1-s2.0-S0957582006713088-main.pdf

A GRID BASED APPROACH FOR FIRE AND EXPLOSIONCONSEQUENCE ANALYSIS

R. PULA, F. I. KHAN1�, B. VEITCH1 and P. R. AMYOTTE2

1Faculty of Engineering & Applied Science, Memorial University of Newfoundland, St John’s, NL, Canada2Department of Chemical Engineering, Dalhousie University, Halifax, NS, Canada

The process area of an offshore oil and gas platform is very compact with a high degreeof congestion and confinement due to space limitations and environmental conditions.Although there are safety systems installed on the platforms, the process area is never

completely safe. Among the loss producing events, fires and explosions are the mostfrequently reported process related accidents. They have potential to cause serious injury topersonnel, major damage to equipment and structure, and disruption of operations. It is there-fore necessary to perform a fire and explosion hazard analysis as a basis for the implemen-tation of appropriate mitigation measures and emergency response plans to protect personnel.

In this paper, we reviewed the existing consequence models, such as source models,dispersion models, ignition models, fire and explosion models, and selected the ones mostsuitable for offshore conditions. These models were then used to perform a consequenceassessment for an offshore platform by simulating four different scenarios. Two mainrevisions were incorporated: (1) a grid-based approach was adopted to enable better conse-quence/impact modelling and analysis of radiation and blast overpressures, and (2) anenhanced onsite ignition model was integrated in the consequence assessment process toobtain better results.

Keywords: consequence modelling; fires; explosions; quantitative risk assessment;offshore risk modelling; grid based approach; onsite ignition model.

INTRODUCTION

The past few decades have seen a wide range of major acci-dents with a number of fatalities, economic losses anddamage to the environment. Examples of accidents in theoffshore oil and gas industry (Spouge, 1999) include thestructural failure and loss of the Alexander Kielland inNorway (1980), the flooding and capsize of the OceanRanger on the Grand Banks (1982), a blowout on theVinland off Sable Island (1984), the process leak leadingto fires and explosions on Piper Alpha in the UK (1988),the explosion and sinking of the P-36 production semi-submersible off Brazil (2001), and the recent helicopteraccident en route to the Bombay High offshore platform.Experience shows that operational failures and human

errors are the most common initiating events for accidentsoffshore. While operational failures could be arrested bysafety-instrumented systems (through monitoring andrestriction to the desirable limits of Safety Integrity Level

or SIL); human errors are difficult to identify and quantify.Recently DiMattia et al. (2004) have developed a uniquehuman error probability calculation index for offshoremustering. The operational failures can be mainly attributeddue to design faults or improper inspection and maintenance.

An offshore development can never be completely safe,but the degree of inherent safety (Mansfield et al., 1996;Khan and Amyotte, 2002) can be increased by selectingthe optimum design in terms of the installation configur-ation, layout and operation. This is done in an attempt toreduce the risk to a level that is as low as reasonablypracticable (ALARP) without resorting to costly protectivesystems. This requires the identification and assessment ofmajor risk contributors, which can be accomplished usingquantitative risk assessment (QRA) techniques early inthe project’s life. If a structured approach of identificationand assessment is not carried out early in the project, it ispossible that the engineering judgement approach will failto identify all of the major risks, and that loss preventionexpenditures will be targeted in areas where there is littlebenefit. This may result in expensive remedial actionslater in the life of the project (Vinnem, 1998).

QRA involves four main steps: hazard identification,consequence assessment, probability calculation, and risk

�Correspondence to: Dr F. I. Khan, Faculty of Engineering & AppliedScience, Memorial University of Newfoundland, St John’s, NL, CanadaA1B 3X5.E-mail: [email protected]

79

0957–5820/06/$30.00+0.00# 2006 Institution of Chemical Engineers

www.icheme.org/journals Trans IChemE, Part B, March 2006doi: 10.1205/psep.05063 Process Safety and Environmental Protection, 84(B2): 79–91

quantification (Khan et al., 2002). Consequence assess-ment, which is central to QRA, involves quantification ofthe likely loss or damage due to anticipated eventualities.Among the various possible loss-producing events onoffshore production facilities, fires and explosions arethe most frequently reported process-related incident(Chamberlain, 2002). These may result in anything fromno damage or loss, up to catastrophic damage or loss,depending upon the severity of their characteristics. Sincethe Piper Alpha disaster, the offshore industry has carriedout a great deal of research work directed towardsunderstanding the characteristics of hydrocarbon fires andexplosions; e.g., Phases 1 and 2 of the Blast and FireEngineering project for Topside Structures, and similarJoint Industry Projects (HSE, 1992a, b, 2004; and Selbyand Burgan, 1998). This research has resulted in thedevelopment of numerous fire and explosion consequencemodels varying from simple empirical models to highlycomplex computational fluid dynamics (CFD) models.The complexity involved in performing consequence

analysis has led to the evolution of computer-automatedtoolkits. There are many commercial software packagesavailable for a detailed analysis of fire and explosionconsequences and associated risk (CCPS, 2000; Spouge,1999), details of which are depicted in Table 1. Thesepackages use empirical, semi-empirical and phenomenolo-gical models for analysing the characteristics of accidents.The advantage of using simple models is that less compu-tational time is required, thus aiding the design engineersto carry out numerous ‘what if’ runs, to test the effect ofdesign modifications. It also aids in conducting a detailedQRA study in less span of time. Complex computermodels (mainly based on CFD) as shown in Table 2, arealso available (Lea and Ledin, 2002; Spouge, 1999). How-ever, the limiting factors in their applicability are related tohigh computational time and user expertise. Further, theinput details required by models are very high, thus beinginappropriate for use in the early design phase.The present work aims to enhance existing knowledge of

hazard assessment through the following advancements:

. Consequence models. Available consequence models,such as source models, dispersion models, and fire and

explosion models, have been reviewed and the modelsmost suitable to conditions offshore have been identified.

