SCHULTE & ASSOCIATES Building Cde Cnsultants 880D Frest Avenue Evanstn, IL 60202 fpeschulte@al.cm 504/220-7475 EXCERPTS Fire Dynamics Simulatr (Versin 5) Technical Reference Guide Vlume 3: Validatin Vlume 3 f the Technical Reference Guide t Versin 5 f the Fire Dynamics Simulatr (FDS) was released n August 4, 2008. The fllwing are excerpts frm Vlume 3: The three vlumes f the FDS Technical Reference Guide are based in part n the Standard Guide fr Evaluating the Predictive Capability f Deterministic Fire Mdels, ASTM E 1355 [2]. ASTM E 1355 defines mdel evaluatin as the prcess f quantifying the accuracy f chsen results frm a mdel when applied fr a specific use. The mdel evaluatin prcess cnsists f tw main cmpnents: verificatin and validatin. Verificatin is a prcess t check the crrectness f the slutin f the gverning equatins. (Page i) Verificatin des nt imply that the gverning equatins are apprpriate; nly that the equatins are being slved crrectly. Validatin is a prcess t determine the apprpriateness f the gverning equatins as a mathematical mdel f the physical phenmena f interest. Typically, validatin invlves cmparing mdel results with experimental measurement. Differences that cannt be explained in terms f numerical errrs in the mdel r uncertainty in the measurements are attributed t the assumptins and simplificatins f the physical mdel. (Page i) Evaluatin is critical t establishing bth the acceptable uses and limitatins f a mdel. Thrughut its develpment, FDS has undergne varius frms f evaluatin, bth at NIST and beynd. This vlume prvides a survey f validatin wrk cnducted t date t evaluate FDS. (Page i) Thanks t Jerry Back, Craig Beyler and Phil DiNenn f Hughes Assciates and Pat Tatem f the Naval Research Labratry fr their cntributin f experimental data fr the HAI/ NRLWall Fire series. Thanks als t Craig Beyler fr assistance with the data fr the Beyler Hd Experiments. (Page v) FDS Technical Reference Guide-Vlume 3 1
Validatin typically invlves (1) cmparing mdel predictins with experimental measurements, (2) quantifying the differences in light f uncertainties in bth the measurements and the mdel inputs, and (3) deciding if the mdel is apprpriate fr the given applicatin. This Guide nly des (1) and (2). Number (3) is the respnsibility f the mdel user. (Page 1) Althugh the FDS develpers cntinuusly perfrm validatin studies, it is ultimately the end user f the mdel wh decides if the mdel is adequate fr the jb at hand. Thus, this Guide prvides the raw material fr a validatin study, but it des nt and cannt be cnsidered cmprehensive. (Page 1) The fllwing sectins discuss key issues that yu must cnsider when deciding whether r nt FDS has been validated. It depends n (a) the scenaris f interest, (b) the predicted quantities, and (c) the desired level f accuracy. Keep in mind that FDS can be used t mdel mst any fire scenari and predict almst any quantity f interest, but the predictin may nt be accurate because f limitatins in the descriptin f the fire physics, and als because f limited infrmatin abut the fuels, gemetry, and s n. (Page 1) When ding a validatin study, the first questin t ask is, What is the applicatin? (Page 1) Validatin studies f FDS t date have fcused mre n design applicatins than recnstructins. The reasn is that design applicatins usually invlve specified fires and demand a minimum f thermphysical prperties f real materials. Transprt f smke and heat is the primary fcus, and measurements can be limited t well-placed thermcuples, a few heat flux gauges, gas samplers, etc. Phenmena f imprtance in frensic recnstructins, like secnd item ignitin, flame spread, vitiatin effects and extinctin, are mre difficult t mdel and mre difficult t study with well-cntrlled experiments. Uncertainties in material prperties and measurements, as well as simplifying assumptins in the mdel, ften frce the cmparisn between mdel and measurement t be qualitative at best. Nevertheless, current validatin effrts are mving in the directin f these mre difficult issues. (Page 2) Keep in mind that fr any fire experiment, FDS might predict a particular quantity accurately (within the experimental uncertainty bunds, fr example), but anther quantity less accurately. Fr example, in the a series f 15 full-scale fire experiments cnducted at NIST in 2003, spnsred by the U.S. Nuclear Regulatry Cmmissin, the average ht gas layer (HGL) temperature predictins were within the accuracy f the experiments themselves, yet the smke cncentratin predictins differed frm the measurements by as much as a factr f 3. (Page 2) FDS Technical Reference Guide-Vlume 3 2
