Application of Adaptive Feature Mode Decomposition Based on Synthetic Index in Fault Feature Extraction of Rolling Bearings

The initial signs of a fault in a rolling bearing are often subtle and can be easily obscured by background noise, making the extraction of fault characteristics challenging. To address this, it is crucial to accurately characterize both the bearing fault features and the accompanying noise. This study introduces a novel approach for feature extraction from rolling bearings, leveraging a comprehensive index known as Composite Feature Mode Decomposition (CFMD). Initially, the Sparse Maximum Harmonic Noise Ratio Deconvolution (SMHD) algorithm is employed to amplify the weak fault signal. A new comprehensive evaluation index combining cosine similarity, kurtosis, and envelope entropy is proposed, which is used to adaptively select the optimal parameter combination for FMD and to filter the Intrinsic Mode Functions (IMFs) based on this index. Subsequently, to extract more detailed fault features, Multi-scale Fuzzy Entropy (MFE) is utilized to quantify the fault characteristics of each effective IMF, which are then classified and recognized using a Support Vector Machine (SVM). The CFMD method's effectiveness is validated through an analysis of bearing failure data. Compared with other commonly used diagnostic techniques, this method demonstrates superior accuracy and effectiveness in extracting bearing fault information.