. Ignition model. An enhanced model for estimating onsiteignition probability has been employed and embeddedin the consequence assessment process to enable betterprediction of impact and risk.

. Radiation and blast consequence modelling. Instead ofpoint/area modelling, a grid-based approach has beenemployed to enable better modelling and analysis ofradiation and overpressure impact at different locationsin the process area, plotting the results as contours.

CONSEQUENCE ANALYSIS

The main aim of consequence analysis is to identify thepersonnel, equipment, plant and structure exposed to theinitial and escalating events, and to assess the likely effectsand failures. The consequences of fires and explosions areusually expressed in terms of thermal radiation intensity,smoke concentration and explosion overpressure.

The analysis of consequences resulting from a small pro-cess leak leading to major fires and explosions is shown inFigure 1. For an unwanted release event, the first stepinvolved in analysing the consequences is to select anappropriate source model based on the type and phase ofrelease. The second step is to select a dispersion model toestimate the dimensions and concentration of the gascloud. The third step is to select an onsite ignition modelto estimate the probability of ignition. The final step is toestimate the heat radiation, blast overpressure and smokeconcentration from fires and explosions, and to evaluatetheir impacts. In the present context, the focus is on impactson human life. An illustration of flash fire consequenceanalysis is shown in Figure 2; the same procedure wouldapply to gas explosions as well, the only difference beingthat human impact arising from flash fires is due to heatload only, whereas from gas explosion it is due to blastoverpressure and heat load. The models used for all thesepurposes are discussed in the following sections.

Source Models

Source models or release models (CCPS, 2000; Crowland Louvar, 2002) are used to estimate the amount offuel released, or the rate of release of fuel. These modelsplay a crucial role in the risk assessment process as therelease rate and quantity of fuel released determine the

Table 2. Software packages that use CFD models.

No.Softwarepackage Purpose Supplier

1 FLACS Gas explosionconsequence analysis

GexCon AS(CMR Group)

2 EXSIM Fire, explosion anddispersion hazard analysis

EXSIMConsultants AS

3 CFX Fire, explosion anddispersion simulation

AEA Technology

4 AUTOREAGAS Gas explosionconsequence analysis

CenturyDynamics Ltd

5 FLUENT Fluid flow, radiation heattransfer applications

Fluent

Table 1. Software packages that use simple models.

No. Software package Purpose Supplier

1 ARAMAS Offshore risk andconsequence analysis

Advantica

2 NEPTUNE Offshore risk management Det NorskeVeritas Ltd3 SAFETI Onshore QRA

4 PHAST Process hazard analysis5 PLATO Offshore risk analysis ERM6 MAXCRED III Consequence analysis (Khan and

Abassi, 1999)7 FIREX Fire and explosion

modellingSINTEF

8 FRED Consequence analysis Shell GlobalSolutions9 SCOPE Confined and vented

explosion modelling10 DAMAGE Consequence analysis TNO11 SUPERCHEMS Onshore QRA ioMosaic12 CANARY Consequence analysis QUEST

Consultants Inc.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B2): 79–91

80 PULA et al.

size of the resulting cloud and hence the probability ofignition. Furthermore, these models are also used to findthe initial sizes of fires and explosions.The initial release rate through a leak depends mainly on

the pressure inside the equipment, the size of the hole andthe phase of release. Offshore hydrocarbon releases areusually gaseous, liquid and two-phase. Among these, thegases can be hydrocarbons ranging from C1 to C4, whileliquids can be crude oil, diesel oil, aviation fuel, andothers. Condensate is considered to be two-phase as it isa mixture of hydrocarbons (mainly C4 to C6) that con-denses from the gas during compression. This material isliquid while it is held under pressure but becomes gas ifthe pressure is released. Identification of the appropriatephase and its corresponding model is essential, as, thisbeing the initial step for risk assessment, it may prove tobe highly sensitive to the risk estimated.

Dispersion Models

The gases (or gas flashed from liquid releases) that arereleased during an accident, form jets and clouds and aresubsequently dispersed by the initial momentum of therelease, turbulence around the obstructions, natural venti-lation and the wind. Dispersion models are used to estimatethe dimensions of these clouds, varying with time andspace in an unobstructed uniform field (CCPS, 1996), ora highly obstructed field (CCPS, 1998).

The main categories of releases encountered offshore arereleases in confined, congested areas of the platform,releases in open areas or outside the platform, and releasesunderwater. The modelling of gas dispersion in the pre-sence of obstacles demands the use of highly complexCFD models, which are supposed to give good results,but are highly sensitive and need extensive validation

Figure 1. Consequence analysis from a process leak.


A GRID BASED APPROACH FOR FIRE AND EXPLOSION CONSEQUENCE ANALYSIS 81

with large scale experimental results (CCPS, 1998).Moreover, there is a necessity for expertise to use theCFD models for simulating the gas dispersion. On theother hand, for dispersion in uniform wind fields withoutany obstructions, box or slab models (Lees, 1996) andthe Pasquill Gifford model (Crowl and Louvar, 2002) canbe used for dense gas releases and neutrally buoyantreleases, respectively. These models are easy to use,demand relatively less computational effort than CFDcodes, and are widely used within QRA studies (Lees,1996). Therefore, these models have been adopted in thepresent study in an attempt to reduce the complexity incomputation for the overall consequence assessment.

