Harnessing Data and Control with AI/ML-driven Polymerization and Copolymerization

Abstract

Creating and curating new data to augment heuristics is a forthcoming approach to materials science in the future. Highly improved properties are advantageous even with “commodity polymers” that do not need to undergo new synthesis, high-temperature processes, or extensive reformulation. With artificial intelligence and machine learning (AI/ML), optimizing synthesis and manufacturing methods will enable higher throughput and innovative directed experiments. Simulation and modeling to create digital twins with statistical and logic-derived design, such as the design of experiments (DOE), will be superior to trial-and-error approaches when working with polymer materials. This paper describes and demonstrates protocols for understanding hierarchical approaches in optimizing the polymerization and copolymerization process via AI/ML to target specific properties. The key is self-driving continuous flow chemistry reactors with sensors (instruments) and real-time ML with an online monitoring set-up that allows a feedback loop mechanism. We provide initial results using ML refinement of the classical Mayo-Lewis equation (MLE), time-series data, and an autonomous flow reactor system build-up as a future data-generating station. More importantly, it lays the ground for precision control of the copolymerization process. In the future, it should be possible to undertake collaborative human-AI-guided protocols for the autonomous fabrication of new polymers guided by literature and available data sources targeting new properties.

---