FIRE AND EXPLOSION CONSEQUENCE MODELI NG IN THE ARCTIC REGIO N by

FIRE AND EXPLOSION CONSEQUENCE MODELI NG IN THE ARCTIC REGIO N

by

A thesissubmitted to thcSchoolof GraduateStudies

in partial fullillment of the requirementsforthcdegree of

MastcrofEngineering

Faculty of Engineeringa nd Applied Scicncc

Memorial Univcrsityof Newfoundland

Abstract

Poolfire sandexplos ions are amongthe mostfrequentaccid entsinproccssfa cilitics.Forpo ol tires,nameimpingementandthermalradiationaremainhazardo uscharacteristic.Whereas, overpre ssur e and negativepulseduration arethe main treatsto hu manandassets in the case of ex plosions.Environmentalvariablessignifi cantlyaffect the behavioroffi res and explo sion s.

However,the effect of environmentalparametersin the coldregion slike arctic hasnot been su llic ientlystud ied. This study presentstwo new models.Asteadystateand fullydeveloped poolfiremodelthattakesinto accountthe effects of allenvironm entalvariab les like temperature.thepresen ceof dropletsandsurface reflexivity onthethcnn al radiation and subseq uentlyon the tireconsequenc e assess ment. Anothermodelhasbeenproposedto account the effectofsnow layers on explosion overpressure.A detaileddescription of model developmentandso lution methodology are present edin the thesis.

Acknowledgments

I would like to thankmythesis supervisorsDr.FaisalKhan and Dr. KellyHawboldt ,lor their patience.guidance andsharing theirimmenseknowledgethroughthisthesi sresearch.I

gratefullyacknowledgethe tinancialsupport providedbytheNatural ScienceandEngineering Resea rchCouncil (NSERC) Canada,and American Bureauof Shipping(ABS) to conduct this

Ilike to thankDr. RonaldHaynes and Dr. RouzbehAbbassilortheirhelps and precise commentsthrou gh this work.

Iwo uld like to thank myfather,mother,brotherandsisters lor the love,motivat ion and friendshipthey have given me everydayof mylife. Finally I like to thankmyfriendsfortheir friendshipsandsupports.

Listof Table s

4.1Listofzone modelswithabriefd eseription 26

4.2 CFD models forfire modeling applieations 34

4.3Comparison between zone models andCFDmodels 36

6.1Listof seleetcde mpirieal models withb rief dcse riptions 77

7.1Constantsfor Eq.(56) 91

Listof Figures

l.lThe areaoffocusforthisthesis 2

2.2 Snowa ndi cc covcredwaters,Arcticr cgion 6

2.3Temperatu revariationduring ayear inpi us 80 N 8

2.4Freq uencyof windvelocities in BarentsSea 9

3.1Ovcrallco nscqucncew ithdifl'ercntcatcgorics 13

3.2Availab le meth odsforconseq uenceassessm ent 16

4.1DilTerentfirecategori es andtheirhazards 21

4.3Layersof amultizonem od el 25

5.1Stepswhic h mustbefollo wedinFay mo de i 48

5.2Solutionstepsoftheproposedpoollire model 56

5.3Temperature distrib ution for Fay model 57

5.4 Temperaturedistributionfordifferentmodcls 58

5.5Co mparisonof predictcd thermalradiations 59

5.6Contrib utiono fplum ezone in thermal radiation tl ux 60

5.7Percentageo fplumezonecontributi onin thet otalradi ationllux 60

5.8DitTerence inthereceivingtherm al radiation 61

5.11Attenuationofthermal radiation lluxfordiffer entdropletssizes 64

5.12 Theellecto fdropleldiameterdistribution onth ermalrad iatione xtinction

5.13EfTectofdroplctmass conccntratio non thc attenuationo f thcnnal radiation.

5.14 DitTcrence in rcceivingthcrmal radiation (CaseI-Case 2) 66

5.15 Differen cein rcceivingtherm alradiationlorarctic conditions(CaseI- Case 2)

5.16EtTectof ditTercnt type ofdropletson the thermalradiatione xtinction

5.17DifTercncci n rccciving thennal radiationatthepresence of iccdroplets

7.1Flow chart toroverpressure calculation by Baker-Strchlow modcl 92

7.2Flo wchartforovc rprcssureca lculation usingt hc newrn odc l...

7.3 Snow comp actionindepth,initial snowden sity=125 kg'm3

andappliedoverpressure=lObar...

7.4 Snowco mpactioni n depth,initia lsnowd ensily=250kg'm3

andappliedoverpressure>lObar...

initialsnowdcnsity=125kglm3 100

7.7Localbchaviorofsnowwhcnsnowdcpth=O.5m 101

7.8Localhchaviorofsnowwhendeplh=0.25m 101

7.9Snowsurfa ce displaceme nt due toshockwave interaction 102

7.10Comparisonbetween overpressureofsevera l casesforcloscdistances

7.11 Pick ovcrpressurc over snowt or scvcral dilTcrcntsnow dcnsiticsin

comparison to Baker-S trehlowmodelformediumdistances...

7- 12Compa riso n between overpress ureof severa lcases in fardistance s...

Nomenclature A Areacross section trrr')

g o

n, E

~Height ofthc co mbustionzone un) p

i'

Superscripts

"."Avera ge value

Subscript s

q

;;'Themnai ratliatio n, intens ity (kwim' )

Contents

Acknowledgement s

Listof Figur es

1.1Motivation andScope I

2.5Fogand Visibility 10

3Conseq uence Assessment 12

3.1Methods for Consequcnce Assessmenl 13

3.2 Convertingto MeasurableImpacts 16

4FireMo de ling 18

4.3.1EmpiricalCorrelations 22

4.3.5CFDModelsRequirements 30

5Mode lingof Pool FiresinArctic Region 38

5.2PoolFire Modeling 41

5.2.2ProposedModel 49

5.3.1 ProposcdM odel versus FayM odeJ 57

5.3.2SurfaceRetlectivity 6\

5.3.3Droplets EfTects 63

5.3.4Arctic RegionCo nditions.. .

6 Exp losion Modehn g...

6.2Eftective Parametcrs inE,plosions 72

6.3.1EmpiricalModels...

6.3.1.2 Multi-EnergyMeth od...

6.3.1.4CongestionAssessmentMethod (CAM).. 75

6.3.2 PhenomenologicalModels ... . 77

7ExplosionConsequence l\1odeling inArcticRegion 87

8.2 Suggcs tions forF uturc Works...

Bibllography...

Chapter I: Introduction

1.1Mot ivati on andScope

Risk -baseddesi gnisa widelyused andyetgrowing practicein mostof industri es . Consequenceassessmentof the dcs ircdfireorexplosionscenarios is among the most crucial steps inthe risk assess ment,Thereare dilTerentmethodsandmodels available to prcfonn conscqucncc analysis.Thesemethods and modelshavebccn developedanduscd ovcryearsfor temperate conditi on s.Fireor explo sionin thearctic regionis auniqueexampleof cnvironmcntallycontrollcdconsequencethat hasnotbeenstudied adcquatcly.The reasonis lack of modelsto includeenvironmental parameterswhichcontroltheimpactpropagation.Snow/Icc coveredsurfacesandthepresenceof waterlice dropletsinthcairaresuchparameters thathave not beenconsideredinpreviousmethodsandmodels.

Theseenvironmental parameterscouldsign ificantly affec ttheconsequenceof anaccidcnt:

therefore,thecontributionoftheseparametersmustbe takeninto account especiallywhen consequenceisestimated torthe arctic region.Hence,studying the effectsof environmental parameters and developingnewmodels tor tire andexplosionconsequenceassessmentare the scope ofthis thesis.Figure 1.1illustratesriskassessmentdiagramandthefocusof thisstudy.

FiJ:urel .I:Th e area o f focusfor thi sth es is.

Chapter2provides abriefdescriptionabout thearctic regionand itsmain characteristics parameters.Consequence assessmentisdiscussedgenerallyinchaptcr3.Thischaptcrcxplains theproc(.-durcofestim atingandconvcrtingthecons(.~uenceof an accident.DifTercntcategories

of fire and difTcrent methods forfireconseq uenceassessment havebecn discussedin detailsin chapter4.Thischapter discussesdifferentmethods,compares themand points out their limitations and applications.Chapter 5 presents lhe proposed modcl for poolfiremodeling inthe

arctic rcgion.This chapterincludes the mathematicalformulationof theproposedmodel and literaturereview of the previousworks on the poolfire. Chapter 6 presentsavailablemethods and modelsfor explosion conseq uence assessment.Thischapterdiscussesthe advantagesand shortcomings o f ditTerent models.Chapter 7 presents the new modei for explosion overpressure calculationin thearctic region .Tnthischapter,jump equation hasbeencoupled with Baker- Strehlow model,Thiscoupling resultedintoorder four partialdilTerential equationswhich have bL'Cn solvL-dusingtinitedifTerence teehniques.Finally,thecontributi ons of thesishave been summarizl-d inchapter8.This chapter includcs thcs uggcstionsfor future works.

