(108c) Methodologies for the Inverse Design of Polyurethane Foam Materials

Informed materials design remains a foremost challenge for scientists and engineers. Having an ability to ideate a composition or process resulting in products with targeted specifications would be highly valuable in the chemical industry.

Recent uses of machine learning (ML) have enabled the connection of design and process to resultant material properties, a process known as forward design. A high-accuracy forward model enables the screening of potential material candidates in silico before spending resources to produce a formulation in the lab. The reverse of this process, going from property to formulation, called inverse design, results in a one-to-many challenge and is hindered by a typically discontinuous property-design space.

Here, we present approaches to inverse design to produce novel formulations having properties predicted to satisfy user-supplied design requirements. These strategies are demonstrated for polyurethane flexible foam materials utilizing real-world component and formulation data and expert-informed constraints.