Title:A machine learning platform for development of low flammability polymers

Abstract:Flammability index (FI) and cone calorimetry outcomes, such as maximum heat release rate, time to ignition, total smoke release, and fire growth rate, are critical factors in evaluating the fire safety of polymers. However, predicting these properties is challenging due to the complexity of material behavior under heat exposure. In this work, we investigate the use of machine learning (ML) techniques to predict these flammability metrics. We generated synthetic polymers using Synthetic Data Vault to augment the experimental dataset. Our comprehensive ML investigation employed both our polymer descriptors and those generated by the RDkit library. Despite the challenges of limited experimental data, our models demonstrate the potential to accurately predict FI and cone calorimetry outcomes, which could be instrumental in designing safer polymers. Additionally, we developed POLYCOMPRED, a module integrated into the cloud-based MatVerse platform, providing an accessible, web-based interface for flammability prediction. This work provides not only the predictive modeling of polymer flammability but also an interactive analysis tool for the discovery and design of new materials with tailored fire-resistant properties.