Abstract
To address the limitations of weak information extraction of rolling bearing fault features and the poor generalization performance of diagnostic methods, a novel method was proposed based on sparrow search algorithm (SSA)-Variational Mode Decomposition (VMD) and refined composite multi-scale dispersion entropy (RCMDE). Firstly, SSA optimized the key parameters of VMD to decompose the fault signal. The time-frequency domain comprehensive evaluation factor algorithm was then employed to select the sensitive intrinsic mode function (IMF) components for reconstruction. Then, RCMDE extracted features from the reconstructed signals to create a state feature set, which was input into the K-means KNN (KKNN) classifier for classification. To verify the effectiveness of the proposed method, comparative decomposition methods were established: EMD-RCMDE, EEMD-RCMDE, CEEMDAN-RCMDE, and RCMDE. Various feature extraction methods were also evaluated, including MDE, MFE, and MPE, along with classifiers such as DT, RF, and SVM. Experimental verification on different types of single and compound faults demonstrated the proposed method's excellent fault identification capability. In order to further assess generalization ability and robustness, noise was artificially added to the single fault signals of the rolling element with varying damage levels. The results show that even under-noise interference, the proposed method maintained high fault identification accuracy excellent anti-noise performance and good generalization ability, which provides a certain reference for the solution of such problems.