Chapter 2: Arctic Characteristics

AsShOWl~inFigure 1.1, environm enta landworkplaceknow ledg eis a vitalstep for consequenceasses smen t.This knowledgecontrib utestohazard identificationandscenario makingandsubscqucntlyaffects the consequence assess ment.Variation in theenvironmental paramctcrsm ayrcsultinto significantconsequcnceditTcrcncl.-s.Forexamplecthermal radiation ofapooltiresignificantly altenuatesdue toprc'Senccofwatcrlicedropletsin the air.Thischapter brieflyindicatesthe maincharacteristicsof arcticregionwhich make this regiondifferent compared to modcratcorwa nn regions.

TheArcticregio nis the regionthat surroundsthe NorthPole whichislocated withinthe ArcticCircle(66.5 N)[I].When definedthisway.the mostfundam ental charac teristicofthe Arctic region is24-hour daylightinsummerand24-hour darkncss duringwintcr.ln sprin g and autumn.duration ofdayhighlydependsonthelatitude.Duringalmostallthe year, Arcticis extremelycoldwithmuchofits landcovered withsnow, icc, and many areasof permafrost,The environm ent rangesfromlarge elevatedmountains totlatplains, spacioustundralands and large bodiesof water,snow,andice.

·••~orl h P O l e. :

"relic~cean

Figur e 2.1:Arctic Circle.amajorpart ofCanadaand Alaska are in ArcticCircle. RetrievedDec 13, 2010,from htlp://www.dom .de/acircle/acircle.htm.

Muchof theArctic environmentisco vered in cnonn ousmassesof ice.TheArctic Oceanis over 1000metersin depth. and thesurface continuouslycoaledwithice[I].The thickness of the icc usuallyranges betweenI to10meterswith a surfaceareaofapproximately9to12million km'.About30percent of theArcticOceanhas veryshallow water;theseparts arc known asthe contincntal shcU:lce shcets arefoundth roughoutthc ocean surfaccaIIyear round.but during summerand the endof spring,theicc usuallymelts at the continentalshelf.

Inccrtain arcas ofArc tic. wann oceancurrcntsin crcaseh.mperature andm oisture contcnt of the atmosphere.causing the amountsofprccipitationto increase aswell.Some regions can

receive morethan3000millimetersofprecipitationperyearsuchastheNorwegiancoast, southern Icelandand Ala ska.Inothe r partsof theArcticwhe rethereisnointluenee ofheat,the temperatures are muchlower,there forereceivinglessthan 150 millimetersperyear.

With in thecentra lareasofthe Arcti cOcean whereit isisol atedfrom surro und ingintluene es thatmay affe ct the climat e,themajorityof the prec ipitati on occursas sno w.Altho ugh the re is rainfall throughout theyear, the mostrainfallsinthew arrnermonths suehasin June,July,and Augus t.

Figure2.2:Snow andiceco veredwaters.Arcticregion.Ret rievedDec13. 20 10. from httn://gall crv.usgs.gov/vid co ta l!s1seatl oo r/list/55/1.

2.3Temperatu re

Theair tempera tureoftheArcticregionisdepend ent on thesolarradiationt hatitr eccives from thesun.Atlocationswithhigher latitudes. solar radiationis very weak and doesnot pro videtoo muchheat totheenvironm ent[2].Tem per ature'Samo ng locat ion s vary becauseof the differencesin radiationand alsofromsurrounding influences. Usuallyasthesunrise sduring the morn ing,theearth's surfaceheat sup.Some timesthewea ther patterncanaffec t the temperatur eduetodrafts of coldair duringtheheatingprocess.The coldairflowmaycausethe

morningtemperatureto rise slowly. remain thesame, or evendecreases . Theoppos iteeffec tcan happen as wellwhen drafts ofwa nnair flowresults inincreaseof air temperatures.

Whensolar radiation interactswith bare land,soilin thegroundabsorb sitandradiatesheat which thenwarmsthesurroundingair.Inthe Arcticregionmostoftheenergy is spent for meltinglarge areas that arecoveredwithice andsnow.Especia llyinthe ArcticOcean,allthe solar radiation isu scdto melt the glaciers;so,thetemperatureremainsfreezingcoldthroughout theday.Duringthe night when there isnoinfluenceof solar radiation.theair temperatureis manipul atedbythe amountofcloudcover present.It hasbeenconcluded thatthetemperatureis colderduring clearconditions andwanne r duringcloudy conditions.

The annual averageof incoming solar radiationin the northern part0fthe Arctic Circle is approxi mated tobe 100W/m'and the areaslocated at mid latitudereceivearangeof l50 t0200 W/m'. Overall,the Arcticregion averagedout tobe -34degreesCelsius in the winteranda range between 3-12 degrees Celsius inthesummerseasons.

Surfaceair temperatures(SATs, approximately2m above the surface)exhibitremarkable regional andseasonalvariability[I,3].January average SATslessthan-40-C characterizes parts ofSiberia. Over the central ArcticOcean, winter temperaturesare somewhat moderatedby

heatfluxes through the icc cover. Januarymean valuesof-25-Cto-32-C are typical [1-3].

DuringJuly, meanvaluesoversnow-free land surfacesare typically10-C to 20-C[1-3]. Over thecentral ArcticOcean.the presence ofamelting ice surface keeps summer temperaturesclose

Figure2.3:Temperatur e variation duringayear in plus 80N,Retrieved Dec l 3,2 010 from https://climatcsanity.wordprcss.com/catc!Jory/arctic/ .

Duringwinter theArcticregionisknownto havehigh windswithsnowstorms betwe en calmed periods[I,4].With high velocities,the Arcticwindsgathersnowasitflowsthrough largeopenspacesand depositsit inshelteredareas. Winterwindspeedsare usually slower than summer windspeeds due tofrequentlyoccurring inversions, andsurface windsareseparated by the inversionlayerfrom the strong upperlayerwinds.Windspeeds that blowfromthewest are also nonmally slowerthan the winds that blow from theeast.

Arcticwindsusuallyblowin from the west as partof the westerlies.Dependingon the topography,there arcinfluences amongwindstrength,direction, and temperature.

flowsthroughcoastalchannelsor mountainpasses,itsstrengthcan be increased.Katabatic windsflowdownward whichcan be warmor coolaccordingto thesituation.forming over glaciers in mountainous areasfromhigh tolowgrounds,acceleratingon the waydown.

Anabaticwinds are usually warmasit flowsupwarddue totherising ofair,frequentlyforming as airflowsoffabodyof water.Anabatic windsareusually very light, moreknownas a

Ncar-surfacewindsare typicallylightinthecentralArctic with mean annualspeeds averaging4-6m/s[I,4].TheNPobservations showmeanspeedsof about5m/s yearround.At stationsinthe CanadianArctic,meanwindspeedsarctypically lessthan in theRussianArctic due to the lower frequencyof cycloneactivity. Nevertheless, when theyoccur,theymaylast several dayswithextremelyhighspeedsexeeeding 25 mls[1,4].

I

DFruho lme ra't rc m sena k e .Bj0 rn0 ya ..Hopen

%Occurrence of Wind Speeds Fi~urc2.4:Frcqueneyofwindveloeitiesin BarentsSea. RetrievedDee 13,2010 from

http://www.barentssea.no/.!n=barentssea&l=en.

2.5 Fog andVisibility

Fogcausesa majorvisibility problemin the Arctic.There arcdilTcrcnt types andsome of them are known as advection. radiation. and iccfog.Fogismost oftenfoundalong the coast parallel tothesho re. Duringthe winter seaso n.the landismuch wanner thanthesurro und ing water.Aswannairfromthe waterflowsoverthe cool land .theairofoppos ite temperaturesmix to formfog.In thesummerseason,thesame thingoccursexcepttheairfrom wannerlands and cooler wate rsmixinstead.Ice fog crea tes fogthe same way exceptthatit consists of ice crystaIs rathcr tha n wa tcr dro plcls. us uallyoccurringat tcm pera turcs bc low-45 dcgrccsCclsius.

Cold airinabilit ytoholdmoisturecauses seasmoke duringwinterseasons.Thetemperature of airis significantlylowcrthanwaterand causes steamtoriscoA blanketof water dropletsis forme..-dbythesteam which is considered tobcseasmoke.Sellsmoke isknowntobe another typeoffog. formedin asim ilar fashi on and havingthesamevis ib ilityeffects,

Anothervisibili ty prob lem intheArcticregio niscalled Arct ichaze .Characteristicsof Arctichaze consist of limitedvisibilit yin thehorizontaldirectionsbutclear visibilityin the verticaldircctions.Itisknowntobe formedfromvcry smallice particles sinceit rctlcctsmany dilTcrcntcolors aslight interact swith thc hazc.As sunligh tshines throughthehaze,it isknown asdiamonddustbecause of allthe rock-like particles,butncarthe ground itappea rstobe mist or smo ke [1-3].

