Physics-Informed Neural Networks for Insulation Power Loss Estimation of Aircraft Wiring Under PWM Voltage

This paper presents a Physics-Informed Neural Network (PINN) methodology for insulation power loss estimation in aircraft wiring under Pulse-Width Modulated (PWM) voltage generated by Wide Bandgap (WBG) semiconductor devices. The higher switching frequencies and shorter rise times of WBG devices create challenges in modeling insulation power loss, especially with the presence of partial discharge (PD) activities. A PINN framework is proposed, which integrates physics-based insulation power loss principles with neural network capabilities to handle complex PWM waveform characteristics. Experimental validation demonstrates the model's accuracy across various PWM voltage conditions, with a training root-mean-square error (RMSE) of 0.1848 W, test RMSE of 0.2178 W, and 5-fold crossvalidation metrics of RMSE at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.2868 \pm 0.0602 ~\mathrm{W}$</tex> and R-squared at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.9930 \pm 0.0028$</tex>. The methodology establishes quantitative relationships between WBG-driven PWM characteristics and aviation wire insulation power loss, contributing to insulation lifetime prediction of aerospace power systems.