Mdel errr tends t be reduced by the averaging prcess, plus mst fire mdels, including FDS, are based n glbal mass and energy cnservatin laws that are expressed as spatial averages. (Page 2) The desired accuracy fr each predicted quantity depends n the technical issues assciated with the analysis. Yu must ask the questin: Hw accurate des the analysis have t be t answer the technical questin psed? Returning t the earlier definitins f design and recnstructin, design applicatins typically are mre accurate because the heat release rate is typically specified rather than predicted, and the initial and bundary cnditins are better characterized at least in the a- nalysis. Mathematically, a design calculatin is an example f a well-psed prblem in which the slutin f the gverning equatins is advanced in time starting frm a knwn set f initial cnditins and cnstrained by a knwn set f bundary cnditins. The accuracy f the results is a functin f the fidelity f the numerical slutin, which is mainly dependent n the size f the cmputatinal grid. (Page 3) If yu are embarking n a validatin study, yu might want t cnsider the fllwing steps: 1. Survey Chapter 2 t learn abut past effrts by thers t validate the mdel fr applicatins similar t yurs 2. Identify in Chapter 3 experimental data sets apprpriate fr yur applicatin 3. Read the specific chapters fr the quantities f interest (Page 3) Mre recently, validatin effrts have mved beynd just transprt issues t cnsider fire grwth, flame spread, suppressin, sprinkler/detectr activatin, and ther fire-specific phenmena. (Page 6) Frmal, rigrus validatin exercises are time-cnsuming and expensive. Mst validatin exercises are dne simply t assess if the mdel can be used fr a very specific purpse. While nt cmprehensive n their wn, these studies cllectively cnstitute a valuable assessment f the mdel. (Page 6) Experiments cnducted slely fr mdel validatin are smewhat rare. Mre cmmn are validatin studies that use data frm past experiments. (Page 7) FDS Technical Reference Guide-Vlume 3 3
In a fllw-up reprt, Vettri [47] extended his study t include slped ceilings, with and withut bstructins. He fund that the difference between predicted and measured sprinkler activatin times varied between 4 % and 26 % fr all cases studied. He als nted that FDS was able t predict the first activatin f a sprinkler twice as far frm the fire as anther; caused presumably by the re-directin f smke by the beams n the ceiling. (Page 9) Althugh FDS simulatins have been cmpared t actual and experimental largescale fires, it is difficult t quantify the accuracy because f the uncertainty assciated with material prperties. Mst quantified validatin wrk assciated with flame spread have been fr small, laminar flames with length scales ranging frm millimeters t a few centimeters. (Page 10) Fr real wd prducts, it is unlikely that all f the necessary prperties can be btained easily. Thus, grid sensitivity and uncertain material prperties make blind predictins f fire grwth n real materials beynd the reach f the current versin f the mdel. Hwever, the mdel can still be used fr a qualitative assessment f fire behavir as lng as the uncertainty in the flame spread rate is recgnized. (Page 10) A significant validatin effrt fr sprinkler activatin and suppressin was a prject entitled the Internatinal Fire Sprinkler, Smke and Heat Vent, Draft Curtain Fire Test Prject rganized by the Natinal Fire Prtectin Research Fundatin [60]. Thirty-nine large scale fire tests were cnducted at Underwriters Labratries in Nrthbrk, IL. The tests were aimed at evaluating the perfrmance f varius fire prtectin systems in large buildings with flat ceilings, like warehuses and big bx retail stres. (Page 10) Hwever, five experiments were perfrmed with 6 m high racks cntaining the Factry Mutual Standard Plastic Cmmdity, r Grup A Plastic. T mdel these fires, bench scale experiments were perfrmed t characterize the burning behavir f the cmmdity, and larger test fires prvided validatin data with which t test the mdel predictins f the burning rate and flame spread behavir [61, 62]. Tw t fur tier cnfiguratins were evaluated. Fr the perid f time prir t applicatin f water, the simulated heat release rate was within 20 % f the experimental heat release rates. It shuld be nted that the mdel was very sensitive t the thermal parameters and the numerical grid when used t mdel the fire grwth in the piled cmmdity tests. (Page 10) FDS Technical Reference Guide-Vlume 3 4
The scpe f the VTT wrk is cnsiderable. Assessing the accuracy f the mdel must be dne n a case by case basis. In sme cases, predictins f the burning rate f the material were based slely n its fundamental prperties, as in the heptane pl fire simulatins. In ther cases, sme prperties f the material are unknwn, as in the spruce timber simulatins. Thus, sme f the simulatins are true predictins, sme are calibratins. The intent f the authrs was t prvide guidance t engineers using the mdel as t apprpriate grid sizes and material prperties. In many cases, the numerical grid was made fairly carse t accunt fr the fact that in practice, FDS is used t mdel large spaces f which the fuel may nly cmprise a small fractin. (Page 13) The FDS results that best replicated the bserved fire behavir indicated that the pening f the basement sliding glass dr prvided xygen t a pre-heated, underventilated fire. Flashver was estimated t ccur in less than 60 s fllwing the entry f fire fighters