2.6 Humidity

Arcticair is expressedtohave verylow temperaturesand moisturecontent. During the winter,air is significantlycolder andd ryer overlandthani t isovcr water.Sincewater iswanner than land in thc wintcr.thcairabovc picksupwatcr partick'S asitllows across thcsurfaceand

retains itshighertemperature valuesandmoisturecontent [5].Onaverage theArcticrelative humidity foundatthesurfaceis rangedfrom50 to 60percent. Ncarthesurface, meanspecific humidity(massof watervapor perunitmassofair. includingthewater vapor)lor winteras averagedforthe regionnorthof 70·N isonlyaboutIg kg'comparedto aboutjtoagkg"in summer[I].

Chapter 3: Consequenc eAssessment

Previou scha pter bric tly pointedout the maincharacteristicsof the arcticregion.Accor di ng tothese cha rac teristic sandworkplacecharactcr isticslgcomctries.scenarios arc defin ed.Once scenarioaredefi ned. they arcana lyzedfor consequenceandlikelihoodassessmcnt.Consl'qucncc asses smentre ferstoattemptstow ardquan tification ofhazard intensity.Amonghaza rdous events;tireandexplos ionare the most commonacciden tsin oiland gasindustries.Fireis more freq uent;incontrastexplosion has higherdamagepotentia land co mm onl yleadsto fatality.

injuryand propc rty loss.Thcconsequc nccsarcusuallycalculatcd intcnnsofprod uction loss.

humanhealthloss.assetsloss, andenvironm ental damages.

Conseq ue nceassessment help s todctcnnincthepotenti aldam ageduetohazard ou s occas ions;how ev er,but docsnot accountfor ho wfrcqucnt the ace identoccu rs.If the result sof consequenceassessmentare not directl yquantifiable,then it has to be convert edtosomc measurableimpacts(monetarytermsusually).

Inlegralionof all identili edio sses (e.g.produclionloss,assel loss.human health andsafely )oss,cnviro nmcnt loss, infonnationand knowlcdgc loss) toobtain the ovcra ll consl'q ucncc o fa hazardou sevent requires spec ialtreatment[6 -8j .Figure 3.1shows the flowch artlor an overa ll conseq uencewith differ ent catego ries.

Fi ~u re3 . 1:0verallconscquentwithdi fferent categories.

3.1 J\1cthodsfor Consequcn ccA ssessment

Scenarios are identified accordingto theconsequences ofhazards,in termsof the probable fatalities,injuries,damages,and failuresthat canbedetenn ined in threedifferent ways[6-9}.

Availablemethods canvary from highlysubjective method slike engineeringjudgment to highly comput ationalmethodsusing computational tluiddynamic( CFD)tools.

Historic aldata can beusedto evalu atecon sequences.The scenario can becompared to accide ntsofthe pastthat had similar hazards.For example.reviewin gthe last accidentswill help toshow thetrend of consequenceand itsintensity. Moreover,it helps forabetter understandingof theco nscquencesfors pccificevents,a nd brcaksdown the damage outco meof eachconseq uentialeffec t, Historicaldata maybespccitic to the environm ent of the incid ent.

spL~ific tostructurcsofas i mi lar typeshari ng thesamegencral locationorowner.

There areadvantagesanddisadvantagestorestima tingconse que nceswith historicaldata The datahclps support the valuesgiven in aspeci fic accide nttocalculatetheconseq uences,if theyaresufficient enough to give preciseresult s,datafromthese pastincidentscanbeuscdto make a reasonab leriskassessment.Tho ugh thismayb epractical foraveragevaluedresultstha t have asetnumberof variables,experimentsthathave alargenumber ofvariablesare not accountedfor. and outputs a sign ificant ditTerence in data.The incidentsusedin comparison sho uld have verysimilarstructures forthebestresults.Anotherprobl emthat may ariseisthe errors or changein data over time which discourages theuseoftheinform ationpreviously

Thesize of databasesis amajor contributioninconsequenceestimation,moredutais

availableasitincreases insize.Though there maybesu m ci entdata. the i n tonnation nc(...~cd to be accurate.Often,thedetailsprovided arc nol allthe detailsrequired forconsequence estimation.Historicaldata israrelycom pletebecause mino r incid ents tha tcouldhave ledto maj or incide ntsarc not alwaysrecorded .tousc historica l data,the relevan cy of the datamustbe

determinedand ifitissuitable tobestudied.(Ifapplicable,Ihenapply, ifpartially,apply judgment tomodifydata, and ifnot, uscanothermethod)

Eng inee ringjudgme ntis another use fulwayto evaluateconsequences.This methodis Iimitcdtojudgmentsor opinionsthatarcof enginccringcx pertise.ino rdertohave consistencyin results. Engineeringjudgments may be basedfrom thepreviousexperienceofanother

practition er orfrom theuse ofanorgan ized andreliable process suehastheDclphi method.This processis very practicalb(...~auseit providcseslimationthatiscompletclyinexistentor lack in other methods,TheDelphi method causesproblem s for most practitioners as itrequiresskilland experiencetobe used.

Another disadvantagefrom the uscof engineeringjudgmentisbias opinions;they arisefrom various individualsaccording to whathas happenedinthe past. Thusalwaysthe opinionof severa l expertsmustbeasked and an individual opinioneannot berelied on lonely.Subsequenll y thisresult s ina rangeof estima tionsand potenti al outco mes. Althoughthismethoddoesnot provide a point value,so metim es the inaccuracyisfine when a rangeofvalues isof intere st insteadofaspcciticestimation.

Modelshavebccndevelopedtoevalu ate accident con sequenc es ;theyeancalcul atethe numberofdeath s or injuries.loss , cost of downtime or busines s interru ptions,and also any environmentaldam age s. Fire modelsare norm allycapableof evaluating fire development, smo ke movement,structuralrespon se,and evacuation respon se, while estimatin gthe time it takestoreach its critica l damagethresh old.Likewiseexplosionmodels arc abletoestim ate dispers ion, tlamepropagationandgeneratedoverpressure,Thesemodels are usedfor quantitativeestimate sbased on rationalizedmethods and any chan gesin the design are relatedto theconsequenceresults;therefore,allowingdesignerstoacknowledgewherechanges should be

Accurateinput valu es are sometim eshardtoacqu ire due to uncert ainti es .Depending on how complexthe problem maybe, onesimple mode lcannot provide allthe requiredresultsand need to be combinedwith other models.When usingone model atatim e,the result sof one may be used asinput data for anotheras a seq uence,and if mistake sare mad einthebeginnin g,thc cntire proc css will beincorrcct.Fif::,'Ure3.2shows available methodsforconseq uence asscss me nt andtheir advantage sanddisadvant ages.

Fi~u re3.2:Availablemethodstorconsl.'quence asscssmcnt 3.2 Conve r tingConseq uence toI\leasurableImpact s

ConSl.'qucnccscaused froma ccidentaltires andexplosions arcmeasurl.~ indifferent\Vays.

Thisincludesthe measure of health andsafetyimpacts, loss ofproperty,businessdisturb ance costs. or environmentaldamages.Theconsequencescan besubjective or objective.andcan be caused byd irect or indircctfactorsand mustbe ident ifiedduringtheriskassessrnent todo a

Evaluatingconseq uence ismorediflicultcomp aredtocvaluatinghazards, ina way that the value of loss or hann maybe unclearand hardtodctennine.\Vhile lifesafetyconscquenccs are

estimated,injuricsand loss of humanlife are includedin thccriteria but a 101ofthctime othcr importantfactors are not included.Facto rs suchaslower qualityof Iife,painand suffering, rehabilitationafteratirerelatedincident.inabilitytoworkafteranaccident. and effects on family support.There are other longtcnnconcerns that include loss ofimageand marketshares foraspec ific busiru..zss.Such impacts arevery compli cated and in most cascs arc notcon sidcn .."d as apart ofcon scquence assessment.

Chapter 4:FireModeling

Asdiscu ssedin previou schapter,therearcsev era l method sto estimatccon sequence ofan accident. Amongthosemethods,mathematical modelin gisrecogn izedto be the mostpowerful methoddue to itsllexib iJity andabi lityto handle differentlevel ofcom plex ityanddeta ils.

Availablemethodsmay bedivided intotwo maincatego ries based on their outputs:physical mod els andeffectmodels.Physical mode lscalculatevariables whichare direc t outcomesofa hazardousevent.For example theymay estimate parameterslike temperature,radiation,toxi c gasco ncentration.Eff ect modelsconverttheoutco mes of physi calmodelsto moresensible imp actslikefatalities,injuryor loss of property.

Modelsmay bec1assilied as emp iricalandcomputationa l.Em pirica l tool s estima te parametersbasedonexperimen tallyobtainedcorre lations;thus,a sho rttimeisnc...xlcdfor calculati on s.Theassumptio nsand limitingcondit ionsduringthe dcrivation o fthc sc corrclation s, limit theapplication tosimplegeome tries.In contrastto empirical tools.computationaltoolstake advant age ofgoverni ng equation sinestima ting desiredparam eter s.Althou ghittakesamuch lon ger com putatio nal timetorunthese models, the result isexpcct cd tobe moreacc urateand reasonable.These modelshavetheability beappliedforcomplexgeometrieswhe rethe applica tionof simple empirical model smightbe meanin gl es s.

