Exploring surface reaction mechanism using a surface reaction neural network framework

Abstract

This paper presents a surface reaction neural network (SRNN) framework designed to effectively explore surface reaction mechanisms. This framework retains complete physical interpretability of the chemical reaction neural network (CRNN) while incorporating the physical constraints of mass conservation and surface coverage correction. As a result, it can be applied more broadly, and the findings can be interpreted with greater clarity. The transient reaction data of various species are analyzed using the SRNN to reconstruct the reaction kinetic parameters. By investigating three types of surface reaction mechanisms—standard Arrhenius form, surface coverage corrections, and surface sticking coefficient corrections—it is demonstrated that the proposed framework can independently analyze and construct surface reaction kinetics mechanisms from datasets of reaction information. Furthermore, it exhibits greater robustness compared to the chemical reaction neural network, achieving an accuracy error of less than 2% and proving suitable for exploring both surface and non-surface reaction mechanisms.