Neural network architectures for fault diagnosis and parameter recognition in induction machines

This paper presents a neural network that is able to give, together with the rotor fault diagnosis, the combined rotor-load inertia momentum of an induction machine. The inputs of the network are the spectral components of machine input currents, speed and torque. A specific neural network architecture containing new fast spline-based neurons with improved generalisation capabilities has been used. The training set is obtained by a faulted machine dynamical model as simulator.