Increasingly methods using robotics and machine learning are applied to discovering new chemical transformations and multiple process parameter optimisation. The task of optimising chemical reactions is challenging, since many simultaneous objective functions such as yield, cost and environmental impact need to be considered. Self-optimisation provides a fully autonomous method for the optimisation of chemical reactions, by creating a feedback loop through the combination of in-line analysis and an optimisation algorithm. The multi-objective optimisation of expensive-to-evaluate functions represents most problems of interest in the chemistry community, yet very few advanced algorithms are available. Furthermore, the scalarisation of multiple objectives fails to identify the complete trade-off between competing performance criteria. We have demonstrated for the first time the use of multi-objective machine learning methods for the self-optimisation of chemical reactions. Integration of the recently developed Thomspson sampling efficient multi-objective (TSEMO) algorithm with our automated platform enabled the complete trade-off curve (Pareto front) between competing economic and environmental objectives to be identified. This approach has been applied to bi- and tri-objective optimisations of exemplar reactions, as well as a pharmaceutical process.

Machine learning meets continuous flow chemistry: Multi-objective self-optimisation

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.