Neural Networks for Detection and Classification of Open Circuit Faults in Electrical Multiphase Drives

Multiphase electrical drives are well-known for their high reliability thanks to a higher number of phases compared to three-phase ones. In the event of a fault, multiphase drives can offer a reconfiguration of the control strategy to achieve an acceptable performance. However, reconfiguring the control requires precise information related to the fault location. With an increased number of phases, it becomes necessary to develop new algorithms for fault detection and localization that exploit the unique properties multiphase drives, which differ from those of three-phase drives. This work aims to develop a fault detection and localization method using neural networks for open-phase faults in a five-phase Permanent Magnet Synchronous Machine (PMSM). To accelerate and simplify the training process, the Hilbert transform is proposed to extract instantaneous phases of currents in the Secondary Machine of a 5 -phase PMSM. Simulation results are presented to confirm the proposed approach.