Fault diagnosis method via one vs rest evidence classifier considering imprecise feature samples

Xiaobin Xu, Haohao Guo, Zhenjie Zhang, Pengfei Shi, Wenguang Huang, Xiaoding Li, Georg Brunauer

Abstract: Highlights•To address the challenge of modeling imprecise feature samples in fault classification, the OvR-REM method is introduced to accurately represent the relationship between fault features and modes.•Based on the ER rule, a two-level fusion mechanism is proposed to integrate the OvR evidence classifiers of all “perfect” and “imperfect” features, the imprecision can be significantly reduced.•Experimental results from rotating machinery fault diagnosis demonstrate the proposed method’s superior classification accuracy compared to traditional models like SVM, BPNN, KNN, FCNN, RF and Bayes.