Recent developments in the use of machine learning in catalysis: A broad perspective with applications in kinetics

Abstract

• A bibliometric and systematic review of catalysis and machine learning is provided. • Explains secondary data, new descriptors & feature engineering in ML applications. • Covers ML model trade-offs and factors influencing model selection in catalysis. • Discusses future directions for ML in optimizing catalytic systems. • Examines both theory-driven and data-driven approaches in catalysis. A thorough grasp of the underlying mechanisms of catalytic reactions is indispensable for furthering our understanding of chemical kinetics. However, traditional phenomenological models present certain difficulties, including the tendency to converge to local minima and a reliance on parameters that are difficult to measure, particularly in complex catalytic systems. These systems frequently comprise intricate feedstock compositions or catalyst structures that are challenging to anticipate through theory-driven approaches. This often results in the utilization of unrealistic models or the allocation of considerable computational resources. While traditional methods offer valuable insights, they are constrained by these challenges and the lack of robust uncertainty assessments. In view of these limitations, data-driven modeling, in particular through machine learning (ML), has emerged as a promising alternative in catalysis in the last five years. This review examines recent advancements in ML applications within the field of catalysis, encompassing a broad range of applications, including data generation, descriptor identification, and feature engineering. While the review takes a general perspective on ML in catalysis, particular attention is given to applications in chemical kinetics wherever relevant, recognizing the interconnection between reaction kinetics, catalyst design, reaction conditions, and reactor configurations. The discussion includes various ML models, including interpretable yet less flexible models and more complex black-box models, and considers their applications in catalysis. It also examines key factors in model selection, such as generalizability, computational efficiency, data quality, and interpretability. Finally, it outlines future directions for ML in catalysis, emphasizing how these technologies can further enhance the optimization, design, and improvement of catalytic systems.