Qiong Tan, De-Yi Wang
Flame retardancy is essential in epoxy resin (EP) applications. Phosphorus-containing flame retardants (P-FRs) have been recognized as one of the most promising halogen-free additives for improving the fire resistance of EP. However, rapid and low-cost screening of high-efficiency EP/P-FR composites and systematic assessment of their flame-retardant performance remain challenging. In this work, a data-driven strategy is developed for the prediction and comprehensive assessment of flame retardancy in EP/P-FR systems. Polymer structural descriptors, formulation, and condition-related features are jointly considered and integrated through feature fusion. Key flame retardancy evaluation metrics, including the UL-94 vertical burning rating and limiting oxygen index (LOI), are predicted using machine-learning models. The proposed models achieve a high accuracy of 0.81 for UL-94 classification and an R² of 0.79 (MAE=1.21%) for LOI prediction on the test set. Feature importance analysis quantifies the contributions of structural and formulation variables, revealing key factors governing flame-retardant performance. Furthermore, a non-compensatory evaluation framework is proposed for polymer flame retardancy assessment. The predicted UL-94 vertical burning rating and LOI values are incorporated as core evaluation inputs within a unified decision framework. Based on this framework, a six-state performance space is constructed, and four flame-retardant semantic levels, denoted as L-1 to L-4, representing excellent, good, moderate, and poor flame retardancy performance, respectively, with suffixes a and b indicating UL-94-prioritized and LOI-prioritized decision strategies. This work not only provides a practical strategy for the discovery and design of high-performance EP/P-FR composites, circumventing extensive trial-and-error, but also enables a more scientific and comprehensive non-compensatory evaluation of flame retardancy in such systems.