Numerical Simulation of Smoke Downdrag due to a Sprinkler Spray using FDS

Transcription

1 Numerical Simulation of Smoke Downrag ue to a Sprinkler Spray using KAI YUAN LI, MICHAEL SPEARPOINT, an CHARLES FLEISCHMANN Department of Civil an Natural Resources Engineering, University of Canterbury, Private Bag, Christchurch, 81, New Zealan ABSTRACT Sprinkler interaction with a moving smoke layer, which might lea to smoke ownrag, is numerically simulate by using a large ey simulation (LES) computational flui ynamics moel, Fire Dynamics Simulator (). The simulations have focuse on computing the gas temperature an the smoke layer epth inuce by sprinkler ischarge into a hot smoke layer. Results are presente for ifferent water flow rates in aition to a simulation of a no sprinkler case. The results of the simulations are compare to measurements from a set of full-scale experiments conucte in previous research an they show that gives fairly goo preictions for the upper layer temperature. However unerestimates the stability of smoke layer an leas to smoke ownrag regarless of the water flow rate so that the lower region temperatures are higher than those measure in the experiments. KEYWORDS: sprinkler, interaction, smoke, moelling, CFD, logging, ownrag NOMENCLATURE LISTING A surface area of roplet (m 2 ) P operating pressure of the sprinkler (Pa) C D rag coefficient T temperature of roplet (K) c heat capacity of water (J kg -1 K -1 ) T g gas temperature (K) C sp coefficient for calculating the mean U smoke flow velocity (m s -1 ) roplet iameter iameter of the roplet for integration (m) U roplet velocity (m s -1 ) iameter of the roplet (m) We Weber number m mean iameter of all roplets (m) Y g water vapor mass fraction at ambient conitions (kg m -3 ) n iameter of the sprinkler nozzle (m) Y water vapor mass fraction at saturation conitions (kg m -3 ) g acceleration ue to gravity (m s -2 ) Greek h convective heat transfer coefficient (W m -2 K -1 ) ensity of surrouning air (kg m -3 ) h m mass transfer coefficient (m s -1 ) ensity of the water (kg m -3 ) h v heat of evaporation (J kg -1 ) log-normal istribution coefficient K sprinkler flow coefficient (L min -1 bar - ) Rosin-Rammler istribution exponent m roplet mass (kg) INTRODUCTION Smoke ownrag, where smoke layer stability is isrupte by a sprinkler spray with the smoke being pulle ownwarly, has been investigate by researchers over the past four ecaes [1 6]. In orer to escribe an then preict the smoke behavior uner a sprinkler spray, numerical an especially computational flui ynamics (CFD) work has been carrie out by Chow et al. [7 9], Gariner [10], McGrattan et al. [11] an O Gray an Novozhilov [12] with ifferent outcomes being obtaine by these researchers. By using a

