(219d) Meso-Kinetic Transient Optimization Via Machine Learning | AIChE

(219d) Meso-Kinetic Transient Optimization Via Machine Learning


Kunz, M. - Presenter, Idaho National Laboratory
Batchu, R., Idaho National Laboratory
Wang, Y., Idaho National Laboratory
Yonge, A., Georgia Institute of Technology
Medford, A., Georgia Institute of Technology
Fang, Z., Idaho National Laboratory
Constales, D., University of Ghent
Yablonsky, G., Washington University in St. Louis
Fushimi, R., Idaho National Laboratory
Understanding the set of elementary steps and kinetics in a given reaction is extremely valuable for making informed decisions about next-generation catalyst materials. Traditionally, this information has been collected through extensive physics-based simulation to construct first-principles models, and/or fitting kinetic models with or without underlying mechanisms to experimental data. However, due to the rough granularity of commonly used steady-state analysis techniques, it is common that multiple first-principle models fit the experimental results equally well. As such, it is often up to the domain experts to infer the kinetics based on chemical intuition. We propose to help bridge the gap between simulation and experiment through the combination of highly dense transient kinetic measurement and chemically-inspired machine learning techniques. Specifically, we leverage non-convex variable selection techniques within a framework constrained by a chemical mechanism to obtain a quasi-equilibrium expression of the reaction mechanism purely based on experimental data. This methodology, referred to as MEso-kinetic TRansient Optimization (METRO), is validated against simulated responses where true kinetic coefficients are known, and is shown to accurately estimate these parameters without any a priori information. Moreover, the method is applied to experimental data from well explored reactions such as carbon monoxide oxidation and ammonia decomposition.The METRO methodology is proposed as a compliment to bottom-up microkinetic modeling, where the top-down fitting can provide insight into the reaction mechanism and extract intrinsic kinetic parameters of elementary processes directly from experimental transient kinetic data.