Non-Invasive Detection of Rotor Inter-Turn Short Circuit of a Hydrogenerator Using AI-Based Variational Autoencoder

This article presents a non-intrusive technique for detecting the rotor inter-turn short circuit (RITSC) of a hydrogenerator using an artificial intelligence (AI) based variational AutoEncoder (VAE). The technique is applied to a large hydrogenerator of 74 MVA and 76 poles, to test its health monitoring and classification potential. The model is trained and validated based on the acquisition of real vibratory data collected in situ from a healthy machine. The frequency pattern of the fault in the vibration signal is obtained based on finite element methods (FEM). Then, to test the sensitivity of the model in early fault detection, the signature is injected into another set of real healthy vibration signals, and the results are compared to those obtained using the traditional vibration monitoring technique. Furthermore, clustering in the latent space of the model is explored. The obtained results prove the ability of this technique and its potential in detecting anomalies at earlier stages as well as its capacity to cluster different degrees of severity of the fault in a 3D user-friendly space.