2 RANS base Euler-Lagrange approach, Chow et al. [7 9] evelope a CFD moel to simulate the interaction between the smoke layer an sprinkler spray. In this moel, the water roplets were assume to be non-evaporating particles which acte as a source term in the smoke momentum an energy equations. The equations were then solve by using the Particle-Source-in-Cell metho. The moel preictions were compare to the experimental ata extracte from the work conucte by Ingason an Olsson [13]. Two parameters, temperature an velocity, were applie in the comparisons. It was foun by Chow et al. that the ratio of the lumpe rag force generate by all roplets an the buoyancy, which ha alreay been suggeste by Bullen [1] to etermine the onset of smoke ownrag, was not a commonly suitable approach. In the UK a quasi-fiel moel name SPLASH was evelope by Gariner [10] an later valiate by Williams [3]. However as the moel was only able to eal with a stable smoke layer, no robust conclusions were mae to the feasibility of preicting the smoke ownrag. As Fire Dynamics Simulator () has become more commonly use, researchers have starte to investigate the capability of the moel to simulate the smoke layer an sprinkler interaction issue [11, 12]. It is claime by the researchers that gives reasonably goo agreement with experimental ata for temperature [11, 12] an the velocity [12]. However it is note that in the above literature all the experimental valiations have been performe outsie of the spray region an that the smoke behavior ue to the sprinkler spray has not been investigate in any great etail. In orer to overcome this limitation, the smoke behavior insie the spray region is stuie in this paper with numerical simulations conucte by using. The temperature an the smoke layer shape are compare to previous experimental results so as to valiate the simulations. EXPERIMENTS Previous experiments [6] have been carrie out in an experimental facility as shown in Fig. 1. The experimental set up consiste of two parts; a burning cabin an a sprinkler cabin. The burning cabin was 4 m long, 2 m wie an m high. Six air supply intakes each with a 0.8 m by 0.4 m opening were locate on both sies of the cabin. Diesel fuelle pool fires were locate in the burning cabin to generate an initial smoke layer in the upper region of the sprinkler cabin. The sprinkler cabin was a cube with ientical length, with an height imensions of 4.2 m. A raft curtain with epth of m was installe to maintain an initial stable smoke layer thickness of m. A 4.2 m high gauge was place in front of the cabin to measure the epth of the ownwar smoke plume as shown in Fig. 1(b). Four thermocouple trees were istribute in a circle of iameter 1.2 m with the sprinkler at the centre. Bare bea K-type thermocouples were use with uncertainties estimate to be of less than ±2ºC an the vertical interval of the thermocouples was 0.3 m. The thermocouples were protecte by waterproofing caps to avoi the influence of the water roplets on the thermocouple bea. In this stuy, the temperatures recore by the ownstream thermocouple tree (on the right-han sie of Fig. 1a) were selecte to compare with the simulation results. For etaile comparisons, the thermocouples are labele as TC 1 to TC 13 from the top own to the bottom of cabin. An open ZSTP-15 Copper Alloys spray sprinkler with a nozzle iameter of 12.7 mm an a flow coefficient of L min -1 bar - was use for the experiments. The sprinkler was installe in the centre of the sprinkler cabin roof in a penant orientation. A pressure reucing valve an pressure transucer were installe in the pipeline to control the sprinkler operating pressure with an accuracy of 02 MPa. The sprinkler spray was manually ischarge at 50 s after ignition when the upper part of the sprinkler cabin was fille with a stable smoke layer. The operating pressure of sprinkler was varie up to 0.13 MPa. A igital vieo camera was use to recor the experiment so as to etermine the length of ownrag smoke (the smoke layer shape) after sprinkler ischarge. Seven experiments which were conucte with an ientical heat release rate were selecte for the numerical simulations. The heat release rate of the pool fires was etermine by the mass loss rate measure by an electronic balance an the calorific value of the iesel which was taken to be 400 kj/kg. The total burning time of each test was about 0 s. Previous research obtaine the burning efficiency of the iesel as 0.8 an the heat release rates for a free burning 0.8 m by 0.8m pan use in experiments was measure to be 476 kw. The efficiency of the iesel was etermine accoring to previous measurements in an ISO 9705 Calorimeter [14].

3 Fire source Draft curtain Burning cabin Air supply intakes Sprinkler cabin TC 2 TC 3 TC 4 TC 5 TC 6 TC 7 TC 8 TC 9 TC 10 TC 11 TC 13 TC 1 Sprinkler Thermocouple for comparison TC 12 (a) Schematic (a) view Depth gauge Draft curtain Burning cabin Air supply intakes Sprinkler cabin (b) Photo Fig. 1. Experimental set up (aapte from ref. 6). MODEL Water roplets solution In, the sprinkler spray is moelle by an ensemble of Lagrangian particles with momentum, mass an energy balances for each roplet being governe by [15] t m t 2 8 m U C U U U U m g T mc t m g D A h Y Y (2) m g T hv A h T (3) t where in Eq. 2, Y is the liqui equilibrium vapor mass fraction, which is etermine by using the Clausius- Clapeyron equation whereas the local gas phase vapour mass fraction, Y g, is calculate with the mass (1)

