Bearing Fault Diagnosis Based on SDWPSO-FMD

Mechanical fault vibration signals often exhibit nonlinear and non-stationary characteristics, making fault feature extraction challenging. Feature mode decomposition (FMD) has emerged as a promising method for extracting fault features of such signals. However, the effectiveness of FMD heavily relies on the appropriate parameter settings of decomposition mode number $(n)$, filter length $(L)$ and filter band number $(K)$. To address this issue, a fault feature extraction method based on improved particle swarm optimization (Sigmoid-DWPSO) is proposed in this paper, and the input parameters of FMD are self-optimized. To enhance the search capability and convergence of the PSO algorithm, the improved PSO combines the nonlinear decreasing inertia weight adjustment strategy with the adaptive learning factor adjustment method based on Sigmoid function. Above all, The Sigmoid-DWPSO (SDWPSO) algorithm employs the minimum envelope entropy as the adaptive objective function to determine the optimal parameter combination $(n, K, L)$ for the FMD algorithm. Subsequently, the vibration signal is decomposed by FMD using the optimal parameters to obtain multiple modal components. Finally, according to the kurtosis value, the optimal component is selected for envelope demodulation analysis. The component with the highest kurtosis value is selected for envelope demodulation analysis. This paper compares SDWPSO with several existing PSO algorithms using standard test functions, and the results demonstrate that the SDWPSO method outperforms other algorithms in terms of optimization performance. Furthermore, the optimized versions of FMD and VMD using SDWPSO are compared using rolling bearing dataset of Case Western Reserve University and Xi 'an Jiaotong University. The results indicate that the SDWPSO-FMD method achieves superior decomposition performance and can effectively improve the accuracy of fault diagnosis.