Although thereare numerousmodelsavailablefor con seque ncemodeling. sel ecting the most

suitable modelisnot aneasytask. Inselecting a modelall workplace•environment conditi on and dilTerentseen ariomustbecon sideredandthesel ectedmodelmusthavethe abi litytotakeinto accountthesecond itions.For example theFRE D maybe usedto estim ate impactsofahazard ou s event in asim plegeometry;howe ver.itisnot applica bleifretlective sur facessuchas snow

present. Fire,Release,Explosion andDispersio n (FRED)isaso ftware developcd by She llfor quick cstimut ionsbascdon thcexpcriment aldata.

4.1FireCatcgor tes

The main cause of injury or damage fromalarge open hydrocarbon lire is thenna l radiation.

Thedifferent typ esof tiresbehave differe ntlyand exhib itmarkedl ydiff erent radiat io n characteristic s.Radiationandconvect ionare theprincipa lmecha nism sfortransferringheatfrorn alire toastructure. The radiation isusuallythe dominantmode of heatlransfer ,although convectiveheat transferbecomesan importantmode forstructures directlyimpinged or engulfed.Typica lly,firescanbe classifie dintofourcategories : pool tires,jettires,tireballsand tlashtires.lnapoollire,thepoolofvaporizingfuelformsbuoyaneyeontro lledturb ulent tlame wherethe fucl vaporhasnegligibleinitial momentum.Poolfiresoccur from the accidental releaseofliq uid fuel during loading of tanks ordueto ruptureand!or frac tu reinpipesand tanks.

The proba bi lityof occurrencefor thistype oflire depe ndson the type off u elsandenv ironm enta l condi tions.Heavyfuelsarc mostlikcly toprodu cepool tire wh ile light fuelmay evaporateand producesvapo r cloud. In warm environments fuel will mostlyevaporate whilethesame fuelmay generate a poolfire in coldenvironmentslike the Arctic.

Jet tires occ urdue to immed iateor quick ignitionofpressuri zedfuel.This fue lcould be a single phaseofgas, gasand liquidandevensome timesliq uid [10-12].Jet tiresarc muchmore dangerousthan pool lire because of the radii of its impacts. Depending on thescenario

spt-cification(rcleasedphase and angle o frclease)jet tires can behorizontalor vertical.

Horizontal jet tires are morehazardous espec ially downwind .Jet firesmayleadto impingement ofstructures, equipmentand vessels,ruptur e ofpipesandsecondaryfailures .

Sincejet fireshavehighradiiof impact, scena rioswhich includcjet fires shouldbespecified prL~isel ybccause ofdominoetfects. Thcdom inocffectisthesi tuationwhen anaccide ntal fire triggersand expandstiresto theother parts workp lace. Thisis anextremi ty hazardousevent when thei nsta llalio n ishighlycongestcd d uc tolim itationofs pacc likcoffshorestructuresand ships.lnthoscplaccs,j ct fircslcadto avcryhigh ralco f hcattrans fcr andrapid

The dornino etfectsof poolfircsorjcttiresmayincreasethe pressureinsideatank whic h contains flamm able com pounds. The inc reasesintemperature andpressure mayft...esultin tank failure duc toboi ling liquid expa nsio nvapor cx plosion (BLEVEl[13,14 ].ln lh issitua lio n,duc to a suddenpressure drop,liquid evaporatcs and ignites. Thisresults inanoverpn...-ssurcand missilehazard s.Furthe rmore.thisresults inahigherthennalradiationcompared to jetandpool

Dueto thethermalradiatio n,overpressureandmissile hazards.BLEVE couldbemore hazardousthan othertypesoffires.If the tank containsgas, thismay resultinaverylargeand rapidfireb all.Suchafireballhasthehigh cst therm alradiati onprolileamon galldifferenttypes oftires .Thefireb allshapc,bc ha viorandsizearestrictlyrelatcdtotypeof fuel.failuremodeand

A tlashfireis atype offire in whichthcrei s a shortdelaybctwecn thereleascandignition times.In this categoryoffire,thegas orvaporcloudisreleased andthen isignited ;thus.there will be anegligibleoverpressure. Flash firesaretransientand resultinatemporaryheat radiation.Howeveritcould trigger other typesof tire especiallyiftlametumsbackto thesource.

Figure 4.1summarizes the categoriesof tires.

Initiation -Liquid relcase

Initia tion -Pressurizedfucl release

Initiation Initiation -Raptureof tank -Release ofgas or

evapo rab lc liquid Impacts

-Thcrmalradi ation -TriggeringBLE VE -Smokeand

structure failure -High radiiof impacts

Impact s -Overpressure -Thermalradiation -Missileand projection

Figu re-t.I : DitTerentfirecategories and theirh azards.

4.2 i\l od elin g

After unde rstandi ngof thescenario's requirements. specifications and characteristics.

mathematicalmodeling isusedtoestimatethe possible scenarioconsequences.Avarietyof tools and models are availabletoperf ormconsequenceasses sme nt.Selectingtheoptimalmodel requires anunderstandingof theirstrengthand limitations.

In tire modelingtheobjectiveisto understandfirebehavioranditsouteomes.Dependingon requiredaccuracy,complexityofsystemand requiredtime therearcnumerousmathematical modelswhich could be usedfor lire modeling.Based on the employedmathematicaltechn iques

andsolution methods,mathematicalmodelscanbe dividedin threemain categories: empirical correlations, zonemodels andCFDmodels.

4.3.1 Emp ir icalCor re lat ions

Empiricalcorrelationsbased modelsare thesimplest models.Theyare basedonempirical relationswhichhavebeendevelopedconsidering the algebraic reiation betweeni nput parameters andoutcomesof fires.Since in firemodelingespeciallyinopenareathermalradiationand convection are the mostimportant parameters,thesemodelsare mostlydevoted to estimate thermalradiation,temperature profile and convection.Forexamplein jettiremodeling, parumctersIiketype offuel,fuel massrateand dischargediameterisused toestimate thenn al rad iation,conve ction andthevisib le length of thetlame.

Although thesemodels are fastanduserfriendly,thereareseveral limitationswhich restricting theirapplication.Firstofall,mo stof thesemodels areonthe base oftinite numberof experimentsconstrained with experimentalconditions.Secondly,thereare inherenterror associate with fittingcorrclationstoarestricted data set.Furthennorc,most of the experiments havebeencarriedoutforopenspacesorvery simpleandnotcongestedenclosures.Sothese modelsmaynotbeapplied forhighly conges ted workplaceslike oITshorestruetures.

The applicationofzonemodels started in mid-1950 with introductionofonezonemodel sfor study ofpost-tlashovcrandspecially movementoftoxic gasesinothercompartments. Later two zonemodcls weredevclopcd forpre-tlashoverbehaviorestimations.Thecon ceptbehindzone models istodivide compartmentsor enclosuresintotwo ormorespatially homogenousvolumes (zoncsj .Thcmodclhasbee nobservedtotak eplaeein realcasesandexperiments [15,16 J.ln

thcsemodel s,the upper layerhashighertemperatur e and temperature dropsins ubscquent lower layers.In thesemodels,the plume concept isusedto take into accountsmokeand hotair

For each spatial volume (zone) ,physicalparameters such as species concentrati on.

tempe rature and density arcassumed tobcunifonn .Thismeansthat all theparticl cs in a specific zone have same properti esand theseproperti eschange only with lime. Although the resultsfrom experimentsshowsome perturbations and variationsfrom these assumptions,ingeneral the resultsof thesemodc1s are in good agreement with experiments.

Mass.momentum and energyconserv ationequationsresultina sctof OrdinaryDifferenti al Equations(ODE) .The conservation equat ions arccoupled with gaslaws (lypiea llytheidea lgas assumption)forcalculatio ns.In zone models the physicalpropertiesinazoneconsidered homogenous;thus. conservation laws areapplied toestimateenergy.mass and momentum transportbetweenadjacentzones. Calculation sarc conducted bymodelingsub-fircproccsscs:

Thcmomentumconservation law isnot appliedexplicitlyasnot all the required information for pressure andvelocityis availableinthe model. Theapplicationof momentum conservatio n lawis governed bythe massmovement from plume firetozonesandfromadjacentzonesto ventilation and open areas.

Zone modelsmaybeclassifiedbasedon the number ofzones in eachcompartment:one zone, two zone and multi zone models.Inonczone modelsthe entire compartme nt hasthe same propertieslike temperatureandspecies concentrati on.Due to plume rising and fire turbulent behavior,thesetypes of models are notappropriate where there is a fire. Such aonezone model

has beendeveloped and used for analysisandestimationofsmoke movementin other

Twozone modelsdivide the volume of enclosureto two volumeswhere the upperzone isat high temperatureand the lowerzone at lower temperature.Thislayering isdue to buoyancy effectand the riseof hotgases from plume.Thesemodelsare morerealisticin comparisontoone zone modelseven when only thesmoke movementin remotecompartments is studied.Thisis due to presenceof colder air in remote compartment. buoyancyetTectandfinallystratification . I3ccauseoflwu zonemodelabililies,lheycouldbcusc't!forprc-llashoverandpost-t1ashovcrfirc modeling. Figure4.2presentscontrolvolumesfor twozone models.

