Analysis of electric field in coaxial cylindrical structures based on physics-informed neural network modeling

This paper explores the application of a Physics-Informed Neural Network (PINN) for modeling the electric field and potential distribution in a coaxial cylindrical structure under corona discharge condition. The proposed model couples the Poisson's Equation with the current continuity equation, enabling the simultaneous prediction of the potential and space charge density. An area-compensated sampling scheme and a regional weighted loss strategy are introduced to improve accuracy near boundary. To analyze the ability of the PINN model to handle the second type of boundary condition, the computational domain is reduced to a quarter-domain exploiting geometric symmetry, which significantly enhances computational efficiency without compromising accuracy. Comparisons between the PINN predictions and analytical solutions for both electric field intensity and potential show excellent agreement, with relative errors below 2% in most regions. This approach offers a promising and efficient alternative for modeling complex discharge phenomena.