4 conservation equation for the gas phase. The rag, mass transfer, an energy transfer coefficients are etermine by a set of empirical relationships [15]. Not all roplets generate by a sprinkler are numerically compute in but instea sample sets of roplets are release into the computational omain at iscrete time steps. The properties of the roplet sample sets are chosen ranomly from the pre-specifie roplet size istribution. The mass an heat transferre for all roplets are calculate by having the sample ata multiplie by a weighting factor of the total water mass flow. Theoretically, it is inferre that more sample roplets injecte into the computation leas to a higher accuracy. However experiments have only ealt with a set of samples for etermining the roplet istributions [16]. The actual number of the roplets generate by a sprinkler is still experimentally uncertain an it is estimate to be 10 5 ~10 7 particles per secon base on the literature [16]. In this stuy 10 5 particles were generate per secon in orer to achieve computational efficiency. Further etails of the numerical moels use in are not iscusse here but can be foun in the technical reference guie [15]. Moel parameters In orer to moel a sprinkler spray in, the sprinkler relate numerical parameters have to be specifie in the input files. Due to the scarcity of etaile escriptions of the spray patterns other than the raius of floor wetting coverage provie by the supplier, the input parameters for the sprinkler use in the experiments ha to be reasonably estimate base on various stuies reporte in the literature. Sheppar [16] has measure the initial rop sizes an velocities by using particle image velocimetry (PIV) an phase oppler interferometry (PDI) for both penant an upright sprinklers. For the sprinklers use by Sheppar, the initial roplet velocity was measure to average at 0.6 of P with an accuracy of 8%. This value was therefore use to specify the sprinkler roplet initial velocity as a moel input. The spray atomisation length, which represents the location at which no further roplet break up occurs ownstream from a specific point, was taken as a istance from the sprinkler orifice of 0.2 m base on Sheppar s PIV measurements. This istance correlates well with the estimate value by Novozhilov an co-workers [17, 18]. The roplet volume meian iameter, which means half the mass of water is carrie by roplets with iameters of m or less, was etermine by using the equation reporte by Yu [19] such that m n C We sp 1/ 3 where We is the Weber number an C sp is a sprinkler constant. In this case, the value of C sp has been taken to be 1.75 for sprinklers with a 13 mm orifice iameter (15 mm nominal iameter) as obtaine by Sheppar [16], which is a more appropriate value in terms of the experimental process to use rather than the 2.33 previously use by Li et al. [6]. Values for C sp vary in the literature, for example Yu [19] gives 2.33 whereas Chow an Cheung [9] suggests 3.2. The roplet istribution was represente by a combination of log-normal an Rosin-Rammler istributions, which is expresse as: 1 e F( ) m 1 e ' ln( '/ m ) ' m m where the empirical coefficients, an have been previously foun to be 0.6 an 2.4 respectively [15]. In terms of Sheppar s work, water ischarge occurs along the range of 0~105 of the spray angle for the sprinklers with an orifice less than 25 mm. Within this range the maximum water flux occurs between 45~75 an the minimum is ischarge between 90~105. In the simulations stuie, the spray angle was (4) (5)