Ignition Model

There is a famous quote from Professor Trevor Kletz: ‘Inthe chemical industry, ignition sources are the only thingswe get for free’ (Kletz, 1998).

Estimation of ignition probability is a key step in theassessment of risk for installations where flammable liquidsand gases are stored. Among the currently availablemodels, a few are based on the assumption that ignitionprobability is solely a function of release rate or size ofthe flammable gas cloud (CCPS, 1998; Spencer and Rew,1997), while the others incorporate some additionalfeatures (Rew et al., 2000; Rew and Daycock, 2004),such as location of the ignition source (whether in hazardousor non-hazardous areas), density, ignition potential and typeof ignition source, multiple ignition sources, and so on.

In the study by Rew et al. (2000), ignition sourceproperty data was obtained for specific off-site scenariotypes (non-hazardous areas) such as car parking lot, trafficlights, road area, kitchen facilities, boiler houses, canteens,and so on. This data was subsequently incorporated into thestatistical framework devised by Spencer and Rew (1997)to produce a working model for ignition probability.However, this model does not consider on-site ignition,particularly ignition in the hazardous areas (whereflammable liquid or gases are stored) of the process plant.Therefore, this model was later revised (Rew and Daycock,2004) for onsite ignition probability estimation and isformulated in such a way that it can be implementedwithin risk analysis models. The main advantages of thismodel are that it is site specific and most importantly, itconsiders the ignition sources within hazardous as well asnon-hazardous areas. These advantages provide theflexibility to revise the model so that it can be applicableto any site specific conditions. Hence, the model wasrevised by incorporating ignition sources (Eckhoff andThomassen, 1994; HSE, 2004a,b) and scenario types(both in hazardous and non-hazardous areas) that aremore appropriate to the conditions encountered on offshoreinstallations. Scenario types such as flames, process area,storage area, accommodation block, kitchen facilities,boiler house, office, and so on are considered andembedded in the model. The revised model was thusemployed for the present study. The equations used in themodel are described below.

For a flammable cloud of area A, containing a randomdistribution of ignition sources with parameters l, p anda, the probability of no ignition at time t in a given typeof scenario is

Q(t) ¼ exp �mA 1� ð1� ap)e�lpt� ��

(1)

where, m is the average number of ignition sources perunit area, P, is the ignition potential of a source (0–1),a is the rate of activation of the source, and l is the pro-portion of time that the source is active.

a ¼ta

ta þ til ¼

l

ta þ ti(2)

If there are n different ignition sources in a given scenario,the probability of no ignition is given as:

Q(t) ¼ Pn

i¼1Qi (3)

Figure 2. Flash fire consequence estimation procedure.


82 PULA et al.

For m scenario types, the probability of the cloud nothaving ignited at time t is given as:

Q(t) ¼ Pm

j¼1Pn

i¼1Qji(t) (4)

Hence, the probability of ignition is:

P(t) ¼ l� Q(t) (5)

The details of ignition sources, scenario types and theignition source data used in the model are listed inTable 3. Most of the ignition source data such as that forignition potential, proportion of time the ignition sourceremains active (a), rate of activation of source (l), and soon have a greater possibility to remain the same irrespec-tive of offshore or onshore conditions. This data has beengathered from Rew and Daycock (2004). However, thedecisive parameter is the ignition source density which issubject to variation depending on a variety of operatingconditions. Since offshore platforms are designed with ahigh quality of ignition control, the ignition source densityoffshore is relatively smaller than the onshore counterpart.Rew and Daycock (2004) provide extensive information onthe density of ignition sources for onshore process plants.Based on this information and keeping an overview onthe offshore conditions, we have assumed that the densitieson offshore platforms can be a fraction of those existingonshore. It should be noted that the data used in thisstudy is specific to the offshore scenario types considered.Following the revisions, the model has been integratedwithin the consequence assessment methodology for bothfire and explosion scenarios.

Fires

Leakage or spillage of flammable material can lead to afire that is triggered by any number of potential ignitionsources (sparks, open flames, and so on). Depending onthe types of release scenarios in the offshore environment,fires are mainly classified into four types: pool fires, jetfires, fireballs and flash fires. Although there are additionalones such as flares, fires on the sea surface and runningliquid fires, they are in one way or the other modeled asone of the four defined types. For example, flares can be

treated as jet fires, and fires on the water surface and run-ning liquid fires can be treated as pool fires. A review ofthe fire models and analysis of their characteristics hasbeen carried out in our earlier work (Pula et al., 2004); abrief description of which is given below.

Pool firesLiquid fuel released accidentally during overfilling of

storage tanks or due to the rupture of pipes and tanks,forms a pool on the surface, vapouries, and upon ignition,results in a pool fire. The probability of occurrence of apool fire is found to be high due to the presence of heavyhydrocarbons on the offshore installations.

Jet firesJet fires usually occur due to immediate ignition of con-

tinuous high pressure releases. They represent a significantelement of risk associated with major incidents on offshoreinstallations, with the fuels ranging from light flammablegases to two-phase crude oil releases. Between horizontaland vertical jet fires, the former is more dangerous becauseof the high probability of impingement on objects down-wind. This can lead to structural, storage vessel, andpipe-work failures, and can cause further escalation of theevent (domino effect). These are considered to be themost dangerous among all the fires, and hence need con-siderable attention.

