Rixin YuDivision of fluid Mechanics

Department of Energy ScienceLTH

Numerical simulation study of turbulent combustion phenomena

-INTEGRATE Advanced Study Group (ASG)

9/27/20161

IntroductionSome combustion related keywords

Energy, heat and power, transportation.Emission, clean combustion, global warming.Fossil fuel, Hydrogen, biomassCombustion efficiency, combustion instabilityTurbulence combustion

9/27/20162

IntroductionTo understand turbulent reacting system, a challenge

Turbulence + Chemistry Wide range of scales

Da Vinci’s sketch

9/27/20163

Research methodologyMathematical equations describing turbulent reacting flow

Flow turbulenceNaiver-Stokes equations(Conservation of mass, momentum, energy), assuming low-Mach number.

Chemistry reactionConservation of N species mass

o Yk , k=1,..N (YH, YO2, YCH3, …)

Reaction rates ωk is given by M elementary chemical reactions

Reactions for CH4/AirH+O2↔O+OHO+H2↔OH+HH2+OH↔H2O+HOH+OH↔H2O+OH+O2+M→HO2+MH+HO2→OH+OHH+HO2→H2+O2OH+HO2→H2O+O2CO+OH↔CO2+HCH4+M↔CH3+H+MCH4+H↔CH3+H2CH4+OH↔CH3+H2OCH3+O→CH2O+HCH2O+H→HCO+H2CH2O+OH→HCO+H2OHCO+H→CO+H2HCO+M→CO+H+MCH3+O2→CH3O+OCH3O+H→CH2O+H2CH3O+M→CH2O+H+MHO2+HO2→H2O2+O2H2O2+M↔OH+OH+MH2O2+OH↔H2O+HO2OH+H+M→H2O+MH+H+M→H2+M

9/27/20164

Research methodologyNumerical simulations of P.D.E governing turbulent reacting flow

Partial-resolution methods (filtered P.D.E., need models)

Reynolds Averaged Navier-Stokes(RANS) Cheap; favored by industrial; Applied approach.

Large Eddy simulations(LES) more accurate, capturing dynamic features

Full-resolution method (no model)Direct(Detailed) numerical simulations(DNS)

Computational expensive, limited to very small sizelog(k)

logE

(k)

resolved LES modelled LES

modelled RANSresolved DNS

-5/3

2π/l0 2π/ηintegral Kolmogorov

2π/δ

9/27/20165

Data analysis : three case studies1: DNS studies of homgenouslycharged compression ignition (HCCI) engine combustion.

2: DNS of turbulent premixed flame propagation

3: Fractal flame front structures due to flame (Landau-Darrieus) instability

9/27/20166

*Animation edited by G. Leung, Original from Mitsubushi Corp

Case 1: Internal combustion(IC) engine

Four strokes Intake CompressionExpansionExhalation

HCCI engineLow NOx emission and high efficiencyControl problem

Types of IC engineDiesel engineSpark ignition(SI) engine

9/27/20167

Case 1: How does in-cylinder flow looks like?Full cycle LES of turbulent flow inside a typical diesel engine with curved-bowl shaped piston

Complicated flow, inhomogeneous mixing9/27/20168

Case 1: In-cylinder turbulent flow and T fieldPiston with a squared bowl

C

n

VR

=Vm+VRV

9/27/20169

Case 1: Some ways of engine-combustion

No spark

Strong turbulence Weak turbulence

Add spark(combining flame propagation + igntion)

_______________________________________

Homogenously charged compression ignition (HCCI)

Sensitive to T fluctuation

Engine combustion inside engine starts from an inhomogeneity in Temperature field and fuel/air concentration

9/27/201610

Case 1: What is HCCI combustion?2D-DNS of H2/Air auto-ignition in a small square (periodic) domain

9/27/2016Initial condition: Averaged T: 1070K; T rms fluctuation: 50K Pressure : 41 bar

11

Case 1: What is HCCI combustion?3D-DNS of Lean H2/Air auto-ignition in a small cubic (periodic) domain

Yu. R et al, Comb. Flame, 2013

Domain size: 53 mm; Result plotting durations: 2 - 3 ms.Initial condition: Averaged T: 1070K; T r.m.s. fluctuation: 50K Pressure: 41 bar

ObservationsMultiple ignition kernels (initial hot spots)Two combustion modes

Slow deflagration waveRapid auto-ignition mode

Complicate topological surface for the reaction front

Saddle surfacesSpherical surfaces

9/27/201612

Case 1: Data analysis (Displacement speed: Sd)

Yu. R et al, Comb. Flame, 2013

= − Diffusion( )+ Reaction( ) ∗

9/27/201613

Case 1: Data analysis (Displacement speed: Sd)= − Diffusion( )+ Reaction( ) ∗

9/27/201614

Case 1: Data analysis Surface topology and its contribution to combustion process

Saddle surface

Spherical surface

Cylindrical surface

1: Both strong saddle and spherical fronts occupy small fraction of total flame area, however they contributes more to the flame stretch, due to the associcated larger Sd.2: In early stage small spherical fronts contribute to a large fraction of total flame stretch, saddles surfaces changes from convex to concave quite early, they are reasonable for reduction of flame front area in late stage of combustion process.

9/27/201615

Case 2: DNS of turbulent, statistical planar, premixed flame

9/27/201616

Case 2: More distributed layer of radical species with increased turbulence intensity

YH*

YHCO*

YHO2*

Ka = 65 Ka = 125 Ka =550 Ka =900 Ka = 3350 9/27/201617

Case 2: DNS of turbulent premixed combustionAt Intensive turbulence (Ka =3350) , two different fuel/air reactants

H. Carlsson, R.Yu, X. Bai, Int. J. Hydrogen ,2014H. Carlsson, R.Yu, X. Bai, Proceeding of combustion Institute, 2015.

CH4/air H2/air

9/27/201618

Case 2: Data analysisObservations: High heat release rate at low temperature region

Non-conventional chemical pathway!Caused by turbulent transport of the most-diffusive H radical !Elementary reaction R(9) is greatly enhanced:

H+O2+M HO2+M

Heat release contributed from R(9) conditioned at T<700K

9/27/201619

Case 3: Intrinsic flame instabilityDarrieus-Landau (DL) instability (caused by density difference)

Cusp shapes or cellular flames shapes,developed with no (or weak) turbulence

9/27/201620

Case 3: Simulations of LD instability in periodic channel

Increasing channel width

Curved flame, but stationary.

Non-stationary flamecusp formation

Fractal flame structure.

9/27/201621

Case 3: Data analysis-Fractal flame

Fractal cascadingstructures:

(1) 4-6 (intermediate) cells on a (largest) cell

(2) 4-7 (minor) cells on a (intermedia) cell

9/27/201622

Case 3: Data analysisThe Fractal flame cascading concept

Koch snowflake

1

4/3

(4/3)2

Gostintsev, et al. 1988

23 9/27/2016

Case 3: Data analysis: Fractal dimension (1+d)power-law fitting of the velocity amplification (Uw)

Power law fitting of Uw(Propagation speed after reaching statiscal staionary)~≈ 0.65 − 0.75≈ . − .

No analytical theory aviable for large channel width9/27/201624

Case 3: Data analysis: Fractal dimensions (1+d’)Box-counting of fractal images

Box counting≈ 0.07 − 0.1249/27/201625

Thanks for you attention

9/27/201626