Remaining Useful Life Prediction of Bearings Using Fuzzy Multimodal Extreme Learning Regression

Remaining Useful Life (RUL) prediction of bearings is one of the crucial conditions for timely maintenance. In this paper, a fuzzy multimodal extreme learning regression is proposed for the RUL estimation. In this method, fuzzy fusion, ensemble empirical mode decomposition (EEMD), and extreme learning machine (ELM) are integrated. The fuzzy fusion is first used to fuse original features for establishing a condition criterion. EEMD is subsequently utilized to decompose the condition criterion into several sub-series of multiple modes. EML is then adopted for predicting the sub-series in each mode. The predicted sub-results in each mode are finally summarized as the final results. The proposed method is assessed by the bearings data from NSF I/UCR center. Experimental results reveal that the proposed approach is able to build the condition criterion to reflect bearings degradation. It performs better in the RUL prediction than benchmark approaches.