Development of a universal diagnostic system for stator winding faults of induction motor and PMSM based on transfer learning

This paper deals with the possibility of using transfer learning (TL) of a deep convolutional network (CNN) to develop a universal stator winding fault diagnostic system for the induction motor (IM) and the permanent magnet synchronous motor (PMSM). The proposed diagnostic system uses direct processing of phase current signals to detect and evaluate the degree of stator damage. The TL idea was based on using the CNN structure pre-trained for the IM diagnostic system in a diagnostic application dedicated to IM and PMSM. Experimental verification carried out on real objects confirmed very high precision of steady-state and transient damage detection and classification for both IM and PMSM. Furthermore, the detection time shown in the study for both types of machines did not exceed 0.07 seconds.