Comparison Of K-epsilon, Les And Des Turbulence Models Towards Numerical Simulation Of Automobile Fires In Enclosed Car Parks

Tez (Yüksek Lisans) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 2013Thesis (M.Sc.) -- İstanbul Technical University, Institute of Science and Technology, 2013Kapalı bir otoparkın yeraltında kalan bir bodrum katında aynı anda çıktığı düşünülen iki binek taşıt yangını sayısal olarak modellenmiştir. Yangın başlangıcından itibaren 9 dakika sonuna kadar geçen fiziksel olay örgüleri zamana bağlı olarak çözdürülmüştür. Çalkantı modeli olarak kullanılan k-epsilon (k-ϵ), LES ve DES’ten elde edilen sayısal sonuçlar karşılaştırılmıştır. Üç farklı çalkantı modelinin ayrı ayrı sonuçlarını elde etmek için probleme ait tüm parametreler sabit tutulmuş olup değişiklik yalnızca çalkantı modeli olmuştur. Zamana bağlı Reynolds ortalama Navier-Stokes denklemleri, süreklilik denklemi, enerji denklemi, kimyasal tür taşınım denklemi, çalkantı denklemleri ve ilişkili oldukları ışınım modeli ticari bir hesaplamalı akışkanlar dinamiği (HAD) yazılımı olan CFX aracılığıyla çözülmüştür. Her model benzetiminde de kullanılan sayısal ağ yapılandırılmamış ağdır. Yangından üreyen dumanın kat tavanından ilerleyecek olması dolayısıyla sayısal taşınırlığın etkin kılınabilmesi amacıyla tavana yakın bölge daha sık bir ağ yapıyla yapılandırılmış olup yangın kaynağının olduğu bölge için karakteristik yangın uzunluğu da dikkate alınmıştır. Taşıt yangınları, yangını temsil eden esas özellikler olan taşınımsal yangın gücü, duman üreme debisi ve alev kaynağının ortama ışınım yoluyla yaydığı enerji bakımından kaynağın yüzey sıcaklığının doğru değerlerle sınır koşulu olarak tanımlanmasıyla modellenmiştir. Bodrum kat otopark hacmine açık havadan besleme yapan taze hava fanları ve oluşan kirli havayı ortamdan dışarı atan duman egzoz fanlarının işlevleri de fan değerlerine karşılık gelen hacimsel debi giriş ve çıkış sınır koşullarıyla tanımlanmıştır. Ortamda gaz sevkine yardımcı olan jetfanlar ise birer hacimsel momentum kaynağı olarak modellenmişlerdir. Taşıt yangını koşullarının zamana bağlı oluşu dolayısıyla problemin benzetimlerinde kullanılan sınır koşulları ticari HAD yazılımına ait bir ifade diliyle özelleştirilmiştir. Elde edilen benzetim sonuçları otopark katının farklı kot düzlemlerinde duman yoğunluğunu temsilen CO2 kütle oranları ve sıcaklıklar bakımından karşılaştırılmıştır. Bunun dışında yangın kaynağından üreyen duman jetine ait analitik çözüm bilgisi benzetimden elde edilen sayısal sonuçlarla da karşılaştırılmıştır. Sonuç görüntüleme düzlemlerinde CO2 kütle oranları bakımından k-ϵ modelinin verdiği sonuçlar LES ve DES ‘ten farklılık göstermiştir ancak sıcaklık değerleri üç modelde de birbirine çok yakındır. Hesaplama uzayında tanımlanan görüntüleme noktalarından alınan verilere göre ise k-ϵ sonuçları hem kütle oranı hem de sıcaklık davranışları bakımından LES ve DES ‘ten belirgin farklılıklar ve her üç modelden de elde edilen yangın kaynağı merkez eksen sıcaklığı değerleri de analitik çözümle neredeyse uyumlu davranış göstermiştir.The last quarter of the past century was the beginning of the rise of fire modelling and simulation by using computers. The rate of influence and the transient character of the behavior of the smoke produced by an automobile fire beginning in an underground car park depending on the boundary conditions of the medium is important to be understood for fire numerical simulations and fire safety engineering. After the beginning of the usage of numerical methods in fire safety engineering possible fire scenarios can be solved numerically as computer simulations for concluding human escape routes and escape durations during emergency, the capabilities of the smoke extraction systems and etc. Additionally numerical fire solutions gives the ability to designers, engineers or architects to make architectural design developments towards fire conditions. The safety of the decisions made in the design period for an enclosed space like an enclosed car park by using numerical simulations is strictly related with the correct turbulence model to be used. Two automobile fires assumed as simultaneously burning which are thought to be occuring in an underground car park storey has been numerically modelled in this thesis to understand the behavior nature of different turbulence models acting on exactly the same conditions in different simulations and to catch some differences between them. Physical phenomena has been solved transiently from the beginning of the fire till the end of the 9th minute. The numerical results obtained from k-epsilon (k-ϵ), LES and DES turbulence models are compared. All the problem parameters were set constant for the case of obtaining the unique effect of each turbulence model seperately. Unsteady Reynolds Averaged Navier Stokes (URANS) equations, energy equation, chemical species transport equation, turbulence equations and related radiation model (radiation transport equation) were solved by the commercial Computational Fluid Dynamics (CFD) software CFX. The mesh used in each of the simulations is an unstructured mesh containing 1,5M elements and 317k nodes. For improving numerical dispersion, because the smoke produced from the fires moves as a ceiling jet, the mesh resolution at the ceiling region was made fine with an element growth rate of 1.1 with no boundary mesh adaptation. Additionally for a well modelling of a fire source, the mesh region around the fire sources were especially made finer taking the characteristic length for fire sources into account. The mesh quality has been checked with different mesh quality parameters. An unstructured mesh in mass formed of tetrahedral elements is used commonly for all the simulations. With this thesis work, the limit and type of mesh independency was not seeked by the author so that the results of this thesis can be concluded as the comparison of different turbulence models (which could be more efficient in a finer or