FireballsWhen a pool fire or jet fire impinges on a vessel contain-

ing pressure-liquefied gas, the container may eventually failas a boiling liquid expanding vapour explosion (BLEVE)and release the inventory due to the phenomenon of internalpressure rise. In such releases, the liquefied gas released tothe atmosphere flashes due to the sudden pressure drop. Ifthe released material is flammable, it will ignite; in additionto missile and blast hazards, there is a thermal radiationhazard from the fireball produced.

Flash firesA flash fire is a transient fire resulting from the ignition of

a gas or vapour cloud, where there has been a delay betweenthe release of flammable material and its subsequent ignition.

Table 3. Ignition source parameters for various sources in different land use types.

Ignition source parameters

Scenario type Ignition sources p ta ti a l m

Flames Continuous 1.00 — 0 1.00 0.000 5Infrequent 1.00 60 420 0.13 0.002 3Intermittent 1.00 5 55 0.08 0.017 2

Accommodation Smoking 1.00 5 115 0.04 0.008 3Kitchen facilities Cooking equipment 0.25 5 25 0.17 0.033 5Boiler house Boiler 1.00 120 360 0.25 0.002 5

Process area Heavy equipment 0.50 — — 1.00 0.028 25Medium equipment 0.3 — — 1.00 0.035 12Light equipment 0.15 — — 1.00 0.056 6

Office area Light equipment levels 0.05 — — 1.00 0.056 6

Storage area Materials handling 0.10 10 20 0.333 0.033 10



When the cloud ignites, the initial damage will be causedprimarily by thermal radiation. However, flash fires maygenerate more damaging ‘knock-on’ events such as a poolfire, jet fire, BLEVE, vapour cloud explosion, and so on,especially if they burn back to the source. Consequencemodelling of a flash fire in a uniform field (no obstacles)where overpressure is not a principal hazard has been welldocumented (Rew et al., 1995, 1998; Cracknell and Carsley,1997). After reviewing the available models, the approachdeveloped by Rew et al. (1998) was adopted in the presentstudy because of the following reasons:

(1) This model was developed after an extensive review ofliterature followed by undertaking full-scale measure-ments to assess the current status of flash fire modelling(Rew et al., 1995). All the typical assumptions made inthe modelling of flash fires were validated against fielddata.

(2) It has been compared with the existing models (Rewet al., 1995, 1998) and found to have obtained superiorresults. Also, to our knowledge, this seems to be theonly state-of-the-art surface emitter model at presentfor flash fires that can model fires from both instan-taneous and continuous releases.

Flash fire modelling is based on release and dispersionmodelling coupled with the probability of ignition, wherethe boundary of the fire is defined by the unignitedcloud’s downwind and crosswind dimensions at flammablelimit concentrations [usually a fraction of the lower flam-mable limit (LFL) is used to account for non-homogeneityin the cloud]. The steps involved in estimating the heatradiation and the subsequent human consequences fromflash fires are shown in the form of a flow diagram inFigure 2. Although this model is applicable only for unob-structed terrains, the incorporation of a grid-based approach(explained later in Grid Based Approach section) in theflash fire consequence modelling makes it suitable toaccount for typical offshore conditions (high congestionand confinement).

Explosions

An explosion is defined as a sudden and violent releaseof energy that causes a blast capable of causing damage.The energy released can be physical, chemical, or nuclearenergy. In the process industries, we deal with explosionswith physical and chemical energy releases. A variety ofexplosions are classified depending on the type of energyand the environment of the release.

Gas explosionsA gas explosion is a sudden generation and expansion of

gases due to rapid burning of a flammable mass. The levelof congestion and confinement in the area covered by thegas cloud usually characterizes a gas explosion. High con-gestion in the form of obstacles causes the turbulence levelin the flow to increase the fluid flow past the objects, result-ing in increased flame acceleration and overpressures.There are several models available for analysing the conse-quences of gas explosions, varying from simple empiricalmodels to complex fluid dynamic models. Informationabout these models is presented in Table 4. Gas explosionsare further divided into three categories, which aredescribed below.

Confined explosions: Confined explosions, or confinedvapour cloud explosions, usually occur in a largely con-fined space, such as inside enclosed modules, or in anoil tank, or a leg of a concrete platform. Overpressure isusually created by the expansion of gas in a confinedvolume as it burns and exceeds the vent capacity of thespace. The presence of obstacles in the path will furtherenhance the overpressure generation and destruction. Thephenomenological models SCOPE (Shell Code for Over-pressure Prediction in gas Explosions) and CLICHE(Confined LInked CHamber Explosion) were speciallydeveloped for confined explosions in offshore modules.SCOPE 3, the most recent version of SCOPE, was selectedfor the present study due to its capability of handling mixedscale objects, rear venting, and an improved combustionmodel. It has been validated against more than 300experiments.

Unconfined/partially confined explosions: Partially con-fined and highly congested conditions are typical of theprocess area of an FPSO (Floating Production Storageand Offloading) vessel and some offshore modules. Ignitionof any vapour cloud under such conditions will lead to anexplosion referred to as a partially confined explosion. Inthis case, overpressure generation is mainly due to turbu-lence generated by the obstacles, such as process equip-ment in the path of the expanding gas. Availableempirical models can be used for modelling this kind ofexplosion as they have been tested and validated for theseconditions. A review of all the empirical models and theircomparison with experimental data has been carried outby Fitzgerald (2001). Subsequently, it was recommendedthat CAM 2 could be used if one wants to study a worst-case scenario and MEM2 if one wants to predict the

Table 4. Gas explosion consequence models.

