As a critical element of the drive motor, rolling bearings are susceptible to localized defects under complex loads and varying operating conditions. Such defects typically generate periodic transient shocks, which reflect bearing fault features. However, the accurate extraction of fault-related transient components becomes challenging due to strong noise influence. To address this issue, a concave sparsity-assisted generalized dispersive mode decomposition (CSA-GDMD) method is developed to enhance fault feature extraction. This method introduces a non-convex sparse model based on generalized mini-max concave (GMC) regularization to preprocess the vibration signal. The GMC penalty effectively suppresses background noise while better preserving the amplitude characteristics of the transient shocks. Subsequently, GDMD is applied to progressively extract transient shock components from the preprocessed signal and reconstruct the signal, resulting in more prominent fault-related transient components. The simulation results show that CSA-GDMD significantly improves the signal-to-noise ratio (SNR), from 6.5905 dB at −15 dB to 9.5122 dB at 5 dB, and reduces the root mean square error (RMSE) from 0.0280 to 0.0196. Consequently, the fault feature frequencies can be identified more clearly in the envelope spectrum, further confirming the accurate fault diagnosis capability of the proposed method for bearing faults under strong noise conditions.
Zhang et al. (Tue,) studied this question.