Application of artificial intelligence to stator winding fault diagnosis in Permanent Magnet Synchronous Machines

• We first show a correspondence between stator winding short circuit faults and the zero current component of the stator current.
• We design an ANN cluster which is able to detect the presence of short circuit faults based on analysis of the zero-current component of the stator of a Permanent Magnet Synchronous Machine (PMSM).
• The system also determines the actual severity of the short circuit problem using an online optimization routine based on Particle Swarm Optimization.
• Results based on an actual PMSM and computer simulations are presented to assess the performance of the method presented in this paper.

This paper proposes a new methodology to solve the problem of fault diagnosis in electrical machines. The fault diagnosis method presented in this paper is, first, able to provide information about the location of a short-circuit fault in a stator winding. Secondly, the method enables the estimation of fault severity by specifying the number of short-circuited turns during a fault. A cluster of Focused Time-Lagged neural networks are combined with the Particle Swarm Optimization algorithm for proposed fault diagnosis method. This method is applied to the stator windings of a Permanent Magnet Synchronous Machine. Each neural network, in the cluster, is trained to correlate the zero-current component to the number of short-circuited turns in the stator windings. The zero-current component, different from the zero-sequence current, are obtained by summing the instantaneous values of current on all phases of the stator winding during the diagnosis procedure. The neural networks are trained offline with the Extended Kalman Filter method using fault data from both computer simulations and an actual Permanent Magnet Synchronous Machine. The use of the Extended Kalman Filter method, for training, ensures that the neural network cluster used can be re-trained online to make the fault diagnosis system adapt to changing operational conditions. Results from both computer simulation and actual machine data are presented to show the performance of the neural network cluster and the Particle Swarm Optimization algorithm.