Neural network based fault detection of PMSM stator winding short under load fluctuation

A negative sequence analysis coupled with a neural network based approach is applied to fault detection of a single phase winding short in a PMSM. A multilayer network provides a near term current prediction as input to the fault detection system. The fault detection is performed using the negative sequence analysis of the residuals (difference between the actual and predicted values of currents). The negative sequence component of the residuals provides the detection of the fault and a measurement of the level of severity of the winding short. The method is validated using a 15 hp PMSM experimental setup.