AI streamlines search for catalysts to clear hydrogen production hurdles

To increase energy efficiency and reduce the carbon footprint of hydrogen fuel production, Fanglin Che, associate professor in the Department of Chemical Engineering at Worcester Polytechnic Institute, is leveraging the power and potential of machine learning and computational modeling. The multi-university team she leads has completed a study that was just published in Nature Chemical Engineering. The study utilized artificial intelligence to identify catalysts with the potential to facilitate cleaner and more efficient hydrogen production.

In the paper, Che and the team present a new strategy to overcome two challenges:

Production hurdles that prevent greater adoption of hydrogen, a fuel that does not emit carbon dioxide
The length of time it takes to identify materials that are optimal catalysts for cleaner hydrogen production

Efforts to improve environmental sustainability and increase the availability of clean energy have long been focused on hydrogen. However, hydrogen is often produced using fossil fuels, which generate carbon dioxide.

An alternative method to produce hydrogen is to use a catalyst to break down carbon-free ammonia into its elements, which include hydrogen. However, this approach as currently designed requires very high temperatures, which are often achieved by using a lot of energy produced by fossil fuels, as well as ruthenium, an expensive rare metal that is used as a catalyst.

Che’s team proposes to reduce the carbon footprint of hydrogen production by decomposing ammonia using plasma technology, which can be done at lower temperatures than traditional chemical reactions. The researchers also propose using more commonly found and affordable metal alloys, such as iron-copper or nickel-molybdenum, as catalysts. Their analysis found this method would use less energy and perform just as well as current approaches to hydrogen production.

Identifying the catalysts

With more than 3,300 bimetallic alloys to consider as possible catalysts, testing each in a laboratory using traditional experiments would take a lengthy trial-and-error period. By leveraging computer models and artificial intelligence, Che’s team developed interpretable machine learning algorithms to identify Earth-abundant metal alloys that outperform ruthenium catalysts in plasma-assisted ammonia decomposition.

This combination of simulations and machine learning streamlined the process by quickly eliminating unsuitable materials and identified six candidates from abundant and easily sourced noncritical minerals. Laboratory tests validated the anticipated performance of the metal alloys and ultimately the researchers selected four alloys as the best catalysts.

Potential applications

Che’s team believes this new approach to producing hydrogen has the potential to be more affordable and cleaner than current methods. Additionally, because ammonia is easy to store and transport, this process could enable on-site hydrogen production on ships, allowing for maritime vessels to be powered by hydrogen fuel cells.

Che’s MAC (Modeling and AI in Catalysis) Lab at WPI combined multi-scale simulations with interpretable machine learning to develop predictions.

“Being published in Nature Chemical Engineering is a milestone for our lab,” says Che. “We are making great progress using computational and AI techniques to make chemical processes more energy efficient and environmentally friendly.”

Researchers at Dalian University of Technology in China conducted laboratory-based validation experiments. Researchers at Northeastern University conducted economic and environmental analysis that suggests plasma technology reduces costs and carbon emissions in hydrogen production when implemented in small, modular reactors.