Wind turbine high-speed shaft bearings health prognosis through a spectral Kurtosis-derived indices and SVR

A significant number of failures of wind turbine drivetrains occur in the high-speed shaft bearings. In this paper, a vibration-based prognostic and health monitoring methodology for wind turbine high-speed shaft bearing (HSSB) is proposed using a spectral kurtosis (SK) data-driven approach. Indeed, time domain indices derived from SK are used and a comparative study is performed with frequently used time-domain features in the bearing degradation health assessment. The effectiveness is quantified by two measures, i.e., monotonicity and trendability. Among those features, the area under SK is utilized for the first time as a condition indicator of rolling bearing fault. A support vector regression (SVR) model was trained and tested for the prediction of the HSSB lifetime prognostics, showing the superiority of SK-derived indices of degradation assessment. We verified the potential of the prognostics method using real measured data from a drivetrain wind turbine. The experimental results show that the proposed approach can successfully detect an early failure and can better estimate the degradation trend of HSSB than traditional time-domain vibration features.