ANALYSIS of BIFURCATIONS in LOW-DIMENSIONAL MODELS of TURBULENT COMBUSTION J. M. McDonough Departments of Mechanical Engineering and Mathematics University of Kentucky, Lexington, KY 40506-0503 E-mail: jmmcd@uky.edu Website URL: http://www.engr.uky.edu/~acfd Acknowledgments: Reported work supported in part by grants from AFOSR and NASA-EPSCoR: Presented at 4 th Joint Meeting of US Sections of the Combustion Institute, Philadelphia, PA, March 20−23, 2005 University of Kentucky Advanced CFD & Immersive Visualization Lab CF D BACKGROUND Low-Dimensional Models (LDMs) Widely Studied Numerous methods for construction Many different kinds of models Especially Appropriate for Use with LES Resources needed for LES generally greater than required for RANS—but there is potential for far better results Details likely to be missing in reduced models usually not resolved with LES Suggests Detailed Understanding of LDMs Could Be Helpful Because LDMs Represent Generally Nonlinear Phenomena, Can Be Expected to Exhibit Bifurcations of Behavior University of Kentucky Advanced CFD & Immersive Visualization Lab C FD BACKGROUND (Cont.) Bifurcations Widely Observed in Combustion Phenomena Implies if particular LDM incapable displaying physical bifurcation sequences, it may be inappropriate Use should be restricted to parameter ranges in which bifurcations do not occur Goals of Present Study Briefly present new form of LES expected to be especially appropriate for use with LDMs Outline derivation of subgrid-scale (SGS) equations corresponding to selected LDM Demonstrate form of analysis that identifies bifurcation sequences University of Kentucky Advanced CFD & Immersive Visualization Lab CF D ALTERNATIVE LES/LDM APPROACH Filter Solutions—not equations Mathematically equivalent to mollification Provides dissipation needed to prevent aliasing of under-resolved solutions Leads to simpler SGS models Model Physical Variables—not their statistics Results in SGS models able to produce DNS-like fluctuations—but, clearly not at DNS spatial resolutions Yields ability to model SGS interactions of turbulence with other physics (combustion in the present case) Directly Use SGS Results—don’t discard them Corresponds (mathematically) to multi-level formalism Hence, provides rigorous foundation, and availability of analysis tools University of Kentucky Advanced CFD & Immersive Visualization Lab C FD GOVERNING EQUATIONS Usual Balance Equations ρt + ∇⋅(ρU) = 0 DU = −∇p + ∇⋅(µ∇U) + ρg ρ —– Dt Ns Ns ρYi DT ρcp —– = ∇⋅(λ∇T) + ∑ c p Di Wi ∇ — ⋅ ∇T − ∑ h i w˙ i Wi Dt i=1 i i=1 ( ) D(ρYi ) ——— = ∇⋅(ρDi ∇Yi ) + w˙i Dt i = 1‚…, N s with w˙i = Wi Nr ∑ (νi,j″ j=1 − ν i,j′ ) wj Ns and wj = kf,j ρYl Wl ν l,j′ Ns ρYl Wl ν l,j ″ ∏ (—) − kb,j∏ (—) l=1 l=1 University of Kentucky Advanced CFD & Immersive Visualization Lab C FD GENERAL SGS/LDM CONSTRUCTION SGS Models (temporal fluctuating part, only) Usual LES decomposition of dependent variables qi (x,t) = q~i (x,t) + q*(x,t) i for i th dependent variable Assume Fourier representation of each dependent variable: qi (x,t) = ∑ ai,k (t) φk (x) k Construct Galerkin ODEs Decimate result to single (relatively-high) wavevector per equation—within part of series corresponding to q* Apply simple numerical integration methods (and some transformations) to obtain multi-dimensional discrete dynamical system (DDS) University of Kentucky Advanced CFD & Immersive Visualization Lab C FD THE DISCRETE DYNAMICAL SYSTEM Equations Initially Derived for H 2 −O 2 Combustion by McDonough & Huang, Proc. 3 rd Int. Symp. Scale Modeling (2000) a (n+1) = βu a(n)(1−a (n)) − γu a(n) b(n) b (n+1) = βv b(n)(1−b (n)) − γv a(n) b (n) + αT