Machine-Learning Approach for Accelerated Design of Low-Flammability Polymer Matrix Composites

New and advanced polymer composites are needed to address the pressing issues of fire safety below the deck on ships. Presently available (natural and synthetic) materials pose significant risks of generating excessive heat, flames, smoke, and toxic/flammable gases, exacerbating the dangers of fire incidents. To mitigate these hazards, flame retardants are traditionally employed as additives, to impede and disrupt the combustion cycle. Despite considerable research and development endeavors, the fire resistance and toxicity standards outlined by the Navy remain unattainable. New composites that can surpass mandated safety thresholds hold the potential to offer substantial benefits, including weight reduction, cost savings, and diminished maintenance requirements, in addition to ensuring the required levels of safety. Here, we propose to utilize advanced machine-learning (ML) based Polymer Informatics methods to accelerate the design of low-flammability composites for below-the-deck Navy use. The flammability of materials, specifically composites, is quantified by a set of characteristics, typically measured using standardized instruments/tests, e.g., those defined by American Society for Testing and Materials (ASTM). To allow usage in naval vessels, composites must meet a set of strict criteria in terms of ASTM and non-ASTM test results such as time to ignition, heat release rates, and smoke density. These criteria make the development of such desirable materials with optimal combinations of base polymers, functional additives, flame retardants, and processing conditions, highly non-trivial. The main challenge is that the traditional materials design approach, largely driven by human experience and intuition, is painstakingly slow, while the available composite space is practically infinite. The Polymer Informatics capability we propose follows certain critical steps. First, we will curate a database of required ASTM test results of neat and polymer composites from the literature. Then, the data will be represented in a form that is digestible by the ML algorithms implemented in PolymRizeTM, our flagship Polymer Informatics platform. Finally, a set of ML models will be developed, tested, and deployed. Within this process, our ML pipeline will be tailored specifically to composites and flammability-related data. At the end of the project, a demonstration of the platform will be provided, including (1) the chemical space spanned by the database, (2) the capability of the ML models to predict the required flammability characteristics, (3) the capability to use Navy data to augment the initial database and build fresh models, and (4) an initial version of low-flammability composite design tools. The foundation laid in Phase I is expected to position us well to address critical Navy needs in Phase II, should we get to that stage, in terms of designing, testing, and validating low-flammability materials, especially suited for below-the-deck deployment.