Abstract:In order to solve the problem that the traditional fault diagnosis method of variable frequency motor bearing is susceptible to low frequency noise and fundamental frequency current interference，a new method utilizing high-frequency oscillations from pulse width modulation （PWM）switches is proposed for diagnosing bearing faults ininverter-fed machine. Firstly，the mechanisms of bearing fault diagnosis in inverter-fed machine and the generation pathways of bearing currents are analyzed. Then，due to the sensitivity of PWM switch dv/dt bearing current to bearing fault states，the measured PWM switch dv/dt bearing current，which is a component of the ground current，is used as a diagnostic variable and data source. Then，the measurement of PWM switch ground current does not require invasive sensors compared to dv/dt bearing current. Finally，the lightweight one-dimensional convolutional neural network is designed for motor bearing fault diagnosis，serving the purposes of feature extraction and fault classification. Experimental results demonstrate an accuracy rate of 96.63%，confirming the effectiveness of this method.