LAUSR

We proposed an appealing method based on refined composite multiscale fuzzy entropy (RCMFE), infinite feature selection (Inf-FS) algorithm, and support vector machine (SVM) for implementing localized defect detection to keep the downtime and extended damage caused by incipient failure of bearing at a minimum. As a useful approach, multiscale fuzzy entropy (MFE) was utilized to measure the complexity and dynamic changes of signals. However, an inaccurate entropy value would be yielded with the increase of scale factor. Here, as an improvement version of MFE, the RCMFE was proposed to address the shortcomings in the case of short time series. For this novel method, we conducted a full investigation of the effects and robustness by comparing the proposed method with two other entropy-based approaches using synthetic signals and real data. Results indicate that the proposed algorithm outperforms the other approaches in terms of reliability and stability. The RCMFE values of bearing signals from one healthy condition and seven fault states are calculated as diagnostic information. Moreover, an intelligent fault identification method was constructed by combining the Inf-FS algorithm and SVM classifier. Experimental results show the usefulness of the proposed strategy.