Machine Learning-Guided Disulfide Modulation of Polyimides for High-Performance and Self-Healable Dielectric Energy Storage.

Conventional PI films exhibit excellent thermal stability; however, their weak self-healing under electrical and mechanical stress limits operational reliability. Herein, a machine-learning-guided screening strategy is employed to predict the Eb of sulfur-modulated PI systems, enabling the rapid identification of optimal disulfide incorporation. As a result, a series of PI films with varying disulfide contents (0-30 wt.%) are fabricated. The results reveal a non-monotonic dependence of Eb on disulfide content, with an optimal composition (PI-15) achieving a high Eb of 700 kV mm-1 and a Wdis of 10.02 J cm-3 at RT, together with 670 kV mm-1 and 6.64 J cm-3 at 150°C. Notably, after self-healing, PI-15 retains high performance, with Eb recovering to 675 and 640 kV mm-1 and Wdis to 8.59 and 6.02 J cm-3 at RT and 150°C, respectively. In addition, the films exhibit excellent operational stability under temperature, frequency, and fatigue cycling at 500 kV mm-1. Mechanistically, disulfide incorporation enables a synergistic coupling of strengthened intermolecular interactions and dynamic bond exchange, suppressing charge transport and local field concentration while facilitating structural rearrangement. This work demonstrates new working mechanisms for designing high-performance PI, providing strong potential for high-temperature capacitive energy storage under harsh operating conditions.