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

The NAVY seeks an active machine learning database to guide the development of polymer matrix resins and composites that meet metrics for time to ignition, maximum heat release rate, and smoke density from a cone calorimeter. The current fire evaluation and certification of composite materials to be used for naval vessels are heavily reliant on tests such as cone calorimeter and TGA tests. Due to the strong thermo-chemo-mechanical interactions on material fire behavior, there is no reliable physics-driven relation between the chemical structure of a resin / additives and the fire performance of its associated composite materials. The presence of limited test data and the high cost in generation of material test data have prevented the development of a machine learning (ML) model to predict the flammability index parameters of a given resin for use in optimizing the design and selection of a resin to achieve the desired fire requirements. To address these challenges and bridge the existing technology gap, Global Engineering and Materials, Inc. (GEM) and its team members proposed to develop a multiscale machine learning (ML) database and a data-driven platform to predict the flammability index parameters for polymer matrix resins from their chemical structures. In this database framework, physics-based digital twins, data-driven models, and machine learning models are combined to generate a machine learning database that relates the chemical structure to the cone calorimeter data. Actual data from the literature will be augmented with synthetic data generated from a digital twin of a cone calorimeter with input based on the chemical structure and a data-driven model. A machine learning model will then be developed to accelerate the discovery of new polymer systems to more rapidly expand the database and allow for inverse design of polymers based on desired flammability index parameters. In Phase I, the research team will demonstrate the capability of the developed ML database using existing test data, generated synthetic data, and new polymer data to predict fire performance based on chemical structure and inverse design for identifying chemical structure based on desired fire performance.