MUSAF II Colloquium Sep. 18-20, 2013, CERFACS, Toulouse Tackling Combustor Design Problems with Large Eddy Simulation of Reacting Flows Wolfgang Polifke Fachgebiet für Thermodynamik Acknowledgements: Joao Carneiro, Patrick Dems, Stephan Föller, Thomas Komarek, Rohit Kulkarni, Luis Tay, Matthieu Zellhuber AG Turbo, Alstom, DFG, DST, KW21, Siemens
Overview Gas turbine combustor design challenges stability emissions This talk mixing and auto-ignition in turbulent flow reacting flow flame dynamics from system identification spray dispersion, evaporation & combustion control & optimization, code coupling multi-physis, multi-scale 2
Sequential combustion in Alstom GT24/26 multi-stream mixing in swirling, turbulent flow auto-ignition & flame propagation 3
Composite PV lookup from PSRs Y c! Yc! 0.3! 0.2! 0.2! 0.1! Yeq! 0.1! t! 0.0! 0! 0.2! 0.4! 0.6! 0.8! 1! Z! Progress variable! source! Progress variable! Mixture fraction! 4
Stochastic fields for subgrid scale fluctuations Turbulence chemistry interaction! Presumed PDF! Deterministic transport equations! Transported PDF! Stochastic partial differential equations! Lagrangian/Particle! method! Eulerian/Fields! method! (ITO formulation)! 11! 5
Stochastic fields, composite progress variable Implemented in Fluent and OpenFOAM Validation: Markides & Mastorakos, Cabra, Delft, SEV H2, n-heptane, CH4 Autoignition length Lign/d! 40! 30! 20! 10! Experiment! Presumed PDF! Stochastic fields! 0! 940! 950! 960! 970! 980! 990! Air temperature [K]! 6
Cabra-H2-Flame with Li-Dryer mechanism see papers by Kulkarni, Zellhuber, Collonval... 7
Thermo-acoustic combustion instabilities from Lieuwen & Yang, 2006 8
Feedback between heat release and acoustics (p, u ) Q (p, u ) Rayleigh s criterion: Instability requires p Q dt > 0, Premix flames are velocity sensitive: Q = Q ( ), System acoustics controls phase p : Z = p. 9
Multi-physics interactions with feedback Fuel Supply Air Supply Flame Combustor Position and Area of Flame u Burning Velocity Q p, u p Equivalence Ratio 10
brute force LES for combustion dynamics? Combustion LES captures (in principle) all relevant effects: but: turbulent, reacting, compressible flow at low Ma-# acoustic boundary conditions!? wide range of length and time scales 11
divide et impera Stability analysis by combined use of system model for combustor acoustics reduced order model (ROM) of heat source dynamics This strategy may be adopted to network models state-space models this is code-coupling, too!! Galerkin models FE/FV-based solvers for Helmholtz / APE / LEE 12
ROMs of input-output systems Matrix-based methods (for state-space models) Truncated Balanced Realization Krylov methods (moment matching) Data-based models PODs & Galerkin Projection System Identification non-parametric model (e.g. spectral analysis) parametric models - Black Box models (e.g. time-delay models) - Gray Box models (e.g. state space models) 13
Time-delay ROMs for linear systems present response depends on present & previous signals and previous responses... and noise, too + trade offs: accuracy - uncertainty - # of coefficients - flow/flame physics 14
Flame transfer function from LES Signal B Q Time Series ( B (kt); Q (kt)) Auto-correlation Cross-correlation c Unit Impulse Response h Frequency Response F s, r, c h F(ω) j k s k s k j c k s k r k h = c Wiener filter equation F(ω) = k h ke ωtk z-transform 15
Turbulent premix swirl burner Komarek et al, 09, 10; Tay et al, 10, 11 URANS LES Solver CFX AVBP Turb./Sub-grid model Combustion Model SST TFC WALE DTFM 1 step Kin. Time step [s] 5e-5 1.25e-7 Spatial/temp. Discretization Second order Second order 16
LES/SI for premix swirl burner 17
Validation: FTF from CFD/SI vs experiment 18
Advanced SI for low signal-to-noise Tychonov Regularization for inversion of Wiener Filter: Daubechies wavelets h α = T + α 2 RR T 1 T c Generation of optimal excitation signals: maximum entropy non-gaussian simulation Föller & Polifke, ICSV 11: Transmission and reflection of sound at duct discontinuity 19
without individual time lag adaption was employed. The aero-acoustical network element representation of the orifice configuration is shown in Fig. 6.2. There are potentially two directions for incident acoustic waves with f u as the corresponding upstream ingoing characteristic wave amplitude and g d as the downstream ingoing counterpart. The two outgoing characteristic wave amplitudes are upstream g u and downstream f d, respectively. Since the orifice configuration represents a 2x2 aero-acoustic Aero-acoustic scattering at an orifice 6.3 Results Figure 6.2: Acoustical network element representation of the cylindrical orifice in a pipe 1.4 1.2 1.2 1 MIMO system, its scattering matrix (Eq. (6.1)) agrees with the acoustic ele189 0.8 0.8! R T+ 1 0.6 0.4 0.4 0.2 0.2 0 0 0.6 0.1 0.2 0.3 0.4 0.5 0.6 0 0 0.7 0.1 0.2 0.3 Sr 0.4 0.5 0.6 0.7 Sr (a) upstream transmission (b) downstream reflection 1.2 1.6 1.4 1 1.2 1! T R+ 0.8 0.6 0.6 0.4 LES/SI 99% Conf. Int. Lacombe, exp. 0.2 0 0 0.1 0.2 0.3 0.4 0.5 Sr 0.8 (c) upstream reflection 0.6 0.7 0.4 0.2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 M = 0.026 M = 0.034 Sr (d) downstream transmission 20
Summary Status Quo LES/SI can give flame dynamics with quantitative accuracy Ongoing work optimal model structures for identification non-linear effects high frequencies, acoustic losses combustion noise LES model for (partially) premixed (spray) combustion 21
Poly-Celerity MOM for polydisperse sprays y y D x0 f(x0,y) f(x0,y;d) 22
Test case: sedimentation of bubbles (Carneiro et al) 23
Polydisperse multi-phase turbulent swirl flow Experiments (Sommerfeld & Qiu, 91) vs. LES (Dems et al, 2012) Axial velocities Swirl air Particles & air D = 200 mm 34(! "5$% &% Gas phase Particles 24
Conclusions Large Eddy Simulation is relevant for R&D in industry Industrial applications imply special modelling requirements Reduced Order Models can be an interesting alternative to code-coupling for multi-physics, multi-scale problems 25
Thank you! 26