LAUSR

Abstract Gearbox is one of critical transmission components in wind turbine (WT) having a high downtime rate among all subcomponents. Fault prognostics and health management (PHM) of WT gearbox is crucial to their high reliability operation. However, presence of background noise in WT signals restricts the applicability of existing PHM approaches in feature extraction. To solve this problem, a novel performance degradation assessment method based on deep belief network (DBN) and self-organizing map (SOM) is proposed to de-noise and merge multi-sensor vibration signals. Minimum quantization error (MQE) is defined as health indicator to detect incipient fault of WT gearbox. After health indicator construction, an improved particle filtering (PF) optimized by fruit fly optimization algorithm (FOA) is employed to predict the remaining use life (RUL) of WT gearbox. To take advantage of dynamic and random operation process of WT gearbox, Wiener-process-based degradation model is developed to improving RUL prediction efficiency. The effectiveness is validated by using simulated as well as experimental vibration signals obtained through a WT gearbox highly accelerated life test. The results illustrate that proposed method can evaluate performance degradation process and predict RUL of WT gearbox effectively.