(697b) Accelerated Design of Flame Retardant Polymeric Nanocomposites Via Machine Learning

Improving the flame retardancy of polymeric materials is an increasingly important strategy to limit the exposure of life safety to fire hazards, especially in this era when polymeric materials are widely used in every aspect of life, including housing, vehicles, commercial products, etc. However, the wide variety of flame retardant polymeric nanocomposite designs prevents rapid identification of the optimal composition for a given application. In this study, we applied machine learning methods, including the benchmark performance multiple linear regression (MLR), k-nearest neighbor (k-NN), support vector machine (SVM), random forest (RF), and gradient boosting regression (GBR), to predict the flame retardancy index (FRI) and related features of cone calorimetry for different types of flame retardant polymeric nanocomposites. In total, 800 experimental data points in literature from 2000 to 2020 were collected for modeling purposes and were split into 75% training and 25% test set in order to conduct cross-validation. The quality of machine learning predictions was evaluated based on the statistical values, such as the coefficient of determination (R²) and root mean square error (RMSE). Among these five machine learning methods, GBR gives the best prediction with R² = 0.95 and RMSE = 0.15. In addition, we studied how physical features of polymeric nanocomposites affected flame retardancy using the correlation matrix of each feature, which in turn was used to guide the design of polymeric nanocomposites for flame retardant application. Following the guidelines deduced from these models, a highly efficient flame retardant polymeric nanocomposite was designed and synthesized, of which the experimental results were compared with the machine learning predictions.