Exploring high thermal conductivity polymers via interpretable machine learning with physical descriptors

This research proposes a high-throughput screening framework for identifying polymer chains with high thermal conductivity (TC) using interpretable machine learning and physical feature engineering. A hierarchical down-selection process optimizes 320 initial physical descriptors to 20, enabling ML models to achieve an R² prediction accuracy exceeding 0.80. Analysis reveals that high-TC polymers are mostly n-conjugated, facilitating strong chain stiffness and high group velocities. The framework connects individual chain characteristics to amorphous polymer behavior, showing that high-TC polymers exhibit strong intra-chain interactions and large radii of gyration in their amorphous states, enhancing thermal transport. This data-driven approach aids in the design of polymers with desirable properties.

Xiang Huang, Shengluo Ma, C. Y. Zhao, Hong Wang, Shenghong Ju