Clean Combustion Technology

Transcription

1 Computational Combustion and Energy Grou p

2 Clean Combustion Technology Agricultural residual burning Aviation Haze in Singapore Power generation Domestic combustion device Ground transportation Decrease pollutant emission (mainly nitrogen oxides NOx) Increase combustion efficiency Innovative combustion Technologies Alternative Fuels 2

3 low NOx (nitrogen oxide) emission Fuel lean combustion Low NOx formation Adaptation of more uels, like natural gas, ropane and realistic hydrocarbons Fuel flexibility Premixed and/or partially premixed flame Combustion instability Mixed and variable combustion modes Close to the fuellean limit? Thermo-acoustic instability and flame extinction 3

4 Turbulence energy spectrum Multi-scale and nonlinearity in turbulence & combustion PRODUCTION by mean flow Energy cascade Large eddies A Hypothetical Chemical Mechanism DISSIPATION by viscosity Small eddies Modelled in RANS Computed in DNS Computed in LES Modelled in LES Nonlinearity in Arrhenius-type reaction rate: Kolmogorov turbulence spectrum Wave number Is Direct Numerical Simulation of Turbulent Combustion Possible? 4

5 Number of reactions, I Opportunity and challenge in numerical combustion CH 4 Ranzi mechanism comlete, ver 11 iso-octane (LLNL) iso-octane (ENSIC-CNRS) 2-methyl alkanes (LLNL) methyl palmitate (CNRS) Biodiesel (LLNL) Gasoline (Raj et al) JetSURF 2.0 CH4 (Konnov) skeletal iso-octane (Lu & Law) USC C1-C4 skeletal n-heptane (Lu & Law) USC C2H4 n-butane (LLNL) 1,3-Butadiene C1-C3 (Qin et al) DME (Curran) GRI3.0 neo-pentane (LLNL) GRI1.2 C2H4 (San Diego) CH4 (Leeds) I = 5K C C16 (LLNL) C12 (LLNL) C10 (LLNL) C14 (LLNL) PRF (LLNL) n-heptane (LLNL) MD (LLNL) before since Number of species, K (Lu & Law, Prog. Energy Combust. Sci. 09) FLOPS of the top 1 supercomputer 1E18 1E17 1E16 1E15 1E14 1E13 1E12 1E11 1 exaflops 1 petaflops Earth Simulator Jaguar Roadrunner ASCI White LLNL Numerical Singapore NSCC Wind Tunnel ASCI Red SNL 1 teraflops CP-PACS Hitachi SR21 SNL Paragon CM Year Sunway TaihuLight Tianhe-2 Titan Tianhe-1A K Computer ARCHER (Data source: Top 500 Supercomputer Website) Is Direct Numerical Simulation of Realistic Turbulent Combustion with Detailed Chemistry Possible? 12 5

6 Conditional Moment Closure (CMC) Model Five-dimensional Eqs. of species mass fraction ( Y α ȁη) and energy ( h ȁη): T(K) Temperature, K YO O 2 mass fraction T1: Convection; T2: Dilatation; T3: Micro-mixing; T4: Chemistry; T5: Turbulent Scalar Flux mixture fraction Sandia flame D Barlow and Frank, Proc. Combust. Inst First-order CMC model Zhang et al., Proc. Combust. Inst. 15. Zhang and Mastorakos, Flow Turb Combust 16. Garmory and Mastorakos, Proc. Combust Inst 15. Modeled Terms : amplitude mapping closure (AMC) : presumed beta-function : first order CMC closure, detailed chemistry, ARM2 mechanism a gradient model used for sub-grid scale scalar flux Sung et al., Proc. Combust. Inst Klimenko and Bilger, Prog. Combust. Energy Sci

7 Percentage Parallelization and computational cost wall time per time step, t wall (s) N p =24 = number of processors, N p 7 LES/3D-CMC, SMA2 LES/0D-CMC, SMA2 LES, N16S % 90% 80% 70% 60% 50% % 30% % 10% 0% MPI Terms in eta-space Terms in physical space Data averaging Number of processors, N P Perfect loading balancing for parallelized CMC solver: round-robin parallel algorithm = (, )+, Good scaling for O(10 3 ) processors on ARCHER Cluster of UK National Supercomputer Zhang, PhD Thesis, University of Cambridge, Cambridge, UK, 15. 7

