Generative Digital Design of Polymeric Coatings with Advanced Properties

Abstract A materials design pipeline consisting of two machine learning models was used to downselect and validate novel corrosion-resistant polymer structures for use in a high-temperature compatible protective coating. The first model is a generative model that creates polymers as candidates for coating applications. The second model is a graph neural network that predicts a variety of properties for the candidate polymers. Combined, these models create a database of over one million polymers with a variety of predicted properties; these polymers are ranked for their ability to serve as a protective coating and cross-referenced with existing literature and patents to validate the conclusions of the AI model. This resulted in producing one commercially available reagent, as well as two brand new structures with straightforward synthetic pathways, demonstrating that the generative AI has potential in the protective coating field as more extreme performance is required.