VALIDATION OF FLACS-FIRE FOR JET FIRES UNDER VENTILATION-CONTROLLED CONDITIONS Deiveegan Muthusamy and Bjørn Lilleberg GexCon AS, P.O. Box 6015, NO 5892, Bergen, Norway ABSTRACT The focus of this paper is to present validation of FLACS-Fire predictions for two confined jet fires under ventilation-controlled conditions: Steckler and SINTEF compartment fire cases. Turbulence is modelled using the standard k-ε model, and the eddy dissipation concept (EDC) is used as the combustion model. The soot is handled with a formation-oxidation model (FOX) and discrete transfer radiation model (DTRM) is used for radiation calculations with weighted sum of gray gas (WSGGM) model for radiation property calculations. Radiation calculations are coupled with the transient simulations of fluid flow. The enthalpy equation is extended with the source term calculated from DTRM, and the flow solver provides the temperature and mole fractions of CO 2 and H 2 O used by the radiation code. Comparisons are made between the experimental data and simulations. The predicted temperature and velocity profiles are found to be in good agreement with experimental data. INTRODUCTION AND MOTIVATION A release of a combustible gas into the surroundings may get ignited. The outcome after the ignition depends on the fuel, amount of fuel, mixing with surrounding air, confinement, and time of ignition. The worst case scenario is a major explosion. Fire is a more frequent event than explosion. Accidental fires may result in enormous damage to property, and endanger people and environment. The potential consequences of hydrocarbon fires on offshore installations were clearly demonstrated by the Piper-Alpha disaster in 1988, and more recently by the Deepwater Horizon disaster in 2010. Experimental investigations of large-scale fire phenomena are inherently expensive, and extrapolation from experimental results is generally not suitable for site-specific safety studies. Numerical simulation of fire represents an attractive alternative. However, it is vital that the numerical modelling is able to capture the key physical phenomena such as the turbulence-chemistry interaction, soot-formation, radiation, and dispersion of smoke and toxic combustion products. Several computational fluid dynamics (CFD) software packages have been developed for numerical simulation of fire. Fire Dynamics Simulator (FDS) 1 and SmartFire 2 (Low-momentum flow), and the Kameleon FireEx 3 (High and low-momentum fires) simulator, are examples of special-purpose software. CFX 4, FLUENT 5 and PHOENICS 6 are examples of general-purpose software used for fire simulations. The CFD software FLACS has primarily been developed to model dispersion and explosion phenomena, and is currently a leading CFD-tool within this field (More details see GexCon s web pages and the FLACS manual 7, 8 ). Models for the simulation of jet and pool fires are under development (FLACS- Fire). The aim is to be able to predict industrial fires efficiently and with good precision. One of the major motivations making FLACS able to numerically model fire is an observed increasing interest for probabilistic fire risk assessments in the oil and gas industry, likely influenced by a recently adopted standard ISO19901:3 9. According to this standard, oil and gas installations shall evaluate the accidental risk from explosion and fire, and if worst-case design is not feasible, it shall be demonstrated that the frequency for escalation (loss of main safety barrier) is less than once every 10,000 years. 485
The implemented models in FLACS-Fire include the Eddy Dissipation Concept (EDC) to model the turbulence-chemistry interactions for non-premixed flames, the Discrete Transfer Radiation Model (DTRM) and Magnussens soot oxidation-formation model. An accurate prediction of the soot level is a key parameter in the calculations of radiative heat loss from the flame. Further, most of the heat impact on objects that not are engulfed in flames is due to radiation. Since the time scales for fire simulations are longer than for explosions, the computational speed is important. The recent development of noncompressible and parallel solvers in FLACS will therefore be important to ensure efficient calculations. However, the presented predictions in this paper are with a compressible and non-parallel solver. Since fire modelling is a complex phenomenon involving interaction between turbulent flow, chemistry and heat transfer, it is important to carry out extensive validation of the CFD models to ensure reliable results. In this paper, we have used jet fire experiments in compartments to evaluate the applicability and validity of the FLACS-Fire. MATHEMATICAL MODELLING This section briefly describes the mathematical modelling in FLACS-Fire. This includes the turbulence-chemistry interaction for non-premixed flames, radiation modelling and soot modelling. FLACS solves, in three dimensions and transient, the equations for continuity, momentum, energy, fuel mass fraction, mixture fraction and turbulence (the standard k-e model). Complex geometries are represented on a structural Cartesian mesh using the distributed porosity concept. Large objects and walls are