int the basement. The resulting fire gases flwed up the basement stairs and mved acrss the living rm ceiling twards the back wall f the twnhuse. These ht gases came in direct cntact with the fire fighters wh were killed. The ht gases traversed the twnhuse in less than 2 s, giving the fire fighters little time t respnd. The mdel shwed that the xygen level was t lw t supprt flaming and, therefre, the fire fighters did nt have a visual cue f the thermal cnditins until it was t late. (Page 14) Hwever, in rder t assess the accuracy f FDS, there must be sme estimate f the cmbined effect f the uncertainty in the reprted input parameters, like the heat release rate f the fire, and the reprted measurement f the quantity f interest, like the ht gas layer (HGL) temperature. (Page 28) A summary scatter plt f the HGL predictins is given in Fig. 4.1. Mst f the predictins fall within the experimental uncertainty bunds. Nte, hwever, that bth f these quantities represent spatial averages. At any given pint in the cmpartment, a specific predictin f temperature may nt fall within the uncertainty bunds. Pint t pint cmparisns f temperature can be fund in the chapters fr plumes and ceiling jets. (Page 46) Nevertheless, temperature measurements near the ceiling can be used t evaluate the mdel s ability t predict the flw f ht gases acrss a relatively flat ceiling. (Page 57) FDS Technical Reference Guide-Vlume 3 5
The thermcuple nearest the ceiling in Tree 7, lcated twards the back f the cmpartment, has been chsen as a surrgate fr the ceiling jet temperature. Curiusly, the difference between measured and predicted temperatures is nticeably greater fr the pen dr tests. Certainly, the pen dr changes the flw pattern f the exhaust gases. Hwever, the predicted HGL heights fr the pen dr tests, shwn in the previus sectin, d nt shw a nticeable difference frm their clsed dr cunterparts. The predicted HGL temperatures are nly slightly less than thse measured in the pen dr tests, due in large part t the cntributin f Tree 7 in the layer reductin calculatin. (Page 57) Test Burner Vent First Act. (s) Ttal Acts. Draft Heat Release Rate N. Ps. Operatin Exp. FDS Exp. FDS Curtains MW @ s I-1 B Clsed 65 53 11 11 Yes 4.4 @ 50 I-2 B Manual (0:40) 66 53 12 9 Yes 4.4 @ 50 I-3 B Manual (1:30) 64 53 12 9 Yes 4.4 @ 50 I-4 C Clsed 60 52 10 11 Yes 4.4 @ 50 I-5 C Manual (0:40) 72 52 9 8 Yes 4.4 @ 50 I-6 C Manual (1:30) 62 52 8 8 Yes 4.4 @ 50 I-7 C 74 C link (DNO) 70 52 10 11 Yes 4.4 @ 50 I-8 B 74 C link (9:26) 60 53 11 11 Yes 4.4 @ 50 I-9 D 74 C link (DNO) 70 55 12 16 Yes 4.4 @ 50 I-10 D Manual (0:40) 72 55 13 14 Yes 4.4 @ 50 I-11 D 74 C link (4:48) N/A N/A N/A N/A Yes 4.4 @ 50 I-12 A Clsed 68 58 14 13 Yes 4.4 @ 50 I-13 A 74 C link (1:04) 69 59 5 11 Yes 6.0 @ 60 I-14 A Manual (0:40) 74 127 7 7 Yes 5.8 @ 60 I-15 A Manual (1:30) 64 59 5 13 Yes 5.8 @ 60 I-16 A 74 C link (1:46) 106 99 4 6 Yes 5.0 @ 110 I-17 B 100 C link (DNO) 58 54 4 6 N 4.6 @ 50 I-18 C 100 C link (DNO) 58 58 4 4 N 3.7 @ 50 I-19 A 100 C link (10:00) 56 60 10 4 N 4.6 @ 50 I-20 A 74 C link (1:20) 54 64 4 4 N 4.2 @ 50 I-21 C 74 C link (7:00) 58 52 10 4 N 4.6 @ 50 I-22 D 100 C link (DNO) 60 54 6 5 N 4.6 @ 50 Table 6.1: Results f the UL/NFPRF Experiments. Nte that DNO means Did Nt Open. Als nte, the fires grew at a rate prprtinal t the square f the time until a certain flw rate f fuel was achieved at which time the flw rate was held steady. Thus, the Heat Release Rate was the size f the fire at the time when the fuel supply was leveled ff. (Page 63) FDS Technical Reference Guide-Vlume 3 6
Figure 6.2 displays graphically the difference between predicted and measured sprinkler activatin times as a functin f burner psitin. Nte that there are n experimental uncertainty bunds n the plt because it is difficult t estimate the cmbined uncertainty related t the varius parameters that are input int the mdel. Fr example, changing the median vlumetric drplet size frm 1000 m t 750 m led t a reductin f apprximately 50 % in the number f predicted sprinkler activatins due t the increased cling f the smaller drplets. Cnsequently, three replicate experiments are cmpared t shw hw much variatin ne can expect in sprinkler activatin times in repeat experiments. (Page 63) FDS treats smke like all ther cmbustin prducts, basically a tracer gas whse mass fractin is a functin f the mixture fractin. T mdel smke mvement, the user need nly prescribe the smke yield, that is, the fractin f the fuel mass that is cnverted t smke particulate. Fr the simulatins f the NIST/NRC tests, the smke yield is specified as ne f the test parameters. Figure and Figure cntain cmparisns f measured and predicted smke cncentratin at ne measuring statin in the upper layer. There are tw bvius trends in the figures: first, the predicted cncentratins are abut 50 % higher than the measured in the pen dr tests. Secnd, the predicted cncentratins are rughly three times the measured cncentratins in the clsed dr tests. (Page 73) * * * * * FDS Technical Reference Guide-Vlume 3 7