(684d) Autonomous Homogeneous Catalysis Enabled By a Self-Driving Flow Reactor

Machine learning (ML) and flow chemistry are rapidly emerging as powerful tools in chemical reaction engineering to optimize complex processes and reduce costs for materials and molecular discovery and development compared to conventional batch experimentation. Recently, there has been a significant increase in the use of ML to model and optimize complex systems. By analyzing large experimental datasets generated from in-house experiments, ML algorithms can rapidly identify patterns in high-dimensional reaction spaces, predict reaction outcomes, and optimize operating conditions.

One of the primary applications of ML is in the development of predictive models. These models can be used to form a digital twin of the chemical reaction and predict process outcomes, allowing for conventional optimization techniques to be performed on the model, reducing the number of experiments needed to achieve a desired target, lowering the cost and time associated with experimental development.

Flow chemistry is another powerful tool for reaction miniaturization and optimization that can reduce the time and cost associated with the discovery and development of chemical processes. Automated flow reactors can provide superior control over the reaction conditions, minimize human error, and intensify heat and mass transfer rates. Furthermore, automation can be used to perform in-line monitoring and characterization of reaction performance in real-time, providing feedback to adjust the reaction conditions and optimize the reaction.

In this work, we present an autonomous flow chemistry lab, specifically designed for rapid exploration/exploitation of high temperature and pressure gas/liquid two-phase catalytic reactions. As a case study, we demonstrate the application of the developed autonomous flow chemistry lab towards accelerated development of rhodium-catalyzed hydroformylation of olefins. The automated reaction execution in flow, leveraged alongside the ML-based experimental selection algorithms, perform autonomous regioselectivity tuning and reaction optimization campaigns as well as developing a digital twin of the reaction for in silico studies.