An integrated methodology for estimating the remaining useful life of high-speed wind turbine shaft bearings with limited samples. (January 2022)

• Record Type: Journal Article
• Authors: Merainani, Boualem; Laddada, Sofiane; Bechhoefer, Eric; Chikh, Mohamed Abdessamed Ait; Benazzouz, Djamel
• Is Part Of: Renewable energy. Volume 182(2022)
• Page Start: 1141
• Page End: 1151
• Publication Date: 2022-01
• Subjects: Remaining useful life -- Wind turbine generator -- Spectral shape factor -- Elman neural network -- Data driven -- Vibration -- Bearing fault; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2021.10.062
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: The wind power industry suffers from unexpectedly high failure rates in wind turbine high-speed shaft bearings (HSSBs). To reduce cost and improve availability, the industry needs an accurate fault prognostic and remaining useful life (RUL) capability. A reliable prognostic allows maintainers to better define when maintenance can be performed, improving availability or allowing for opportunistic maintenance. Wind turbines operate under harsh conditions and condition monitoring data shows the environment to be both non stationary behavior with high noise. This paper proposes a practical and effective data-driven methodology that can be applied for RUL prediction of HSSBs. A new health indicator (HI) is constructed based on the entropy measure of the so called "spectral shape factor" (SSF), after strengthening the original signal by Teager energy operator (TEO). An Elman neural network (ENN) is used for RUL estimation. Furthermore, prediction intervals of the RUL estimates are computed based on the trained ENN model in order to quantify the errors associated with the prediction. The methodology is validated using a real word data collected form 2 MW Suzlon S88 wind turbine. Highlights: A spectral shape factor (SSF) approach is developed for feature extraction. SSF derived time-domain features give better degradation trending. Critical value was determined for scaling the bearing health. RUL prediction of high-speed shaft bearing based on Elman neural network. Prediction interval of RUL estimates are computed based on trained ENN model.