Fault Feature Extraction for RV Reducers Based on IWHO-VMD and Effective Mode Reconstruction

Fault feature extraction for rotate vector (RV) reducers is hindered by weak impulsive components masked by noise, empirical parameter selection in conventional variational mode decomposition (VMD), and mode redundancy. To address these issues, an improved wild horse optimizer-based variational mode decomposition (IWHO-VMD) framework with effective-mode reconstruction is proposed. A two-stage search strategy and a composite fitness function integrating squared envelope spectrum (SES) negentropy, reconstruction error, and an effective-mode penalty are used to optimize the VMD mode number and penalty factor. Fault-related intrinsic mode functions are then selected using SES negentropy and normalized energy ratio, followed by effective mode reconstruction and Hilbert envelope spectrum analysis. The method is validated on two self-built RV reducer datasets involving rolling-element wear and planetary gear tooth-surface wear, together with a public crankshaft wear dataset, and compared with several optimization methods. It achieves an average characteristic-frequency identification accuracy of 98.95% across the three datasets. On the comparative dataset, IWHO-VMD achieves the lowest fitness value and reduces the number of convergence iterations by 25.0-50.0%; with effective-mode reconstruction, its average identification accuracy reaches 99.03%. The proposed framework improves VMD parameter adaptivity and enhances fault-feature representation, providing a reliable basis for RV reducer condition monitoring.