Direct Numerical Simulations of Turbulent Combustion Jacqueline H. Chen Combustion Research Facility Sandia National Laboratories [email protected] Data.

Direct Numerical Simulations of Turbulent CombustionDirect Numerical Simulations of Turbulent Combustion

Jacqueline H. Chen

Combustion Research Facility

Sandia National [email protected]

Data Management All-Hands Meeting

March 3-4, 2005

Sponsored by the Division of Chemical Sciences Geosciences, and Biosciences, the Office of Basic Energy Sciences, the U. S. Department of Energy

Challenges in combustion understanding and modelingChallenges in combustion understanding and modeling

Diesel Engine Autoignition, Laser IncandescenceChuck Mueller, Sandia National Laboratories

Stiffness: wide range of length and time scales

– turbulence

– flames and ignition fronts

– high pressure

Chemical complexity– large number of species and

reactions

Multi-physics complexity – multiphase (liquid spray, gas

phase, soot)

– thermal radiation

– acoustics ...

Direct Numerical Simulation (DNS) Approach

High-fidelity computer-based observations of micro-physics of chemistry-turbulence interactions

Resolve all relevant scales

At low error tolerances, high-order methods are more efficient

Laboratory scale configurations: homogeneous turbulence, v-flame turbulent jets, counterflow

Complex chemistry - gas phase/heterogeneous (catalytic)

Turbulent methane-air diffusion flame

HO2

CH3O

CH4

O

Oxidizer

Fuel

. S3D0: F90 MPP 3D

. S3D1: GrACE-based

. S3D2: CCA-compliant

Software design developments

. IMEX ARK

. IBM

. AMR

Numerical developments

. Thermal radiation

. Soot particles

. Liquid droplets

Model developments

CFRFS

CCA

Post-processors: flamelet, statistical

CMCS DM

MPP S3D

Arnaud Trouvé, U. Maryland Jacqueline Chen, SandiaChris Rutland, U. WisconsinHong Im, U. MichiganR. Reddy and R. Gomez, PSC

High-fidelity Simulations of Turbulent Combustion (TSTC) http://scidac.psc.edu

3D DNS Code (S3D) scales to over a thousand processors

Scalability benchmark test for S3D on MPP platforms - 3D laminar

hydrogen/air flame/vortex problem (8 reactive scalars)

Ported to IBM-SP3, SP4, Compaq SC, SGI Origin, Cray T3E,

Intel Xeon Linux clusters

Office of Science INCITE award provides 2.5 million cpu-hours at NERSC for combustion science simulationOffice of Science INCITE award provides 2.5 million cpu-hours at NERSC for combustion science simulation

Direct simulation of a 3D turbulent flame with detailed chemistry (200 million grids, 12 species, 5 TB raw data, 5 TB derived data, 3000 cpus)

• Extinction-reignition dynamics

• Among largest simulations

• Benchmark data for testing models

• FY05 BES Joule PART goal

3D DNS performed at NERSC, ORNL, PNNL – preparatory runs of up to 40 million grid points, 20 dof

Extinction-Reignition DynamicsExtinction-Reignition Dynamics

Mechanisms for reignition: Edge flame propagation, flame propagation normal to isosurface, self-ignition

TNF Workshop: International Collaboration of Experimental and Computation Researchers TNF Workshop: International Collaboration of Experimental and Computation Researchers

• International Workshop on Measurement and Computation of Turbulent Nonpremixed Flames (since 1996)

– Framework for detailed comparison of measured and modeled results– Identify what does not work, define research priorities– Core groups: Berkeley, Cornell, TU Darmstadt, Imperial College, U Sydney

• Adds leverage and impact to BES Combustion Program– Built around Sandia experiments and CRF visitor program– New opportunities for numerical benchmarks – highly resolved LES and DNS

Reacting Turbulent Jet flow Simulation Reacting Turbulent Jet flow Simulation

Heat release rate

3D Turbulent Reactive Jet Flames – 40 Million Grids, 1 TB data, 480 cpus on MPP2 at PNNL3D Turbulent Reactive Jet Flames – 40 Million Grids, 1 TB data, 480 cpus on MPP2 at PNNL

Vorticity magnitude OH mass fraction

Volume Rendering by Kwan-Liu Ma

Motivation: Control of HCCI combustion

Overall fuel-lean, low NOx and soot, high efficiencies

Volumetric autoignition, kinetically driven

Mixture/thermal inhomogeneities used to control ignition timing and burn rate

Spread heat release over time to minimize pressure oscillations

Experimental evidence of ignition front propagation Experimental evidence of ignition front propagation

PLIF of OH in HCCI engine at TDC, Richter et al. 2000

Hultqvist, et al. 2002 – chemiluminescence and fuel LIF imaging of time-

resolved sequence in a single cycle

Volumetric combustion early on, kernel evolution at discrete locations later

(discrete edges between burned/unburned, reaction fronts spreading at 15

m/s.

