999.
[7] Paik JK, Czujko J, Assessment of hydrocarbon explosion and fire risks in offshore installations. The IES Journal Part A: Civil
& Structural Engineering, in press.
[8] Spouge J. A guide to quantitative risk assessment for offshore installations. Aberdeen: The Centre for Marine and Petro-
leum Technology, UK; 1999.
[9] IMO. Guidelines for the application of formal safety assessment (FSA) for use in the IMO rule-making process. Marine
Safety Committee MSC/Circ. 1023, MEPC/Circ. 392. London: International Maritime Organization; 2002.
[10] ISO. Risk management vocabulary, guidelines for use in standards. ISO/IEC Guide, ISO Guideline 73. Geneva: International
Organization for Standardization; 2002.
[11] HSE. Fire and explosion guidance: part 1 avoidance and mitigation of explosions. London: Health and Safety Executive; 2003.
[12] Bjerketvedt D, Bakke JR, van Wingerden K. Gas explosion handbook. Journal of Hazardous Materials 1997;52:1–150.
[13] Franssen J, Real PV. Fire design of steel structures: ECCS Eurocode Design Manuals. Berlin: Ernst & Sohn; 2010.
[14] Paik JK, Thayambali AK. Ship-shaped offshore installations: design, building, and operation. Cambridge: Cambridge
University Press; 2007.
[15] Nolan DP. Handbook of fire and explosion protection engineering principles for oil, gas, chemical, and related facilities.
Park Ridge, NJ: Noyes Publications; 1996.
[16] Smith PD, Hetherington JG. Blast and ballistic loading of structures. Oxford: Butterworth Heinemann; 1994.
[17] Czujko J. Design of offshore facilities to resist gas explosion hazard: engineering handbook. Oslo: CorrOcean; 2001.
[18] Paik JK, Kim BJ, Jeong JS, Kim SH, Jang YS, Kim GS, et al. CFD simulations of gas explosion and fire actions. Ships and
Offshore Structures 2010;5(1):3–12.
[19] Kim BJ, Seo JK, Park JH, Jeong JS, Oh BK, Kim SH, et al. Load characteristics of steel and concrete tubular members under jet
fire: an experimental and numerical study. Ocean Engineering 2010;37(13):1159–68.
[20] Ye KQ. Orthogonal column Latin hypercubes and their application in computer experiments. Journal of the American
Statistical Association 1998;93(444):1430–9.
[21] FLACS. User’s manual (Version 9.0). Norway: GexCon AS; March 2009.
[22] Lund JK, Christensen JA, Wiklund J. Ignition modelling in risk analysis, vol. 19. Norway: Scandpower Risk Management AS;
March 2007.
[23] NORSOK Z-013 Risk and emergency preparedness assessment. 3rd ed.; October 2010.
[24] Paik JK, Czujko J. Explosion and fire engineering of FPSOs (EFEF JIP): definition of design fire loads. Berkshire, UK:
Newsletter of Fire and Blast Information Group (FABIG); February 2011.