Empirical models Phenomenological models Numerical models

TNO multi-energy method (MEM) 2D and 3D gasexpansion (Van den Berg and Mercx, 1997;Van den Berg et al., 2000)

SCOPE (Puttock et al., 2000) FLACS, EXSIM, AUTOREAGAS, CFX, COBRA,Imperial College Research Code. (Lea andLedin, 2002)

Baker Strehlow 1D, 2D and 3D gas expansion(Baker and Tang, 2000; Baker et al., 1998, 2004)

CLICHE (Catlin, 1990)

Congestion assessment method (CAM 2) 2D and 3Dgas expansion (Puttock, 1999)


84 PULA et al.

average result for a set of explosion conditions. CAM 2 waschosen for explosion analysis in the present study.

External/vented explosions: This kind of explosionoccurs in conjunction with confined explosions. During aconfined explosion, unburned gas is released through thevents. On ignition, the release will lead to an external/vented explosion. The consequence modelling of externalexplosions is usually combined with confined explosions.Therefore SCOPE 3 (used for both confined and ventedexplosions) model has been employed to study the ventedexplosions.

Boiling liquid expanding vapour explosion (BLEVE)When a fire such as a pool or jet fire impinges on a vessel

containing pressure-liquefied gas, the liquid vaporizes andexpands, thus raising the pressure in the vessel. Further,the external heating causes the vapour wetted walls torise in temperature (due to poor heat transfer) and thuspotentially weaken. The weakened wall may then rupturelocally—giving a jet-fire, or catastrophically resulting in aBLEVE or total loss of containment. The total energyreleased during a BLEVE is shared for overpressure andfragment generation (CCPS, 1994; Venart, 2001). Theamount delivered to the fragments by the blast wavecauses the fragments to become airborne and to act as mis-siles. The missiles are characterized by velocity, weight andpenetration strength. The cumulative effect depends uponthe mass of material involved in the explosion. There areempirical models (CCPS, 1994; Lees, 1996) as well asdynamic response simulation codes (Salzano et al., 2003)for peak overpressure prediction. The model proposed byCCPS (1994) has been selected here for source pressureprediction. The fragments generated by explosion cancause an escalation when the missiles hit the neighboringequipment. Modelling of this effect is beyond the scopeof the current paper.

IMPACT MODELLING

The consequences of fires and explosions are usuallyexpressed in terms of thermal radiation intensity; smokeconcentrations and explosion overpressures received bythe personnel, equipment or structure. In order to estimaterisk, it is useful to convert these effects into impacts caus-ing damage. Dose response evaluation is used to quantifythe damage (fatality) from thermal radiation and overpres-sure. To facilitate this analysis, personnel harm is expressedin terms of probit functions (Khan and Abbasi, 1998),which relate the percentage of people affected in a boundedregion of interest by a normal distribution function. Further,a grid-based approach has been used to facilitate betterimpact modelling and analysis.

Probit Equations

The probit function, Pr, for heat radiation lethality isgiven as

Pr ¼ �36:38þ 2:56 ln(tq4=3) (6)

where q is the radiation heat flux and t is the time ofexposure.

The probit function for likelihood of death due to over-pressure (lung rupture) caused by explosions is given as

Pr ¼ �77:1þ 69:1 ln(P0) (7)

where P0 is the overpressure.Finally the probability (percentage), P, of damage is

quantified using the formula (CCPS, 2000):

P ¼ 50 1þPr � 5

jPr � 5jerf

jPr � 5jffiffiffi2

p

� �� (8)

where erf is the error function.

Grid Based Approach

A grid-based impact modelling approach has beenemployed to enable better modelling and analysis of radi-ation and overpressure impacts at different locations inthe process area. During grid-based computation, the pro-cess area under study, shown in Figure 3, is divided intoa specific number of computational grids as shown inFigure 4, and the hazard potentials and consequences arethen evaluated at each end of the grid. This leads to atwo-dimensional matrix of hazard and consequence resultswhich can be finally plotted as contours. Contour plotting isa more user-friendly representation than the ordinary lineplots obtained by other software packages. A questionthat arises is ‘into how many grids should the processarea be divided?’ We have performed extensive trial simu-lation runs on a (100 � 100 m2) process area (Figure 3) tostudy the effect of number of grids. Consequently, the com-putational time and the precision of results for plottingwere found to be the most important as well as highly sensi-tive parameters on the application of a grid system;50 � 50 ¼ 2500 grids turned out to be the optimal settingthat could ensure an optimal balance between precisionand computational time.

Figure 3. A typical process area.



Additionally, the analysis of radiation heat and the blastwaves is uncertain if the presence of obstacles (partialbarriers, e.g., process equipment and solid barriers, e.g.,passive fire protection walls, explosion proof walls) is nottaken into account. This issue does not arise when usingCFD models, as application of appropriate boundaryconditions will solve the problem. However, it appears tobe difficult to resolve this issue while using semi-empiricalmodels unless a grid-based approach is used. The capabilityof this approach is apparent from Figures 5 and 6 , whichshow the effect of partial barriers and solid barriers onthe heat radiation from a jet fire. It is clear that the solidbarrier totally blocks the radiation while the partial barrierreduces the effect to some extent. Also, the damagecontours obtained permit the development of a clear pictureof potential impact zones. This can facilitate properselection and specification of safe separation distances toprevent injury to people and damage to nearby equipment.The grid heat load and overpressure load obtained from theanalysis can also facilitate the design of protective layers(barriers between accident and receptors). In addition, agrid-based approach seems to be most valuable for model-ling of dispersing or expanding clouds (i.e., in dynamicsimulation), and in defining risk.

RESULTS AND DISCUSSION

The layout of the process area (100 � 100 m2) of a typi-cal offshore platform is shown in Figure 3. In this section,four different scenarios are considered, and the results areplotted as radiation contours, blast waves, and fatality con-tours over the process area. The effect of obstacles has beenconsidered for all the scenarios.

