Enhanced hierarchical grey entropy and dynamic fusion particle filter for rolling bearing degradation trend prediction

Abstract

The service of rolling bearings generally experiences a process from normal state to failure, during which a series of degradation of vibration performance usually occurs. To ensure the high reliability and safe operation of mechanical equipment, accurate prediction of the degradation trend of rolling bearings is essential. However, the vibration signals collected from rolling bearing exhibit complex nonlinear and non-stationary characteristics. In addition, the vibration signal of early fault of rolling bearing is very weak. To address this issue, this article presents a novel rolling bearing degradation trend prediction framework that integrates enhanced hierarchical grey entropy (EHGE) and dynamic fusion particle filter. First, EHGE is proposed as a new entropy to measure complexity, utilizes the moving-averaging procedure and moving-difference procedure to improve the statistical reliability. Then, the dynamic fusion particle filter method is developed to predict the degradation trend, and introduce the interval prediction functionality and the dynamic prediction reliability indicator to enable multi-dimensional evaluation of the prediction results. Finally, the validity of the proposed method in the prediction of bearing degradation trend is verified by the experimental result. When compared to hierarchical grey entropy, hierarchical fuzzy entropy (HFE), and hierarchical sample entropy (HSE), the EHGE demonstrates reduced dependence on data length and achieves remarkable extraction accuracy. In contrast to the traditional particle filter (PF) method, GM method, extended kalman filter (KF), and recurrent neural network (RNN), the dynamic fusion PF estimation method has better prediction accuracy and higher prediction reliability.