A Two-Step Method for PMSG Bearing Fault Recognition under Varying Speed Condition

Permanent magnet synchronous generators (PMSGs) are widely used in wind turbines to convert the mechanical energy to electrical energy. Bearing is a vulnerable component in a PMSG and its condition monitoring and fault diagnosis are important to guarantee the safety operation. As the PMSGs always rotate at varying speed conditions, the bearing signals are complicated and how to accurately recognize the bearing conditions is a challenge. In this study, a two-step method is proposed to recognize the bearing fault types in varying speed conditions. In the first step, the current and vibration signals are synchronously sampled. The amplitude fluctuation of the current signal and the rotating angle of the rotor are estimated from the current signal by using the peak detection technology and zero-phase filter, respectively. In the second step, the amplitude of the vibration signal is compensated according to the current amplitude fluctuation, and then the vibration signal is resampled according to the rotating angle curve. The features are extracted from the vibration signals and then fused by using back propagation neural networks (BPNNs). The bearing fault types can be finally recognized from the well-trained BPNN model. The proposed method improves the recognition accuracy by introducing the amplitude compensation and resampling procedures.