Physics-Informed Neural Network with Adaptive Loss Weighting for Parameterized Surface Potential Prediction on Polymeric Dielectric Interfaces

Abstract Surface charge accumulation on polymeric dielectrics is widely encountered in electronic packaging, triboelectric energy harvesting, and aerospace insulation, with direct consequences for device performance and insulation reliability. Surface potential provides an accessible indicator of surface charge accumulation, but its rapid decay makes it difficult for sensors to capture the complete spatial distribution. Here, we propose a physics-informed neural network (PINN) integrating adaptive loss weighting for the prediction of surface potential distributions on polymeric dielectric interfaces. Specifically, a corona discharge setup with a needle–plate electrode configuration and an automated scanning system equipped with an electrostatic microelectromechanical systems (MEMS) sensor was developed to capture the surface potential distributions under different voltages. A numerical model consistent with the experimental setup was then developed to supplement the spatial field information required for model training. Using sparse spatial data at the target conditions, the model predicted the complete surface potential at different voltages, with relative L2 errors (RL2Es) from 0.03% to 0.15%. The predicted and measured potentials showed a high degree of consistency, with a coefficient of determination of R2 = 0.99991. This study demonstrates the capability of the proposed framework for rapid prediction of surface potential, with potential applications in intelligent condition assessment of polymeric dielectric materials.