Bearing Fault Diagnosis Based on Ensemble Feature Selection Method with Combination of Sorted Subsets

Rolling bearings are critical components in rotating machinery, where their failure can lead to abrupt system shutdowns and significant financial losses. A prevalent strategy for bearing fault diagnosis involves analyzing features extracted from vibration signals. To address the challenges posed by irrelevant and redundant features in the original feature set, as well as the instability in outputs from individual feature selection algorithms, this paper introduces a bearing fault diagnosis method based on Ensemble Feature Selection with combination of sorted subsets (shorten for EFS-CSS). This method integrates the outputs of multiple feature selection algorithms across various feature numbers to determine ensemble feature combinations that exhibit low dimensionality and robust fault characterization capabilities. Following feature identification, fault diagnosis is performed using four machine learning algorithms. The effectiveness of the proposed approach is validated through classification experiments involving 24 and 48 fault conditions using the Case Western Reserve University (CWRU) benchmark dataset. The results demonstrate that the proposed method achieves superior diagnostic performance compared to techniques reported in existing literature.