represented on-grid, while objects smaller than the grid-cell size are represented sub-grid. Sub-grid objects contribute to flow resistance, turbulence generation, and flame folding in the simulation. Combustion Model The transport equation for fuel mass fraction can be represented as: where the source term ω fuel and the turbulent diffusion/flux have to be modelled. The turbulent diffusion is modelled by using a gradient model 10 and the standard k-e turbulence model. The source term for the reaction rate is modelled using the Eddy Dissipation Concept (EDC) 11, 12 with the assumption of fast chemistry. The EDC model postulates that most of the reactions occur in the smallest scales of the turbulence, the fine structures. The source termω fuel can be written as: m χ ω fuel = Y min (2) * 1 γ χ Here, Y min = min [Y F,Y O2 /s], m is the transfer of mass per unit of fluid and unit of time between the fine structures and γ * is the mass fraction of the fine structures. For further details on EDC can be found in literature 11,12. Soot Model In FLACS-Fire the formation of soot has been modelled by Magnussens soot model 13. This model assumes that soot is formed from a gaseous fuel in two stages, where the first stage represents formation of radical nuclei, and the second stage represents soot particle formation from these nuclei. In FLACS, the soot level must be determined from known scalars, such as the mixture fraction, the fuel composition and the local equivalence ratio. Hence, models based on some intermediate species in the combustion process cannot be used. Furthermore, to limit memory requirements and computation time, soot progress will be (1) 486
described by one variable field only. The limitations stated above give two possibilities for modelling of soot, which are both implemented in FLACS-Fire: 1. A fixed soot conversion factor model (CFM), where a certain amount of carbon in fuel is converted to soot directly. The amount of carbon transformed to soot depends only on the fuel composition. 2. A formation-oxidation model (FOX), where there is a formation source term and an oxidation source term in the soot transport equation. Radiation model The accuracy of radiative heat transfer calculation depends on accurately calculating radiative intensity and also accurately modelling the properties of the radiating media. The Discrete Transfer Radiation Method (DTRM) of Lockwood and Shah 14 is used to calculate radiative intensity and the properties of the radiating media are calculated using the weighted sum of gray gas model 15. The DTRM is one of the widely used methods to solve radiative transfer problems with participating medium. Rays are sent from surface elements into a finite number of solid angles that cover the radiating hemisphere about each element and the main assumption of the DTRM is that the intensity through a solid angle is approximated by a single ray. The number of rays and directions are chosen in advance. The governing equation for describing radiation intensity field in an absorbing, emitting and scattering medium is the Radiative Transfer Equation (RTE), which is of the integro-differential type 16. The radiative transfer equation is given by: 1 2π di ( τ, µφ, ) ω µ = I( τ, µφ, ) + ( 1 ω) I [ ] + ( τ µ φ ) Φ µ φ µφ µ φ B T τ π µ = φ I,, (, ;, ) d d (3) d 4 1 = 0 where µ is the cosine of the polar angle θ, φ is the azimuthal angle, I(τ,µ,φ) is the intensity along direction µ, φ at optical depth τ measured perpendicular to the surface of the medium, I B is the spectral black body intensity at temperature T, ω is the single scattering albedo and Φ ( µ, φ, µ, φ) is the scattering phase function. In the DTRM method, the RTE is solved for each ray from one solid boundary to another solid boundary in the geometry. Rays are sent from solid surface boundaries and traced through the volume. The calculation of radiation source term is based on the distance travelled in each control volume. At the boundaries radiative heat transfer boundary conditions are used to determine the intensity of rays fired from that surface area. As the correct initial intensities are unknown at the start of the calculation, the procedure becomes iterative until correct radiative intensities are resolved. In the present model, the products of combustion like CO 2 and water vapor, H 2 O, have been considered as the participating gases, which absorb and emit radiation depending on local mixture temperatures. For these gases scattering is insignificant. Exact solutions can be obtained with spectral calculations but are not practical for industrial applications. Solution of the RTE through DTRM ray tracing mechanism provides the source term in the energy equation at every nodal point in the domain. VALIDATION Since FLACS-Fire is a tool still under development, the amount of validation results is limited. In connection to the release of a test version of FLACS-Fire, Melheim 17 performed a validation study of jet fires of various fuels and jet velocities. Some of the cases with jet fires in no wind and cross wind were compared with experimental results. A qualitatively good comparison was found. Nilsen 18 performed a more extensive validation study. The main conclusions of these studies were that flame lengths and shape seemed to be modeled well, but the too simple six-flux radiation model led to incorrect heat loads 18. This radiation model has later been replaced by the DTRM. 487
