Speaker
Description
Abstract: Wind turbine high-speed bearings play a critical role in ensuring operational reliability and efficiency, making accurate remaining useful life (RUL) prediction essential for predictive maintenance. Several remaining useful life prediction algorithms have been discussed by many researchers, but there is a lack of degradation trend extraction based on deep learning algorithms. Additionally, however, several forecasting algorithms are available and it’s important to study their accuracy and effectiveness to select the most effective one. The purpose of this study to analyze the RUL prediction accuracy for different combinations: Prophet, Arima, LSTM, and GRU. An LSTM-AE is first employed to extract the degradation trend from vibration signals and then feed to the forecasting combination algorithms. The utilized dataset in this study an open-source data about the progressive failure of the wind turbine high-speed bearing due to an inner race fault. The results show that the proposed combination of LSTM-AE and prophet forecasting is the most effective one with an accuracy of 82%. This research provides valuable insights into a data-driven prognosis approach by identifying the most effective algorithm for RUL prediction and optimizing maintenance strategies for wind turbine high-speed bearings.