The work was enabled by a recent award of 60 million core hours on Mira— the fifth-fastest supercomputer in the world—located at the Argonne Leadership Computing Facility (ALCF), a DOE Office of Science User Facility. (Earlier post.)

Presently, engineers exploring new engine designs can do a small number of simulations—perhaps 100—on cluster computers, which can take weeks to complete. Each change of a variable in an engine design, such as piston bowl or fuel injector configuration, requires a new simulation, and changing multiple variables increases the number of required simulations exponentially, quickly using up the computing power available to industry.

At the end of this lengthy simulation process, the engineers have data to select promising designs to go to a prototype hardware build, but only based on a limited number of simulations with a limited number of variables.

This new computing capability is a quantum leap from what anyone was doing before, and it holds the potential to unlock major breakthroughs in engine efficiency, as well as very substantial cost savings. In the past, doing 10,000 simulations was unthinkable. Now we can do that in a matter of days.

—Sibendu Som, principal investigator and principal mechanical engineer at Argonne’s Center for Transportation Research

Dr. Som is presenting a paper on Monday at the 2016 SAE High Efficiency IC Engine Symposium on the team’s work. The presentation, “Enabling High Throughput Calculations on Supercomputers for Efficient Engine Design”, will highlight two specific applications of high throughput computing:

1. Global sensitivity analysis on various uncertain inputs to an engine CFD simulation by running ~10K calculations to identify most sensitive parameters effecting LTC (low temperature combustion) engine performance; and

2. Capturing cyclic variability in engines by running ~2K calculations per operating condition for an SI engine application.

Other applications of this high throughput computing include engine design and strategy optimization.

Mira is often used to do a few huge simulations, but the VERIFI team wants it to do large numbers of smaller, engineering-type simulations simultaneously. To get there, they had to optimize the source code of CONVERGE—engine simulation software from Convergent Science, Inc.—and use an Argonne-developed, high-level programming language known as Swift to manage the massive workflows involved.

This massive simulation capacity has opened up a new capability for industrial partners seeking new advanced engine designs. VERIFI is already working with a major auto manufacturer, a leading company in energy and transportation and a global fuel supplier to put this unique capability to work.

We’re talking about bringing the power of supercomputing to engine design, which will accelerate deployment of new technologies. You’re letting the computer do the heavy lifting. When you can ask a family of questions and get answers in hours, rather than months, you change your approach to asking questions.

—Janardhan Kodavasal, Argonne mechanical engineer

The power of supercomputing will not only increase the quality and quantity of simulations while reducing the development costs, it will also broaden the number of vehicle systems that can be simulated at once. The smaller computer systems VERIFI has been working with have forced engineers to focus on finite aspects of engine design, such as fuel injectors or the fluid dynamics of combustion. With this new approach, they can broaden their inquiry to the entire powertrain.

The modern internal combustion engine is an extremely complex system, so it needs these types of computing resources to do simulations on the scale required to enable real breakthroughs. This is a very exciting area of computer science that could have real-world impacts through greater transportation efficiency and a reduction in harmful emissions.

—Kevin Harms, senior software developer at ALCF

Funding for the project is provided by DOE’s Office of Science and Office of Energy Efficiency & Renewable Energy Vehicle Technologies Office.

Also assisting in the research was Marta García, assistant computational scientist at ALCF.