Hydrogen fuel cells provide clean, renewable, reliable, low-emission electricity. A stable domestic energy source, they also offer flexibility and support other renewables like wind and solar.
Even with these advantages, significant obstacles remain to broad adoption, including the high cost and material constraints of platinum, the catalyst for the oxygen reduction and hydrogen oxidation reactions.
Searching for cheaper, more abundant catalyst alternatives, a team of materials scientists led by Corey Oses, an assistant professor of material science and engineering and a member of JHU’s Ralph O’Connor Sustainable Energy Institute, looked to high-entropy alloys, or HEAs. “We believed they were the best option because they’re made of many components that might be able to mimic or even exceed the emulator, outperforming a material like platinum,” Oses says.
To predict which HEA combinations would work best, the team created an AI framework. “Our goal was to quickly screen more compositions and generate rapid predictions. We wondered if we could teach an algorithm these properties and then integrate it into a machine learning framework,” says Oses. “And the answer is yes, we can.” Their study was published in Nano Futures.
“Our goal was to quickly screen more compositions and generate rapid predictions. We wondered if we could teach an algorithm these properties and then integrate it into a machine learning framework. And the answer is yes, we can.” — Corey Oses
While this project focuses on one common type of atomic arrangement—body-centered cubic structures—future work will explore other arrangements found in HEAs and close the loop between computation and experiment. Top AI predictions will be synthesized and tested, with results fed back to refine the model.
“That cycle could speed and economize the hunt for platinum alternatives,” Oses says. By combining computational depth with practical shortcuts, the approach aims to accelerate discovery of catalytic materials that could lower costs and help hydrogen fuel cells reach wider commercial use.
— CONNER ALLEN