CFD Modelling of Compartment Fires

Transcription

1 CFD Modelling of Compartment Fires A validation study of a Large Eddy Simulation CFD code using compartment fire test data from Cardington By Nicholas Pope

2 Presentation summary Introduction Fire modelling Basic theory Practice Cardington compartment fire tests Setup and results Modelling the tests Setup and results Sub-grid-scale modelling parameters Smagorinsky, Prandtl, Schmidt numbers Sensitive dependence on initial conditions Summary 23rd September

3 Fire Engineering Introduction Analysis of the building Structural members Materials Escape strategies Analysis of the fire Combustibles Fire and smoke movement and spread Effect on humans Analysis of fire-fighting measures Passive Active 23rd September

4 Hand calculations Zone models Fire modelling E.g. OZone Simple one/two zone mass/energy balances Not considered further Field models E.g. FDS, SOFIE, SMARTFire, JASMINE Equations of CFD solved Turbulence models Combustion and radiation models 23rd September

5 Field models (CFD) Advent of higher-powered computers Computational domain split into cells Equations of mass, energy, momentum, and species conservation solved for each time step Navier-Stokes equations Turbulence models LES, k-omega, k-epsilon Combustion models Radiation models 23rd September

6 Fire Dynamics Simulator FDS is the field model being developed at NIST Solves equations of energy, mass, momentum and species conservation Large Eddy Simulation code (LES) Smagorinsky SGS eddy viscosity model Acoustic waves filtered out Low Mach number equations (NIST) Fast Fourier transforms used in solving technique Mixture fraction combustion model 23rd September

7 FDS validation Most NIST validation exercises plume orientated Cheaper Faster Bench-scale More accurate Large scale fires have a different nature Enclosed Fully engulfed Flashover Higher computational power now available 23rd September

8 Cardington 12m x 12m compartment Steel frame, concrete ceiling, blockwork walls Simultaneous ignition of 49 cribs Timber Polypropylene/timber mix Ventilation Front Front and back Linings Insulated Highly insulated 120 minute blaze per fire test 23rd September

9 Results from Cardington Time (mins) Temp (deg C) Test 2 Test 3 Test 4 Test 5 Test 6 Test 7 Test 8 23rd September

10 Test 6 details Fuel source wood only Ventilation front and back Linings insulated 1400 Temperature-time curve peaking just over 1200 O C Temp (deg C) rd September 2003 T im e (m in s ) 10

11 Modelling Cardington 23rd September

12 Grid resolution Tacit rule for users of CFD: The more cells in the domain (the finer the underlying grid), the more accurate the solution. Analysis of effect of grid resolution on resulting output a key study carried out Considerations Higher density of grid = longer run times Higher density of grid = more accurate geometry Results of all tests showed above rule may not always hold 23rd September

13 1400 Results of simulations Grid resolution variation (Test 6) Temp (deg C) Time (mins) Test Test Test data data data 23rd September

14 Conclusions of grid resolution analysis Higher density grids may not give more accurate solutions Convergent optimum grid resolutions appear for different simulations: Plumes Compartments Industrial User should be aware of the differences relating to grid size How this also effects sub-grid-scale parameters 23rd September

15 Sub-grid-scale modelling In FDS the dynamic eddy viscosity is modelled as µ = ρ ( ) S 2 ijk ijk C s Smagorinsky constant Filter width This is a static eddy viscosity model and is hence fairly simple to implement The numerical value of the constant is subject to debate taking values between rd September

16 Smagorinsky variation The default value included in the FDS software is 0.20 Test 6 was analysed further Varying the Smagorinsky constant between 0.01 and 0.35 Grids of 13,500 and 28,800 cells were used 23rd September

17 Results of Smagorinsky variation ,500 cells Max temp attained with Smagorinsky constant of 0.15 Temp (deg C) Time (mins) rd September

18 Results of Smagorinsky variation ,800 cells Max temp attained with Smagorinsky constant of 0.20 Temp (deg C) Time (mins) rd September

19 Smagorinsky conclusions The grid density has an effect on what value of Smagorinsky should be taken µ = ρ ( ) Unsurprising Analyses carried out for all Cardington tests Varying the Smagorinsky constant has no major effect Adequate results obtained even with a constant eddy viscosity in the momentum equation Little attention paid to other two SGS constants 2 ijk ijk C s S 23rd September

20 Further SGS parameters Eddy viscosity linked to the thermal conductivity via the Prandtl number k ijk c p, 0 ijk Eddy viscosity linked to the material diffusivity via the Schmidt number ( D) = ρ = ijk µ Pr µ ijk Sc 23rd September

21 Multi-parameter variation Variable (adjustable) parameters in the sub-gridscale modelling: Smagorinsky number Prandtl number Schmidt number Grid size (affecting filter width) Varying one SGS parameter very constrictive Method required of varying more than two parameters at once 23rd September

22 Multidimensional method Imaginary compartment set up in the domain Four thermocouples in hot layer 700 combinations of Smagorinsky, Schmidt and Prandtl numbers 2.5 MW fire 400 second simulation 23rd September

23 700 combinations Each point represents a particular combination of SGS constants Each output will deliver a value of maximum average compartment temp 23rd September

24 Results of multiple parameter variation Temp (deg C) Default SGS values used in FDS 23rd September

25 Test 6 simulated with ideal SGS parameters Temp difference 85 O C 1000 Temp (deg C) Smag = 0.2 Prand = 0.5 Schmi = 0.5 Smag = 0.2 Prand = 0.8 Schmi = Time (mins) Test data Test data Smag = 0.2, Pran = 0.5, Schm = 0.5 Smag = 0.2, Smag Pran = 0.2, 0.5, Pran Schm = = 0.8, 0.5Schm = rd September

26 Chaos Theory Plume analysis Sensitive dependence on initial conditions Something very rarely addressed in the published literature Simple input 23rd September

27 10 Second simulation 23rd September

28 HRR output HRR (kw) Time (secs) 600 kw exactly 600 kw exactly kw 23rd September

29 Detail of bifurcation HRR (kw) Time (secs) 23rd September kw exactly 600 kw exactly kw 29

30 Chaos conclusions The Navier-Stokes equations are a complex system nonlinearity dominates solutions No surprise at finding sensitive dependence in CFD model Fundamental limitations of field models? Flashover stochastic phenomenon User awareness Model limitations Physics limitations Further work to be done 23rd September

31 Overall conclusions Subject of sub-grid-scale modelling an important one Implementation of dynamic eddy viscosity model? Linking Prandtl and Schmidt parameters to model? SGS model that encompasses all aspects Further work, modelling further large-scale tests Subject of chaotic nature of model potentially important Further work Testing against nature rather than imaginary modelling User awareness 23rd September

32 By Nicholas Pope Thank you for listening Any questions? 23rd September