Advanced Computing in the Age of AI | Tuesday, March 19, 2024

HPC and Code Modernization May Deliver Car ‘Super Engine’ 

shutterstock_206781265

With 53 miles-per-gallon CAFE requirements bearing down on the auto industry for 2025, car makers are investigating any number of new engine designs – electric, hybrid and alternative fuel, among others – that will enable them to comply with the government mandate.

But in the end, the engine that may win is based on a 135-year-old concept and runs on gasoline. It’s called an “opposed-piston” engine and it could improve fuel efficiency by 50 percent. The key to converting the opposed-piston engine from a potentiality into mass production – and significantly reducing greenhouse-gas emissions – is forcing gasoline to do something it stubbornly resists: auto-ignite. Also called “gasoline compression,” it’s a complex process that could revolutionize internal combustion and make possible the “super engine” of the future for the light-duty transportation market.

Getting to a production-ready, opposed-piston engine that runs on gasoline compression is the goal of a $13 million project combining the efforts of Delphi Auto, Achates Power and the supercomputing might of Argonne National Laboratory. The research is being conducted under a three-year project funded by a $9 million award from Department of Energy's Advanced Research Projects Agency-Energy (ARPA-E) and an additional $4 million from the two companies.

"Conventional spark-ignited engines have improved so dramatically over the past few decades that there is little room to make big efficiency gains," said Steve Ciatti, mechanical engineer in Argonne’s Energy Systems Group who will be the lab’s experimental lead. "You need a game-changer to get into large double-digit efficiency gains, and we believe this engine is capable of doing that."

Argonne researchers have been using Fusion - an Intel-based cluster with 320 nodes and 2560 cores – and Surveyor – one rack of an IBM Blue Gene/P system with 1,024 nodes and 4,096 cores – to model and predict the complex fluid dynamics and fuel/air combustion in the cylinders to optimize the design of the engine and fuel injectors.

According to Sibendu Som, principal mechanical engineer at Argonne and computational lead for the project, a critical aspect of the effort is code modernization – updating Converge CFD software from Convergent Science to run on Fusion and Surveyor as well as, Mira, comprised of 738,000 processors and ranked No. 5 on the Top500 list of supercomputers.

Som told EnterpriseTech the code modernization work and gasoline compression research have been going on for five years. Converge CFD is a general-purpose CFD code for calculating the chemical reaction in fluid flows, including turbulence, spray injection, liquid drop dynamics and combustion of hydrocarbon fuels.

According to a white paper co-authored by Som, he and his team found that the original load balance algorithm performed poorly when the CFD software ran on more than 128 processors, which was required due to the large size of the compression ignition engine simulations.

“The big challenge is to take software not designed to run on supercomputers and make it run on supercomputers,” he said, adding that Converge CFD was originally written to run on no more than 16 processors. “If you’re trying to run 1,000 cases at one time, how do you handle the IO issues? How do you resolve the load balancing issues when you’re using so many cores?”

After researching several load balancing libraries, METIS was chosen to replace the original load balancing routines because it “is widely known for its efficiency in partitioning complex geometries and its capabilities to minimize the connectivity and to enforce contiguousness between partitions,” according to the white paper.

Converge CFD code modernization

Converge CFD code modernization

The code modernization work has paid off, according to Som. When Converge CFD is run on the Fusion and Surveyor clusters, the team can run between 20 and 25 cases at a time. With Mira, researchers are able to run 3,500 cases simultaneously, “changing the paradigm for engine simulation and design,” said Som, and enabling examination of high-throughput simulations of operating conditions, fuel injection conditions and piston design “that people haven’t been able to explore in the past.”

While the opposed-piston engine project is infused with high hopes, DoE officials note they don’t expect the auto industry to adopt the new engine – assuming it comes to fruition – until 2025 or 2030. This is because a new engine architecture would be disruptive to car makers’ manufacturing, car design and service maintenance processes, and the companies lay out their production strategies up to 10 years in advance. The auto makers’ plans for meeting the 2025 CAFE standards are already in place.

Many engines have been proposed that show benefits over conventional spark-ignited gasoline engines, but the efficiency gains have not been dramatic enough to convince auto manufacturers to retool in order to incorporate the new approaches. The notion behind Achates’ opposed-piston design – an engine concept dating back to 1882 – is to set two pistons moving in opposition in one cylinder. As the crowns of the pistons slide toward each other, they compress a mixture of air and gasoline to such extreme pressures that the mixture auto-ignites without the need for spark plugs. The design eliminates cylinder heads — which are a major cause of heat loss and inefficiency in conventional engines — and allows the engine to run with diesel-like efficiency and power combined with gasoline's emissions advantages.

“Proper fuel-air preparation will be a key factor for clean, efficient gasoline combustion in the Achates engine, requiring a systems approach to design of the engine and fuel injection system,” said John Kirwan, chief engineer, Advanced R&D for Delphi’s Powertrain Division.

“Argonne’s modeling work will provide us with detailed predictions of the in-cylinder behavior for a given engine and injection system design,” he said. “This information will be an important complement to experimental testing to enable the project team to work quickly and efficiently. We will be able to optimize the fuel injection system.”

EnterpriseAI