+-Q••

It/mu,

I

Eu,R! thefirecompartment:

E istheintErnalenergy of gas

Upperlayer _ mum isthemasstErms

LO\'Ier

laye~\

~

~nr:~~ ~~~~:~:s

m,tt,h+- : Q} -mfi ,«,l.

~ :~~ ;~u~:P3rature

p isthegasdensity

Figure~.2 :Controlvolumes andzones fora twozone model. Retrieved 12 January 2011 from:http://www.mace.manchcster.ac.uklprojt..Ctlrcsearch!structur cslstruetireJDesign/perfonnanc

c/fireModelling/zoneModelsllwoZoneModel.hlm

Finall ymultizone model sweredevclopcd to improveaccuracyandalso predict vertical temperaturevariationsand toxic gas concentrations [17,18].Inthismethod. the compartme nt space isdivided into an arbitrary number of zonesandthe physical properticsofeach layc ra re assumedto beuniform ,The tire plumenowdocsnotmix withadjac entlayersandtire plum e flowis upward untilit reachesthe ceiling.Therefore,inthesemodelstheassumpt ionisthat all the heatistransferredto thetoplayer.

Conservationslaws aresolvedin theboundary oflayersto estimate the averagevaluesof parameterssuchastemperaturein layers.Duetothepresence ofmultiplezones.computational timeislongerthantwo zone models.Figure4.3showssegmentsofamultizone model.

i-th layer

2ndlayer

Ist Iayer_{. . . . . . ._} . . . . .iiiiIiiii . . .

Fil:u re 4.J:Layers of amultizonemodeI.

Althoughdifferent zonemodelsapplydifferentmethodologyand techniqueslor calculations,they have someassum ptions in common whichdistingui shthem from CFD model s:

Uni form physicalproperti esinsideeach zone.

Diffusioncan occur betweenzonesandthe enclosurewall.Althoughin mostmodeledit isneglected(due to very sm allvalue).

Plume tlow risesto theceili ngcontinuo usly.

Plume tlowdocsnot mixora fTect moder atelevelzones in itsrising .

In mostof models,the horizont alcross sec tion (area) of the compartmentiscon stant.

Numerous zone modelshave been developedand validatedwith experimentalresults.The belowtable presents mostcommo nlyusedmod els.

TabI04.I: List o fzon e mod eiswith a brief descri tion19, 20.

Model Country Description

ARGOS Denmark Multi-compartmentzone model ASET US One roornzone,no venti lation

ASET-B US ASET in BASICIanzua'0 insteadof FORTRANlanuuare BRANZFIRE New Multi room zone model,includingflame speed,multi fire and

Zealand mechanical ventilation

BRI-2 Japa n/ US Two layerzone model formultistory,multi compartmentsmoke

CA LTEC H US

~:~:~over

CCFM .VENT SUS Multi roornzone withventilation CFIRE-X German Co m artmentfire,seciallIiuidh drocarbonpoolfire CiFi France Multi roomzone rnodcl

COM PBRN -IIIUS Com artmentzoncfire

FFM US Pre-flashover zone model

FIRAC US Includ e com lex ventsstems

FIREWI ND Australia Multi roomzone model withseveralsmallersub-model MRFC Germa ny ~~~t~~o~::t:~~smodclfor smoke movementand temperature NAT France Singlecompartment zone model withattention to respo nseof

structures

OZONE Bel uium Zone model wi tha ttention tos tructure resoonse

RFIRES US Pre-flashovcr zoncm odcl

SFIRE-4 Swed en Post-flashove rzonc modcl

WPIFIR E US Multiroomzone fire

Zonc modclsarc limitcd duc toassumptions during dcrivation.ronnulationandapplication.

Oneofthemostimportant assumptionsin thesemodelsisthat a compartmentistobedivided into small numbcrofunifonnedpropertyzones andcalculationsarc to be donewith ineach layer.Since ineachlayerall the physicalproperties arc thesame, there isnotemperature or co ncentrationgradients in the layer. This meansthat in whole layer,temperature andgas concentration isthesam e regardlessto the locationof fire orvent. Thisisproblematicwhen the gasconcentrationor temperature at aspeci ficlocation iscritical,for examplewhen determining thcoplim all o cati onof smok edctcclOrs.

Theother limitation isdue tozone transparenc y and radiation.Theradiationmagnitud eis directly a func tionofsurro undingenvironme ntspeciesconccntra tion . abso rption coefficientand temperature.The assum ptio n ofa uniformhotlayer abov e thecold Iayerov cres timates radiative heat from hotlayerto cold layer.This in tum couldresult in underestimatingthetemperature of thehotlayer.

Oneofthe mainshortcomings ofzone mod elsisthe inability to accountfor turbulent plume behavior. Especia lly inmultizonemodels where therearcsevera l laye rs.the assumption thathot

gasesmo veupwardtoward stheceiling without anyeffect on intennediate layers.Thisresult in an under est imationfor lower layers temperatures andoverestimation forupperlayers

Most of zone model sneed therate ofheatreleasebefo re sta rting the simulati onproce ss. The rateof heat releaseisa functiono f combustionc fficiencywhich in tum is rela ted to air to fuel

rates.For examplewhenoxygenconcentrationdecreases,forasolidfuel, flamelength may decrease and change to a smoldering process. Since theh eatr elease ratecha ngeswithtime and space.using itas an input param eter willim pactthe mod el acc uracy.

4.3 .4Co m puta tionaIFluidDynamic (C FD) l\I od els

Rapidgrowthof both hardwareandsoftwarecomputerknowledgeandtechnologyhave moved models from correlation basedto compute based onthegoverningequa tions(CFD mode ls ).Altho ughdescriptionofCF D isbeh indthe se opc ofthi s thesis,a brie fdescrip tio nis helpful. A CPDmodelis a determi nis tictoolfor simulatingrealtime scenarios. CPDmodels approximatethesolutionof a setofalgebraicand differentialequations. Howeverhere, equationsaresolved based on localapprox ima tio nof differe ntiaIequa tio ns.

Inthesemode lsa set ofnon-linear, timeandspatialdepend en t equations (Navier-Stok es) are solve doverthe doma inofinterest. These mode ls can cap tureturbulence throu gh intro duc tionof turbulentequatio ns.Thewayinwhichturbul en ce istak enintoaccountis important;basically CFD models arc class ified base on themethod ofdescrib ing turbulent.

Turbulence is a stateofflowmotionthat involves threedimensional randomvortices,higher energydiss ipa tio n,mixing and drag effects.CFD mod elshavedifferentmethod s forturb ulent modelin g:Reynold s AveragedNavier -S tokes (RA NS),Large EddySimulatio n(LES).Direct

NumericalSimulation (DNS),DetachedEddySimulation (DES) and Turbulencencarwall mode ling. Thefirstandsecondmethodsarc the mostpopular,appli cable andfrequencyused.

It isnotsurprisingthat the DNS is the mostaccurate model [21-24J. The DNSmodel was introduced in 1970tomodelisentropicturbulenceup to Re=35.Thismodelsolves Navier- Stokcscquationsin averyfinegri danda verysmall timestepsto capturc theturbulence. vortex and fluctuations.The reasonwhyDNSmodels are not usedin mostof engineering applications

isdueto theirvery long computationa lrequireme nttime. Furthermore,DNSarc needed lobe solved in threedimensionalspaces where turbulence andvorticesexist.Bt.-c3USCofthese limitations,DNSarcmostlyused asaresearchtooltounderstand thefundamental turbulence

The purposeofRANS istoestimate Reynolds stresses. Thismay bedonebythree different methods:LinearEddyViscositymodel s, NonlinearEddy Viscositymodelsand Reynold s Stress models(RSM).In lineareddyviscosity model s,the Reynolds stress ismodeled through introducing alinear constitutiverelationshipwith maintlowstrainfield.However.lineareddy viscos ity model s arc knownto fail inanumberofflow situations.For example thesemodel s over-predict turbulence energy levelinstagnant regions.More..zovcr these modelsusuallyover- predictK (Kinematicenergy)nearthewallswhereactually Kcan be neglected.Becauseofthcse shortcomings,atprcsent,lineareddy viscositymodels are rapidly rcplaced bynonlin ear eddy visco sity modclsin engineerin gapplications.

With respectto accuracy and complexit y,LESliesbetweenRANS and DNSmodels.Inthis model,smalleddiesarc removed andmodeled by Sub-Gri d-Sca le(SGS)models.Kolmogro v's theory(self-similarit ytheo ry)statesin turbulent now largeeddies aredependentongeometry

while small eddies arc unive rsalwhich allowsforan explicitsolutio noflarge edd iesand implicit solutionofsma ll eddi esby SGS model s.Thisfi ltration enableshigherReynol dsnumber(mor e turbu lenttlows) tobemodeled .Howe verthis tiltratio nmayresultin low er accuracyespecia lly ncar wallboundaries.To compe nsate this situation,a muchfincrme sh is req uired.

