Machine learning guided development of high-performance nano-structured nickel electrodes for alkaline water electrolysis

Abstract

Utilizing a human in the loop Bayesian optimisation paradigm based on Gaussian process regression, we optimized an Ni electrodeposition method to synthesize nano-structured, high-performance hydrogen evolution reaction electrodes. Via exploration-exploitation stages, the synthesis process variables current density, temperature, ligand concentration and deposition time were optimized influencing the deposition layer morphology and, consequently, hydrogen evolution reaction activity. The resulting structures range from micrometre-sized, star-shaped features to nano-sized sandpaper-like structures with very high specific surface areas and good hydrogen evolution reaction activity. Using the overpotential at 10 mA cm−2 as the figure of merit, hydrogen evolution reaction overpotentials as low as -117 mV were reached, approaching the best known technical high surface area electrodes (e.g. Raney Ni). This is achieved with considerably fewer experiments than what would have been necessary with a linear grid search, as the machine learning model could capture the unintuitive interdependencies of the synthesis variables.