Rolling Bearing Early Fault Detection Method Based on Feature Clustering Fusion Degradation Index

Abstract

Abstract The research on rolling bearing early fault detection is mainly focused on degradation index extraction and adaptive setting of alarm threshold. The mainstream methods are to extract degradation indicators based on adaptive features and set adaptive alarm thresholds based on the Shewhart control chart. However, the adaptive feature extraction method does not consider the correlation between features, and the Shewhart control chart is not sensitive to small fluctuations caused by early faults. In this study, a rolling bearing early fault detection method based on a feature clustering fusion degradation index is proposed. The multidomain statistical features are extracted to form the initial feature set, and the improved hierarchical clustering algorithm is combined with the feature evaluation index to select features to form a preferred feature subset, to ensure the richness of index information and reduce redundancy. After the construction of the degradation index, to suppress the interference caused by nonstationary and abnormal shocks in early fault detection, the accurate evaluation method and anomaly determination strategy of control chart parameters are studied, and an improved exponential weighted move average control chart is designed to monitor the degradation index. The effectiveness and superiority of the proposed method are verified by public data sets. This research provides a rolling bearing early fault detection method, which can provide comprehensive degradation indicators, eliminate interference caused by random anomalies and running in periods, and achieve an accurate detection of early bearing failures.