Scenario 1: Flash fire due to ignition of a gas cloudformed by. an instantaneous release (IR) of 1000 kg of natural gas

over the process area;. a continuous release (CR) of natural gas at 10 kg s21 due

to leak in a storage tank.

The gas that is released (instantaneously or continuously)forms a cloud and disperses with initial momentum in lowwind conditions. The dispersion of these clouds was simu-lated using a SLAB dispersion model to estimate thedimension of the cloud. Probability of ignition (P) was esti-mated using an onsite ignition model when the cloud con-centration is within the flammability limits. The variationsof P and cloud size with time, as well as P, and cloud con-centration with time, were plotted for the IR scenario inFigures 7 and 8, respectively. It is clear from the figures

Figure 4. Grid system over the process area.

Figure 6. Jet fire heat radiation in the presence of solid and partial barriers.This figure is available in colour online via www.icheme.org/psep

Figure 7. Variation of P and cloud concentration with time.Figure 5. Jet fire heat radiation in an open space without any obstacles.This figure is available in colour online via www.icheme.org/psep


86 PULA et al.

that an increase in the cloud size with time would lead to anincrease in P and decrease in cloud concentration.The results obtained from the SLAB model (cloud

radius, height), and the ignition model (P, time of ignitionafter release) were used as an input to the Flash FireRadiation model to obtain the fire dimensions and itscharacteristics. These results are presented in Table 5 forboth IR/CR scenarios. Further, radiation contours andfatality contours were plotted over the process area andare represented in Figures 9 and 10 for IR and in Figures11 and 12 for CR. Radiation and fatality contours obtainedas a result from the IR scenario show that most of the plantarea is affected, whereas the contours from the CR scenarioshow that relatively a very small part of the plant isaffected. With this we can conclude that flash fire froman IR is the more dangerous event. However, the effectsof escalation events (e.g., pool fire, jet fire, BLEVE,VCE, and so on) that result when the cloud burns backduring a CR, are likely to be more severe than that forthe flash fire itself.

Scenario 2: BLEVE followed by a fireball inan oil separatorHigh-pressure development in the separator causes the

unit to fail as a BLEVE. The vapor cloud formed due toBLEVE on ignition causes a fireball.The simulation results are represented as blast waves

obtained from BLEVE and as radiation contours from

the resulting fireball, shown in Figures 13 and 14 , respect-ively. It is clear from the figures that the effect of BLEVEis concentrated in the near field, whereas the fireball radi-ation is widespread over most of the process area. Thehuman impact (%fatality) due to the cumulative effect ofoverpressure and heat load is then represented inFigure 15. It was found that the impact due to the fireballis greater than the blast waves and encompasses the entireprocess area.

Scenario 3: Partially confined vapour cloud explosionA 1000 kg of natural gas released into a process area that

is highly congested and partially confined, resulting in avapour cloud explosion. The simulation of this scenariowas carried out using CAM2.

The simulation procedure is similar to that used for flashfires except that here, the results obtained from the SLABand ignition models were used as input to the CAM2explosion model to estimate the source overpressure, andthen an overpressure decay model was used to obtain theblast waves. The results from the simulation are presentedin Table 6, and the decay of the source pressure is plottedas contours over the process area in Figure 16. It is evidentfrom the results (Figure 16) that the lethal overpressureengulfs the complete process area. The human impactdue to the blast waves is represented as fatality contoursin Figure 17, which further strengthen the earlierprediction.

Figure 9. Radiation contours from a flash fire (instantaneous natural gasrelease). This figure is available in colour online via www.icheme.org/psep

Figure 8. Variation of P and cloud radius with time.

Table 5. Flash fire simulation results.

Type of release

SLAB dispersionmodel Ignition model Radiation model

Cloudradius (m)

Cloudheight (m)

Prob. ofignition

Time forignition (s)

Flameheight (m)

Flame speed(m s21)

Maximum‘q’ (kW m22)

Cloud burntime (s)

Instantaneous (1000 kg) 36.83 1.13 0.52 19 44.82 6 25.54 12Continuous (10 kg s21) 21.60 1.24 0.44 24 24.14 6 26.33 8



Figure 11. Radiation contours from a flash fire (continuous natural gasrelease). This figure is available in colour online via www.icheme.org/psep

Figure 14. Radiation contours from a fireball. This figure is available incolour online via www.icheme.org/psep

Figure 10. Lethality contours from a flash fire (instantaneous natural gasrelease). This figure is available in colour online via www.icheme.org/psep

Figure 13. Blast waves from a BLEVE in oil separator. This figure isavailable in colour online via www.icheme.org/psep

Figure 12. Lethality contours from a flash fire (continuous natural gasrelease). This figure is available in colour online via www.icheme.org/psep

Figure 15. Lethality (overpressure and radiation) contours from aBLEVE—fireball scenario. This figure is available in colour online viawww.icheme.org/psep


88 PULA et al.

Scenario 4: Confined vapour cloud explosionAn offshore module (10 � 5 � 5 m3) is completely filled

with a stoichiometric amount of natural gas. The presenceof obstacles at distances 1, 2, 4, 7 and 9 m (seeFigure 18) from one end of the wall, causes the gases toburn rapidly leading to a confined vapour cloud explosion.This particular scenario was simulated using SCOPE 3 and

the results are shown in Table 7. The results in the tableshow how the turbulent intensity, drag coefficient, flamespeed and overpressure vary after passing over each gridof obstacles. Turbulence, flame speed, and overpressureincrease as the unburned gases pass over obstacles, whichare reflected in the tabulated results. Precisely, the profileof variation of overpressure with time is shown inFigure 19, which shows that the gas in a 10 � 5 � 5 m3

volume is burnt in just 0.13 s leading to the developmentof high pressures. However it was found that only 28%of the gas was burned inside the module while the restwas vented out and combusted as an external explosion.Also, the variation of overpressure with flame position inthe module is shown in Figure 20. The overpressure dueto the gas burned in the module was 4.14 bar, whereasfrom the gas vented out was 2.04 bar. Thus the appro-ximate maximum source overpressure generated by thecombination of confined and vented explosion (Pmax ¼Pmoduleþ 0.7Pvented) is 5.56 bar.

