Physics-guided machine learning for co-design of charge-pumped triboelectric nanogenerators at critical air-breakdown

Operating charge-pumped triboelectric nanogenerator (TENG) at the critical air-breakdown state is essential for maximizing power output, yet it intensifies the fundamental trade-off with durability, creating a complex co-design challenge. Herein, this study first systematically investigates electromechanical mechanism in charge-pumped rotary freestanding TENG (RF-TENG) at the critical air-breakdown state. Experiments reveal that the dynamic evolution of the transfer film governs the competition between the output charge density and the wear loss, enabling establishment of the quantitative relationships among wear loss, transfer film coverage ratio and output charge density. Then, leveraging this insight, an electromechanical model is established and further a physics-guided hybrid surrogate model is constructed through integration of conditioned-modulation Gaussian process regression (CMGPR) and artificial neural network (ANN). The CMGPR-ANN framework provides efficient performance evaluation, where CMGPR enables stage-aware wear prediction while ANN captures nonlinear electrostatic induction. Ultimately, by combining this hybrid model with multi-objective optimization, the Pareto frontier between electrical performance and durability is mapped, extracting quantitative design principles for key parameters (e.g., the gap ratio) and delivering optimal designs for high-performance charge-pumped RF-TENG. This work establishes a co-design paradigm bridging physical mechanisms with computational optimization for advanced TENG development.