Green hydrogen: a prime candidate among eco-friendly alternatives to fossil fuels. When burnt, all it emits is steam; when produced from renewable energy sources, it causes far less greenhouse gas emissions than oil and gas production. These features make green hydrogen a promising clean means of storing energy, powering vehicles and feeding factories.
Catalysts are key to making green hydrogen as they facilitate the water-splitting reaction, which shatters water (H2O) molecules into hydrogen (H2) and oxygen (O2). Currently, most catalysts for the purpose use some of the Earth’s rarest metals, such as platinum, palladium and iridium. Their scarcity drives up costs and complicates supply chains, limiting green hydrogen’s wider adoption.
Yee-Fun Lim, a Principal Scientist at A*STAR’s Institute of Sustainability for Chemicals, Energy and Environment (ISCE2), points to layered double hydroxides (LDHs) as a more accessible option for water-splitting catalysts. Essentially made of clay, LDHs aren’t just structurally versatile—holding a variety of elements while keeping their unique layered and charged structure—but are also easily manufactured at scale from chemical solutions.
“Most research to discover new LDH compositions with higher catalytic activity still relies on lab-based trial and error,” said Lim. “We decided to test if machine learning (ML) optimisation methods, combined with high-throughput synthesis and standardised electrochemical testing, could more efficiently find optimal LDHs for green hydrogen electrocatalysis.”
A team comprised of Lim, ISCE2 Senior Scientist Albertus Denny Handoko and collaborators from ISCE2, A*STAR’s Institute of Materials Research and Engineering (IMRE) and the University of Toronto, Canada, set out to develop an ML-based method that could not only predict the best-performing LDH electrocatalysts within a certain range of raw materials, but also the optimal parameters required to craft them.
The team initially synthesised and tested a small sample of 53 LDH catalysts to produce a training dataset for their ML algorithms, which were built based on gradient boosting and neural network frameworks. The trained algorithms were then paired with Bayesian optimisation algorithms to suggest a new set of LDH material combinations. An automated cycle of synthesis, testing and optimisation was repeated until the catalysts hit a performance ceiling.
“While we could have used more complex molecular structural simulations, those would take hours to days to run even on supercomputers, due to the structural flexibility of LDHs,” said Lim. “However, our approach relies solely on understanding a catalyst’s compositional ratio, not its structure; as such, model training takes just about an hour on a personal computer.”
The team's final optimised catalyst was a novel composition based on cobalt and iron with added molybdate (CoFeMo LDH), which efficiently split water with minimal energy loss and prolonged stability. To their surprise, the catalyst lacked nickel—a core element in the first documented LDH with good water-splitting performance.
The team plans to scale up their CoFeMo LDH recipe for commercial applications, and also to explore how their automated synthesis and ML optimisation workflow can optimise other catalysts or industrial chemical processes, broadening the impact of their work.
The A*STAR-affiliated researchers contributing to this research are from the Institute of Sustainability for Chemicals, Energy and Environment (ISCE2) and the Institute of Materials Research and Engineering (IMRE).