A Hybrid Physics-Based and Data-Driven Approach for Monitoring of Inverter-Fed Machine Stator Insulation Degradations Using Switching Oscillations

Inverter-fed machines are widely used in many important industrial applications. However, the machine stator groundwall and/or turn insulation are prone to failure suffering the inverter's high dv/dt. It is necessary to identify them as early as possible but challenging due to their coupled and weak symptoms. In this article, a hybrid physics-based and data-driven approach is proposed to monitor insulation degradations. First, the physical mechanism analysis is carried out to obtain high-quality data sensitive to stator insulation conditions, i.e., the high-frequency common-mode switching oscillations. Then, the continuous wavelet transform is used to extract the time-frequency features of switching oscillations. Finally, an improved convolutional neural network is designed for the groundwall and/or turn insulation faults diagnosis. The experimental results on a 3 kW permanent magnet synchronous motor drive system demonstrate the effectiveness of the proposed method for multiple faults diagnosis with excellent sensitivity, accuracy, and robustness.