5 efine to be 0~105 in all simulations, which is foun to cover the majority of the experiments by Sheppar [16]. As the water flux istribution is not uniform within the range of the spray angle, the actual spray pattern will be slightly ifferent from the simulate one. The spray angle is an issue which might nee further investigation. A summary of the input parameters for the sprinkler spray are liste in Table 1. Table 1. Key input parameters for sprinkler moelling Key parameters Simulation values K-factor L min -1 bar - Approximate istance from ceiling 0.1 m Range of operating pressure 0 ~ 0.13 MPa Atomisation length 0.2 m Droplet volume meian iameter Calculate by Eq. 4 (μm) Droplet initial velocity 0.6 P (m/s) Spray angle 0~105º Droplets per secon 10 5 particles version 5. was use for the simulations in this stuy. Figure 2 shows the computational omain use to represent the experiment an the mesh use. The size of a cube gri was set to 50 mm with up to 1.43 million cells which reache the memory limitation of the available computers. Initial simulations using a coarser mesh with a cell size of 75 mm (ientical to that use by O'Gray an Novozhilov [12]) gave similar results to the finer 50 mm gri. The characteristic fire iameter for a 0 kw fire is 0.65 m an therefore a 50 mm mesh was finer than 1/10 of the characteristic fire iameter to meet the requirement for simulations [15]. To reuce overall simulation run times the numerical moel was ivie into two meshes so as to cover the burning an sprinkler cabins. Comparisons between single mesh an multi mesh simulations on heat release rate an temperature shows that the multi mesh strategy has little influence on the simulation results, as shown in Fig. 3. Two 2.33 GHz processor cores were use in parallel with 2 GB RAM for the numerical calculations on a Winows XP platform. RESULTS AND DISCUSSION Smoke temperature preictions Fig. 2. Computational omain an numerical meshes in. Recors of thermocouples TC 1, TC 8 an TC 11 have been selecte to be plotte in Fig. 4 where the outputs were given by the TEMPERATURE parameter. These particular thermocouples have been chosen as the smoke layer epth was 2 m or slightly more ue to the spill plume, the thermocouples

6 Heat release rate (kw) Temperature ( C) selecte therefore respectively represent the top of the smoke layer (TC 1), the interface region between the smoke layer an the fresh air zone (TC 8) an the lower part of the sprinkler cabin (TC 11) which is usually the fresh air zone. The simulation ata were outputte at each time step with an interval of 0.3 s an were smoothe using a 50 point averaging an also plotte in Fig Experiment TC 1 Experiment TC 8 single mesh TC 1 single mesh TC 8 multi meshes TC 1 multi meshes TC Single mesh Multi meshes (a) Heat release rate (b) Comparison of TC 1 an TC 8 Fig. 3. Comparison of ifferent mesh setup strategies. Since the burning cabin was slightly ifferent from the ISO room, the actual heat release rate therefore iffere from the measure value. In orer to create a similar smoke layer uner no sprinkler spray compare to the experiment, the moelle heat release rate was ajuste so that a steay-state peak heat release rate of 4 kw was applie to the numerical simulations. It shoul be note that the pool fire nees a certain time to get to the peak heat release rate an ecays when the pan began to run out of fuel. In terms of Reference 14, for a 0.8 m square pan, it took about s to reach the peak heat release rate the fire starte to ecay at 0 s as inferre from the temperature curves. In orer to get a similar result which is comparable to the raw ata, the fire source in was esignate as a gas burner whose heat release rate linearly increases to the peak at s an linearly ecays after 0 s until 270 s at which the heat release rate becomes zero. As shown in Fig. 4(a), the temperatures erive from for the no sprinkler case agree well with the experimental results as woul be reasonably expecte since the peak steay-state rate of heat release was ajuste to give similar layer conitions to the experiment. Once the sprinkler is operate, it is seen in Fig. 4 that the preictions at the top (TC 1) agree reasonably well with the experimental results. However as to the other thermocouples, the conclusion becomes much less clear. The gap between preiction an experiment expans for the TC 8 cases particularly for the 3 MPa an 9 MPa experiments. Overall, the preictions for TC 11 give the most scatter in the results. Typically the preicte values are higher than the experimental ones other than the 0.13 MPa experiment where the two curves agree quite well. In terms of the experimental observations, the lower thermocouples were less affecte by the steam compare to the upper ones an remaine quite ry uring the experiments. Therefore the impact of water on the low temperatures measure coul be ignore. In terms of the simulate fire an the temperature curves in the no sprinkler experiment, the perio between 100 ~ 0 s coul be regare as a steay state in the experiments. Consequently, the average temperature rise extracte from this time perio is plotte as a vertical profile for each case in Fig. 5. The overall impression is that the temperature preictions above the height of 2 m give a relatively goo agreement with the experimental recors compare to those beneath this height. Typically in the upper region the ifferences in the temperature rise between the preicte an the experimental values are less than 10 C. On the other han, the ifference in the lower region goes up to 30 C (in the 3 MPa experiment). In terms of the experimental curves, the temperature ifference between the smoke an the air layers are much clearer in the experiments with relatively low operating pressures, so that a robust smoke layer coul be efine in these cases. However it is impossible to fin a sharp temperature ifference which might be use to efine a smoke layer in the simulation curves. The lower temperatures preicte are apparently higher than the ambient temperature in all simulations, which partly inicates that the upper smoke originally in the smoke layer before sprinkler operation has been ragge own to the air layer. As a result, the preicte temperatures in the lower region are higher than the experimental values in all cases