Objectives Objectives

Gain fundamental insight into turbulent autoignition with compression

heating

Develop systematic method for determining ignition front speed and

establish criteria to distinguish between combustion modes

Quantify front propagation speed and parametric dependence on

turbulence and initial scalar fields

Develop control strategy using temperature inhomogeneities to control

timing and rate of heat release in HCCI combustion

deflagration

spontaneous ignition

detonation

Chen et al., submitted 2004, Sankaran et al., submitted 2004

Temperature skewness effect on heat release rateTemperature skewness effect on heat release rate

Heat release, HighT, positive skewness

2.0 ms

2.4 ms

2.6 ms

2.8 ms

Symm Hot core Cold core

Ignition front tracking methodIgnition front tracking method

cd

DtDss

od

*

YH2 = 8.5x10-4 isocontour – location of maximum heat release

Laminar reference speed, sL based on freely propagating premixed flame at local enthalpy and pressure conditions at front surface

Density-weighted displacement speed (Echekki and Chen, 1999):

Species balance and normalized front speed criteria for propagation mode Species balance and normalized front speed criteria for propagation mode

Black lines – s*d/sL < 1.1 (deflagration)White lines – s*d/sL > 1.1 (spontaneous ignition)

A – deflagration B, C – spontaneous ignition

A C

B

Heat release isocontours

SummarySummary

Addition of hot fluid parcel (temperature skewness) slows down heat release, so does increasing temperature variance – effective control of HCCI

Both spontaneous ignition and deflagrative propagation present for initial spectrum of ‘hot’ spots modulated by turbulent mixing

Significant effect of heat conduction and dissipation of temperature gradients along with front annihilation – increase propagation rate

New method for determining the speed of ignition fronts and criterion for deflagrative versus spontaneous front propagation (s*d/sl > 1)

Detection and tracking of autoignition features

FDTools (Koegler, 2002): evolution of ignition features

Hydroperoxy mass fraction

Feature graph tracks evolution of ignition featuresFeature graph tracks evolution of ignition features

time

Feature-borne analysisFeature-borne analysis

800

1000

1200

1400

1600

1800

2000

2200

2400

2600

2800

0 0.02 0.04 0.06 0.08 0.1Time (msec)

Max

Tem

pera

ture

(K) #5

#46#39#47

#45#40 #11

#18,#52#27

#41,#68

Terascale virtual combustion analysis facilityTerascale virtual combustion analysis facility

Data Management Challenges for CombustionData Management Challenges for Combustion

• Parallel data-analysis tools for combustion analysis– 3D iso-level set analysis normal and tangent to surface for thin flames– Conditional statistics– Reduced representations of combustion data (POD, PCA, topology of

vector and scalar fields) for model development and viz.– Tracking flame elements or fluid particles in time - interpolating

• Parallel feature detection and tracking of TB-scale data• Quantitative viz. coupled with analysis of TB-scale – vol. rendering• Mid-range platforms for preparing runs, analysis and visualization (10-fold

smaller than leadership class – 1 Tflop, $300-600K Opteron cluster, raid storage systems 1-10 TB)

• IO issues for postprocessing phase when temporal analysis is required.• Further remote analysis and viz. of numerical benchmark data and

comparison with experimental data by modelers at different locations – Framework or Virtual Facility??

• Jointly funded activities (?? FTE’s combustion; ?? FTE’s from Data Management ISIC both for research and deployment).

AcknowledgmentsAcknowledgments

SNL Postdoctoral fellows: SNL collaborators:

Evatt Hawkes Jonathan Frank

Shiling Liu John Hewson

Chris Kennedy Wendy Koegler

Ph.D. Student:

James Sutherland

External collaborators:

Prof. Stewart Cant (Cambridge U.) Prof. Heinz Pitsch (Stanford)

Prof. Hong Im (U. Michigan) Prof. Tarek Echekki (NC State)

Prof. Arnaud Trouve (U. Maryland) Ramanan Sankaran (U. Michigan)

Prof. Chris Rutland (U. Wisconsin) Reinhard Seiser (UCSD)

Prof. K. Seshadri (UCSD) R. Reddy and Wang (PSC)

Computing ResourcesComputing Resources

DOE NERSC – IBM SP

ORNL – IBM SP

PNL – Linux cluster

SNL – Intel Linux cluster, SGI Origin, Compaq Sierra Cluster