Enabling Catalyst Discovery through Machine Learning and High-Throughput Experimentation

Abstract

Machine learning is an avenue to unravel multidimensional relationships present in catalytic systems. We describe a novel framework that incorporates machine learning algorithms with experimental high-throughput catalytic data and elemental properties to discover new materials. The framework uses a small experimental dataset coupled with chemically descriptive features to predict future catalyst performance and guide synthesis. This led to the discovery of several novel catalyst compositions for ammonia decomposition, which were experimentally validated against “state-of-the-art” ammonia decomposition catalysts and were found to have exceptional low-temperature performance at substantially lower weight loadings of Ru.