COMPUTATIONALFLUID DYNAMICSMETHODSOF CALCULATING GAS DISPERSION IN A BUILDING OR OUTSIDEAS A RESULTOF GAS LEAKS

Ron Pape,Ph.D. Inc. EngineeringSystems Avenue 3851 Exchange Aurora, IL 60504 And Kim R. Mniszewski, F'X Engineering,lnc. 244 Ogden Avenue, Suite I l4 Hinsdale.IL 60521

ABSTRACT NIS'| Fire DynamicsSimulator(FDS) computerprogramhas beenusedto predictthe dispersionof gases insidebuildingsand outdoorslbr a variety of scenarios. 'l'hc of present olol-applications Ir[)S tbr dispcrsion gascs. l'he authclrs examplcs spill in a room,a naturalgasleak in of includedispersion vaporfiom a gasoline examples gas lcakage, comparisonof a naturalgas apizza restaurant, plumesfrom underground of area,and a simple assessment leak to a propanetank rupturein a shipping/receiving odoranttransportin naturalgasdispersion. 'l'he

INTRODUCTION The Fire DynamicsSimulator(FDS) computer model hasbeendeveloped the National by Instituteof Standards and Technology(NIS'f). Although FDS was developed primarily for simulation of fire phenomena(l), (2), (3), (4), and (5), it is a more general computationalfluid dynamics computer code suitable for application in a much wider range of fluid dynamicsproblems. The FDS model is unique in that it is liee to the public.and it enjoyscontinualfundingsupportand technical evolutionthroughthe ef-forts of NIST and other researchers. In recent years the FDS model has increasedour understanding complex fire phenomena of and has even beenusedto model the early fire (6). development the 911 incident in Although F'DSwas developedfor predictingfire dynamics,it has wider applicability. Of specific importancewith respectto this paper,FDS has been usedto model dispersionof gases insideand outsideof buildings. Mniszewskiand Pape(7) have usedFDS fbr pure without combustionand comparedpredictedresultsto experiments dispersion analyses and analyticalpredictions. FDS has also been used for indoor air quality calculationsby Musser, et al. (8) In the area of ComputationalWind lrngineering,Rehm, McGrattan, and Baum,havetested FDS (9) ( l0). The authors use the l.'DS model extensively lor applicationssuch as these. I'he F'DS model has been testedagainstexperimentaldata fbr a variety of situations,however the preponderance such validationtestshave becn conccrned with the prcdictionclf firc of with predictionof gas dispersion effects. This paperis concerned using FDS. Several examplesof dispersionmodeling using FDS are presented. With proper validation, this modeling technique should be of value in many gas industry applications,particularly thoseinvolvingpublic salbtystudies. As FDS is free to the public, continually being improved and updatedand user-lriendly, and powerful computationalfluid dynamicsmodeling tool is available this sophisticated to fbr anyonewho takesthe time to leam it. in contrast other CIFDmodelswhich may be more costlyto learnand use. DESCRIPTION OF THE FDS MODEL at fluid dynamics(CFD) model developed the NationalInstitute FDS is a computational (NIST). Computational is f'luiddynamics the mathematical and of Standards Technology solution of the equationsthat describe fluid motion. [rDS stands for Fire Dynamics Simulator. Although the FDS code has beendevelopedfor predictionof fire behavior,it is applicable for general fluid dynamics computations.such as wind f'low around solverfbr low Mach numbers.The equation The FDS code is a Navier-Stokes obstacles. FDS equationsdescribethe motion of a fluid, such as air, and include conservationof mass, momentum, and energy relations. Analyses presented in this paper were usingVersionsI .0 through4.0 of the code. accomplished

The FDS computercode is describedin detail in a number of references (l), (2), (3), (4) and (5). As describedin thesedocuments,FDS solvesthe fluid dynamicsequationsfor low Mach number flows. The governingequationsare the equationsfor conservation of mass,momentum,and energy.where energyconservation included in the equationfor is flow divergence.The simplified equations that are solvednumericallyare given below: Mass:
) :c * o
r1 |

. p u

(1)

Momentum:

