Dynamic normalized health indicator construction and Bayesian recurrent state estimation for remaining useful life prediction of high-speed bearings in wind turbine drivetrain

Accurate remaining useful life prediction of bearings in wind turbines is beneficial for safe operation. However, conventional health indicators rarely consider monotonicity, normalization and dynamic degradation simultaneously. Also, traditional recurrent methods lack attention to trend constraint in long-term rolling prediction. Both aspects are significant for prognosis. To resolve these challenges, this paper proposes a dynamic degraded feature normalization model for the monotonic health indicator construction. A hybrid binary ternary tree is presented for finer band division of vibration signal, and a nonlinear correction operator is designed for normalization. Besides, a dual dimensional attention block is designed to strengthen the feature at the elemental level. A Bayesian recurrent state estimator is developed to incorporate the enhanced historical trend into recurrent prognosis for robust prediction. The proposed method innovatively introduces the degradation trend constraint into the indicator construction and state prediction. Several bearings from accelerated degradation platform and on-site wind turbine are utilized to validate the effectiveness and superiority of the proposed model. Experimental results demonstrate that the proposed method achieves more than 16% reduction in prediction error in root mean square error compared to several state-of-the-art methods.