Formost ofengineering applica tionsthe accuracy ofRANSmodels is sufficient,Recently hybridmodelshavebeenintro duced.These models arcacom bination ofLES and RA NS meth ods.Thesemodels havethe adva ntagesofbothcategoriesofmodeIs:spl'<.'dof RANSand accuracyofLES.

4.3.5CFl> i\Io dc ls Rcq u irc mc nl

Lots of engineeringproblemsareaddressedbyCFD models;however,forfire modeling applica tions,amodelsho uld have certainfeature sto be used.Turb ulcnce modelingis oneof the mostimportant features.Thisismore important for jet fire modeling since the flow ishighly turbulent.One other importantcriterion isheat releaseand radiationmodeling.Inc tli ci cnt

modelin gofsoot mayre sultincompletelydifferentoutputs; thus,especialtreatmentofsoot particlesismandatory. Combustion modelingisof importancedue to itsdirecteffectson comb ustion process outputssuc h asheatrclcase and smo ke.Bound ar y and initialcond ition arc employedto obtai n unique solution. Beside allofthese.amodel sho uld beab letointak e complexgeometries withoutrest riction s onspaceor tim e steps.

Fir esandespec iallyjet tires arc highl y turbulence phenomen a.Afterthecombustio n process.numerous cddics in variabl e si zes are generatL~.\Vhil einili allyeddiesareon large size.

they breakd owninto sma llereddiesandvorticesas theinitialcncrgy diss ipates.To takeinto accountsuc h beh avio rs.theturbulencemodelplays a critica l rule.Most ofCFDtoolsemploy

RANSturbulencemodelin g(especi all yK-candK-w).Results arereliableandles s com putationa l timeis requiredincompariso n toLESbased CFD model s. For more congeste d areaslike eng inerooms,due tohighturbulenc etlowinthecaseoffirc,LES based modeis may

Ingeneralradiation modcling of aliquidfueltireis easierthan gasfucltiresbecause gas fuel tiresmore complex, turbulent and unpredi ctable.The rad iationequatio n is a combinat io nof integral and differentialterms(integro -diITercntial equa tion)which makes solving morediffi cul t.

Rad iationdep end s on par ame tersliketemp erature andcom pos ition. there fo re,to solveradiation equation,assumptions fo rradiativeproperties ofmedium arc needed.Typical models for radiation modeling arc point sourceandsolid frarnemodcls.Tnpoint sourcemodclsvthe total release d heat offireandfractionwhich is converted toradiationarc required.Point source models arcaccuratefor fardistancesbutover-predict forneardistancesdue tothe fact that the thermalradiation is emitted from a single point.Incontrast, insolid framemodel sheatis em itted from an idealized sha pe (e.g.con eor cylind er).Although thismod elis simple.it requiresthe estimationofflameheight and diameter.Inconventionalsolid framemodels, allthe height of tlame was considered to participate in radiation .Howevermodified solid tlamemodelshasbeen developcdth at onlyth elum inoustl ame zon ep articipatesin calcul ations.

Wherein models suchasISIS-3D ,the onlytlameedge zone radiates and hot gases andsoo t outside thetlam edonot contributein theradiation .ISIS-3Dis abletoaccura telycalculate the charactcristics ofcngulfed objcctsa nd hasa rcIativelys hort computational time. This assumption mayresultin under-prediction of radiation calculations and heatloadon human or structure due the neglecting ofhotgases radiation above the tlame.

OthermodelsmayemployditTerent radiatio nlransport algorithmsIike ray tracing.Thisisa simple method which usesrandomlygeneratedrays betweensourceandtargetsto repr esentthe radiationphenomenon.Such asimple modelisnot ableto rcprc..escnt thcetTect of surrounding environmcntandthe effectsof hot gases ,wall sand equipmen tarcnegIccted insuch modcls.

Anothermore adva ncedmodels tohandle radia tionare Radi ati ve Transport Equ ation( RT E) mod els.Althou gh in practicalsimulationsthis model cannot be solvedaccurate ly,theaccuracy is muchbetter thanraytracing, pointsourceandsolid framemodels.RTE modelsestimate radiation intensitybasedon wavelength banding.Here.the radiat ionspectrum isdivided tosmall number of band s anda separate RTE isso lved loreachband.The wholeintcnsityi sobtainedby summation ofall bands; asthe numberof bands increases.accuracyincreases andalso the computationaltime increases .In thismodel,thelim itsof bandsarcselec tedcar efully to representthe mostimportant radiat ionbandsof impo rta ntcomponents,rnostlyCfh and11,0.

Finallyitshould bcnotcdthat radiationfractioninrcal firc sccnario sdep end s onfucl type. soot andoxygenconce ntratio nand tempe ratur e.Hence,usin g a consta ntvalue (default valueof tool)

Combustionistheprocessinwh ichspl.~iesarefonnedanddcstroyedbychemicalreactio n.

For most ofcascswhich have been discussedin this thesis.combustion isnon-premixed phenomenonlike a jet fire.poolfire or fire ball. Combustionconsistsof two coupled phenomena:therma landchemical.The chemistryof combustionishighlyexothermie(high releaserateofhcat).Therefore combustionisaself-accelerating process,inmost modelsis cons ide redas an irreve rsi blephenom en on.Thechem ica l reaction s arc dependenton thetypeof fuel. Thisis even more com plex whenthe fuclisin a liquid phase wh ichneed sto evapora te, diffuseandfinally react. Combustionoffuelsresultsingenerationof numerous produ ctsfrom

stablecompo undstolessstab leradicals.Tosimplify thecombustio n process,the onlymajor products(st able) arc considered ; CO"H,OandCO.The classi calapproachforcombustion mod elin g isbased on prob ab ili tydensityfunctionswbichallow thedeterm ination of mean reactionrate.Amorcadvanccdrcp rcscntationofcombustionrclics onFlamclct conccpt.

Touniquelydefine aproblemusing CFD.asetof bound ary and initialvalues arc need ed. Boundar y condi tio nsspecify the limitation s on the spaceand initial valucsdetincp ropcrticsin

thebeginn ing of simu latio n.Genera llythe rearctwotypes ofbo undary conditions:thermaland velocity.Thcnnalboundary conditionsincludeconstanttemperatur ewherethetempera tureof wallassumedconstant,adiabatic where a zeroheat flux isassumed.Other boundaryconditions like constant heattluxm ayb cu scd also.Settin gthe appro pria te thcnn albound ar y cond itio nis importa ntsinceitdirect lyaffectsthetem pe ratu reprofi le,smokebchavior and final conscqucncc.

Velocit yboundarycondit ionsincl udeasnoslip,slip. no penetratio nandpenetratio nboundaries.

AllCFD modelsdiscrct izetimeandspace intofinite numberofmeshes andestimate parameter s in thecente r oredgesofmesh es.Theaccuracyof approximationdependson thesize ofmcs hesand toacertai n level. asmcshisfiner.the result ismoreaccurate.Therefore,toobtain accurateresults.usersmayneed toreduce thesize ofmesh esand timesteps.Asthenumberof meshesincreases,computationaltimeincreasesaswell. Spacediscretizingmustbebasedon the geometryandcomplexityof scenario.Whilemost oftool susually applya Cartesian discretizati on.rad ial isrcquiredinsomegeometries . More adva nced toolsmay use body fitt ed coordinate(BFC) :as theobjec t deforms,thegrid deform.Application ofBFCim provesthe accuracyand may save computationaltime since the onlypartwhich isneededis calculated.

Table4.2 shows the mostcommonlyused CFD toolsand a brief description.It isuser responsibilityto comprehen siv elybe familiarwith a tool.beaware of assumptions and

Model Country Dcscriotion

ALOFT- FT US

CFX UK,

FDS US SIT e;;dhot

FIRE Australia fa

FLACS US

FLOW3D

FLUENT/AirPak US

JASMI NE UK movement

KAMELEO N Norwa

KOBRA·3D German

MEFE Portugal

ments 'omi

OpenFoam UK

PHOENICS UK

RMFIRE Canada

SMARTFIRE UK

SOFIE US/Sweden

1O01 ions

SO LVENT US

SPLASH UK ra

STAR-CD UK Generaluroosetool

UNDSAFE US/Ja an

The most significant limitation inCFD modelingiscomputationallime.Detailedmodeling maytake fromhoursto days depending on the complexityofgeo metryand used soflwarc.

Moreo ver.some lim itationsarc related tothe methodologywhich hasbeen applied.Forexample.

UNDSAFE has been developed basedon a finite differencescheme.Thus, thismodelhas an inh eren t d itli cu lty forcomplexgeometries.

Sub- mode lswh ich havebeen usedinamod elusuallyhavelimitation s andaffec t theresult s.

