Designing flame retardants for epoxy resins with generative AI for preventing battery thermal runaway propagation

• As-trained descriptive AI model can describe fire properties of fire retardants. • The AI model can predict how new FRs impact of tensile strength and T g of EP. • As-trained GAI model can generate new FR molecules with better performances. • Fire-retardant EP-coated wood can prevent inter-battery thermal propagation. Artificial Intelligence (AI) has demonstrated great potential for discovering new flame retardants (FRs) for polymers. While existing AI can describe flame retardancy of FRs, it has been unable either to generate new FR molecules with desired performances or predict their impacts on mechanical strength and glass transition temperature ( T g ) of polymer matrices. To fill this knowledge gap, we, herein, propose a generative artificial intelligence (GAI)-based de novo molecular design approach (GAI4FR) to generate new FR molecules for commercially important epoxy (EP) resins. Also, a descriptive AI model is trained using existing works to predict impacts of AI-generated FR molecules on flame retardancy, tensile strength ( σ t ), and T g of EP, enabling the identification of three FR molecules with better overall performances. The as-identified FR molecules are then synthesized, and their predicted performances are well- validated. We further demonstrate the application of as-prepared flame-retardant EP-coated wood sheets as heat shields for preventing thermal runaway of lithium-ion battery (LIB) packs. The coated wood shows equally desirable thermal protection for LIB packs to commercial counterparts. This work offers a groundbreaking GAI4FR framework for creating next-generation high-performance flame retardants for various flammable polymers and opens the door to discovering many other functional materials.