Machine learning-enhanced optimal catalyst selection for water-gas shift reaction

• Published in: Digital Chemical Engineering Vol. 12; p. 100165
• Main Authors: Golder, Rahul, Pal, Shraman, C., Sathish Kumar, Ray, Koustuv
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-09-2024; Elsevier
• Subjects: Bayesian optimization; Catalysis; Machine learning; Screening and prediction; Water-gas shift reaction; Catalysis; Bayesian optimization; Water-gas shift reaction; Machine learning; Screening and prediction
• ISSN: 2772-5081; 2772-5081
• DOI: 10.1016/j.dche.2024.100165

The water-gas shift (WGS) reaction is pivotal in industries aiming to convert carbon monoxide, a byproduct of steam reforming of methane and other hydrocarbons, into carbon dioxide and hydrogen. Selecting an effective catalyst for this transformation poses a substantial challenge, as it requires a delicate balance between conversion, stability, and cost. We combine machine learning-driven prediction models with Bayesian optimization to explore and identify novel catalyst compositions. The proposed method efficiently explores the catalysis composition space for a predefined set of active metals, supports, and promoters to identify the most promising catalyst formulations. We assign weights to different performance metrics of catalysts, enabling tailored optimization according to specific industry needs. Our screening system streamlines catalyst discovery and facilitates the screening and selection of catalysts that balance conversion performance, stability, and cost-effectiveness. This approach holds significant promise for advancement in heterogeneous catalysis to meet the growing demands of efficient industrial processes.
[Display omitted]
•Catalyst performance metrics like activity, stability modelled using LightGBM.•Physical interpretation of predictive results explained using SHAP analysis.•Activity. stability and cost parameters accounted with novel catalyst coefficient.•Catalyst screening performed using Bayesian Optimization over predictive ML models.•A flexible multimodal screening code provided for users to screen catalysts based on needs.