Carter cluster aids in probing turbulent combustion in engines

  • March 28, 2013
  • Science Highlights

Consider two vital processes that go on in your car engine. One is turbulence, a kind of chaotic, random flow involving fuel and air. The other is combustion, a chemical process by which fuel ignites to provide the oomph to move the vehicle forward.

Either process presents a complicated problem and Haifeng Wang, aeronautics and astronautics professor, models both in an effort to better understand how the processes unfold and interact. The payoff could be knowledge leading to more efficient engines that burn less fuel.

"We want to improve the design of combustors so that they are clean and they are efficient," Wang says. "Clean in the sense that they produce less pollutant emissions. Efficient, we're burning fossil fuels that are nonrenewable, so given the limited amount we want to use it as efficiently as possible."

To capture both processes, Wang's lab combines large eddy simulations (LES) to capture the turbulent flow and a probability density function (PDF) method to capture detailed chemical kinetics and interactions between chemistry and turbulence. The researchers approach the probability problems stochastically, factoring in both predictable actions and random elements to provide more of a real-world picture.

"It's a state-of-the-art modeling strategy," Wang says.

For Wang, simple jet flames serve as, in a biology research sense, a model organism. Jet flames are well characterized experimentally and by other modeling methods, offering a solid starting point and basis for comparison in Wang’s modeling.

"To some extent, we know what's going on in those flames," Wang says. "To measure a realistic gas turbine engine, or internal combustion engine, it's not practical — yet. It's very difficult to make a detailed measurement. There's no experimental data available. We're trying to move to those more realistic problems in time."

The work involves demanding computational fluid dynamics (CFD) and Monte Carlo simulations. To get the length scales and resolutions he wants, Wang breaks the CFD problems into a million or more grid cells. The Monte Carlo simulations capture interactions among as many as 30 million particles.

"A third reason we need supercomputing is because of chemistry," Wang says. "We are treating very detailed chemical kinetics. Even for the simplest fuel, like hydrogen, the chemistry involves on the order of 10 species (atoms, molecules, molecular fragments and the like) and 20 reactions. Chemistry also has a wide range of scales. The smallest time scales may go down to 10 to the minus nine seconds and the larger time scales may be up to one second."

Even the simplified jet flame problems take considerable computational muscle. Wang has run simulations on some of the largest high-performance computing systems at national laboratories and, since joining the Purdue faculty, on Purdue’s Carter community cluster supercomputer. Carter was the fastest system serving a university campus when it went online in 2012.

Carter has proven to be "very fast," says Wang, who came to Purdue from Cornell in July 2012. "The performance doubled for some codes."

Wang says the University's community clusters were a selling point in his coming to Purdue. Another school planned to build him a cluster but discussions with its technical people left him feeling a bit lost.

"With Carter, I just worry about how to make use of it," Wang says. "I don't need to worry about the purchasing, installation, maintenance, anything like that. We're very busy and those things are time consuming."

More Information

Originally posted: July 1, 2014 4:18pm EDT