Misalignment Severity Classification in Rotor-Bearings using the mRMR Feature Selection Method

The accurate detection and diagnosis of mechanical faults are critical for predictive maintenance and operational safety. While signal processing techniques, particularly entropy measures, are widely used for this purpose, the comparative performance of newer entropy methods and their synergy with traditional features remains an active area of research. This study presents a comprehensive evaluation of nine entropy measures and nine time-domain statistical features for fault classification using an optimized Support Vector Machines (SVM). The optimization of the SVM is done using the Genetic Algorithm (GA). The analysis is conducted on an acoustic dataset comprising normal conditions and four distinct fault severities. Results indicate that while individual entropy measures like Bubble Entropy achieve strong fault detection (۹۵.۷۵% test accuracy), their performance for multi-class fault diagnosis is limited (۴۷.۷۰%). Conversely, a combined set of time-domain features provides a robust baseline, achieving ۹۵.۰۳% and ۸۵.۵۵% accuracy for detection and diagnosis, respectively. To overcome the limitations of individual features, a hybrid framework integrating Minimum Redundancy Maximum Relevance (mRMR) for feature selection is proposed. This mRMR-GA-SVM model demonstrated superior performance, achieving near-perfect fault detection (۹۹.۵۵%) and excellent fault diagnosis (۹۵.۲۱%) accuracy. The findings conclusively show that a strategically selected hybrid feature set in mRMR method, significantly outperforms any single feature type, establishing a powerful and reliable methodology for complex fault diagnosis tasks.

کلیدواژه ها:

Fault Diagnosis ، Entropy Measures ، Feature Selection ، mRMR Feature Selection