For materials scientists, optimising chemical catalysts is like solving a Rubik’s Cube: each twist and turn represents an adjustment in chemical composition or process parameters, with only a precise combination of elements and conditions yielding the optimal catalyst.
“The traditional approach to materials discovery uses ‘Edisonian’ trial-and-error, with one-variable-at-a-time manual experimentation,” explained Yee-Fun Lim, a Principal Scientist at A*STAR’s Institute of Sustainability for Chemicals, Energy and Environment (ISCE2). This method not only demands significant time and effort but also stifles innovation by its sheer inefficiency.
In the energy sector, for example, there's a critical race to find the best catalysts for generating clean hydrogen—a potential game-changer for global decarbonisation efforts. Hydrogen is seen as a clean alternative to fossil fuels, but the metals typically used to catalyse its production are rare and expensive. The demand for these metals far exceeds their current production rates, presenting a substantial barrier to scaling up hydrogen technologies.
Lim and IMRE colleagues have turned their focus to alternative catalysts that are both cost-effective and abundantly available as potential solutions. Enter copper antimony sulfides (Cu-Sb-S), a group of materials that have emerged as strong candidates with their excellent light absorption, stability in water and effective electrical properties for facilitating hydrogen generation.
Recognising the need for a more streamlined approach to optimise these promising materials, Lim and co-corresponding author, Kedar Hippalgaonkar, worked with the team to pioneer a revolutionary automated workflow that drastically accelerates the optimisation process.
Their new system employs an advanced robotic platform capable of rapidly assembling and assessing various configurations of the Cu-Sb-S catalyst. By quickly sifting through numerous possibilities, the robot identified the most effective catalyst setups.
Complementing this, a specially designed machine learning (ML) algorithm analysed performance data to predict the optimal catalyst configuration. With three targeted optimisation goals, this dynamic robot-algorithm duo pinned down the ideal catalyst parameters in less than 20 iterations—transforming months of manual labour into mere days of automated precision.
“We hope our findings can encourage more researchers in the field to embrace high-throughput experimentation, ML techniques and multi-objective optimisation as tools to expedite research progress,” commented Lim.
The team is now poised to expand this method to other materials that traditionally require extensive manual testing, while also building a comprehensive database of material properties. This resource can help train future ML models, ensuring a faster, more fruitful journey from concept to creation across various scientific fields.
The A*STAR-affiliated researchers contributing to this research are from the Institute of Materials Research and Engineering (IMRE) and the Institute of Sustainability for Chemicals, Energy and Environment (ISCE2).