+ ,++(u'V)uv

V - ( p - p , ) g = f + V

(2)

[)iversence Constraint:
t / t/ t V . u = - 1 . t . k v T ' * v .lIc n , d T ' o - y,.y ! ,,r o ' l * - : v D y

p',r\

- L l \+
P,,)tlr

)\r*',T

(3)

p, wherep is massdensity, is the ambient density, is time. u is velocity(bold indicates t vector quantities), I is the perturbationpressure.p,, is the backgroundpressure,cn is k D; specificheat.T is temperature. is thermalconductivity. is diffusivity for speciel, Y1 is the mass fraction for specie I, g is gravitational acceleration,f is the body force (excludinggravity),and r is the shearstress.["orthe low Mach Numbcr fbrm. prcssure is pressureadded to the hydrodynamicpressureand a written as the averagebackground flow-induced perturbationpressure,V. The cquation of state in the model is the ideal gaslaw. The FDS model has two solution options for capturing turbulence. These are direct presented numericalsimulation(DNS) and largeeddy simulation(LES). All simulations in this report used the LIJS approach.l,argc cddy simulationallows direct numerical solution of the larger scale lluid motion (including large turbulenteddies)but greatly reducesthe requiredcomputationaltime by calculatingthe sub-grid scaleviscosity using viscositymodel. (ll) (12) The an eddy viscositybased(in FDS) on the Smagorinsky ability to use larger grid elementsand account fbr the sub-grid mixing using the problems in Smagorinskyviscosity allows solution of complex three-dimensional Large eddy simulation is now practical computationaltimes and memory requirements. an option in many CFD codes,as well as FDS. in similarlyfbr all gasspecies the model.wherediffusioncoefficients Diflusion is treated put into question usefulness the model in quiescent of the arethe same. This tendsto the that in most fuel-gasleakagescenarios, mass applications. However,it is reassuring the subjectgastendsto be dominatedby the bulk movementfrom the gas transferof that in not eff'ects, diffusion. Thus.it is expected release, turbulentmixing and buoyancy well. the of the most cases. physics the modelwill approximate realworld reasonably

The primary limitation of the code is its validity for low Mach numbers. This requires that the maximum flow velocitiesbe well below the speedof soundin the medium being analyzed.When considering fact that the speedof soundin air at ambientconditions the is about340 m/s (1100 fils). this is not reallya practical restriction a wide varietyof for problems,particularlyinvolving lire development, dispersionof gases and wind flow aroundstructures.For dispersionanalyses, low Mach numberrestrictionmay be the significantif it is necessary modelthe detailsof a high velocitygasjet very ncara to leak,but generally detailed the leak flow right at the source not importantas long as the is overall leak flow rate is accounted for. geometries ol-structures IrDS simulations in must bc represented composites as made of rectangular solidblocks.This is sometimes listedas a linritation the FDS code of because curved surfaces approximated a stair-step are by arrangement.However,when this restrictionis balanced againstthe simplificationsand consequent efficiencythat results. block structure well worth it. the is F'igures md 2 are examplesof resultsfrom IiDS analyses I o1'lire (what FDS was original designedfor). Figure I is the simple burning of an itcm undera hood, showing thc flow of hot gases usingtraccrparticlcs and hcatingof a ncarbysurface.Figurc2 showsa kitchenrangefrre in a townhouse.Flameplumesare shownspreading through the rooms. Many optionsare availableto allow for displayingquite a numberof parameters, gas including temperature, smokedensity,heatrelease, concentrations, etc. FDS VALIDATION A greatdeal of effort has beenexpended NIST in the last severalyearsin the by validation of the FDS model involving fire phenomena.The currcntedition of the FDS 'l'cchnical a liom many sourccs citing Rel-crencc Guidc (4) provides long list ref'ercnccs Much of the validationwork hasbeen validationcases and application techniques. concemedwith fire phenomena.Mniszewski and Pape(7) have testedF'DSfor pure without combustionand comparedpredictedresultsto experiments dispersionanalyses predictions.Musser. al. (8) haveusedFDS tbr indoorair quality et and analltical F of Rehm,McGrattan, and Baum,havetested DS (9) ( l0) lbr prediction calculations. wind flow aroundbuildings. (iuidc (13) states "AlthoughfrDSwasdesigned that 4 The FDS version lJser's that do not it fbr specifically frre simulations. can be usedfor otherfluid flow simulations an kind". This guide also presents exampleof a include fire or heataddition of aury filled with air. Dr. McGrattenof NIST is the main release helium into a compartment of thc in architect the FDS software.IIe hasencouraged authors the useof thc modelfor of guidance how to implementit on and applications, hasprovideduseful dispersion properly. 'l'he