8 SMA2 Sydney swirling flames Swirl air Fuel Annulus Fuel pipe Coflow Atmosphere Outlet Downstream domain One of the target flames in TNF workshop, measured by researchers from University of Sydney and Sandia Swirl number S N = W S /U S = 1.59 Oxidizer: swirling air; fuel: nonswirling CH 4 /air (1:2) Approximately 8,0,000 tetrahedral LES cells, About 1,000 polyhedral CMC cells Strong turbulence, increased local extinction Cases U S (m/s) W S (m/s) U j (m/s) U j / U j,sl (kw) SMA % ,0 SMA % ,800 Sydney swirl burner SMA % ,300 Masri et al, Combust. Theor. Model. 07. Masri et al, Combust. Flame 04. Al-Abdeli and Masri, Combust. Theor. Model

9 Y OH, [-] Y CH4, [-] T, K Reactive scalars x/d b =0.4 x/d b =0.6 x/d b =1.0 x/d b =0.4 x/d b =1.0 LES/3D-CMC Experiment r/d b r/d b r/d b Symbols : Experimental data (Masri et al. CNF 04) Solid lines : LES Dashed lines: 0D-CMC calculation with low scalar dissipation T, K Y O2 Y OH Unconditional scalars Conditional scalars 1. Unconditional means of temperature and methane mass fraction agree well with experimental data. 2. The conditional temperature and O2 mass fraction are computed well, which are close to the equilibrium flame structures, but the OH is over-predicted in LES. 3. Overall, the current LES/CMC solver demonstrates good accuracy in predicting the reactive scalars of this simulated flame. Experimental data: Masri et al. Combust. Flame 04 9

10 Time evolutions of OH and heat release rate SMA2 SMA2 SMA3 SMA4 SMA2 SMA3 SMA4 SMA3 Resolved OH mass fraction (OH-PLIF-like) (black iso-lines: stoi. mixture fraction 0.25) Line-of-sight integrated heat release rate (MJ/m 2 s) (OH* chemiluminescence-like) Experimental flame photos from:

11 Cambridge swirling flames IRZ 95 x 150 y (a) Chamber (b) Bluff body Schematic of experimental setup 11 CRZ Blow-off curve Fuel/oxizider: pure methane (non-swirling) and air (swirling) The swirl number S N is calculated following Beer and Chigier s method (Applied Science 1972): Air Fuel Air Air Fuel Air (a) Outlet (b) Fuel inlet LES and CMC computational domain Farfield inlet Air inlet Walls Experimental data from: Cavaliere et al. Flow Turb. Combust

12 Prediction of global extinction (blow-off) condition in Cambridge burner 45 ~1.6U BO,exp 35 BO3 ~1.25U BO,exp U a, m/s LES,ST EXP,BO LES,ST EXP,BO BO2 BO1 LES,ST S2 S3 S1 LES,ST EXP,BO Blow-off Range from LES/3D-CMC LES,ST LES,ST EXP,BO 1.0U BO,exp Prediction of the full blowoff curve is still one of the targets of combustion CFD. Capturing the blow-off condition with LES has not been demonstrated yet U f, m/s 12

13 Blow-off transient: LES vs. experiment Simulation Experiment Heat release rate OH* chemiluminescence OH OH-PLIF Swirling air Swirling air non-swirling CH 4 (Cavaliere et al. Flow Turb. Combust. 13) 13

14 f,ext, [-] Localized extinction during blow-off burning (a) extinction, m (b) (c) t, s Metrics for quantifying flame extinction: Area of the stoichiometric mixture fraction iso-surface Extinguished fraction Stable flame Blow-off Zhang and Mastorakos, Flow Turb Combust

15 DLR (German Aerospace Center) model gas turbine combustor Experimental DLR Case Fuel mass flow rate (kg/s) Air mass flow rate (kg/s) Swirl number Thermal power (kw) Global equivalence ratio A C (a) (b) Computational domain Experimental visualizations of flow pattern and reaction zone of Flame A case Dual swirlers Weigand et al. Combust Flame 06. Stohr et al. Combust. Flame

16 Inner recirculation zones Corner ecirculation zones Axial mean velocity distribution Resolved field Mean field Contours of resolved (left) and mean (right) axial velocity for Flame A case Lines: simulation; rcircles: experiment r 34 x=1.5mm x=2.5mm x=5mm x=7.5mm x=10mm x=15mm x=mm x=30mm x=mm x=50mm 16

17 OH mass fraction vs. OH-PLIF Flame A (a) Experiment (Weigand et al. Combust. Flame 06) Flame C (a) (b) Simulation Height, m (b) Localized extinction? Lean blowout? Partially premixed? Radius, m 17