Unbiased Dataset for Methane Dry Reforming and Catalyst Design Guidelines Obtained by High-Throughput Experimentation and Machine Learning

Traditional solid catalyst design is a highly labor-intensive and post hoc process, involving repeated experimental trials and errors based on hypotheses derived from prior knowledge. Recently, integrating high-throughput experimentation (HTE) with machine learning (ML) aims to achieve a more systematic catalyst design without relying on specific knowledge or assumptions about the target catalysis. As a first step, this study constructs an unbiased HTE dataset for dry reforming of methane (DRM) at 500 °C on 256 γ-Al2O3-supported multi-element catalysts, prepared by randomly combining 17 elements selected from the periodic table without any preconceptions. The obtained data and selected catalysts are analyzed in various ways to gain insights into catalyst design and catalysis. It is found that the inclusion of Ni or platinum group elements does not necessarily lead to DRM activity; rather, careful combinations of elements are crucial. Specifically, the catalysts that exhibit the highest activities are not only based on Ni as the main active element but also frequently contain Li, Al, and Nb, which are hardly regarded as promoters in the literature. Data scientific analyses also reveal that the coexistence of specific elements such as Al, Nb, and Hf with Ni or Rh promotes catalysis. Studying the best-found Ni-based catalyst elucidates individual elements’ roles in improving activity and suppressing carbon deposition. In particular, the ternary combination of Al, Nb, and Hf reduces carbon deposition while enhancing activity. Overall, this study demonstrates the validity of unbiased exploration in providing a foundational dataset for ML and in discovering catalyst design guidelines.