WATER MIST SPRAY MODELLING WITH FDS

Transcription

1 PROCEEDINGS, Fire an Evacuation Moeling Technical Conference 0 Baltimore, Marylan, August 5-6, 0 WATER MIST SPRAY MODELLING WITH FDS Simo Hostikka, Jukka Vaari, Topi Sikanen, Antti Paajanen VTT Technical Research Centre of Finlan P.O.Box 000, Espoo FI-0044 VTT, Finlan simo.hostikka@vtt.fi ABSTRACT Numerical simulation of water mist systems is gaining interest among fire safety engineers ue to the increasing range of water mist suppression an protection applications, an the lack of simple an general esign rules. The existing Eulerian- Lagrangian framework of FDS can be use for water mist simulations, but the publishe reports on the simulation process an valiations are few. In comparison to traitional sprinkler technology, water mist has a wier range of physical suppression an cooling effects. An accurate moeling of such systems requires that all the relevant input parameters can be prescribe. From the moeling viewpoint, an important ifference is the stronger exchange of momentum between the gas an liqui phases. The FDS capabilities an improvements concerning the spray ynamics an heat transfer, incluing the experimental valiation, are presente. The experimental work inclues the measurements of rop size, rop spee an mist flux profiles, gas phase entrainment spee in a channel geometry an raiative heat flux attenuation. INTRODUCTION Over the past 0 years, emonstrating the suppression performance of fixe water-base fire fighting systems has been one through full-scale fire testing, although tentative computational capabilities an tools for preicting the suppression system performance have existe all along. Water mist systems represent a recent evelopment for water base fire suppression technology. The 0-year commercial history of water mist systems has seen a large number of experiments, but no general esign an installation rules have emerge. This is primarily because there are several extinguishing mechanisms for water mist, the most important being gas-phase cooling, surface wetting, an scattering an absorption of heat raiation. The relative importance of the mechanisms is ifficult to quantify as it epens on the technical etails of the water mist system, as well as the application. The R&D of large fire suppression systems calls for abunant resources both in terms of time an money. It can be expecte that the nee for full scale experiments can be reuce by making use of stateof-the-art fire simulation software in the R&D process. Such tools are in everyay use in the fiel of Fire Safety Engineering. Yet, these tools have not been applie in simulating the performance of active fire suppression systems. This evelopment is seen as evient in the near future, an inee it is recently recognize as the top priority by the International Forum of Fire Research Directors (Grosshanler 007). In this presentation, we present enhancements to an valiation of Fire Dynamics Simulator (FDS) (McGrattan et al. 007) in escribing water spray ynamics, air entrainment an raiation attenuation. SPRAY COMPUTATIONS IN FDS The motion of a single spherical roplet is governe by the equation of motion, consiering acceleration by gravity an rag between the gas an a roplet. m t u m g gcdr u u () where m is the roplet mass, u is the roplet velocity, g is the gravity vector, C D is the rag coefficient, r is the roplet raius an u rel is the relative velocity between roplet an gas. Recently, Ranall McDermott an Howar Baum of NIST have evelope an analytical solution for the roplet an gas velocities in a multi-roplet-gas system uring a short time interval. This analytical solution is use to upate the roplet positions an velocities, as well as gas velocity, uring each FDS time step. In ense sprays, the iniviual roplets can start to influence each other through aeroynamic interactions. These aeroynamic interactions become important when the average roplet spacing is less than 0 roplet iameters. This correspons approximately to a roplet volume fraction = 0.0, achievable in some water mist suppression systems. The reuction of hyroynamic forces to the secon (trailing) sphere ue to the wake effect is calculate using the correlation evelope by Ramírez-Mnoz et al. (007). The rag coefficient is calculate as C D = (F/F 0 ) C D0, where C D0 is the rag coefficient of a single roplet. The ratio of hyroynamic forces on trailing an single roplets is rel rel

