The findability of microkinetic parameters by heterogeneous chemical reaction neural networks (hCRNNs)

Finding microkinetic parameters for heterogeneously catalyzed processes with conventional methods is a challenging task. Recently, the use of artificial neural networks has been described as a promising and flexible tool for kinetic parameter estimation. In this work, an extension to the methodology of chemical reaction neural networks (CRNNs) to heterogeneously catalyzed reaction networks (hCRNNs) is proposed. The developed network architecture encapsulates physically interpretable layers for the Arrhenius expression, coverage dependency, and power-law terms encountered in a typical microkinetic model and accounts for possible reversibility of all elementary step reactions in the mechanism. Thus, it is fully interpretable and acts as a drop-in replacement for a conventional kinetic expression. The methodology is further examined on a prototypical heterogeneously catalyzed reaction mechanism under transient conditions and various operational and kinetic regimes. This work offers a framework for quantifying network errors and interpreting its predictions as well as a systematic overview assessing its ability to identify kinetic parameters. It is found that kinetic behavior is generally described very well by the network. Additionally, kinetic discovery is possible for the fastest reaction in the mechanism, if observed. A link between the results and the transient regime is established. With this, the design of suitable hCRNNs training strategies becomes possible. • hCRNNs are an extension of Chemical Reaction Neural Networks to heterogeneous catalysis. • Fully physically interpretable neural network architecture based on microkinetics. • hCRNNS discover kinetic behavior from differential transient kinetic data. • Kinetic parameters identified for the dominating reactions observed.