Rotor Fault Diagnosis of Large Hydrogenerators Using Vibration Data and VAE-Desirability Function

In electrical equipment maintenance, precise diagnosis is critical for dependability and efficiency. Variational Autoencoder (VAE) approaches show promise for fault classification but encounter hurdles due to their non-deterministic nature, which results in inconsistent fault grouping across machines. To address this, a novel method incorporates a desirability function into the VAE’s cost function, resulting in standardized fault classification. This approach, evaluated with vibration data from large hydrogenerators, demonstrates that the enhanced model can cluster 86% of comparable faults into predefined zones, compared to 27% without including this term, while accounting for two severity levels of three rotor faults.