The Complexities of Combustion
Titan proves vital for design of next-generation engines
Despite the rush to green energy technologies, traditional combustion engines and devices will continue to be a major economic reality for years to come, in part because they have the potential to employ low-carbon and renewable fuels.
This reality has made improving the efficiency of these devices, from internal combustion engines to gas turbine power plants to industrial boilers, a priority for the Department of Energy. The effort would help reduce both our reliance on fossil-based fuels and the amount of CO2 being released into the atmosphere.
As a point of reference, consider the fact that Americans use two-thirds of their petroleum for transportation and one-third for heating buildings and generating electricity. “If low-temperature compression ignition concepts are widely adopted in next-generation automobiles, fuel efficiency could increase by as much as 25 to 50 percent,” said Sandia National Laboratories’ Jackie Chen, who uses the Oak Ridge Leadership Computing Facility’s (OLCF’s) Titan supercomputer to study the combustion of a wide variety of fuels. OLCF is a DOE Office of Science User Facility.
Specifically, Chen’s team member Ankit Bhagatwala recently employed the direct numerical simulation (DNS) code known as S3D on Titan to simulate a jet flame burning dimethyl ether in an attempt to match the conditions of a companion experiment at Ohio State University (OSU).
The jet flame configuration is used to probe fundamental turbulent flame physics associated with local extinction, where parts of the flame burn out, and reignition that may occur in diesel jet flames. If researchers can figure out strategies to minimize flame extinction, this will greatly enhance efficiency and minimize undesired emissions in combustion devices such as engines.
But even with computers as powerful as Titan, performing a direct numerical simulation of a diesel jet flame matching all of the aero-thermo-chemical conditions and with all of its chemical reactions is out of the question. Researchers must therefore simulate a jet flame configuration at lower Reynolds number (a measure of the mixing intensity and dynamic range of turbulence) that matches the critical thermo-chemical conditions associated with combustion, using around 30 chemical molecules known as species to model the combustion of dimethyl ether. This enhanced number of species, only possible on systems such as Titan, advances the team toward simulating the behavior of more realistic fuels, including biofuels.
While Chen’s team has simulated jet flames in the past, the latest simulations on Titan were a breakthrough for two reasons: the inclusion, for the first time, of dimethyl ether (DME) and the highest Reynolds number ever achieved by the team, 13,050. The increased Reynolds value allows the team to resolve a wider range of turbulence scales in space and time, a major breakthrough when trying to match experimental conditions and also evaluate turbulent mixing and combustion models.
“These simulations represent the first time we’ve incorporated DME and the highest Reynolds number ever achieved in a fully resolved reacting direct numerical simulation,” said Ramanan Sankaran, an OLCF scientific liaison who assists Chen’s team with their simulations on Titan.
Chen’s team will make the data available to the international modeling community for the development and validation of more accurate mixing and combustion models, alongside benchmark experiments, when feasible. In fact, Xinyu Zhao, another member of Chen’s group, is using this data to evaluate a newly proposed turbulent mixing model.
These models will ultimately be used in engineering-scale computational fluid dynamics (CFD) simulations, which run on desktops and computer clusters to optimize designs of combustion devices using diverse fuels. Because industrial researchers must conduct thousands of calculations around a single parameter to optimize a part design, individual calculations need to be inexpensive.
The goal is a shorter, cheaper design cycle for US industry. The work addresses DOE mission objectives to maintain a vibrant science and engineering effort as a cornerstone of American economic prosperity and to lead the research, development, demonstration, and deployment of technologies to improve energy security and efficiency.
And Titan, a Cray XK7, is the perfect platform for such research. Currently the nation’s fastest supercomputer for scientific research, it is uniquely geared to handle large, complex problems such as the turbulent combustion that takes place when fuel is burned in an engine.
In fact, Titan and S3D pair so well together that the application is six times faster on Titan than on the OLCF’s previous CPU-only system known as Jaguar. This increase in speed is due almost entirely to the code’s porting to the GPUs with the incorporation of OpenACC, a programming standard that makes it easier for users to adapt their applications to take advantage of GPU technology.
S3D was the only OLCF Early Science Project (a group of applications granted early access to Titan to test the new architecture) that used the OpenACC community standard programming environment. “In that sense,” said OLCF Director of Science Jack Wells, “it was an important experiment.” Part of Titan’s promise is performance portability, or the idea that improvements made to applications for the sake of running on Titan will translate to other systems. S3D’s strong acceleration is a testament to OpenACC, Wells said.
A flame reborn
A burning flame can manifest chemical properties on small scales from billionths of a meter up to thousandths of a meter, whereas the motion of an engine valve can exert effects at scales from hundredths of a meter down to millionths of a meter. This multiscale complexity is common across all combustion applications, hence the need for a higher Reynolds number, and therefore Titan.
As a flame burns, portions of it die out, or “extinguish.” While there is still fuel being released, there is no burning. However, it will eventually reignite, at least partially, as the turbulent mixing weakens downstream of an injector.
Specifically, the team wanted to know the dependence of reignition on the local mixing rate, or the rate of fuel and air mixing during the combustion process. “We found that oxygenated fuels such as DME generate considerably more stable intermediates such as formaldehyde, rendering the flame more robust to local extinction than conventional hydrocarbons such as methane,” Bhagatwala said.
Furthermore, the simultaneous imaging of formaldehyde and hydroxyl radical, a neutral form of the hydroxide ion, was evaluated to determine its effectiveness at measuring the peak heat release rate. The simulation data verified that this imaging method performed extremely well at predicting maximum heat release rate in the flame. This method was subsequently applied to the experimental data as a diagnostic to measure the extent of local extinction and reignition.
The simulations were a team effort to be sure. The University of Connecticut’s Tianfeng Lu, a combustion chemistry expert, was charged with taking the detailed mechanisms and deciding which species are important enough to be included in the simulations (even with computers as powerful as Titan, approximations still have to be made). And Jeffrey Sutton and Han Shen of OSU were tasked with the experiments and with detailing all of the observable properties possible.
Ultimately, more accurate combustion models, analyzed by DNS and experiments, will be used to optimize engine design, bringing us one step closer to more efficient combustion devices, from the engine in your car to the one propelling airplanes.
“These numerical results compliment the limited number of observables possible in experiments,” Sankaran said. In the future, he added, the team is gearing up to perform DNS at high pressures found in engines with realistic fuel surrogates, requiring up to 116 species representing a gasoline blend.
And for those simulations, Titan will no doubt come in handy.
Related Publications: A. Bhagatwala, Z. Luo, H. Shen, J. Sutton, T. Lu and J. H. Chen, “Numerical and experimental investigation of turbulent DME jet flames,” Proc. Combust. Inst. (2014), http://dx.doi.org/ 10.1016/j.proci.2014.05.147.
S. Pope, “A model for turbulent mixing based on shadow-position conditioning,” Phys Fluids 25, 110803 (2013); doi:10.1063/1.4818981.
J. M. Levesque, R. Sankaran, and R. Grout. 2012. “Hybridizing S3D into an Exascale Application Using OpenACC: An Approach for Moving to Multi-Petaflops and beyond.” In International Conference for High Performance Computing, Networking, Storage and Analysis (SC12), 1–11. doi:10.1109/SC.2012.69.
The Oak Ridge National Laboratory is supported by the Office of Science of the U.S. Department of Energy. The Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit science.energy.gov