CPOS SEMINAR: "Quantifying Experimental Uncertainty in Catalysis for Machine Learning Development"
Speaker: Selin Bac, Postdoctoral Researcher, Christopher Group, Department of Chemical Engineering, UC Santa Barbara
Heterogeneous catalysis is fundamental to a wide range of industrial processes, yet predicting catalyst behavior, particularly under long-term reaction conditions, remains a significant challenge. A primary obstacle is catalyst deactivation, influenced by operating conditions and the catalyst's physicochemical properties. Machine learning (ML) models offer the potential to address this challenge by leveraging experimental data to predict catalytic performance and deactivation over time. However, the effectiveness of such models hinges on the quality and reproducibility of the underlying data, issues that are particularly difficult to address in heterogeneous catalysis due to the scarcity of interlaboratory reproducibility studies.
In this talk, I will present the results of a round-robin study conducted across four independent laboratories to assess the reproducibility of Rh/TiO2 catalysts in CO2 hydrogenation. This reaction yields CO and CH4 at distinct active sites: CO formation is primarily associated with isolated Rh single atoms via the reverse water-gas shift reaction, while CH4 formation predominantly occurs on Rh nanoparticles via the exothermic methanation reaction. The relative abundance of these sites, determined by Rh loading and synthesis method, dictates the product selectivity. By systematically varying reaction temperature, Rh loading, and synthesis method, we quantified variability in performance metrics across different experimental setups. Despite following standard protocols, substantial interlaboratory variability in CH4 production was observed, which was sensitive to reactor design and heat management. In contrast, CO production rates were more consistent. This demonstrates that not all catalytic metrics exhibit the same degree of reproducibility, making some more suitable for ML model development than others. To ensure the reliability of such models, rigorous standardization and uncertainty quantification in catalytic experimentation are essential.
