Rotor Fault Diagnosis in a Hydrogenerator based on the Stator Vibration and the Variational Autoencoder

A non-intrusive method for the detection of the Rotor Inter-Turn Short Circuit and Static Eccentricity (SE) of a large hydrogenerator (74 MVA, 76 poles), by the means of real vibratory stator data, Finite Element Model (FEM) and Deep Learning technique, is presented in this paper. The Variational AutoEncoder is implemented for fault diagnosis: it is trained and validated on healthy vibratory signals. The signature of both faults is obtained from FEM, injected into another set of healthy real vibratory signals, and tested in the VAE. The classification in the latent space is investigated. The results prove the capability of the method in fault classification in a user friendly space, where each colored zone corresponds to a different case.