Automated Feature Engineering and Model Aggregation for Data-Driven Oxidative Coupling of Methane Catalyst Design

Area of Science:

• Catalysis
• Materials Science
• Chemical Engineering
• Machine Learning Applications

Background:

• Identifying active species and mechanisms in Oxidative Coupling of Methane (OCM) catalysis is crucial but challenging due to complex dependencies on catalyst properties and reaction conditions.
• In-situ characterization of catalysts during OCM operation is often infeasible, hindering mechanistic understanding and rational catalyst design.
• Machine Learning (ML) offers a promising approach to predict catalytic activity based on catalyst composition and operating parameters, aiding in the discovery of new materials.

Purpose of the Study:

• To develop and apply an ML-driven framework for discovering novel metal-support catalyst combinations with high activity for the Oxidative Coupling of Methane (OCM).
• To utilize engineered compositional features that encode both active metal and support information to improve regression model performance.
• To identify specific metal-support formulations that exhibit superior C2 yield in OCM reactions.

Main Methods:

• Engineered compositional features were created to represent both the active metal components and the support material of potential catalysts.
• Multiple regression models were aggregated using these engineered features to predict OCM catalytic activity.
• A systematic search within a large materials space was conducted to identify promising catalyst candidates.

Main Results:

• Three metal-support combinations, namely (Na, K, W)/CeO2, (Cs, Ba, W)/TiO2, and (Na, Cs, W)/SiO2, were identified as highly active for OCM.
• These identified catalysts demonstrated a C2 yield exceeding 20%, indicating significant performance.
• The study successfully demonstrated the efficacy of an automated framework in discovering active catalyst formulations.

Conclusions:

• The developed ML framework, incorporating engineered features, effectively accelerates the discovery of high-performance OCM catalysts.
• The identified catalyst formulations represent promising candidates for further investigation and optimization in methane conversion.
• This approach highlights the potential of automated feature generation and ML-based screening for exploring vast materials spaces in catalysis.