Forexampleifathermal elementsub-model hasbeen usedtorcombustion model ing, the outcome issignifi cantly ditTerent from a Flamcle tsub-model. The complexityof thecombustion processhasa direct effect on both n..esults and computationaltime.Species consideredin the simulationimpact radiatio nand calculatedheatloadon thcstruc ture.Furthe rmore,thegeneral techniquetorturbul en cemod elin g results in diver se outco mes. Dctcnnin ation oni f a boundary has aconstant temp erature ora constantheat nux isa complicatedtask.For example intire modeiinginsidean enclosure,the wall mayshow constant heatflux at the beginning and then switch into a constanttemperature boundary asthe temperature increases .Thisismore complicated when the propertiesof wall andits behavio risnot known complete ly.The other limitationm ay arisedue to requircdd ata andt raining.Some so llware needsspecial train ing to acquire theability to workwith its sub-models,inter-connecti on of sub-modcisandobtain

The main limitationemergeswhen the appropriatesub-mod elstosimulate thespec ifications are notpresent .For instant, sno w andice have great capacityto absorbconvcctionaIheat and refl ect rad iatio n.Thusmelti ngandevaporatio nofsno wlicehave significant effectsonthe near field due toheat abso rption. Thepresence of more vapo rsin the nearfieldaffec ts the concentrationofspec ies,stoichiometry of reactions,hot gases heat capacityand radiation parametcrsofmcdia.For farfield,thereareopposingfaclors:snow/icc refcctio n may increase the radiationintensity,however, transparencyof the nearfield air,andradiation indexof fire mayhave decreased due topresen ce ofmore vaporand less tem pera ture.

Zone modclsweredevelopedandosedtoinvesti gatethe tire andsmoke behaviorinside an enclosure .In thes emodelswhole compartmentisdivided into twoor morezonesandave rage paramctcrscalculatcd for eachzone. Asthe numbcrof zoncsincreases.the resultsimproveif geo metryis sim ple. Howe ver,asthegeometry bec ome smore complex,the accuracyislo st.In contr ast . CFDmodcl s are abletotakeinto accountmorcdctails and the so lution chan gesbased on thevaria tio nof dctail inputted.This genera lityandada ptation for differentcases and scenarios make CFDs as very usefullools[25 J.Moreover,the app licationdom ainforzo ne modelsislimited tocompartmentswhere dividing ofspace intosevera lsepara te volumes co uld bedone.lneonl rast C FDmodelscouldbeapplic'dloopcn space s.

Zone models are easy tolearnandalsorequiredveryshort computationaltimein compa risontoCFDmod els. Zone models arc effectivewhen theobjectiveistheunderstanding ofphen omena,andgeneral beh avior. Thus,they maybeapplied forquick approx ima tionandin the beginni ngofdesign.Incontras t,CFD model s req uire lon g compu tatio nal time andthuscould beused in thefina lsteps ofdesign : when delai led designisnel'dl'd.Table 4.3 comparesZone

Accurac y Acceptable forsome Dependingon the resolutionof mesh,

~:~~~~~~:2~~;::~~del ~~~I:~~~:r~~~~~~~~~~~;delS;

Applicati on s Indoorfire behaviorand smoke movementsuitable torpreliminary calculation s

~;~en~~~ageneralbehavior

Shorttime.is suitable whenLon'com utationaltime,de endson

Input data

Operation knowledge

Output data

Chapter 5: Modelingof Pool FiresinArctic Region

In the previouschapter,difTercnt typesof tiresandmodelingmcthodwercdiscusscd. Fires andespec iallypool fires arcamong the mostfrequentaccidents in process faci lities,Flame

impingement and thermalradiation arc the mainhazard ou scharacteristic ofpoolfires .Poolfires have beenthe subjectof numerousmodclingsandexperimentstudies coveringverity of areas suchastireandtlamestructure,cmissivitypowcr ,temperaturedistrib ution andfrcdistinction.

Theeff ectsofenviro nm ental paramete rs suchaswindvelocity, humidity and water/icedroplets inthcairhavcnotbcen studil~cxtcnsivcl y.Furthcr.thectTcctof SUITOundingsurface rctlectiv it y

hasnot beenstudicd.Thisissueisveryimporta ntfor coldregions like thearcticwhereoutdoor surfaccsarcco vcrt,,~ wi thsnowandicc forsevcral months o f thcycar.Furthennore,thereisno comprehensivetire consequencemodelingtoolthat includespoolfire development.

environmentalcharacteristicseffects and thermalradiation.This study proposes anew comp rehe nsivemodel forsteadystateandfullydeve lopedpoolfire.Th isnew modeltakesinto account the effectsof environme ntalvariables such astemp erature.the presence o fdroplctsand surface re tle xiv ityo n the thermal radiationandsubs equentlyo n thetireconsequcnce assessmcnt.

Instorag e facilities. the design and implementofsurround ing systemswhich collectand drainacc identalreleasedliquidhydro carbonis ausualpractice [26].Ignitio n o f theeollee ted anddrained liquid resultsinapoolfire.A pool lire is definedas aturbulencediffusionl1ame contro lled by buoyancyforces[27-29]. Herethe combus tio nphe nom eno nischaracterizedby lowmomentumdiffusionl1ames.A pool fireisdividedinto threeditTerentregion s:l1ame base

(persistenlzone),inlermitten t zone andfinaJly plume zone[ 28-30].Thepersistentregion isrich in flammablevaporsand flow islaminar andasa resultitmaintainsitsshape andstructure. The intermittentzone ischaracterizedwith lluct uatingandturbu lent 11ameandllow.In thiszone, fuel vapors arc consumedcompletely anddue to turbulcncc,thctcmpcraturcand radiationarc higher comparedto the persistentzone [31].In the plumezone no reaction0ccurs,Turbulentllowand smoke characterizethezone.Due to presenceofsmoke andsoo t. radiation Icvcl islower.

Thebebaviorofapoolfireistigbtlyboundtothepoolsize[32].Thesizeofpooleontrols thehcightofdilTcrcntzones, thermal radiation and temperature;therefore. extrapolationofthese parametersmayresultintolargeerr ors.Temperature,thermalradia tion anddurationarc themost important parametersin consequenceassessmentof pool fire [33].Dependingof these parameters. consequenceof firemay be immcdiate(e.g.the personals are exposedto radiation) or delayed(c.g.heating upthestruct uresand dom inoeff ects). Thehazard s associatedwithpool fires arc relatedtoIhethermalradi ation [34].The thenn al radiation depend s on the type o f fucl,

sootyieldand flam etemper ature s[35] . Numerousstudies havebeenconducted toinvestigatethe radiation phenome nonof poolfires.Modak[36]presented atheoretical studyfortherma l radiationofhorizontalandaxisymme tric pool tires.Orloff [37Jintrodu ced asimplemodel10 calculatetheradiatio nof pooltiresby simplifyingnon-homogenous and non- isothermalfiresto equivalent isothermal andhomoge noustires.Hami nsetal.[38]conducted aset of experime nts forditTerentfuelsto developand modifymethodologiesfortherma lradiationmeasurements.

Later,Rew etaJ. [28]proposcdascmi-empirica lcorrclatio nforlhermalra diation of hydroearbon poolfires. Chun et aJ. [34] have conductedbothexperimentsand CFDsimulation tostudy radiation of poolfires. Jensen et aJ.[35] andKrishnamoorthy[39]studieddifferentapproaches10 radiationmodeling,etTectsofsootandsmoke cfTcctson theestimatedradiatio n level. However,

environmentalvariables likesurface reflexivityorthepresenceofwaterlice dropletswerenot

Kimetal.[40]experimentallyinvestigatedvariabl es suchastheetfeet s ofdirect-downw ard water/icespray onthe burningrate,behavior and extinctionofpoolfires. Theyhave reported thatsmallwater/ice droplets areineffective and mayinverselyincreasethe burningr ate. Later studics suggcstcd that direct-downwardsprayisnot the optimum dircctionand theetlicienc y of extinctionincreasesas spraydirectiongoeslow ardhori zontal[41].Chenetal.[42 ]studiedthe effectof initialfue l tempe rature on the burning rateand buming stateso f pool lires. Their expcrimentalstudys howsthatthe durationofsteadyburning decreases asinitialtemperature of fuelincreases.Furthermore,ifinitial fuel temperaturereachesthe boilingtemperature,there wouldnotbesteadybumingperiod.Raviguru rajan[4 3]hasproposedamethodtocalculatethe etfectofwater/icedropletsont he thcrnlalradiationatt enuation.

Despitenumerousworksinvestigatingpooltires,theeff ectofenviron ment alvariableshas notbeencomprehe nsively studied.In mostofstudies justasingleenvironmentalparameterh as bccn considercd.Thecombination ofth csecn vironmcntal variableswillatTect the consequcnce ofany pooltire especially inarcticregion astheseparameterscoexist andaffectthe behaviorand thennalradiationofpooltire.In this studya newmodelisdeveloped whic htakesintoaccount environmentalparametersincludingtemperature,wind,thepresenceofdroplets andsurface rellectivity.TheFaymodel[33]wasusedasthebaseforthe model.Thismodel hasthe advantageofa continuoustemperaturedistribution.Furthcnnore,italso includestheeffectof water/icedroplets andsurface reflexivity.Therefore,the proposedmodelaccounts for all

5.2PoolFtrcModeling

Thetlam e o fpool ti resistypi call ynonuni formin tempe ratur e andspeciesconce ntra tion distribu tion intermsof three dimensional calcul ations;therefore,mode ling of pooltires is challenging.Pooltiremodelsaregenerallyclassifi edinto twogroups.The simplestmodel assumestherm alrad iat ioncontro lsfuclevaporation. Thesemodelsdonot accountfor comb ustionandair entrainme nt in their calculationsrathercalculatingrad iat ionbased onflame tempe rature. sizeandshape. OrlotT[37]presentedsuchamodeltocalculate the radiationofpool tires.Incontrasttosim plemodels,funda mentalmode lsprescribe thetirebehaviorbased onair entrainment, mixin g,comb ustion,flo wandplume rising.Fay[33] hasrecentlyprop osedsuch a mod el fo rawidevarietyo f poolsizesand wind vcl ocit ies.

