Select Page

ACCOLADES for a Computational Auto Fleet

By Rob Mitchum // December 8, 2015

Everyone in the automotive industry wants to build a better engine, one that gets higher gas mileage, produces less emissions, and stills provides the power that drivers want. But actually building and testing different engines in the wide range of conditions it experiences is a costly process, and prohibitively complex. Dozens of variables, including RPM, fuel type and temperature, weather conditions, and city vs. highway driving, can affect an engine’s performance, and trying to test every single permutation of those variables quickly grows to millions of trials — more than any company would be able to conduct.

To solve this problem, CI Senior Fellow Shashi Aithal and CI Fellow Stefan Wild created a computational “fleet” of automobile engines, capable of simulating the fuel consumption and emission of thousands of engines simultaneously. The name the researchers chose for this framework, ACCOLADES (Advanced Concurrent Computing for Large-scale Dynamic Engine Simulations), proved prophetic, as last month they were awarded the HPC Innovation Excellence Award at SC15, the supercomputing conference recently held in Austin.

The award, given by International Data Corporation, specifically honors work that connects supercomputing research with industry and demonstrates the value of high-performance computing to funding agencies and the public. The use of supercomputers, such as Argonne’s world-class Mira machine, for designing cleaner and more efficient engines made ACCOLADES a prime candidate for the award.

“The power of large-scale computing enables ACCOLADES to open up new vistas in engine design, development ,and product planning,” Aithal told Argonne. For instance, engine designers can use ACCOLADES to optimize engine design for the next-generation of fuels and fuel additives while policy makers can evaluate their effect on emissions and their role in smart city planning.”

In the auto industry, the rigorous measurement of engine performance under different conditions is called dynanometer, or “dyno,” testing. Researchers run each trial for an entire “drive cycle” of 25-30 minutes, sampling data for several measures as often every second. Computer simulations can complement these experiments by simulating the results under different conditions, but even for a computer, it’s not a simple task. Testing just four different parameters for four variables over sixteen drive cycles each would require more than 4,000 simulations, each using complex, computationally-demanding physics models that can each take hours or days to complete.

Aithal and Wild developed ACCOLADES to large-scale parallel computing to speed up this process, producing simulations in less than real-time (~4-20 minutes per 25-minute drive cycle, according to their recent paper). The framework incorporates two components: pMODES, a fast engine simulator, and TADA, a data analysis toolbox applicable to both simulation outputs and real-world experimental data.

“ACCOLADES can be viewed as running hundreds or thousands of dynos simultaneously,”  “It greatly streamlines the workflow management of conducting large-scale parametric sweeps for generating performance maps of engines, thus accelerating the design and development process.”