About 30%-40% of AC motor faults are related to stator insulation. It is necessary to study a motor winding fault diagnosis (WFD) method for the safe operation of the entire motor system. In recent years, WFD methods using artificial intelligence (AI) technology have received more attention. However, the existing methods usually collect the steady-state current for fault diagnosis, which has low timeliness. In addition, since motor winding fault data is difficult to obtain, it is urgent to propose a WFD method suitable for small samples. Therefore, in this paper, a permanent magnet (PM) motor WFD method based on image recognition and transfer learning applicable to small samples is proposed. Firstly, the motor current signal is collected and 1-D Park’s vector modulus (PVM) operation is performed. Then, the current time-varying signal is processed by recurrence plot (RP) and converted into an image signal. With the help of the ResNetl8 pre-trained model and transfer learning, the new motor WFD model is established. After verification, it was found that the method proposed in this paper can achieve high diagnostic accuracy and fast diagnostic speed in the case of small samples, effectively protecting the safe operation of the motor system.
A permanent magnet motor winding fault diagnosis method based on image recognition and transfer learning
10.10.2024
1111617 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
New Split-Winding Doubly Salient Permanent Magnet Motor Drive
Online Contents | 2003
|