Select Page

Simulating Engines & The Need for HPC Speed

By Rob Mitchum // January 24, 2014

The secret weapon of the modern automobile engine is a tiny nozzle with an opening less than one millimeter across: the fuel injector. Since fuel injections systems replaced carburetors in the 80’s and 90’s, they have helped automakers build cars with higher fuel efficiency, lower emissions, and higher performance. Because of this important role, manufacturers are in fierce competition to find better designs or systems that can utilize a new generation of biofuels. But the complex physics and chemistry happening at the very tiny tip of a fuel nozzle are extremely difficult to study with standard experimental methods.

Engineers increasingly overcome these obstacles — and other challenges in combustion engine design — through the use of detailed simulations performed on powerful supercomputers and computing clusters. Researchers from the Computation Institute and Argonne National Laboratory are among the leaders in this field, collaborating with companies such as Chrysler, Cummins, and Caterpillar to discover and test new engine features.

“We’re taking advantage of the amazing computational power at Argonne to create more robust fuel spray and combustion models for predictive engine simulations,” said CI fellow Sibendu SomArgonne Principal Mechanical Engineer, in a 2013 article on the Argonne website. “These simulations provide unprecedented insights into the complex processes taking place inside engines.”

engine-sim.png

Som’s group in Argonne’s Transportation Technology R&D Center pulls together resources from across the laboratory’s campus to study engine design from computer simulation to practical use. High-resolution measurements of the spray from the fuel injector nozzle at the Advanced Photon Source, high-fidelity chemical kinetic models from the Chemical Science and Engineering group, modeling runs on supercomputers in the Argonne Leadership Computing Facility, and test runs at the Engine Research Facility all contribute to the research. Together, the experimental and computational methods deployed by the team approach the incredible complexity of the combustion engine from many angles.

Engine simulations must replicate both the physics of fuel flow and phase transitions and the chemistry of combustion. The fuel injector, for instance, atomizes liquid fuel into fuel vapor, changes that require complicated fuel spray, two-phase flow, and heat transfer modeling. Turbulence produced by the chaotic motion inside engine cylinders due to the fuel spray necessitates the inclusion of these high-fidelity computational fluid dynamics approaches to describe the spray of the fuel into the combustion chamber. The actual release of the fuel’s energy through combustion is a chemical process involving thousands of chemical species and tens of thousands of reactions.

Som and his team create a virtual, three-dimensional combustion chamber to simulate all of these interwoven factors. Auto industry engineers use a software called CONVERGE to conduct small-scale engine simulations on a relatively modest system of 50 computer processors. But, using a load-balancing algorithm called METIS, the Argonne team has dramatically scaled up the software to run on large, state-of-the-art computers containing over 1,000 processors (as reported in this 2013 paper).

engine-sim2.pngThe additional computational horsepower allows Argonne researchers to simulate the activity inside an engine at a much finer scale. The space simulated in a computational fluid dynamics models is divided into cells — a process called meshing — and the smaller the cell, the more accurate the model. Industrial models typically use millimeter resolution, but the scaled-up high resolution models created at Argonne can simulate at a micrometer scale — a thousand times more precise. With these methods, the Argonne team has run simulations using 50 million cells, believed to be the largest diesel engine simulation ever.

“The finer mesh sizes allows for simulations to be grid-convergent, which is a critical parameter that needs to be verified to ensure the accuracy of the simulations,” Som said. “It basically means that further changes in mesh resolution will not impact the simulations; hence, the mesh resolution does not become a tuning parameter.”

The more powerful models may also create a tool that engineers have long hoped for: models that are truly predictive, instead of just responding to experimental data fed into the system. These new models will allow researchers to stretch uses of the models beyond what can currently be done experimentally, testing out theoretical innovations such as low temperature combustion, Som said.

Indeed, while improving the scale at which physics and chemistry can be simulated and observed, the Argonne models will also be valuable testing ground for the future of alternative fuel. As the traditional fossil fuels gasoline and diesel give way to biofuels based on corn, soy, and other new sources, their performance and emissions in different engine designs can be simulated virtually before they are tested in the laboratory.

“With the ability to run more detailed simulations at a faster rate, manufacturers will have valuable new information that will ultimately help them design better engines more quickly and at a lower cost,” Som said.