Case Description The two test cases Steckler compartment 19 and SINTEF compartment 20 are chosen for validation of FLACS-Fire for confined geometries. The Steckler compartment-fire case involves experiments performed within a square compartment with sides of 2.8m and height of 2.18m. A circular gas burner, fuelled with commercial grade methane and having a diameter of 0.3m, was placed on top of a insulating ceramic fibre board. The layout of the room is illustrated in Figure 1A. The present work concentrates on the condition where the burner is positioned centrally in the room and ventilation is provided by a room opening (doorway) with a height of 1.83m and width of 0.74m, see Table 1. The fuel flow rate of 0.001258 kg/s corresponds to a heat output of 62.9 kw. In the Steckler experiments measurements of temperature and velocity were performed at the compartment door opening. The schematic of the SINTEF experimental setup is shown in Figure.1B. The dimensions of the compartment were 12.750 5.5 5.9. The fuel nozzle (with diameter of 0.025 m) is centrally located in the enclosure, 0.52 m above the concrete floor. In this experiment propane is used as fuel. The locations of the thermocouples, gas sampling measurement and heat flux gages are also shown in Figure 1 and Table 1. Temperatures were measured at three vertical lines (EF, CD and the vent). C E D F A- Steckler Compartment B- SINTEF Compartment Figure 1 - Schematic of fire compartment Table 1 Summary of the fire experiments Configurations Case A:Steckler Compartment Case B: SINTEF Geometry Dimensions of the compartment 2.8 2.8 2.18 m (17.09 m 2 ) 12.750 5.5 5.9 m (145 m 3 ) Ventilation 0.74 1.83m (1.35 m 2 ) 3.45 2.81 m (9.69 m 2 ) Location of the fuel nozzle 1.4 1.4 0.3 m 6.375 2.75 0.62 m Diameter of the fuel nozzle 0.3 m 0.025 m Fuel Methane Propane Fuel Rate 0.001258 kg/s 0.84 Kg/s [21] 488
Computational Details In Figure 2 the geometry configurations, with grid, for the Steckler and SINTEF compartment cases are shown. Front side of the compartment the computational domain is extended. The number of grid cells used is 31920 and 219816 respectively. The number of grid cells used at the release is four. Fixed soot conversion factor model (CFM), with 5 % soot yield is used. For radiation calculations the boundaries are assumed to be diffuse and gray. The total number of rays fired is equal to 100 (10x10). A- Steckler Compartment B- SINTEF Compartment Figure 2 Calculation domain and grid distribution RESULTS AND DISCUSSION Steckler compartment Figure 3 and 4 shows comparison between FLACS-Fire predictions and the experimental measurements of the Steckler compartment case. The figures show very good agreement for both temperature and velocity for all the measurement points at the compartment door opening when radiation is accounted for. When the radiation modelling is excluded from the calculations the temperature field is over predicted especially in the high temperature area. Presented results on the Steckler compartment case in the literature show the same good agreement as FLACS-Fire 22. 489
A. Door opening y=-0.212 B. Door opening y=0 C. Door opening y=+0.212 D. Corner of the Room Figure 3 - Predicted and measured vertical temperature profiles A. Door opening y=-0.212 B. Door opening y=0 C. Door opening y=+0.212 Figure 4 - Predicted and measured vertical velocity profiles at door opening 490
SINTEF compartment fire Figure 5 show comparisons between predicted and measured temperature for the EF, CD and vent opening in the SINTEF compartment-fire case. Differences between predictions and measurements are seen for the region close to the floor level. At higher elevations the predictions show a better trend compared to the measurements. The burning rate is overpredicted close to the floor. Figure 6 show temperature contours with and without radiative losses being accounted for. The impact of including radiation modelling is clear and should not be omitted, although using DTM is relatively computationally expensive (approximately 20 % extra time). A. Centre of the Room B. 3.25 m from door opening C. Centre of the door opening Figure 5 - Predicted and measured vertical temperature profiles A. without radiation 491
