Current Sensor Fault Diagnosis of MUPMSM Drive System Using Similarity Learning

For the multi-unit permanent magnet synchronous motor (MUPMSM) drive system, the current sensor is the most sensitive device that determines the control performance and safe operation. Therefore, fault diagnosis of the current sensor is significant for fault tolerant control and further maintenance. In this paper, a current sensor fault diagnosis method based on similarity learning (SL) is proposed. Innovatively, for the N-unit motor system with two current sensors in each unit, the proposed method can realize the fault diagnosis of 35*N kinds of current sensors without needing permanent magnet flux information and a heavy calculation burden. Furthermore, since the proposed method is based on structural characteristics, the transient system status has little effect on diagnosis. Particularly, the exponential Euclidetic distance (EED) and the polar Euclidetic distance (PED) are respectively proposed to reflect the difference in the reference voltage between the healthy and the faulty unit. Then, the EED and PED values under different states are theoretically analyzed to ensure the diagnosis speed and reduce the false alarm rate. Finally, different from the traditional SL, four parameters with anti-false alarm functions are designed to realize the fault sensor diagnosis rather than complex classifiers. The experiment results on four-unit PMSM verify the effectiveness of the proposed fault diagnosis method.