Partial discharge monitoring in twisted pair using non-intrusive sensors: Numerical analysis

This paper deals with partial discharge (PD) measurements in twisted pair fed, in parallel, by a PWM inverter driving an electric motor. PD detection is carried out while the motor is running using a non-intrusive antenna sensor. Data are randomly acquired while the powertrain is running. Then, two alternative numerical signal processing methods based on energy spectral density are proposed to classify signals with or without PDs. The first method uses spectral features chosen by human experts. The second method automatically derives optimum features by using linear discriminant analysis.