5.2.1TheFayl\lodel

Thismodelisatwozonepoolfiremodelwhic h describ esflam eprop erties,combustio nand plume zones.Thecomb ustionzone beginsfromthe base of fireandgoesuptotheendofvisib le tlame.lnth is zon ethecvaporatcd fuelvaporreactswithairinstoichiometricprop ortions and fonns combustio nproducts.When fuelevaporatesfrom poolsurface,itente rsacirculation region.Fuelvaporflowsradiallytowardthe edgesof pool fire;the n,movesupward andinward toward thenam e tilt in thecenterline.Aportion of fuelvapor movesdown ward to ward the liquidsurface and hen cethecirc ulationiscompleted.This fuelcirculationprovidesfuelvapor fortl am esurface.Fuelvaporandair di ffuserespectively ou tward andinwardandintersect each

temper atureincrease anddensityredu ctionandsubs eq uently upward diffusionandbulk moveme ntofproducts.The lowdensit y and hightemp eratu reproductsenter plumezone.The

plumezone isabove combustion zoneand risesuntil temperaturesignifi cantlydecrea..o;;es and product gases dilutes.

Temperatureand fuelconcentration distributionarefunctionsofthemixingproces s.If mixing isdifTusiondrivenand lam inar, oxygenand fuclvapor difTuscand reacti on occurs.For this case.thename surface isthinand temperatureis almost equal to adiabatic flame temperature.Since thephenomenon islaminar,thechangesoftemperature and combustion

prod uctsconcen trationis sudden wherebothtemperature and combu stion products co ncentration increaserapidlyina thin layer.Incontrast tolamin ar and diffusiveflames,therapid mixing ensuresthetemperature change issmooth due toturbulence.Hence the flamesurface temperatureisless and flame surface thicknessis greatercompared tothe laminarregime.

Takin ga hori zontalplancwithcross sectionA ata heightzabove poo I fireand considerin gthe verticalvelocity10be equal10oi;mass(M),momentum(1')and energy(E) can be written as

(1)

p={(PW)WdA

E=

f

f

(2)

(3)

\Vhcre. Cp.p andpaarec onstant specificheat.flowden sityand ambient aird ensity respt.."Clively.

Massfl uxgrowth depends on theair enlrainment which isrelatedtodiameter and height of combustionand plume zones. Hencemassfluxvariationsfor these zones maybe shown byEqs, (4and5) respect ivel y.

(4)

(5)

WhereUeandUparcdimensionl ess constantstorcomb ustionandpiumezonerespec tivclyand D isthe pool diameter.Eq.(4)shows that therateof massfl uxgrowth decreasesrapid ly asmass tluxincrcascsin combustionzone .lncontrast, masstluxgrowthisju stexplicitlyproportional to momentumin the plumezone.Dueto buoyantforce andgrowthof massflux, itisexpectedthat verticalmomentumincreaseswith heightas shownbyEqs. (6and7)forco mbustio nand plume zonesrespective ly.

~ ⁼ fg ^t»;-

⁼ ~c (c!r.)(~)

(7)

Where.11.:and llrarc dimensionlessconstants.Theseequationsarcsimilarandonly vary in their constant s. Finallyvariationof energywithheightcanbeshown by Eqs.(8and 9) for combustion and plumezo nesrespectiv ely.

(8)

(9)

Whe...e,h.o,f,ritand0carefuelheatingvaluepermass offuel.massratioof prod uctstofucl in a sto ichiometric mixture,fuelburni ngrateper unitand thecombustio nzoneequivalenceratio ,

Eq.(9) shows that energy isconstantinplumezoneand isdirectly rclatedtopoollirebuming ratc.lnthiscquation,F ,rcpresentsthefuclFroudcnurnbcrandi s show n as

(10)

In theintegrat ionof Eqs.(4, 6and8),it hasbeenassumed thatmass fl ux,mome ntum and energy arczeroatthe baseofpool fire.This assumptionis validsince rele asedenergy and verticalvelocityarevery small in thebaseof fi recompared to the top of combustionzone.

Integrated form of mass.momentumand energyequationsin thecombustionzonearc

(11)

(12)

(13)

P,andEcinEqs.(12 and 13) arefunctionsofMe.however.thescaling relationtorPcandE, versus zcanbefoundtobe2andI.5respectivcly.BascdonEqs.(I I-13),averagevertical velocityand tempera tureare estimatedas

(14)

(15)

Where, w'eand T",are avera getemperatu reandvertica lvelocityin combu st ion zonc.Whilew·_e isproportionalto zo.s•'I",isindependent ofzin comb ustion zoneand isconstant. Flo warea at any heightisobtained usin g:

(16)

Finally the heighto fcombuSl ionzone can beobtai ned replaci ngE=Epat z~L"as

(17)

Forthe plumezone,Eqs.(5 and7)arc integratedto achieve massflux.momentumand encrgyrcl ation srespe ctivel y.To integrateandso lvetheseequat ions,initialcond itioncab be set equalto valuesat the end of combustionzone.Thisresultsin a newset of unknowncoetlicicnts.

Thealtemativemethodistointroduceanimaginary sourceatZobclo wthe poo lfireandtindt he plumeso lutionfor thislocation.Thisinitialcond itio nresultsinto discont inui tyofsolutionat Z=L",thisistheo nlypracticalmethodtoget reasonableequa tionsandresults.Eqs.(18and 19 ) presentthe integratedfonnto r massnux andmomentuminplumezone, respectively.

M

p

[~~pp~:~apT33

(19)

Based on thes eequationsaveragevelocity,temperature and area of the plumezo nearc

(20)

(21)

(22)

These setofcquationsrevea l thattemp eratur edecreasesrapidlyin the plumezoneasplume areaincreases dueto air entrainment.Tocompl etethe above equationscaformul awhichrelates l11assbumingratctothetire andfucl charactcri sticsi s ncL~.Thcfuclbum ing ratc hasbccnusL-d

in thefuelFroudc numbcr. Fayprcscntt-d.a sct of simplccquat ionsfor two different cases.Tf'thc thermalradiationis smallcompared toheat convection.thenthe burningratecan beexpress ed

m=(1±0.19) 1.30x

W3 (~:~;))

Where,h.undhearcheat of evaporationandfuelheating valuerespectively.(±)shows the unccrtainty ofu singthe sccquations .F orlargcp oollircswhicharc usuall yturbulent,radiation is the mainmechanism of heattransferfrom flameand hot gasesintothe liquidsurface. The burningratecanbe expressed as Eq.(24)under these conditions.

m=1.0x

1O-3(~)

Unitof'the burningrate is(kg/m's) in Eqs. (23and24 ). When pool fireeharacteristics Iike l1amctemp eratur e.diameter.burningrate , visible height arc known,the thcrmalra diationo f thc

pool lire can be simulated.Gray gas model[17] andsurface emissivepower are tbe most widely usedmodels.Fay hasuseda gray gas sub-model in hispool lire model. The main assumptionin thismodel isthatsoo t concentration is proportionalto the localconccntrationo fproduct s.

Forlargepooltires.due to signifi cantgenera tedsoots, only the baseof tire maybe assumed tocontributc intothennalradiation.The heightof thisregion isinvcrscly proportionaltosoot emissivity (k).Considering a cylindricaland untilted lire,the thermal nuxto areceiverat distancexfrom the edgc of apool fire canbecalculatedas

4 X (

~+l ~-!)

q=yaTf- - - + - ar ct an- -a rcta n -

rry~ y2_1 y-! y+! (25)

x '"kii;

'2

Atmospheric instabilityhas beenconsideredbyincluding a crosswindsub-model.Although

thecross windusuallydoesnotchangethesize of visibleflame,itcausesthe flame totilt through anangle0 from theverticalax is.Thereforein the caseof windtheheightof visible

(27)

include both vertical and horizontalmotionsofthe plume gaswith respecttothecrosswind.

Therefore,massflux and momentumforthe plumezo ne can be rewritten as

(28)

(29)

\Vhcrc. sist hc ccntcriine o ft he piumc.Forthccascofaxis)mmctricplumes,thevaluesofpv and Bhhavc bc'Cn rcportcd to be O.5and O.lrcspcclivcly [24].By inlegratingE qs.(28and29) ncwcquationsforthcplumczonecanbeobtaim..'lI.Figurc5.1presentsthesolutionstepsofthe Faymodcl.

Fil(ureS.I:Steps which mustbc followc-d in the Fay model.