Figure l. IdealizedFire Experiment Simulation.

ffiffir

,Y.f*{ffi

wry*R
Analysis. Figure2. Exampleof FDS for Fire Development SuggestedAdditional Experimental Validations. Although therehas beensome testingof the FDS computercode for dispersionanalysis,for gas industryapplicationsit validations experimental to would be desirable conducta numberof highly instrumented might approach One suggested naturalgasleak scenarios. involving sometypical include: home/2 storywith basement leak single-family l. Indoor/typicalresidential at involving a failed flexibleconnector the kitchenrange scenario

2. Indoor commercialfacility - leak scenarioinvolving a rupturedI %" gaspipe pipe ruptureat involving underground 3. Outdoor Scenario - leak scenarios variousflow rates While thereare safetyconcerns using undilutednaturalgasfor suchtesting,the useof in diluted gastestingor useof an appropriate simulantgasmay help qualm public concem in real buildings. With naturalgassource concentrations below kept and allow testing shouldbe the LEL at perhaps50% LEL for safety,the masstransferphenomena reasonably similar to sources undilutedconcentrations.If a simulantgaswere used, at would have to be considered and proper molecularweight and diffusion characteristics would be neccssary A arrayof gassensors scalingof results evaluated. reasonable in throughout eachtestingvolume. Infiltrationwould needto be considered indoor validationanalyses testingto somedegree. Suchadditionaltesting,with corresponding using FDS, would increase in conf.idence the predictionsof FDS for suchcases. Propane leak scenarios be considered similarly. can

EXAMPLES OF DISPERSIONANALYSES USING FDS previously Mniszewski werediscussed by simulations Several examples gasdispersion ol 'I'he here)includethe discussed that time (not repeated at and Papein (7). cases following: o Mixing Theory with One Room Model Perl-ect Comparison

Perfbctmixing theory predictionswere comparedto FDS simulationresultslbr lbr werecompleted dispersion dispersion gasinsidea singleroom. Calculations of perl-cct mixing within thc room with and without room fbrcedventilation.Whcreas FDS predicts more gradual a two distinctlayers. on theoryis based formationof within the upperspaces the room and the lower of transitionfrom the concentrations spaces. o in fbr with'l'heory VaporDispersion a Wind Comparison

leaking fiom a componentin Predictions dispersionin an imposedwind o1'vapors of (e.g.a valve or pump) were comparedto concentration plant a chemicalprocess point source vapors. which therewas an for of profilesproduced fiom an idealized effbctare significant. wheregeometry analyticsolution. Exceptnearto the source, the FDS resultscomparedquite good to the ideal point sourcesolution. o Migrationin a Room fbr with Experiment Propane Comparison

FDS was usedto simulate an experiment fbr which there appearedto be a good descriptionof the setupand resultsin a paperin the literature. The experiment and carbondioxidein a room with of involveddispersion a 50:50mixtureof propane

gas air infiltration. FDS over-predicted experimental concentrations, the which may haveresultedfrom an oversimplificationof the distributionof air infiltration. Dispersion Vaporsfiom a Gasoline of Spill in a Room Severalexperiments were performedinsidean enclosure simulatinga closedgaragewith a water heaterin a corner. In the centerof the floor, therewas a gasolinespill or a release of a simulant(carbon dioxide). Initially.the enclosure was kept closed. eitherwith or without room forcedventilation. six minutesinto the experiment, At fbrcedventilation was introducedthrough a slot at f'loor level at the oppositeend of the enclosure, simulatingcrackingopen a garagedoor with outsidewind. Figure3 showsthe concentrations gasoline vaporsbeforeand afterthe outsidewind was introduced of throughthe slot. The FDS predictions time lbr flammable of concentrations reachthe to water heatercombustionchambermatchedthe time to explosionin the gasolinetests reasonably well.

