A Robust AI-Driven Multisensory Framework for Bearing and Gear Fault Diagnosis Based on VMD, HES, and RFE

In the field of signal analysis for machinery health monitoring and fault diagnosis, this paper presents a comprehensive methodology that combines Variational Mode Decomposition (VMD), Hilbert Envelope Spectrum (HSE), Recursive Feature Elimination (RFE), and advanced machine learning techniques. The primary goal is to establish a robust and precise approach for signal decomposition and feature extraction. Initially, VMD is used to decompose the signal into Intrinsic Mode Functions (IMFs). The HES of each IMF is then calculated, and the IMF with the highest Spearman coefficient correlation with the HES of the original signal is selected. Key indicators are computed from this selected IMF, and RFE is employed to identify the most relevant features. The methodology begins with VMD-based signal decomposition. The performance of each IMF is assessed by its correlation with the HSE of the original signal. The IMF with the highest Spearman coefficient is selected as the primary diagnostic feature. These indicators are standardized to ensure robustness and comparability. The standardized features are then used for fault diagnosis with various machine learning algorithms, including support vector machines, random forests, and discriminant analysis. The proposed methodology is validated using five datasets comprising three vibrational, one acoustic, and one electrical dataset. Experimental results demonstrate the effectiveness of the approach in accurately detecting and diagnosing faults, enhancing the reliability and maintenance efficiency of industrial machinery.