Machine learning-guided design of organic phosphorus-containing flame retardants to improve the limiting oxygen index of epoxy resins

The addition of organic phosphorus-containing flame retardants (OPFRs) has greatly improved the fire resistance of epoxy resins (EPs). Developing the relationship of the fire resistance with the structure of OPFRs and their addition amount will help discover high-performance EP composites, which was achieved in this work by ma-chine learning (ML). By combining descriptors encoded from OPFR molecules and the addition amount as fea-tures, an ML model with the limiting oxygen index (LOI) as the target was developed with a coefficient of determination (R2) of the ML model on the test set of 0.642. The trained ML model indicated that fire retardants containing conjugated systems with penta-substituted phosphorus containing a P--O bond and the nitrogen element can significantly increase the LOI of EPs, which led to the synthesis of a 9,10-dihydro-9-oxa-10-phospha-phenanthrene-10-oxide derivative (BDOPO) in this work. Furthermore, the accuracy of the ML model was validated through experiments. The predicted LOI values of the EP/BDOPO composites followed the same trend as the experimental values, with an average error of 5.1 %. The model can also illustrate the molecular structure required for synthesizing an OPFR and predict the amount of this OPFR to be added into EPs for enhanced LOI of the EPs.

Automated summary

By using machine learning, a high-performance EP composite with enhanced fire resistance was developed by incorporating OPFRs. The ML model identified fire retardants with specific molecular structures that significantly increased the LOI of EPs. Experimental validation confirmed the accuracy and reliability of the ML model.