. , i

Figure 3. GasolineSpill in Room With Water Heater. in NaturalGas Dispersion a PizzaRestaurant 'l'his naturalgas leak causedby extemal forces, exampleinvolves an underground adjacentto an old commercialbuilding with a graniteblock foundation. Iterative into the basemenl. usingthe known the modelingwas utilizedto establish rateof leakage ignition source andthc probable leak initiationand thc explosion. timing between (furnacepilot) locationas constraints.Figure4 is a vertical slice of gasconcentrations distributionof naturalgasthroughout for one leak rate scenarioshowingconcentration 'l'he of gasconcentration shownto be highestnearthe source the outside is the building. while openstairwaydoorsallow gasto riseand lill gradelevel leak in the basement, areas.

Figure 4. Gas Dispersionin a Pizza Restaurant. Gas Plumesfrom Undereround Leakage exampleinvolvesthe possible of modelingto estimate abovcgroundgas use the plumeavailable fiom a varietyol-undcrground leaksizesand wind conditions. Figure5 showsan exampleof resultsfrom a 1270CFtl naturalgas leak distributedover a square meter,with a wind speed 2.24 mph (0.1 m/s). Suchresults of may allow a field engineer to estimate leakage ratebelow by simply measuring somepointswithin the gasplume and wind conditions. Natural Gas VersusPropaneDispersionin a Shipping/Receiving Area An explosion area occurrcdinsidethe shipping/rcceiving of a fbod processing facility. The shipping/receiving was covered a peaked arca roof, but it was otherwise by opento the outdoorsand wind penetration.Figure6 showsthe facility with the white roofed at the middle coveringthe shipping/receiving area. Figure7 showsthe layout of the facility with the roof removed.T'heshipping/receiving had two lorklift trucksin the rearleft area cornerand the propane tank on one was beingchanged.One possible cause scenario involved an overfilled propanetank rupturing. A secondcausescenariothat was involved a naturalgasleak emanatingfrom a floor crack at the right-handside evaluated of the shipping/receiving area,behinda plywoodpartition. Figure8 showsthe results for the naturalgascase,and Figure9 showsthe resultsfor thc propanctank rupturc. Clearly, the propanetank rupture was capableof producing a large flammable cloud whereasthe naturalgas leak producedconcentrations the spacewell below the lower explosion in limit. 'fhis

s m o t e v r e w 4 0 5 F e b 1 42 0 0 5

il

0r5

014

0 t2

0rl

009

008

!6

1$;,

005

003

,I
0 0"1 Frrne5i6 l L n ei J J

Figure 5. Dispersionof Gas From An Underground Leak.

Facility. Figure6. FoodProcessing

i l

r'-----J-,
.3..-

,,.,

q--

Figure 7. Facility Layout.

tLllllu i N"G"

NaturHff 'n@as &0

h f$r

,rVh

40

zu*
*:.

fl"t0)
o

Figure 8. Natural Gas Leak Scenario.

l0

\ t

qY.
!'l'

miflm LEL*
'I
l l

f,

ll"

'Pnolgam ilm-frn

Figure 9. PropaneTank Rupture Scenario. OdorantTransport final caseto be discusscd was done to evaluate propensityfor odorantto separate the from naturalgasduringdispersion air. A simplecubicalenclosure in was considered. A small sectit-rn the f'loorwas given a100o/o of concentration methane, no lbrcedflow of but into the enclosure. Odorantwas appliedto the methane concentrations to l% by at up volume. Irigure l0 showsthe result fbr this case,with the methane rising to the gas ceiling as a plume by naturalconvectiondue to its low dcnsity. When the methane source was at thc ceiling it.iustcollccted bcncath cciling. Whenthc sourccwas on the the sidewall, a wall plumeroscto the cciling. In all cases, odorantconccntration was 'l'he monitored. odorantconcentration remainedidenticallyat the fraction of the gas 'fhere concentration which it was injected,without deviationin any case. at was no separation the odorantliom the carriergas. of CONCLUSIONS Basedon the varietyof gasdispersion analyses havebeenconducted. following that the conclusions reached: are
o o

