Two-Stage Strategy-Guided Improved Feature Mode Decomposition for Mechanical Incipient Fault Diagnosis

Bearing incipient fault diagnosis presents significant challenges due to the weak fault features in incipient fault signals, which are often obscured by harmonics and random shocks. Recently proposed feature mode decomposition (FMD) performs well in extracting fault features. However, the initial number of iterations in FMD significantly affects both the efficiency and effectiveness of the decomposition. To this end, this paper proposes an improved feature mode decomposition (IFMD) guided by a novel two-stage strategy. IFMD ensures the elimination of redundant iterations and enhances the accuracy of decomposition. First, a filter bank is initialized using a method where adjacent filters overlap by 50%, ensuring that each filter contains more fault feature information. In the two-stage strategy, autocorrelation theory is adopted for period estimation, successfully extracting fault features. The CK value after the first stage is retained for selecting the fault frequency band. Secondly, the iteration stop criterion (ISC) in the second stage ensures that the value of the objective function, i.e., the CK, continues to increase. Finally, the effectiveness of the IFMD is validated using two sets of fault experimental data. The results demonstrate that IFMD achieves superior performance for incipient bearing faults compared to conventional decomposition algorithms.