Early Detection of Rotor Faults in Large Hydrogenerators Using Vibration Measurements, Variational Autoencoder, and Euclidean Distance

In this paper, the authors present an Artificial Intelligence (AI) based Variational AutoEncoder (VAE) technique for detecting rotor faults in a large hydrogenerator. The proposed technique is applied to assess health monitoring and classification capabilities on an existing <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$74 MW$</tex-math></inline-formula>, 76 poles hydrogenerator. The study uses real vibratory data collected in situ from a healthy machine and faulty signals generated using Finite Element Model (FEM) that are combined to the healthy signals as to constitute a database used for the AI model. The latter is then trained and validated using only one single severity pattern of each fault. The proposed method is tested using a third dataset that revealed the technique's proficiency in clustering various health cases within the model's latent space, effectively clustering them in a 2D user-friendly space. Furthermore, a novel health monitoring metric based on the squared Euclidean distance in the latent space, as well as the statistical law (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\chi _{2}^{2}$</tex-math></inline-formula>) is presented. The obtained results show the model's ability in detecting rotor anomalies at early stages of their occurrence, thus underscoring its advantages in health monitoring.