Leveraging explainable AI in the design of carbon composites with high thermal conductivity for thermal management

Achieving high thermal conductivity in polymer composites with low filler loading is crucial for thermal management. Constructing continuous and directional carbon networks has emerged as a promising strategy to enhance thermal conductivity. However, challenges remain, including the absence of accurate predictive models and explainable optimization strategies. In this study, we propose a design methodology leveraging explainable AI to optimize geometric patterns formed by carbon fillers, with a focus on sub-10 wt% carbon content. High-throughput finite element simulations are employed to generate datasets for prediction models, which incorporated variables including filler content, interfacial conductivity, connectivity, and gap distance. Symbolic regression yields predictive equations for thermal conductivity with high accuracy (R2 > 0.99), and SHAP analysis identifies the filler connectivity and content as key determinants. As a proof of concept, graphene-based composites with continuous and directional networks achieve thermal conductivities exceeding 33 W m− 1 K− 1, demonstrating that sub-10 wt% carbon fillers can reach high thermal conductivities, achieving high directional conductivity without requiring high global loading in composites. This work establishes an explainable AI-driven strategy for enhancing thermal conductivity in low-content carbon composites, expanding their applications in thermal management.