2 F Re Re W exp, () F0 6 / L / 6 L where Re is the single roplet-reynols number, L is the istance between the roplets an W is the nonimensional, non-isturbe wake velocity at the centre of the trailing roplet C D0 Re W exp. (3) 6 L / In above, is the roplet raius. This moel assumes that the trailing roplets are perfectly in line, an thus provies an upper boun estimate for the rag reuction. On the other han, a comparison against the results by Prahl et al. (009) suggests that the above correlation unerestimates the rag reuction when L/ < 3. Droplets are introuce in to the computational omain on a segment of a spherical surface with origin at the sprinkler nozzle location an raius etermine by the offset parameter. The spray angle outlines the soli angle of this sphere segment. The insertion points of the particles are selecte as follows. The longitue is uniformly istribute between 0 an. The latitue is ranomly chosen from a probability ensity function P f sin (4) where f() is a istribution function. In the following simulations, a Gaussian profile was assume max f exp x, x (5) max Here min an max are the angles that outline the spray cone. For full cone sprays, the efault parameters are = 5 an = 0. The cumulative volume fraction of roplet iameters follows a istribution that is a combination of lognormal an Rosin-Rammler istributions: F 0 e e ln / m m min m m (6) By efault =.5/ so that the probability ensity function is continuous. In simulations the particle size is boune from below. Smaller roplets are assume to vaporize instantly. The meian roplet size epens on the operating pressure use. Since the experimental roplet size istribution is etermine at certain pressure, this variation in roplet size is taken into account by scaling the meian roplet size as m p -/3. The spray ynamics involves strong coupling between the Lagrangian an Eulerian phases. For the simulation of the water mist sprays, the accuracy of spray turbulence is very important. The inclusion of ynamic sub-gri scale turbulence moel in FDS6 provie a significant improvement in the spray preictions of the spray with an velocities. The main features of the raiation-spray interaction moel are escribe in Hostikka & McGrattan 006. The most important ifference between the current moel an the original version escribe in the references is the use of single roplet iameter for the spray raiation property calculation instea of averaging over the rop size istribution. During the simulation, the properties are foun from a look-up table using the local Sauter-mean iameter 3. This metho is expecte to be better justifie in situations where the local rop size istribution has not the same functional form as the original size istribution. EXAMPLE NOZZLES The example nozzles were Spraying Systems LN- an three high pressure micro nozzles A, B an C from Marioff Corporation. They were moelle using the knowlege of the operating pressures, experimentally etermine flow rates an roplet size istributions. The roplets were introuce to the simulation omain on a section of a spherical surface meters away from the nozzle location. The spray angles were etermine from close-up photographs. The initial roplet velocities were calculate from the pipe pressure P P v, C (7) 0 where is water ensity an C was taken to be 0.95 to account for friction. The roplet size istribution parameters were foun by least squares fit of Eq. (6) to the experimentally etermine cumulative volume istribution. Fitting the FDS cumulative number istribution to the experimentally measure cumulative number istribution was also teste an it was iscovere that these two methos resulte in significantly ifferent istribution parameters. For nozzle C, this metho was use. An example of size istribution is shown in Figure for nozzle A. The moel parameters are liste in Table.

3 Cumulative volume fraction Diameter (m) Experiment FDS Figure : FDS roplet size istribution fitte to the experimental ata for nozzle A. 50 Table : Moel parameters for nozzles. nozzle K (l/min/bar ½ ) (eg) m (m) LN A B C profiles were etermine with a simulate phase Doppler particle analyzer collecting roplet statistics from a -cm spherical volume aroun each measurement location. A comparison between measure an simulate LN- profiles at istance 40 cm below the nozzle is shown in Figure. Pressure was 0 bar. Simulation results are compute at, an 4 cm spatial resolutions. The simulate roplet spee profiles are not fully converge even at cm resolution, as a mesh refinement tens to make the profile wier, reucing the peak value. The mist flux an mean iameter profiles are quite well converge, showing a reasonable agreement with experimental ata. The experimental mist flux at spray axis is lower than at 5 cm istance, inicating that the assumption about full-cone profile with Gaussian istribution may not be correct. The multi-orifice spray heas use in this stuy are summarize in Table. They are constructe by attaching single orifice nozzles of types A, B an C into a spray hea boy. The assemble spray hea has a centre nozzle spraying in the axial irection, an a number of orifices istribute evenly at the perimeter, each spraying at an angle with respect to the axial irection. In the simulations the multiorifice spray heas were moele using several iniviual single orifice nozzles place at the same location with but with ifferent orientations. Table : The multi-orifice spray heas. SH SH SH3 SH4 SH5 Centre nozzle A C B B B Perimeter nozzle A B A B B Number of perimeter nozzles Perimeter angle (eg) RESULTS AND DISCUSSION Spray profiles Raial profiles of the mean roplet spee, mist flux an mean rop size were measure using irect imaging (DI) technique. In the simulations, the spray Figure : Experimental an simulate roplet spee, mist flux an mean iameter profiles for LN- nozzle.