CONCLUSIONS

Consequence models most suitable for offshore con-ditions have been selected and used to perform a conse-quence analysis for an offshore platform for four differentscenarios. The present consequence analysis methodologyhas two major revisions:

. a grid-based approach was adopted to enable better mod-elling and analysis of radiation and overpressure impact;

. an enhanced onsite ignition model was incorporated inthe analysis.

Four fire and explosion accident scenarios were inves-tigated in detail using the proposed methodology and

Table 6. CAM2 simulation results.

SLAB dispersion model Ignition model CAM2 explosion model

Cloud radius (m)Cloud

height (m)Prob. ofignition

Time forignition (s)

Sourcevolume (m3)

Sourceradius (m)

Sourcepressure (bar)

36.83 1.13 0.503 19 37392 26.14 7.73

Figure 16. Blast waves from a partially confined VCE. This figure isavailable in colour online via www.icheme.org/psep

Figure 17. Lethality (overpressure) contours from a partially confined VCE.This figure is available in colour online via www.icheme.org/psep

Figure 18. Geometry of an offshore module with a vent and obstacles at1, 2, 4, 7 and 9 m.



models. In most scenarios, the fatality contours engulfedthe process area. The grid heat load and overpressure con-tours obtained from the analysis help to depict the potentialimpact zones. This can facilitate the design of protectivelayers (barriers between accident and receptors) and effec-tive emergency response plans.

NOMENCLATURE

m average number of ignition sources per unit areap ignition potential of a source (0–1)a proportion of time for which source is activel rate of activation of the source, s21

t time, sta average time for which source is active, sti average time between source activations, serf error function

P percentage damage, %P0 overpressure, N m22

Pr probit numberq radiation heat flux, kW m22

Q net heat released by combustion, kWu0 root mean square burning velocity, m s21

REFERENCES

Baker, Q.A. and Tang, M.J., 2000, Comparison of blast curves from vaporcloud explosions, Journal of Loss Prevention in the Process Industries,13: 433–438.

Baker, Q.A., Doolittle, C.M., Fitzgerald, G.A. and Tang, M.J., 1998,Recent developments in the Baker-Strehlow VCE analysis method-ology, Process Safety Progress, 17(4): 297–301.

Baker, Q.A., Pierorazio, A.J., Thomas, J.K. and Ketchum, D.E., 2004, Anupdate of the Baker-Strehlow-Tang vapour cloud explosion predictionmethodology flame speed table, 38th Loss Prevention Symposium,AIChE.

Catlin, C.A., 1990, CLICHE—a generally applicable and practicableoffshore explosion model, Process Safety and Environmental Protec-tion, 68(B): 245–253.

CCPS, 1994, Guidelines for Evaluating the Characteristics of VaporCloud Explosions, Flash Fires and BLEVEs (American Institute ofChemical Engineers New York, USA).

CCPS, 1996, Guidelines for Use of Vapor Cloud Dispersion Models(American Institute of Chemical Engineers, New York, USA).

CCPS, 1998, Estimating the Flammable Mass of a Vapor Cloud (AmericanInstitute of Chemical Engineers, New York, USA).

CCPS, 2000, Guidelines for Chemical Process Quantitative Risk Analysis(American Institute of Chemical Engineers, New York, USA).

Chamberlain, G.A., 2002, Controlling hydrocarbon fires in offshorestructures, Proceedings of Annual Offshore Technology Conference,Houston, TX, USA, 1211–1218.

Cracknell, R.F. and Carsley, A.J., 1997, Cloud fires—a methodology forhazard consequence modelling, IChemE Symposium Series No. 141,139–150.

Crowl, D.A. and Louvar, J.F., 2002, Chemical process safety: fundamen-tals with applications, Prentice Hall International Series in the Physicaland Chemical Engineering Sciences (Prentice-Hall Inc., NJ 07458,USA).

DiMattia, D.G., Amyotte, P.R. and Khan, F.I., 2004, Determination ofhuman error probabilities for offshore platform musters, InternationalConference on Bhopal and its Effects on Process Safety, IIT-Kanpur,India.

Eckhoff, R.D. and Thomassen, O., 1994, Possible sources of ignition ofpotential explosive gas atmospheres on offshore process installations,Journal of Loss Prevention in the Process Industries, 7: 281–294.

Fitzgerald, G.A., 2001, A comparison of simple vapor cloud explosionsprediction methodologies, Second Annual Symposium, Mary KayO’Connor Process Safety Center, Texas A&M University, CollegeStation, Texas, USA.

HSE, 1992a, Oil and gas fires: characteristics and impacts, BFETS-Phase 1, Research report OTI 92 596.

HSE, 1992b, Behavior of oil and gas fires in the presence of confinementand obstacles: characteristics and impacts, BFETS-Phase 1, ResearchReport OTI 92 597.

