Prognostic study of ball screws by ensemble data-driven particle filters

The prognostic study of the ball screw is critical to increase the reliability of manufacturing system, which has drawn great attention in the field of Prognostics and Health Management (PHM). The particle filters (PF) method is a powerful tool for prognostic study because of its capability of robustly predicting the future behavior. However, lack of analytical ball screw measurement model limits the application of PF. In this paper, an ensemble GRU network is designed to extend PF to the case where the analytical measurement equation is not available. The proposed hybrid GRU-PF method integrates the data-driven model and the physical model into the particle filters network to realize the prognostic and remaining useful life (RUL) prediction of the ball screw. The effectiveness of the proposed method is validated by designing a ball screw accelerated degradation test (ADT), and the results of this experimental study demonstrate the satisfactory performances in terms of prognostic sensibility.