4 It is interesting that the simulations can accurately preict the formation of the V-shape iameter profile, cause by relatively higher proportion of large, high momentum roplets on the outer parts of the spray an accumulation of smaller roplets to the central spray by the air entrainment. Similar profile was also observe at a istance of 6 cm. The experimental an simulate results for micro nozzles A, B an C at 70 bar pressure are shown in Figure 3. The experimental uncertainty of the results is high because sprays were sometimes unsymmetrical. The simulations were performe with - cm spatial resolution. Velocities show a reasonable agreement with experimental results. Mist fluxes on the spray centreline are significantly higher than the experimentally etermine values. Further away from the spray centreline the iscrepancy between the simulation an experiment iminishes. It shoul be note however that there is consierable measurement uncertainty associate with the experimental mist flux values. Qualitatively the mist flux profiles are correct: there is a relatively thin ense core with a less ense outer spray as observe uring experiments. The raial rop size profiles are foun to be flat an quite ifferent from the V-shape profile observe for LN- nozzle. This may be ue to the significantly higher initial spees ( m/s vs. 60 m/s) an the resulting ifference in spray turbulence. The gri sensitivity stuy for nozzles A, B an C inicate that the an cm resolutions gave practically same results but 4 cm resolution showe significant ifferences. The results were not sensitive to the offset istance that was varie between 5 an 5 cm, but the velocity an mist flux profiles were very sensitive to 50 % variations in spray angle. In most simulations, 0 5 roplets per secon per nozzle were introuce into computation. Increasing the number of particles to 0 6 or 0 7 per secon tene to yiel larger velocities on the spray centreline but i not otherwise affect the results. The minimum iameter min was set to m in most simulations. Sensitivity analysis showe that this parameter i not have a noticeable effect on the results. Restricting the global time-step or increasing the number of sub-time step iterations i not improve the results either. The effect of aeroynamic interactions on water mist properties was investigate by running the simulations with the aeroynamic interaction moel turne on an off. The rag reuction by aeroynamic interactions ha a very moest effect on the results. The most noticeable effect was the slight flattening of the roplet iameter profile. The roplet volume fractions in the ensest parts of the spray are just slightly over 0.0 for all nozzles. These results inicate that the inclusion of roplet-roplet aeroynamic interactions is not necessary for the simulation of practical water mist systems. However, ae cost of the rag reuction computation is small. Figure 3: Experimental an simulate roplet spee, mist flux an mean iameter profiles of A, B an C micro nozzles. Air entrainment The capability of the water mist spray hea to transfer the water momentum into the gas phase an to retain the entraine air flow is one of the main characteristics that etermine how well the spray can penetrate into hot plumes an how efficiently the system can mix up the gas in close space applications. Air entrainment to the multi-orifice spray hea water sprays was measure using a.0 m long a rectangular plywoo channel having a cross section of 0.6 m by

