1. Perspectives from the TNF Workshop
Robert S. Barlow
Combustion Research Facility
Sandia National Laboratories
Support by:
DOE Offices of Basic Energy Sciences
Multi-Agency Coordinating Committee on Combustion Research (MACCCR)
Workshop on
Next Steps in Using Combustion Cyberinfrastructure

2. Perspectives from the TNF Workshop* : Outline
Background
History (>10 years)
What, Why, Who, When
Use of the web
Past use
Types of content
Current status
Future directions
Expanded types of flames
Expended types of data
Expanded size of content
What might Cyber Infrastructure do for TNF?
Needed functionality
Chicken-egg problem
* International Workshop on Measurement and Computation of Turbulent Nonpremixed Flames

3. Validation of Models for Turbulence/Chemistry Interaction
TNF Workshop – Focus
Validation of Models for Turbulence/Chemistry Interaction
spray
pressure
scaling
particulates
kinetics
complex
geometry
turb/chem
practical
combustion
systems
Simple Jet Piloted Bluff Body Swirl
Progression of well documented cases that are consistent with to the range of current capabilities of current models

4. TNF Workshop – Basics Genesis (early 1990’s)
Incomplete experiments, poorly defined comparisons with computations
Internet offered opportunity for rapid communication, collaboration, data sharing
TNF1 in Naples (1996) established the ground rules
Collaboration of experimental and computational researchers
Core groups: Sandia, Berkeley, Cornell, TU Darmstadt, Imperial College, U Sydney
Framework for detailed comparisons of measured and modeled results
Identify gaps in data and models, define research priorities
Emphasis on fundamental issues of turbulence-chemistry interaction
Nonpremixed and partially premixed flames of simple fuels (H2, CO, CH4)
Progression in complexity of flow field and kinetics
Public (web-based) availability of data sets and comparisons
Primary basis (so far)  point statistics from Raman/Rayleigh/LIF and LDV
Naples
Heppenheim
Boulder
Darmstadt
Delft
Sapporo
Chicago
Heidelberg
1996
1997
1998
1999
2000
2002
2004
2006

5. TNF Workshop – Themes
“We emphasize that this is not a competition, but rather a means of identifying areas for potential improvements in a variety of modeling approaches.”
“This collaborative process benefits from contributions by participants having different areas of expertise, including velocity measurements, scalar measurements, turbulence modeling, chemical kinetics, reduced mechanisms, mixing models, radiation, and combustion theory.”
TNF8 Organizing Committee
R.S. Barlow, R.W. Bilger, J.-Y. Chen, A. Dreizler, J. Janicka, R.P. Lindstedt, A.R. Masri, J.C. Oefelein, H. Pitsch, S.B. Pope, D. Roekaerts, and L. Vervisch.

6. Gallery of Turbulent Flame Examples
Figure 1 (Barlow, ProCI 2007)
a, b Simple jet flames
c, d Piloted jet flames
e Bluff-body
f Bluff-body/Swirl
g Lifted flame in vitiated coflow
h Opposed jet flame
i Unconfined swirl flame
j Enclosed swirl flame
k Premixed low-swirl flame
l Premixed swirl, bluff-body
m Enclosed premixed swirl flame
n Premixed jet in vitiated coflow

7. TNF Workshop – Current Mode of Operation
People
1 main organizer
a few (~6) active members of the organizing committee
loose ongoing scientific collaborations (roughly people in “clumps”)
~ 80 participants at each workshop (every 2 years)
Budget
no targeted funding for TNF Workshop activities
Sandia (DOE/BES) pays for some admin. support
TU Darmstadt has also contributed admin. support (SFB 568)
Sponsorship used to reduce registration fees for faculty & students
Web Contents: (
very simple site
data sets (compressed ascii files) and submodels (chem, mixing, radiation)
proceedings (PDF)
bibliography
total < 1GB (only point data; no 1D, 2D, 3D; no simulation data)

8. TNF Workshop – Web Issues?

9. TNF3 Workshop (1998) – Classic results for Flame D
laser
axis
laser axis
x/d =45
x/d =30
x/d =15
x/d =7.5
x/d = 2
Premixed Pilot Flame
PDF, CMC, ILDM steady flamelet, …

