Producing Hydrogen: Unlocking Green Hydrogen Production with AI and Mathematical Model

• Artificial Intelligence
• Batteries
• Carbon Neutrality
• Data-Driven Approaches
• Electrochemistry
• Hydrogen Materials
• Hydrogen Society
• Research Center for Energy and Environmental Materials (GREEN)

Green hydrogen, produced by splitting water using electricity from renewable energy sources, emits no CO₂ during production. As such, it is drawing attention as a key technology for realizing a sustainable energy cycle. A research team led by Ken Sakaushi is striving to enable the large-scale deployment of water electrolyzers by developing high-performance, low-environmental-impact electrocatalysts and accelerating the understanding of fundamental electrochemical phenomena through data-driven approaches.

From Six Years to One Month: Accelerating Catalyst Discovery With AI

To realize a hydrogen-based society, growing attention is being paid to a technology called “water electrolysis”, which splits water (H₂O) into hydrogen (H₂) and oxygen (O₂) by applying electricity. The principle here is simple: when two electrocatalysts (an anode and a cathode) are immersed in water and an electric current is applied, H₂ and O₂ are generated at the respective electrodes. Because the mechanism is so straightforward, the performance of the electrocatalysts has a direct impact on the efficiency of green hydrogen production.

Currently, the oxygen-evolving electrocatalysts used in water electrolyzers rely on expensive and scarce platinum-group elements such as iridium and ruthenium, posing a significant barrier to wider adoption in our society. As in response, attention has turned to multicomponent materials created by mixing multiple abundant elements. It is known that even abundant elements is able to exhibit dramatically different properties when they are combined.

However, the possible combinations of elements and their mixing ratios result in an enormous number of composition variations. To efficiently explore this vast space, Sakaushi’s research team is leveraging artificial intelligence (AI) to rapidly identify multicomponent materials with high catalytic performance. Sakaushi explains:

“Here is one example of our strategy. First, based on factors such as the abundance and availability of elemental resources in Japan, we select about 10 candidate elements for multicomponent materials—such as iron, copper, and zinc. From this pool, we randomly generate 10 different compositions, each consisting of five elements, and fabricate electrodes accordingly. These electrodes are then tested in water electrolysis experiments, and the resulting electrochemical property data is used to train an AI model. Based on the AI’s suggestions, new electrodes are fabricated, and the cycle of prediction and experimentation is repeated.”

“Using this approach, in 2023 we identified an electrocatalyst composed of manganese, iron, nickel, zinc, and silver. The AI employed was a relatively simple model based on Bayesian statistics, developed in consultation with Ryo Tamura, Team Leader of the Data-Driven Algorithms Team. Remarkably, even with a limited training dataset, the AI successfully suggested the optimal composition. What would normally take up to six years through conventional experimentation was accomplished in just one month (NIMS NOW Vol. 22, No. 1: Cover Story and Leaders of the Future).”