Remaining Useful Life Prediction of Bearings Based on Jensen-Shannon Divergence and Dynamic Multi-Time Scale Aggregation

Remaining useful life (RUL) prediction is a key research task in intelligent maintenance of mechanical equipment. However, there are still shortcomings in bearings RUL prediction: 1) Due to the different degradation stages of rolling element bearings, health indicators (HI) also exhibit different degradation characteristics. However, the two-stage nature of the HI is often overlooked in existing research. 2) Model-based methods in current use typically employ a single model to describe the bearing degradation process, making it difficult to comprehensively describe the various potential degradation trends of bearings in the future. To deal with the aforementioned issues, this study constructs a health indicator, named HI-JS, using the self-organizing map (SOM) clustering and Jensen-Shannon (JS) divergence methods. The effectiveness of HI-JS is validated using comprehensive evaluation criterion. A new bearings RUL prediction process is proposed using JS divergence and dynamic multi-time scale aggregation for bearings RUL prediction. The proposed approach is evaluated using XJTU-SY bearing datasets. Experimental results demonstrate the effectiveness of the proposed approach in improving the accuracy of RUL prediction of rolling element bearings.