10. TNF4 Workshop (1999) – Example Comparisons
T scatter plots
Conditional means: T, O2, CO2, CO, H2O, H2, CH4, OH, (NO)
Axial and radial profiles: U, u’, F, F’, T, T’

11. Current Status on Piloted Flames
Thorough parametric studies of model sensitivity in RANS/PDF
Pope’s group at Cornell
Combinations of different mixing models and mechanisms
100’s of pages of figures as “supplemental material”
Ten years to really understand the details for set of 3 flames
Other RANS approaches still trying to get extinction right
Only one LES/FDF of flame E (w/ extinction) by V. Raman

12. TNF8 (2006) Target Flames and Phenomena
Bluff-Body Stabilized
Sydney University, Sandia
High velocity coflow, central fuel jet
Flow recirculation
CH4/H2, CH4/air, CO/H2
Swirl/Bluff-Body
Sydney Univ., Sandia
Large-scale instability
Vortex breakdown
CH4, CH4/air, CH4/H2
Comparisons of measured and modeled results on the Sydney Bluff-Body and Swirl flames coordinated by Andreas Kempf at the TNF8 Workshop

13. TNF8 Workshop (2006) – Comparison Table for BB and Swirl
Prepared by Andreas Kempf

14. TNF Workshop – Successes
improved quality and availability of experimental data for “simple” flames
higher “standards” for combustion model validation
better understanding of model capabilities and experimental uncertainty
a few very productive collaborations
collaboration & rapid feedback  accelerated progress for same $$
impact (journal references, use by industry), high visibility
11/20 nonpremixed modeling papers in PCI used TNF data

15. TNF Workshop – Failures
still very few “appropriate, complete” data sets
comparisons limited to easiest quantities (point statistics)
little information from calculations has been preserved
funding for collaborative experimental work in the US is very limited
currently no model for growth of TNF Workshop beyond “cottage industry” (larger data sets, more complete comparisons, …)
web content editor is way behind!

16. TNF Workshop – Future Directions and Challenges
Larger experimental data sets
beyond single point statistics
imaging data (PIV, stereo PIV, 2-D and 3-D imaging data)
time domain (high-speed imaging, multi-frame, time series…)
more complex flames (fuel and flow geometry)
Larger simulation data sets
highly-resolved LES of lab-scale flames
basis for comparison?? – lots of work to do here
feature extraction, vis, etc.
Connections to other areas
premixed, soot, pressure effects, multi-phase
educational possibilities
more formal “publication” of results
comparison “engine”

17. Turbulent Combustion Laboratory
Instantaneous thermochemical state and local flame structure
Combined measurement:
T, N2, O2, CH4, CO2, H2O, H2, CO
220-mm spacing, 6-mm segment
state of mixing (mixture fraction)
progress of reaction
rate of mixing (scalar dissipation)
local flame orientation
8 laser
5 cameras
7 computers

18. New Data Sets Can Take Years to Fully Analyze
Line-imaged Raman/Rayleigh/CO-LIF and Crossed OH PLIF in the “DLR-A” and “DLR-B” flames
CH4/H2/N2 (22.1%, 33.2% 44.7%)
Nearly constant Rayleigh cross section
8 mm jet exit diameter, 0.3 m/s coflow, ~60 cm flame length
DLR-A: Re = 15, DLR-B: Re = 22,800
High resolution Rayleigh (40 mm sample spacing, 50 mm optical)
Conditional Mean c
DLR-B, x/d=10

19. More Piloted Jet Flames
Piloted CH4/H2/air jet flames
Collaboration with Henri Ozarovsky and Peter Lindstedt (Imperial College)
High Reynolds number (60,000 and 67,000); f = 3.2, 2.5, 2.1; total of 18 cases
Increase local extinction in multiple steps
Lindstedt, Ozarovsky, Barlow, Karpetis, Proc. Comb. Inst. 31 (2007)

20. Simultaneous Planar Imaging: Scalar Dissipation
Temperature, T
Mixture fraction, x
Forward reaction rate, [CO] + [OH]  [CO2] + [H]
Scalar dissipation, c
J.H. Frank, S.A. Kaiser, M.B. Long, Combust. Flame 143 (2005)

21. High-Speed Multi-Frame OH PLIF Imaging + Stereo PIV
CH4/H2/N2
DLR-B
Six Frames of OH PLIF, Dt = 30 ms
Strain Rate OH
outline
J. Hult, U. Meier, W. Meier, A. Harvey, C.F. Kaminski, Proc. Combust. Inst. 30 (2005)