7 Temperature ( C) Temperature ( C) Temperature ( C) Temperature ( C) Temperature ( C) stuie. These results are contrary to the experiments where it was foun that the smoke layer remaine stable in cases where the operating pressures were relatively low TC 1 TC 8 TC 11 TC 1 TC 8 TC TC1 TC8 TC11 TC1 TC8 TC (a) With no sprinkler (b) 3 MPa 70 TC1 TC8 TC11 TC1 TC8 TC11 70 TC1 TC8 TC11 TC1 TC8 TC (c) 5 MPa () 9 MPa TC1 TC8 TC11 TC1 TC8 TC (e) 0.13 MPa Fig. 4. Temperature curves preicte by at specifie sprinkler pressures compare to the experimental recorings. In orer to further investigate the ifference between preictions an the experiments, the photographs taken uring experiments are compare to the simulation results. In this case, the temperature profiles are use to represent the smoke flow pattern in. The contours are overlai on the photographs to compare the smoke layer shapes. Comparisons are plotte in Fig. 6 an analysis further confirms that gives a smoke ownrag result regarless of the operating pressure of sprinkler hence the water flow rate. The smoke flow patterns are ifferent between an the experiment. In, the smoke is pulle own to the floor by the sprinkler spray however in the experiment the smoke stops at certain height, which gives a bowl shape of the smoke layer. The most similar case from is the 0.13 MPa experiment where the smoke was ragge almost own to the floor. The temperature comparison also gives a similar result in that the temperature curves are close to each other as shown in Fig. 5(g). Qualitatively, it coul therefore be euce that the preictions from are more accurate as the operating pressure increases.

8 TC 9 TC 1 TC 10 TC 11 TC 12 TC 13 TC 8 TC 2 TC 3 TC 4 TC 5 TC 6 TC (a) With no sprinkler (b) 3 MPa (c) 5 MPa () 7 MPa Tmeperature rise ( C) (e) 9 MPa (f) MPa (g) 0.13 MPa Fig. 5. Comparison between an experimental average temperature rise at steay state.

9 (a) 3 MPa (b) 5 MPa (c) 7 MPa () 9 MPa CONCLUSIONS (e) MPa (f) 0.13 MPa Fig. 6. Comparison between visual smoke observations an temperature contours. is applie to numerically moel smoke ownrag ue to a sprinkler spray. The simulation results are compare to a set of previous experimental ata with the temperature an smoke layer shape consiere. By appropriately ajusting the peak steay-state rate of heat release it is shown that performs well

