5 June 2026 in SSRN Electronic Journal
Feature Mode Decomposition (FMD) is effective for extracting impulsive fault features, but its random filter initialization may cause fault-induced resonance components to deviate from the filter center or fall into transition bands, leading to feature attenuation. To address this limitation, a center-frequency-guided and bandwidth-constrained Modified Feature Mode Decomposition (MFMD) method is proposed. First, a Spectral Structure Information Detector (SSID) is constructed to localize fault-sensitive center frequencies through a band-limited variational detector and a sparsity-based criterion. Then, a cumulative-energy-guided strategy is introduced to determine the corresponding filter layout range, and the Newton–Raphson-Based Optimizer (NRBO) is employed to adaptively select the cumulative-energy threshold. In this way, the proposed method replaces random frequency-band partition with a fault-sensitive filter-layout scheme. Simulation results show that the mean center-frequency error is reduced from 216 Hz to 14 Hz, while the impulsive-energy coverage is increased from 53.48% to 97.46%. Comparisons with NRBO-FMD, NRBO-VMD, IFMD, and SIVMD demonstrate that NRBO-MFMD provides clearer compound-fault mode extraction under strong interference. For the measured wind-turbine vibration signals, the results from 50 randomly selected samples show that NRBO-MFMD achieves the highest fault-characteristic energy ratio (FCER) and spectral envelope autocorrelation index (SEAI) values for both the 162 Hz and 106 Hz fault features. The results indicate that the proposed method improves fault-sensitive mode extraction under both simulated and real operating conditions, while SEAI is used only as a supplementary assessment indicator.