22. Coupling Experiments and Simulations
Direct comparisons with matched bc’s
High-fidelity (expensive)
Combined experimental & computational benchmarks
Turbulent Flame Experiments
Address key flame phenomena, but provide only partial information
TNF Workshop flames
Large Eddy Simulation (LES)
Resolve energetic scales, but model subgrid scales

23. Photograph of Experimental Flame
CH4/H2/N2 Flames
Nozzle: dINNER= 8.0 mm, tapered to sharp edge
Fuel: 22.1% CH4, 33.2% H2, 44.7% N2
Coflow: 99.2% Air, 0.8% H2O
Stoichiometric mixture fraction: Zst = 0.167
DLR A:
UJET= 42.2 m/s
TJET = 292 K
Red = 15200
UCOFLOW= 0.3 m/s
TCOFLOW = 292 K
Flame Luminosity (800 s)
Photograph of Experimental Flame

24. Cross-section of 6-million cell 3D curvilinear grid (80cm x 40cm)
State of the art (high-quality) grids
Distributed multiblock domain decomposition
Adaptive mesh capability (R-refinement technique)
CH4/H2/N2 Flames
Cross-section of 6-million cell 3D curvilinear grid (80cm x 40cm) 20
Field of view for Kaiser and Frank imaging experiments 10 5

25. Scatter Data (x/d = 10)
Experiment
LES

26. Data Capacity Requirements for LES and DNS
Unformatted data:
Primitives: Q = [ρ, ρu, ρv, ρw, ρet, ρY1, …, ρYN-1]T
8 Bytes/[(Variable)(Grid-Cells)(Time-Step)]
10 variables, 1-million grid-cells  80 Mbytes/Timestep
Variables:
10 – 100 (spatial-coordinates + primitives + composite)
Grid size (presently):
O(1 – 10-million) … LES
O(1 – 100-million) … DNS
Typical animation uses approximately 200 frames
Tradeoff between amount of data stored and cost of post-processing on-the-fly
Capacity requirements on the order of terabytes rapidly approached
from J. Oefelein

27. LES-FDF Simulation of Sandia Flames
Simulation details
256 x 128 x 32 points
30-50 particles/cell
More than 5K time steps
Data stored
75 Mb of LES data/timestep
3.75 Gb of FDF data/timestep
20 time-stations stored
Mining operations
Sub-filter FDF conditioned on filtered scalar fields
Conditional mixing plots
Flame D
Flame E
from V. Raman, UT Austin

28. LBNL Laboratory-scale DNS O(10 cm)3
Lean premixed turbulent flames (V- and slot-flames, and low-swirl burners)
Very large scale simulations, current datasets O(1-10+) TB
Complex detailed chemical mechanisms w/detailed transport, 100’s of reactions
Map simulations onto theory / experiment.
Use simulations to analyze flames beyond traditional theory/experimental approaches
Time-scale of research:
Code/algorithm development O(10 yrs) Highly specialized enabling algorithms, ongoing development
Data production O(6 mos) Remote supercomputers, high-speed networking, parallel filesystems
Validation O(6 mos) Significant interaction with experi- mentalists/theorists
Extraction of “new science” O(1 yr) Analysis framework requires multi-year interactions to relate simulations to experimental diagnostics, theory and beyond
Requirements for data analysis
- Computing resource intensive, supercomputer CPU/disks/archiving
- Complex, specialized data structures specific to the simulation
- Data manipulation consistent with simulation (physics databases, discretization)
- Manpower intensive (specialized analysis driven by multi-disciplinary interactions)
- Coordinate efforts amongst math/computer science, theory, experimental diagnostics

29. What might Cyber Infrastructure do for TNF Workshop
Collection, packaging, and distribution of large data sets:
Experiments and high-fidelity simulation
Reduce labor by standardizing(?) and automating the upload process
Allow for archiving of model results in digital form
Include functionality to interactively view (plot) data in various ways
Promote formation of more collaborative groups
Pay for people to build/maintain custom tools for data archiving/analysis
Infrastructure is not (currently) the bottle neck for progress in TNF
Interagency collaboration to promote research on targeted problems could be very useful