Graphene has emerged as a promising support material for Cu-Zn catalysts in CO2 hydrogenation to methanol due to its high surface area and potential for functionalization with heteroatoms like nitrogen and oxygen, with nitrogen believed to contribute to the reaction. In this study, we combined machine learning and data analysis with experimental work to investigate this effect. Machine learning (using a decision tree model) identified copper particle size, average pore diameter, reduction time, surface area, and metal loading content as the most impactful features for catalyst design, while nitrogen doping showed negligible influence on methanol space-time yield. However, experimental results indicated that nitrogen doping on graphene support improved the space-time yield by up to four times compared to pristine graphene. This improvement is attributed to nitrogen's role in lowering the catalyst's reduction temperature, enhancing its quality under identical reduction conditions, though nitrogen itself does not directly affect methanol formation. Moreover, machine learning provided insights into the critical features and optimal conditions for catalyst design, demonstrating significant resource savings in the lab. This work exemplifies the integration of machine learning and experimentation to optimize catalyst synthesis and performance evaluation, providing valuable guidance for future catalyst design.
Read full abstract