5 0.6 m. A picture of the channel with a nozzle operating is shown in Figure 4. The spray heas were installe at the mipoint of the channel an they were spraying along the channel axis. The air velocity in the irection of the channel axis was measure with a bi-irectional probe. Measurements were taken 0.5 m behin the nozzle on the channel axis an 0.06 m from the channel wall. The bi-irectional probe an the associate ifferential pressure transucer were calibrate using a hot-wire anemometer. In each test, the water pressure was measure immeiately outsie the channel wall using a capacitive pressure transucer. In the experiments, a wooen obstacle was place close to the channel inlet, isturbing the entrainment airflow. To capture this effect, the computational omain was extene outsie the channel. An overview of the simulation geometry is shown in Figure 4 (b). The spatial resolution in the simulations was cm. in the amount of momentum injecte in to the simulation. The SH5 spray-hea has the perimeter nozzles at 30 egree angle, giving the highest momentum of all the spray heas consiere in this paper. In general, we can conclue that the moel is able to capture the effects of the pipe pressure an the main characteristic ifferences between the spray heas in terms of the air entrainment. The offset parameter ha a significant effect on the simulation results. Offset value of 4 cm was use here instea of the 0 cm. When using a smaller offset value, a reverse flow on the channel axis behin the spray hea was observe. Appropriate offset value was foun to epen on the numerical resolution use: The offset shoul be large enough to istribute the roplets within more than one computational cell. In the case of multi-orifice spray heas, it is also important to ensure that the gri is fine enough to resolve each of the spray jets. This may become a problem with small perimeter angles. a) a) b) Figure 4: Air entrainment measurement channel picture (a) an FDS moel (b). Comparisons of the center line an close-to-the-wall velocities for the multi-orifice spray-heas are shown in Figures 5 (a) an (b), respectively. The agreement with experiment is goo on the center line of the channel. While the spray heas SH4 an SH5 are both constructe from the same B type orifices, the velocities on the channel axis are slightly over preicte for SH4 an significantly uner preicte for SH5. The ifference between these spray heas is b) Figure 5: Air entrainment velocity for multi-orifice spray heas in (a) channel center an (b) 6 cm from the wall of the channel. Raiation attenuation The attenuation of thermal raiation by the micro nozzles A, B an C was measure using a raiant panel heat source an heat flux gauge. A schematic iagram of the experimental arrangement is shown in

6 Figure 6. The heat source was a LPG gas heater unit with a total heat output of 3.5 kw, raiating area of 0 cm x 30 cm an maximum surface temperature of 950 C. The experiments were performe at 50, 70 an 00 bar pressures. The experimental uncertainty of the attenuation measurements was less than 7 % for nozzles A an B, an less than % for nozzle C. Figure 6: Experimental arrangement for the measurement of raiation attenuation. The simulations were performe using two ifferent sets of numerical parameters, liste in Table 3. For raiation relate input parameters, the number of Mie-angles was set to 30 an the raiation source temperature to 950 C. From the viewpoint of typical engineering applications, both parameter sets represent very well resolve simulations. For example, the spatial resolutions of.0.0 cm are selom possible in full scale fire engineering applications. The better-than-usual resolutions are necessary ue to the special characteristics of the valiation simulation. To evaluate the preictive capability of the raiation moel it is necessary to ensure that the numerical aspects of the solution o not ominate the errors. Table 3: Numerical parameters in the raiation attenuation simulations. Numerical parameter Set Set Spatial resolution (x).0 cm.0 cm Angular resolution N Droplet insertion rate 0 5 /s 0 6 /s Droplet CFL-conition.0 The high angular resolution is necessary because the heat source represents only about % of the full 4 soli angle for the point observer at 0.6 m istance. Using the efault angular resolution of 00 angles woul mean that the whole source woul be represente by practically one control angle, i.e. one iscrete intensity solution. Using the 000 angles means that at least 0 angles can contribute to the raiation at the measurement location. Naturally, this argument shoul be kept in min when using the FDS coe to simulate the local thermal raiation from a relative small source, such as local pool fire. A comparison of the simulate an measure attenuations is shown in Figure 7. Average relative experimental uncertainty of E = 0.04 was assume. In case of parameter set, all the simulations preict less attenuation than what was is observe in the experiment. The average bias factor is 0.79 an the relative uncertainty of the simulations is M = 0.. In case of the parameter set, the moel bias is 0.93 an many of the iniviual results are within the experimental uncertainty. A clear exception is the Nozzle C, for which the attenuations are still significantly below the experimental values. The explanation can be relate to the ifference between the ways to prescribe the roplet size istribution between Nozzle C an nozzles A or B. For C, the istribution was base on the number istribution, which resulte in better agreement of the number ensity in the small range of iameters but increase mean iameter. As the spray absorption an scattering coefficients are base on the Sauter mean iameter 3, the overestimation of the mean iameter is likely to cause unerestimate attenuation. Set Set Figure 7: Comparison of simulate an measure attenuations with parameter sets an.

