Wiener-based remaining useful life prediction of rolling bearings using improved Kalman filtering and adaptive modification

The remaining useful life (RUL) prediction of rolling bearings is a key process for prognostics and health management (PHM) of machines. In RUL prediction, the wiener-based models are widely used to describe the degradation process of machinery equipment. However, the random noises in monitored signals, caused by operating conditions and measurement errors, can seriously decrease the stability and accuracy of wiener-based RUL prediction. Although some filtering algorithms have been used for noise processing and state estimation, there are still some problems that need to be solved in current studies. First, the quantification of noises is usually based on the historical data while the information of the online monitored data is not fully utilized. Second, the difference between the measurement value and the filtered value can cause huge error in RUL prediction. To overcome these drawbacks, this paper proposes a wiener-based RUL prediction method using improved Kalman filtering and an adaptive modification algorithm. The wiener-based exponential model is utilized to establish the degradation process. An improved Kalman filtering (KF) is proposed to minimize the interference of random noises, in which the noise quantification is achieved with Bayesian updating. To further improve the accuracy of the prediction results, an adaptive modification method is proposed to reduce the prediction errors. To investigate the performance of the proposed method, the comparisons are conducted with five commonly used RUL prediction methods on a simulation example and an experimental bearing vibration signal dataset. The results demonstrate the superiority of the proposed method on stability and accuracy in RUL prediction.