Autonomous kinetic model identification using optimal experimental design and retrospective data analysis: methane complete oxidation as a case study

Abstract

Automation and feedback optimization are combined in a smart laboratory platform for the purpose of identifying appropriate kinetic models online. In the platform, model-based design of experiments methods are employed in the feedback optimization loop to design optimal experiments that generate data needed for rapid validation of kinetic models. The online sequential decision-making in the platform, involving selection of the most appropriate kinetic model structure followed by the precise estimation of its parameters, is done by autonomously switching the respective objective functions to discriminate between competing models and to minimise the parametric uncertainty of an appropriate model. The platform is also equipped with data analysis methods to study the behaviour of models within their uncertainty limits. This means that the platform not only facilitates rapid validation of kinetic models, but also returns uncertainty-aware predictive models that are valuable tools for model-based decision systems. The platform is tested on a case study of kinetic model identification of complete oxidation of methane on a Pd/Al₂O₃ catalyst, employing a micro-packed bed reactor. A suitable kinetic model with precise estimation of its parameters was determined by performing a total of 20 automated experiments, completed in two days.