文献类型：期刊文献

英文题名：Fault Diagnosis of Rotating Machinery under Imbalanced Samples Based on Outlier-Removal Hybrid CAB-SMOTE-LOF

作者：Liao, Zhiqiang[1,2,3]; Yan, Zhijia[1]; Jia, Baozhu[1,2,3]; Song, Xuewei[1,2,3]

机构：[1] Naval Architecture and Shipping College, Guangdong Ocean University, Zhanjiang, 524088, China; [2] Technical Research Center for Ship Intelligence and Safety Engineering of Guangdong Province, Zhanjiang, 524088, China; [3] Guangdong Provincial Key Laboratory of Intelligent Equipment for South China Sea Marine Ranching, Zhanjiang, 524088, China

年份：2026

外文期刊名：SSRN

收录：EI(收录号：20260394908)

语种：英文

外文关键词：Data handling - Electric fault currents - Signal processing - Statistics

外文摘要：In rotating machinery fault diagnosis, the original feature space often suffers from anomalous outliers, limited sample availability, and imbalanced class distributions. These issues severely hinder the accurate recognition of minority-class samples and lead to suboptimal diagnostic efficiency. To tackle these challenges, this paper proposes a hybrid method named HCAB-SMOTE-LOF, which integrates the Hybrid Clustered Affinitive Borderline Synthetic Minority Over-sampling Technique with the Local Outlier Factor (LOF) algorithm. To extract the fault-relevant time-frequency features, the high-pass filtering is applied to the raw vibration signals to remove low-frequency components. The LOF is introduced to eliminate samples that are clearly abnormal or mislabeled in the feature space. On this basis, the HCAB-SMOTE algorithm is employed to synthetically generate minority-class fault samples, thereby improving the quality and distributional plausibility of the generated samples and alleviating the adverse impact of data imbalance on classifier performance. The effectiveness of the proposed method is validated using both a laboratory dataset and the Ottawa motor dataset. Furthermore, a comparative analysis is conducted between the proposed method and several oversampling strategies applied during data processing, including the use of original data, SMOTE, Borderline-SMOTE, and HCAB-SMOTE. Experimental results demonstrate that the proposed method consistently outperforms all existing oversampling methods across all evaluation metrics, confirming its marked advantages in outlier removal and sample-distribution optimization and thus its capability to effectively detect rotating machinery faults. ? 2026, The Authors. All rights reserved.