more proper mesh) in the same conditions of which the one is mesh resolution. xxii Car fires were modelled using the primary source properties as the boundary conditions which are convective heat flux of the source, smoke production mass flow rate and the fire source surface temperature for well defining the radiative properties of the source. The activity of the smoke extraction fans which are located in the exhaust shafts of the car park are defined as outlet volumetric flow formed of uniform velocities in the outward normal direction of the exhaust shaft ends. Similarly the activity of the fresh air supply fans are inlet volumetric flow boundary conditions in the numerical side at the fresh air supply shaft ends formed from uniform inlet velocities directed inward normally. There is additionally one type of fan more acting in the flow solution which are the jetfans. Jetfans are co-working with the smoke extraction system helping the smoke extraction fans by injecting the smoke in the medium to near the attraction border of the exhaust (extraction) shafts. Jetfan activity has also been modelled by volumetric momentum sources in a transient caharacter. Not only the jetfans, all the boundary conditions of the problem are time dependent so the transient character of the boundary conditions were defined with a specialized expresion language within the solution software named as CFX expression language (CEL). Fire sources’ heat release rates has also been defined time dependent to obey the t2 fire growth curve rule and exponential decay rule during the decay period after the sprinkler system activation. Exponential decay behavior is assumed to be initiated coherent with the time which the growth curve reaches a fire heat release rate peak value of 5 MW when the sprinkler system is activated. As the radiation properties for the simulations, the radiation transport equation and the related blackbody radiation equations are solved by CFX with independency from spatial direction (isotropy) for the radiation intensity or the P1 Radiation Model. In addition to radiation properties frequency independency or the Gray Gas approximation is also used. All the walls of the computational domain are assumed to be adiabatic walls because of the location of the 3rd underground storey which is assumed to be surrounded by earth which is a perfect isolator. For the case of boundary layer solution, standard wall functions are used with logarithmic inner layer. The 3rd underground storey’s medium is connected by car ramps to the upper floors. It was presumed to take the fluid medium of the upper floors into account when constructing the solid model of the computational domain but this was going to load a big charge on computational efficiency and solution time so that ramp exits were designed to supply similar conditions to the reality by linking to a semi-spherical free entrainment boundary with 20oC temperature and 1 atm static pressure conditions. The algebraic multigrid solution (AMG) method with finite volume discretization of the flow equations has been used in the flow simulation software. SIMPLE has been the pressure-velocity coupling algorithm during solutions. In the simulation software, AMG solution is accomplished by a modified algorithm named as Multigrid Accelerated Incomplete Lower Upper Factorization or shortly MG-ILU. For spatial iterations 1st Order Forward Differencing or Central Forward Differencing schemes are used blendly and for time iterations First Order Backward Euler scheme is used. In the simulations a time step size of 0,25 second is used with total physical simulation duration of 540 seconds. The convergence criteria was to achieve a residual of 10-9 or to pass to the continuing time step after 20 inner iterations. There has been no residual value observed to stand bigger than 10-5. An HP Z600 xxiii workstation containing an Intel Xeon CPU with 6 physical and additionally 6 imaginary cores with 16 GB of rams included was used for simulations. The obtained simulation results has been compared for mass fractions of CO2 representing smoke density for different height levels and temperatures in the underground storey. Additionally the analytical solution made for the fire plume and the smoke jet moving towards the ceiling emerging from the fire source was also compared with the numerical results. The results obtained from the simulations in the result monitoring planes from the k-ϵ model are typically different than LES and DES but the same can’t be mentioned for temperature distributions transiently. Temperatures are always near the same values depending on the contour plots for each turbulence model. When the results obtained from the monitoring points defined previously in the solution domain are compared the k-ϵ model differs distinctly in mass fraction distribution and also temperature distribution from LES and DES models both. All the fire plume centerline temperature distribution results for each turbulence model are nearly the same in character and values with the analytical solution. Literary survey has shown that the fire temperature at the engine compartment of an automobile gradually reaches nearly 900oC at 420 seconds and linearly reaches the design fire source surface temperature of nearly 500oC at nearly 300 seconds. The numerical simulation results for the turbulence models has shown some differences at the fire source location in temperatures which could be concluded depending on the mesh resolution. For all the three turbulence models the adiabatic flame temperature for methane combustion (2210oK) which has been chosen as the guide chemical mechanism for the design fire simulation and behavior has been observed at a near top location from the fire source. Without using a reactive flow computation, by using the appropriate boundary conditions it is accepted that fire modelling as made in this thesis is acceptable and enough for smoke production and diffusion simulations.Yüksek LisansM.Sc