HSE, 2004a, A critical review of post piper alpha developmentsin explosion science for the offshore industry, 2000, Research ReportRR 089.

Table 7. SCOPE 3 simulation results.

Gridnumber

Gridposition (m)

Turbulenceintensity (m s21)

Dragcoefficient

Flamespeed (m s21) u0�(m s21)

Overpressure(bar)

Reynoldsnumber

1 1 0.64 4.09 1.16 0 0.02 742 2 1.10 4.08 7.94 0.23 0.04 1593 4 3.12 4.08 44.87 1.62 0.51 11054 7 7.47 3.98 107.61 9.21 1.03 62775 9 11.24 3.48 161.85 23.37 1.97 159186 10 13.84 3.08 199.25 38.19 4.14 26019

�Normally mean turbulence intensity.

Figure 19. Overpressure variation with time.

Figure 20. Overpressure variation with the flame position in the module.


90 PULA et al.

HSE, 2004b, Fire and explosion strategy, Issue 1. Hazard InstallationsDirectorate, Offshore Division, www.hse.gov.uk/offshore.

Khan, F.I. and Abassi, S.A., 1998, Rapid quantitative risk assessment of apetrochemical industry using a new software package MAXCRED,Journal of Cleaner Production, 6: 9–22.

Khan, F.I. and Abassi, S.A., 1999, Assessment of risks posed bychemical industries—application of a new computer automated toolMAXCRED-III, Journal of Loss Prevention in the Process Industries,12: 455–469

Khan, F.I., and Amyotte, P.R., 2002, Inherent safety in offshore oil and gasactivities: a review of the present status and future directions, Journal ofLoss Prevention in the Process Industries, 15: 279–289.

Khan, F.I., Sadiq, R. and Husain, T., 2002, Risk based process safetyassessment and control measures design for offshore process facilities,Journal of Hazardous Materials, 94: 1–36.

Lea, C.J. and Ledin, H.S., 2002, A review of the state-of-the-art in gasexplosion modeling, HSL Report 2002/02, London, UK.

Lees, F.P., 1996, Loss Prevention in the Process Industries (Butterworths,London, UK).

Mansfield, D., Poulter, L. and Kletz, T.A., 1996, Improving inherentsafety: a pilot study into the use of inherently safer designs in the UKoffshore oil and gas industry, HSE Offshore Safety Division ResearchProject Report, HSE Office, London, UK.

Pula, R., Khan, F.I., Veitch, B. and Amyotte, P.R., 2004, Revised fire con-sequence models for offshore quantitative risk assessment, InternationalConference on Bhopal and its Effects on Process Safety, IIT-Kanpur,India.

Puttock, J.S., 1999, Improvements in guidelines for prediction of vaporcloud explosions, International Conference and Workshop of Modelingthe Consequences of Accidental Releases of Hazardous Materials,San Francisco, CA.

Puttock, J.S., Yardley, M.R. and Cresswell, T.M., 2000, Prediction ofvapor cloud explosions using the SCOPE model, Journal of LossPrevention in the Process Industries, 13: 419–431.

Rew, P.J. and Daycock, L., 2004, Development of a method for thedetermination of on-site ignition probabilities, HSE Contractor ReportWSA/226 (HSE Books, London, UK).

Rew, P.J., Spencer, H. and Daycock, L., 2000, Offsite ignition probabilityof flammable gases, Journal of Hazardous Materials, 71: 409–422.

Rew, P.J., Spencer, H. and Madison, T., 1998, Sensitivity of risk assess-ment of flash fire events to modeling assumptions, IChemE SymposiumSeries No. 144, 265–278.

Rew, P.J., Spencer, H., Hockey, S.M. and Lines, I.G., 1995, Review offlash fire modeling, HSE Contractor Report WSA/RSU8000/015(HSE Books, London, UK).

Salzano, E., Picozzi, B., Vaciaro, S. and Ciambelli, P., 2003, Hazardsof pressurized tanks involved in fires, Industrial and EngineeringChemistry Research, 42: 1804–1812.

Selby, C.A. and Burgan, B.A., 1998, Blast and fire engineering for topsidestructures—Phase 2: final summary report (SCI Publication no. 253, TheSteel Construction Institute, Ascot, UK).

Spencer, H. and Rew, P.J., 1997, Ignition probability of flammable gases(HSE Books, London, UK).

Spouge, J., 1999, A Guide to Quantitative Risk Assessment for OffshoreInstallations (The Center for Marine and Petroleum Technology(CMPT), UK).

Van den Berg, A.C. and Mercx, W.P.M., 1997, The explosion blastprediction model in the revised CPR 14E (Yellow Book), ProcessSafety Progress, 16(3): 152–159.

Van den Berg, A.C. Mercx, W.P.M., Hayhurst, C.J., Robertson, N.J. andMoran, K.C., 2000, Developments in vapour cloud explosion blast mod-elling, Journal of Hazardous Materials, 71: 301–309.

Venart, J.E.S., 2001, Boiling liquid expanding vapour explosions(BLEVE); possible failure mechanisms and their consequences,IChemE Symposium Series No. 130, 491–505.

Vinnem, J.E., 1998, Evaluation of methodologies for QRA in offshoreoperations, Reliability Engineering and System Safety, 61: 39.

ACKNOWLEDGEMENT

The authors gratefully acknowledge the financial support of the NaturalSciences and Engineering Research Council of Canada.

The manuscript was received 17 March 2005 and accepted forpublication after revision 21 October 2005.



1-s2.0-S0957582006713088-main.pdf

Documents

Transcript of 1-s2.0-S0957582006713088-main.pdf