Optimal experiments for hybrid modeling of methanol synthesis kinetics

Abstract

The transition of the chemical industry towards the utilization of feedstocks based on renewable energies results in a more dynamic process behavior. Advanced mathematical methods are a key factor to handle this complexity. In this contribution, methanol synthesis from hydrogen, carbon dioxide and carbon monoxide is investigated as promising power-2-X technology. Optimal experimental design is used to recalibrate an existing mechanistic kinetic model. Subsequently, the most uncertain sub-model, namely the reversible catalyst dynamics, is partially replaced by neural networks. Several architectures were evaluated, and optimal experimental design was applied to enhance the performance of a chosen architecture. All experiments were realized in an experimental set-up able to acquire time-resolved data. A commercial CuO/ZnO/Al₂O₃ catalyst was used in a well-mixed Berty type reactor. The combination of optimal experimental design with hybrid modeling led to an improved quality of the kinetic model needed for process control and optimization.