Author(s): Xiao Lang; Hakan Nilsson; Wengang Mao

Linked Author(s):

Keywords: No Keywords

Abstract: This study employs machine learning techniques to model bearing vibrations for predictive maintenance within a hydropower plant, utilizing over three years of full-scale vibration measurement data. Operational parameters, including turbine speed, guide vane opening, and generator active power, serve as input features to predict vibrations in both upper guide and turbine guide bearings. The models, developed from datasets across different periods, aim to predict and analyze discrepancies in future monitoring data to evaluate potential performance degradation. When the statistical distribution of the future monitoring data closely aligns with the training data, the models demonstrate a capacity to predict gradual bearing performance degradation effectively. However, when future monitoring data diverge significantly from the training set, traditional machine learning models produce irrational predictions, leading to unreasonable trends. To overcome these challenges, the adoption of more sophisticated machine learning approaches is recommended to enhance the reliability of predictive maintenance in the face of unseen data scenarios.

DOI: https://doi.org/10.1088/1755-1315/1411/1/012046

Year: 2024