Automated Generation of Mechanistic Models for Chemical Process Digital Twins using Reinforcement Learning – Part I: Conceptual Framework and Equation Generation

The derivation of versatile and robust mechanistic models from experimental data is regarded as a key challenge in engineering and the natural sciences. This is particularly true in the field of chemical reaction engineering, where the development and maintenance of digital twins by reactor manufacturers and operators is increasingly pursued. These digital twins rely on frequent model updates and necessitate the automation of the modelling process. In this work, an automated workflow is proposed through which accurate mechanistic reactor models are generated from experimental concentration data obtained from a given reactor. At the core of this workflow, an interpretable reactor model is assembled by a reinforcement learning agent through iterative simplification of general differential balance equations, and the resulting candidate models are fitted to the experimental data. The performance of the workflow is demonstrated through two case studies. In an in silico case study, the underlying model of a synthetic dataset is correctly reconstructed, and robustness against noise in the input data as well as favourable scaling properties are observed. The model derivation process is significantly accelerated by the agent in comparison to an exhaustive enumerative search. In a second, experimental case study, a Taylor–Couette prototype reactor is employed. A liquid-phase esterification reaction between (2-bromophenyl)methanol and acetic anhydride is used as the test system. Based on the experimental data, meaningful mechanistic models are derived by the workflow, with the most accurate model yielding a normalized root mean squared error of 2.4%. Future efforts are expected to focus on the integration of automated experiments into the workflow and the extension of its application to process units beyond chemical reactors.

• Heyer, M
• Zhang, J
• Sugisawa, N
• Laub, J
• Lapkin, A

• AVT Process Systems Engineering, RWTH Aachen University, 52074 Aachen, Germany

• Request a Brochure
• Request more Info
• Request a Quote