Presented by

High Fidelity Numerical Simulationsof Turbulent Combustion

Jacqueline H. Chen (PI)Chun S. Yoo

David H. LignellCombustion Research Facility

Sandia National Laboratories, Livermore, CA

Ramanan SankaranNational Center for Computational Sciences

Oak Ridge National Laboratory

2 Chen_S3D_SC07

Direct numerical simulation (DNS) of turbulent combustion

DNS approach and role

Fully resolve all continuum scales without using sub-grid models.

Only a limited range of scales is computationally feasible.

Petascale computing = DNS with O(104) scales for cold flow.

DNS of small-scale laboratory flames. Investigate turbulence-chemistry interactions relevant in devices. Provide numerical benchmark data for predictive model validation and development.

Combustor size ~1m

Molecularreactions ~1nm

Turbulent combustion involves coupled phenomena at a wide range of scales

O(104) continuum scales

Turbulent combustionis a grand challenge

3 Chen_S3D_SC07

S3D—First principles combustion solver

Used to perform first-principles-based DNS of reacting flows

Solves compressible reacting Navier-Stokes equations

High-fidelity numerical methods

Detailed reaction kinetics andmolecular transport models

Multi-physics (sprays, radiation and soot) from SciDAC-TSTC

Ported to all major platforms

DNS provides unique fundamental insight into the chemistry-turbulence interaction

DNSPhysical models

Engineering CFD codes

(RANS, LES)

4 Chen_S3D_SC07

Efficient parallel scaling

Results from weak scaling test on various Office of Science platforms

Scales to 23,000 Cray XT processors

1000

100

101 10 100 1000 10000

Number of processors

Co

st p

er g

rid

po

int

per

tim

e-st

ep (

µs)

SP3 (NERSC)

SP5 (NERSC)

Itanium(PNNL)

XT(ORNL)

Opteron(Sandia)

X1E (ORNL)

5 Chen_S3D_SC07

Combustion science enabled by NCCSCO/H2 non-premixed flames (2005) 500M grid points

Flame-wall interaction (2006)

Lifted flames (2007) 1B grid points

Ethylene non-premixed flames (2007) 350M grid points

Lean premixed flames (2006)200M grid points

6 Chen_S3D_SC07

DNS of turbulent lifted H2/air jet flames in heated coflow Determine stability and

characteristics of a lifted flame

Understand flame stabilization mechanism Effect of degree of fuel-air pre-mixing Effect of turbulent flow Effect of preheating and auto-ignition

Simulation performed on Jaguar on 9000 cores and 2.5 million cpu-hrs ~1 billion grid points Detailed H2/air chemistry with

14109 DOF 9 resolved species and 21 elementary

reaction steps Jet Reynolds number = 11,000

Instantaneous volume rendering of the local mixing rate, by H. Yu and K. L. Ma of the SciDAC Institute for Ultrascale Visualization

7 Chen_S3D_SC07

Flame stabilization primarily due to autoignition

Flame stabilizes in fuel-lean mixture where the temperature is high (toward the heated air coflow) and mixing rates are low.

Hydroperoxy radical (HO2): Precursor of auto-ignition in hydrogen-air chemistry. Builds up upstream of OH and other intermediate

radicals (H and H2O2). Indicates auto-ignition should be primary

stabilization mechanism.

Instantaneous volume rendering of HO2 (ignition marker) and OH (flame marker) by H. Yu and K. L. Ma of the SciDAC Institute for Ultrascale Visualization

Temperature versus mixture fraction at various axial positions

8 Chen_S3D_SC07

Stabilization depends on competition between autoignition and large-scale eddy passage

Isocontours of temperature with mixture fraction (dotted white line) and streamlines (arrowed line) averaged in time and z-direction

Correlation of stabilization point and axial velocity

and ignition

Near flame base, slightly negative or small positive axial velocity is observed (recirculation assists stabilization of flame base).

Stabilization point movement is cyclic with passage of large-eddies

9 Chen_S3D_SC07

Extinction and reignition in a nonpremixed ethylene/air turbulent jet flame Goal: study extinction reignition processes in hydrocarbon flames

Autoignition Premixed flame propagation Edge flame propagation

3D DNS of a slot jet at Re = 5120

Reduced ethylene mechanism consisting of 19 transported and 10 quasi-steady state species, with 167 reactions (Lu and Law 2007)

340 million grid points (8.16x109 DOF)

2.0 million cpu-hours; 14,112 cores on Jaguar

10 Chen_S3D_SC07

Temperature history in a spanwise slice reveals extinction and reignition

Stoichiometric mixture fraction (black line)

(mm)

(m

m)

(m

m)

(mm) (mm) (mm)

11 Chen_S3D_SC07

Excessive mixing quenches the flame and flame reignites by premixed flame propagation

Propagation speed of ignition front compared with laminar flame speed

Conditional mean and variance of scalar dissipation rate (mixing rate) and temperature at the flame

0.2Time (ms)

0.3 0.4 0.5 0.6 0.7

7000

T

st

(K)

st,

(1

/s)

st

T

6000

5000

4000

3000

2000

0

1000

0 0.1 0.2 0.3 0.4 0.5 0.6Time (ms)

800

1200

1600

2000

2400

SD*SL

Me

an

YC

O2 s

urf

ac

e S

pe

ed

(c

m/s

)

300

200

100

0

-200

-100

12 Chen_S3D_SC07

Contacts

Jacqueline H. ChenPrincipal InvestigatorCombustion Research FacilitySandia National LaboratoriesLivermore, CA 94550jhchen@sandia.gov

Ramanan SankaranScientific Computing liaisonNational Center for Computational SciencesOak Ridge National Laboratorysankaranr@ornl.gov

12 Chen_S3D_SC07