Decoding Non-Linear Electromechanical Coupling in Triboelectric Nanogenerators via Spatial Covariance Machine Learning

Battery-dependent data logging configurations in structural health monitoring (SHM) introduce substantial long-term maintenance overheads, toxic electronic waste footprints, and operational vulnerabilities within civil infrastructure monitoring. Transitioning green, self-powered Triboelectric Nanogenerators (TENGs) from highly controlled laboratory benches to outdoor civil environments represents a viable alternative; however, it is fundamentally bottlenecked by the slow, static, and manual nature of empirical electrical impedance matching sweeps. To resolve this challenge, this work establishes a rigorous, physics-informed data-driven framework capable of mapping and optimizing the complete continuous electrical power output landscapes of multi-material harvesters. Utilizing a structured dataset of eight TENG devices spanning distinct materials such as polytetrafluoroethylene (PTFE), Fluorinated Ethylene Propylene (FEP), Nylon, and Polyvinylidene difluoride (PVDF)-Hexafluoropropylene (HFP), raw closed-loop voltage signatures from resistor sweeps (1–200 MΩ) and capacitor charging tests (1–10 μF) under low-frequency scenarios (1–4 Hz) are analysed. The data are parsed under a non-overlapping slicing protocol to eliminate temporal window data leakage. A customized anisotropic Gaussian process regression (GPR) engine fused with a Matern 5/2 spatial covariance kernel is established, utilizing logarithmic compression to handle wide-ranging electrical resistance boundaries up to open-circuit limits. GroupKFold cross-validation splits based on independent file groups confirm that the physics-informed model achieves good predictive generalizability across unseen hardware setups, reaching an R² fitness score of 0.885 and an RMSE of 28.99 mW/m². This self-adaptive digital twin platform lays a potential foundational scenario for intelligent power optimization in regions like India and Southeast Asia by reducing over-reliance on lithium chemistries.