Non-Invasive Anomaly Diagnosis for Hydro Electrical Generators Rotor Inter-Turn Short-Circuit Detection Using Stray Flux and the VAE

This paper presents an overview of a new approach for the diagnosis of synchronous generators based on the stray flux along with the Variational AutoEncoder technique (VAE). A full study on the use of the VAE to detect the rotor inter-turn short-circuit fault of the salient-pole synchronous machine has been conducted. Real data were measured on a salient-pole synchronous hydro-electrical generator with a rating of 325 MVA and 56 poles. Synthetic data from simulated defects were added to the real data and used to train and test the VAE. Moreover, it is shown that adding faulty signals with different severities into the training database had led to a better clustering result. Furthermore, the sensitivity of the VAE to faulty signals and its ability to separate them from healthy signals have shown that the VAE is a promising technique for fault diagnosis based on the stray flux.