'fhe

o o

FDS is valid for conductingdispersionanalyses Someof the validations FDS for dispersion examples dispersion of and of in applications usingFDS havebeenpresented this paper. problems needed. is More validationof FDS fbr dispersion validation. Suggestions provided fbr more detailedexperimental are Benefitsto the Gas lndustry from theseefforts will includebetteranalysistools for public safetystudies.

1t

SmnkPViPw4n4 Nnv16 Tnnd

SlieP

'.{i
, l

nnq

nn4 0.04 nn? 0.03 nn2 nn2 0.01 q nF-3 4.5E-3 nnn iu
!

F r a m e9 5 0 :
Timp 342 tl

Figurc 10. Methane PlumeWith OdorantInjected

REFERENCES

l . McGrattan, K.B, et al. "Fire Dynamics Simulator-TechnicalReference Guide",

NISTIR 6467,National Instituteof Standards Technology,January 2000 and "liire DynamicsSimulatorVersion 2) -'l'echnical[tefcrence 2. Mc(irattan, K.B, et al, November and Technology. Guide", NISTIR 6783. National Instituteof Standards 2001 J. McGrattan, K.B, et al, "Fire Dynamics Simulator (Version 3) -Technical Reference and Technology, Guide", NIS'flR 6783 2002 Ed., National Instituteof Standards November2002. 4 . McGrattan, K.B, "fiire Dynamics Simulator (Version 4) -Technical Reference Guide", NIST Special Publication 1018, National Institute of Standardsand Technology. July 2004. 5 . McGrattan. K., "Computational Fluid Dynamics and Fire Modeling," Fall (ftp.nist.gov/pub/bfrl/mcgratta/fds3/MANUALS/class2002.pdf) 2001. "ProgressReport on the FederalBuilding and Fire Salbty Investigationof the World June2004. NIST Special Publication1000-5. TradeCenterDisaster."
1

t2

7. Mniszewski, K.R., and R. Pape, "The Use of FDS for Estimation of Flammable GasAy'apor Concentrations," Technical Symposiumon Computer Applications in 3'o Fire Protection Engineering, Societyof Fire Protection Engineers, 12-13September 2001. 8. Musser, A., K.B. McGrattan, and J. Palmer, "Evaluation of a F'ast, Simplified Computationalfluid DynamicsModel for Solving Room Airflow Problems,NISTIRn 6760. NationalInstituteof Standards Technology. and Gaithersburg, Maryland,June 2001. 9. Rehm, R.G., K.8., McGrattan,and H.R. Baum, "An Efficient Large Eddy Simulation Algorithm for Computational Wind Engineering:Application to Surface Pressure Computations a SingleBuilding."NISl'lR 6371. on 10.Rehm, R.G., K.8.. McGrattan, and H.R. Baurn,"L.arge Irddy Simulationo1'Flow Over a wooded Building Complex," Wind and Structures, Vol. 5 no. 2-4 (2002) 291300. 1l. Smagorinsky.J.. "General Circulation Experimentswith Primative I,.quations," M o n t h l yW e u t h eR e v i e w . 9 1N u m b e r , 1 9 6 3 p p . 9 9 - 1 6 4 . r . . 3 12.Smagorinsky, S. Manabe,and J.L. Holloway, "Numerical resultsfiom a NineJ., Levef GeneralCirculation Model of the Atmosphere,"Monthly ll/eatherReview,93, N u m b e r1 2 .1 9 6 5 , p . 7 2 7 - 7 6 8 . p 13.McGrattan, K.B. et al "Fire DynamicsSimulator(version4) lJser'sGuide",NIST Special Publication 1019, National Institute of Standards and T'echnology, February 2005

1 a I J