B with radiation Figure 6 Temperature contours predicted by FLACS-Fire DISCUSSION In general the importance of including radiation modelling in the calculations was clearly illustrated in both cases. For the Steckler compartment case very good agreements were seen. In the SINTEF compartment fire case, however, differences between the predictions and measurements were clearly observed. The reason for the discrepancy was not fully understood and need further investigation. As the SINTEF compartment case is a very valuable experiment for the validation of FLACS-Fire this will be further investigated in the development of FLACS-Fire. Numerical modelling of fire involves complicated physics and a lot of models that are coupled to each other. This contributes to the modelling uncertainty when it comes to validating numerical results. Another uncertainty comes from the numeric itself in the form of discretization and iterative convergence errors. A third source of uncertainty comes from the experimental data the calculations are validated against. Validation of the CFD code FLACS-Fire is in its initial stage and the differences in the presented results needs further investigation. CONCLUSION Two compartment fire experiments have been simulated with the CFD code FLACS-Fire which is under development. Very good agreement was seen for the Steckler compartment fire case. On the other hand, in the SINTEF compartment case differences between predicted results and measurement were significant close to the floor, but followed the trend in the experiment for higher elevations. In both cases the inclusion of radiation modelling, although being relatively computationally expensive, gave significantly better results which justified its expense. The results showed that radiation should always be a part of such calculations to give more realism to the calculations and avoiding giving to conservative results. Despite giving very good agreements for the Steckler case it was clearly seen that the SINTEF case, with its inside measurements, showed clear differences between predictions and measurements. It is clear that validation of a fire code with vent temperature distributions, and/or from measurement points far from 492
where the combustion occurs is questionable. Very little information about the combustion modelling accuracy can be drawn from such measurements. Hence, more detailed measurements inside the compartment are needed to improve and fully validate CFD codes intended for fire safe design. ABOUT THE AUTHOR Deiveegan Muthusamy gained his PhD from Indian Institute of Technology Madras in 2007. Working as senior research engineer at GexCon AS in Bergen, Norway. Bjørn Lilleberg, PhD in combustion physics from Norwegian University Science and Technology in 2011. Working as senior research engineer at GexCon AS in Bergen, Norway. REFERENCES 1. Kevin B, McGrattan, et al, Fire Dynamics Simulator (FDS) Manual, Technical Reference Guide, 2002. 2. Vembe, B. E., Lilleheie, N. I., Holen, J. K., and Magnussen, B.F. Kameleon FireEx, A Simulator for Gas Dispersion and Fires, 1998 International Gas Research Conference. 3. J Ewer, E R Galea, et al, An Intelligent CFD Based Fire Model, Journal of Fire Protection Engineering, V10 (1), 1999, pp12-27. 4. Simcox S, Wilkes et al, Computer Simulation of the Flows of Hot Gases from the Fires at King s Cross Underground Station, Fire Safety Journal, V18, P49-73, 1992 5. D. Barrero, B. Ozell and M. Reggio, On CFD and graphic animation for fire simulation, The Eleventh Annual Conference of the CFD Society of Canada, Vancouver, BC, May 2003 6. Glynn DR, Eckford DC & Pope CW, Smoke concentrations and air temperatures generated by a fire on a train in a tunnel, The PHOENICS Journal of Computational Fluid Dynamics and its Applications, Vol. 9 No. 1, pp 157-168, 1996 7. http://www.gexcon.com/flacsoverview, Accessed 10 th April, 2011. 8. FLACS v9.1 user s manual. (2011). 9. ISO 19901-3 (2010) Petroleum and natural gas industries - Specific requirements for offshore structures - Part 3: Topsides structure, the International Organization for Standardization 10. Pope. S. B., Turbulent Flows. Cambridge University Press, Cambridge, United Kingdom, 2000. ISBN 0 521 59886. 11. Magnussen B. F., and Hjertager. B. H., On mathematical modelling of turbulent combustion with special emphasis on soot formation and combustion. In 16th Symp. (Int.) Combustion, pages 719 729, Pittsburgh, PA, 1976. The Combustion Institute. 12. Ertesvåg. S., Turbulent strøyming og forbrenning. Tapir akademisk forlag, Trondheim, 2000. ISBN 82-519-1568-6 (in Norwegian). 13. Magnussen B. F., The Eddy Dissipation Concept, ECCOMAS Thematic Conference on Computational Combustion, Lisboa, 21-24 June 2005. 14. Lockwood, F. C., & Shah, N. G., A New Radiation Solution for Incorporation in General Combustion Prediction Procedures, Eighteenth Symposium (International) on Combustion/The Combustion Institute, 1405-1413, 1981. 15. Truelove, J. S., A Mixed Grey Gas Model for Flame Radiation, AERE Harwell, Oxfordshire, UK, 1976. 493
16. Siegel, R., & Howell, J., Thermal Radiation Heat Transfer, New York: Hemisphere, 2001. 17. Melheim, J.A., Introduction to FLACS-Fire, GexCon Internal report, 2007. 18. Nilsen, C., Jet diffusion flames in FLACS Fire, GexCon Internal report, 2010. 19. K.D. Steckler, J.G. Quintiere and W.J. Rinkinen, Flow induced by fire in a compartment, NBSIR (1982). 20. Blast and fire engineering for topside structures, test programme F3, confined jet and pool fires. Technical Report for test JF4, SINTEF Energy, Norwegian Fire Research Laboratory, June 1996. 21. J. Wen, L. Y. Huang, CFD modelling of confined jet fires under ventilation-controlled conditions, The Fire Safety Journal, 34, 1-24. (2000). 22. YEOH GH and YUEN RKK, Computational Fluid Dynamics in Fire Engineering - Theory, Modelling and Practice, Butterworth-Heinemann, Elsevier Science and Technology, ISBN: 978-0-7506-8589-4 (2009). 494