Using Unsupervised Machine Learning for Online Monitoring of Turbine-Generator Bearings in Hydro-Electrical Generating Stations

Electric utilities undertake a significant cost to maintain the health of their assets every year. In recent years, in BC Hydro, one of the strategies in maintaining generating stations has been to shift focus from preventative maintenance to predictive maintenance. This approach could help utilities to reduce their expensive preventative maintenance expenditure and further rely on predictive measures to assess the health risk of assets and only execute physical interventions when the assessed risk compels. In this paper, the authors share BC Hydro's experience in using machine learning for predictive online monitoring of turbine and generator bearings in hydroelectrical generating stations. In addition, the paper presents two successful examples where predictive monitoring prevented forced outages and saved the company financial loss and downtime.