Machine Learning Database to Guide Development of Low Flammability Polymer Matrix Composites

The NAVY seeks an advanced active learning-based machine learning (ML) database to support the development of polymer matrix resins and composites with low flammability. Although composites can save weight and reduce maintenance, their use below deck on ships is limited because polymer matrix resins can fuel fires. In particular, commonly used epoxy matrix resins are too flammable to meet Navy flammability requirements. Polymer resins with reduced flammability tend to leave more char when burned, are highly crosslinked and brittle, and require higher curing temperatures, making them more expensive than metal pressure vessels. Adding flame retardants to epoxy resins can also reduce their properties. To address these challenges, Global Engineering and Materials, Inc. and collaborators propose to develop a physics-informed ML framework and an active learning database to guide the design and generation of low-flammability polymer matrix composites. This ML framework leverages physics-based methods and simulations to augment the generally limited cone calorimeter data and trains the developed ML models to make accurate predictions of composite flammability metrics. The proposed toolkit features generative modeling capabilities to create hypothetical polymers and composites from learned material fingerprints, coupled with an optimization scheme to produce polymer and composite structures with the lowest flammability. Additionally, the framework allows for straightforward extension from polymers to composites, with and without flame retardants, by encoding additional composite and flame retardant information in the final ML model.