c (n) c (n+1) = Ns [( ∑ αTd d i i=1 (n+1) di (n+1) i − γuY a (n+1) i = −(βY i +γuYi a (n+1) + γvY b (n+1) − γvY b i (n+1) i )c (n) Ns − ∑ Hi ω⋅ i ] ⁄ (1+βT ) + c 0 i=1 ) d i(n) + ω⋅ i + d i,0 , N N N j=1 l=1 l=1 i = 1,2,…,Ns r s s ′ ν j,l ν ″j,l ⋅ ωi = ∑ [Cf,ij ∏ dl −C b,ij ∏ dl ] Produce temporal fluctuations for SGS model DDS very general—can be applied to any reaction mechanism—simplifies significantly for elementary reactions Contain numerous bifurcation parameters—all related to physics University of Kentucky Advanced CFD & Immersive Visualization Lab C FD REDUCED MECHANISM AND LDM General Features of LDM Based on DDSs DDS used for each product species of each elementary reaction Number of iterations of each DDS proportional to Kolmogorov-scale Damköhler number, DaK Automatically provides subcycling during velocity time steps Nine-Step Mechanism for H 2 −O 2 Combustion At least one step each for initiation, propagation, chain branching, recombination H2 + O2 → HO 2 + H O 2 + H → OH + O Ordered such that species do not appear as reactants earlier than they have been produced H 2 + O → OH + H Make following identifications d1 ~ H 2 d 4 ~ OH d7 ~ O d2 ~ O2 d 3 ~ H 2O d 5 ~ HO2 d6 ~ H d8 ~ N2 HO 2 + H → OH + OH OH + M → H + O + M OH + O → O2 + H OH + H + M → H 2 O + M H 2 + OH → H 2 O + H H 2 O + H → H2 + OH University of Kentucky Advanced CFD & Immersive Visualization Lab C FD LDM (Cont.) Example Equations Consider initiation step: contains two products: HO 2 and H, corresponding to d 7 and d 5 , respectively Results in following contributions to DDS (n+1) d7 (n) = −(βY + γuY + γvY )d 7 7 7 DDS-Based LDM Experiment 7 +ω ˙ 7 + d 7,0 with ω ˙ 7 = Cf,7,1 d1 d2 and W W1W2 7 k ″ ——– Cf,7,1 = ν7,1 f,1 Similar equations hold for all species Computed results compared with experiments of Meier et al. (case H3) in figure at right University of Kentucky Advanced CFD & Immersive Visualization Lab C FD BIFURCATION DIAGRAMS OF LDM Recent Detailed Studies of Specific Reduced Mechanisms Further Illuminate Model Behaviors For example, bifurcation diagrams display sequences of transitions accessible to model part (a): temperature regimes as functions of velocity strain rates part (b): H2O concentration vs. velocity strain rate and Re part (c): H 2O concentration behaviors as function of H2O concentration gradient table shows nominal values of bifurcation parameters not being varied University of Kentucky Advanced CFD & Immersive Visualization Lab C FD SOME STATISTICAL PROPERTIES OF MODEL Two Important Statistical Quantities: scalar dissipation and skewness of scalar derivatives: χ = 〈D(∇ξ)2 〉 s = 〈(∂T⁄∂y) 3 〉∕〈(∂T⁄∂y) 2 〉 3 ⁄ 2 ξ is elemental hydrogen mixture fraction: YH − YH,o ξ = ————— YH,f − YH,o Spatial derivatives obtained from time series using Taylor’s “frozenflow” hypothesis Model Results Well Within Range of Expected Experimental Values for These Statistical Properties University of Kentucky Advanced CFD & Immersive Visualization Lab C FD SUMMARY/CONCLUSIONS Alternative Approach to Construction of LES/LDM Briefly Described Outline for Deriving DDS Providing Temporal Fluctuations on SubGrid Scales Presented Specific Example of LDM Based on DDS Corresponding to 9-Step H 2 −O2 Reduced Mechanism Analyzed Via Bifurcation Diagrams Results Display Many Different Bifurcation Sequences and Wide Range of Possible Bifurcation Pameters Statistics of Results Generated with DDS Consistent with Physical Observations Suggests Potential Use of DDS Form of LDM for SGS Models in LES of Turbulent Combustion, and Possibly in Real-Time Control Applications University of Kentucky Advanced CFD & Immersive Visualization Lab C FD