Next time you’re presented with a birthday cake, try blowing gently on the candles before giving the big, out-they-all-go puff.
If you do it right, they should burn more intensely. But blow too hard, of course, and the candles go out.
That light puff of air is a form of turbulent mixing. Turbulent mixing produces more power from combustion, but – like blowing hard on a candle – too much causes problems, says Jacqueline Chen, a researcher at Sandia National Laboratory’s Livermore, Calif., site. Sandia is a Department of Energy facility.
“Key chemical reaction rates can’t keep up with the mixing rate, and local quenching occurs” – pockets where fuel remains unburned, Chen says. “If it’s pervasive enough, total blowout may occur” – a bad thing if it happens in an airplane engine. Even local quenching cuts efficiency and increases pollution as unburned fuel goes out the exhaust.
Designers are turning to turbulent mixing and combustion at lower temperatures as they strive to make more efficient and cleaner-burning engines. That’s “pushing combustion to the ragged edge,” Chen says, where flames are nearly unsupportable and combustion occurs by other means, such as autoignition – ignition without an external source like a flame or spark. Combustion at these limits is poorly understood.
Chen and her fellow researchers are casting new light on the issue with a massive computer simulation of a turbulent, non-premixed jet flame with detailed chemistry. Their work could lead to more efficient, cleaner engines in cars, jets and other vehicles and devices. The Department of Energy’s Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program made the simulation possible with a grant of 2.5 million computer processor hours. It was the largest such grant announced in 2005, and the simulation ran on some of the world’s most powerful computers:
- The IBM SP3 RS/6000, named Seaborg, at the National Energy Research Scientific
- Computing Center (NERSC) at Lawrence Berkeley National Laboratory in California
- The IBM p575 POWER 5, named Bassi, at NERSC
- The Cray X1E and CrayXT3 at Oak Ridge National Laboratory in Tennessee.
The simulation called for that kind of power because of its incredibly high resolution. It calculated reactions at each of up to half a billion points in the simulated flame at 120,000 time increments. It tracked 11 different kinds of molecules and 21 different reactions. The simulation generated about 30 terabytes of data – enough to fill the 100-gigabyte hard drives on 300 desktop computers.
That kind of detail would have been impossible without INCITE’s massive allocation of high-performance computer time. With less processor time and less-powerful computers, the simulation would have taken decades to run. Instead, it took just a few weeks.
The results question some basic assumptions about combustion. The researchers found:
- The simulation revealed that how reactive molecules mix in combustion is different from how non-reactive molecules mix. The process depends not just on the balance between how turbulence moves molecules and how those molecules diffuse on their own, but also on how the molecules react. That was surprising, because scientists usually have assumed that the time it takes to mix is the same for all kinds of molecules and corresponds to how turbulence changes over time. The simulation also found this dependence was present even at high rates of turbulence.
- Greater turbulence causes greater intermittency – small fluctuations in which combustion stops and restarts – and a longer lag time for reignition.
- Locally extinguished portions of the flame most often are reignited when turbulence folds burning flame sections onto non-burning ones, rather than by the slower process of smooth, nonturbulent flame propagation along the edges of the extinction holes.
This fundamental understanding could lead to new, more accurate combustion models and better engine designs, says Evatt Hawkes, a Sandia postdoctoral researcher who worked on the simulation. “The biggest impact will be felt if this will translate into a model that can be used at an engineering level,” he adds.
The researchers continue to analyze mountains of data the simulation generated, seeking clues to local extinction and reignition. They’re also focusing on fundamentals of the turbulence-chemistry interaction. Because of the amount and complexity of the data, the researchers are working closely with computer scientists to:
- Automatically find and focus on interesting details in the simulation
- Analyze events in the simulation across multiple scales of time and space – from billionths of a second to minutes and from atoms to whole flames
- Create images to understand the simulation results
- Manage the huge amounts of data these simulations produce.
It couldn’t be done without INCITE and DOE’s world-class computers, says Ramanan Sankaran, another Sandia postdoctoral researcher on the team.
“Leadership computing is enabling this kind of science,” Sankaran says.