10 with a goo agreement for the temperature measurements without sprinkler spray. appears to give a reasonably goo preiction of the heat transfer when the sprinkler is ischarging. As a result, the temperature preictions above a height of 2 m in the spray region, which is mainly the hot smoke, agree reasonably well with the experiment measurements. However over-preicts the smoke ownrag in the current simulations, which actually leas to smoke being pulle own to the floor regarless of the operating pressure of the sprinkler spray. Therefore the temperatures preicte in the lower part of the spray region are typically higher than the experimental values. The current spray representation assumes the water flux istribution to be uniform over a spray angle of 0~105 an this might not be the actual spray pattern for the sprinkler. Sensitivity analysis on water flux istribution an spray angle will be carrie out later on for further investigation. ACKNOWLEDGEMENTS This work was supporte by the Opening Fun of State Key Laboratory of Fire Science of University of Science an Technology of China uner Grant No. HZ09-KF01. Kai-Yuan Li is currently the Arup Fire Post-octorate Fellow at the University of Canterbury. We woul also like to acknowlege the New Zealan Fire Service Commission for their support of the Fire Engineering programme at the University of Canterbury. REFERENCES [1] Bullen, M.L., (1974) The Effect of a Sprinkler on the Stability of a Smoke Layer Beneath a Ceiling. Fire Research Note 1016, Fire Research Station, Borehamwoo, Herts, UK, pp [2] Heskesta, G., (1991) Sprinkler/hot Layer Interaction, National Institute of Stanars an Technology, Technical Report NIST-GCR , Gaithersburg, MD, USA. [3] Williams, C., (1993) The Downwar Movement of Smoke ue to a Sprinkler Spray. PhD issertation, South Bank University, Lonon, UK. [4] Cooper, L.Y., (1995) The Interaction of an Isolate Sprinkler Spray an a Two-layer Compartment Fire Environment. Phenomena an Moel Simulations, Fire Safety Journal, 25: , [5] Cooper, L.Y., (1995) The Interaction of an Isolate Sprinkler Spray an a Two-layer Compartment Fire Environment, International Journal of Heat Mass Transfer, 38: , [6] Li, K.Y., Hu, L.H., Huo, R., Li, Y.Z., Chen, Z.B., Sun, X.Q. an Li, S.C., (09) A Mathematical Moel on Interaction of Smoke Layer with Sprinkler Spray, Fire safety Journal, 44: , [7] Chow, W.K. an Yao, B., (01) Numerical Moeling for Interaction of a Water Spray with Smoke Layer, Numerical Heat Transfer, 39: , [8] Chow, W.K., an Fong, N.K., (1991) Numerical Simulation on Cooling of the Fire-inuce Air Flow by Sprinkler Water Spray, Fire Safety Journal, 17: , [9] Chow, W.K. an Cheung, Y.L., (1994) Simulation of Sprinkler-hot Layer Interaction Using a Fiel Moel, Fire an Materials, 18: , [10] Gariner, A.J., The Mathematical Moelling of the Interaction Between Sprinkler Sprays an the Thermally Buoyant Layers of the Gas from Fires, PhD issertation, South Bank Polytechnic, Lonon, Unite Kingom, [11] McGrattan, K.B., Hamins, A. an Stroup, D., Sprinkler, Smoke & Heat Vent, Draft Curtain Interaction Large Scale Experiments an Moel Development. National Institute of Stanars an Technology, Gaithersburg, MD, USA, 1998.

11 [12] O'Gray, N. an Novozhilov, V., (09) Large Ey Simulation of Sprinkler Interaction with a Fire Ceiling Jet, Combustion Science an Technology, 181: , [13] Ingason, H. an Olsson, S., Interaction Between Sprinklers an Fire Vents, SP Report , Sween Sweish National Testing an Research Institute, [14] Yi, L., Stuy on Smoke Movement an Management in Atrium Builing. PhD issertation, University of Science an Technology of China, Hefei, Anhui, China, 05. [15] McGrattan, K., Fire Dynamics Simulator Technical Reference Guie, version 5, NIST special publication 1018, 04. [16] Sheppar, D.T., Spray Characteristics of Fire Sprinklers. PhD issertation, Northwestern University, Evanston, USA, 02. [17] Novozhilov, V., Harvie, D.J.E., Green, A.R. an Kent, J.H., (1997) A Computational Flui Dynamic Moel of Fire Burning Rate an Extinction by Water Sprinkler, Combustion Science an Technology, 123: , [18] Novozhilov, V., Harvie, D.J.E., Kent, J.H., Apte, V.B. an Pearson, D., (1997) A Computational Flui Dynamics Stuy of Woo Fire Extinguishment by Water Sprinkler, Fire Safety Journal, 29: , [19] Yu, H.Z., Investigation of Spray Patterns of Selecte Sprinklers with the FMRC Drop Size Measuring System, Fire Safety Science Proceeings of the First International Symposium, International Association for Fire Safety Science, 1986, pp ,