7 The sensitivity of the attenuation results to the numerical parameters of the moel was further investigate by varying the numerical parameters one at the time for nozzle B at 70 bar. The results are summarize Table 4. For most of the parameters, the improve resolution leas to better agreement with the experimental ata, consistently with the results shown above. However, the improvement of spatial resolution from.0 cm to.0 cm ecreases the attenuation. The explanation is relate to the statistical accuracy of the Lagrangian spray an the ratio of spatial resolution an roplet insertion rate. When the spatial resolution is ouble, the same number of roplets is istribute among 8 times higher number of cells. As a result, some of the cells may have few or zero roplets, an the spray istribution becomes very non-uniform. For the raiation, these regions appear as holes in the water shiel an the attenuation is ecrease. Table 4: Results of the attenuation sensitivity stuy for nozzle B at 70 bar pressure. Parameter Value Attenuation Notes Experimental result Spatial resolution (cm) base case Angular resolution N base case Droplet insertion rate (/s) Droplet CFL-conition Combine effect t ins = 0.0 s base case base case x =.0 cm NRA = 5000 DPS = 0 6 /s CFL = CONCLUSIONS The increasing use of high-pressure water mist systems for fire protection applications has mae it necessary to investigate the capabilities of FDS coe in the simulation of such systems. This presentation summarizes some of the moel enhancements that have been mae to improve the accuracy of the high pressure spray simulations in the light of the internal spray properties an the capabilities of the spray to entrain air an attenuate raiation. The results with one inustrial spray nozzle an three fire suppression nozzles showe that the spray properties can be accurately preicte. The simulations were able to preict the V-shape iameter profile of the LN- nozzle, an on the other han, the flat iameter profiles of the high pressure micro nozzles. The effects of the pressure an multiorifice spray hea characteristics on the air entrainment were correctly preicte. The raiation attenuation results were within the experimental uncertainty when a very goo numerical resolution was use, but somewhat uner preicte in case of coarser resolution. The conclusion from the raiation sensitivity stuy was that the spatial resolution shoul always be in balance with the statistical representation of the spray (number of roplets per secon). The current results form a goo basis for the attempts to moel complete water mist systems in the conitions of intene applications. We are currently working on publishing the simulation results concerning system activation, cooling performance, flame suppression an effectiveness in the fire protection of complete roa tunnels. More work is neee in scaling-up the spray simulations for engineering applications where the spatial resolutions are typically much coarser than what was use here. ACKNOWLEDGEMENTS We woul like to thank all the members of the FDS evelopment team, an Kevin McGrattan, Ranall McDermott an Jason Floy in particular, for the fruitful co-operation. The work was sponsore by the Finnish Funing Agency for Technology an Innovation, Marioff Corporation Oy, Rautaruukki Oyj, YIT Kiinteistötekniikka Oy an Insinööritoimisto Markku Kauriala Lt. REFERENCES Grosshanler, W. L. (007), A research agena for the next generation of performance-base esign tools. Interflam 007. International Interflam Conference, th Proceeings. Volume. September 3-5, 007, Lonon, Englan, pp

8 Hostikka, S., McGrattan, K. (006), Numerical moeling of raiative heat transfer in water sprays. Fire Safety Journal 4, pp McGrattan, K.B., Hostikka, S., Floy, J.E., Baum, H.R., Rehm, R.G., Mell, W.E. an McDermott, R. (007), Fire Dynamics Simulator (Version 5), Technical Reference Guie, Volume : Mathematical Moel. NIST Special Publication 08-5, National Institute of Stanars an Technology, Gaithersburg, Marylan. Prahl, L., Jaoon, A. an Revstet, J. (009), Interaction between two spheres place in tanem arrangement in steay an pulsating flow. International Journal of Multiphase Flow 35, pp Ramírez-Muñoz, J., Soria, A. an Salinas-Roríguez, E. (007), Hyroynamic force on interactive spherical particles ue to the wake effect. International Journal of Multiphase Flow 33, pp