Fault Feature Extraction of Rolling Bearings Based on Ordered Singular Spectrum Decomposition-Multipoint Optimal Minimum Entropy Deconvolution Adjusted.

In the early fault stage of rolling bearings, the fault-induced impact signals in vibration data are often extremely weak and easily obscured by strong noise, making effective extraction and analysis challenging. To address this issue, this paper proposes a novel weak fault impact signal feature extraction method combining Ordered Singular Spectrum Decomposition (OSSD) and Multipoint Optimal Minimum Entropy Deconvolution Adjusted (MOMEDA). First, OSSD is employed to decompose the raw vibration signal, progressively extracting signal components across different frequency bands. The optimal signal components are adaptively selected based on mutual information criteria, effectively avoiding mode mixing issues. Subsequently, MOMEDA is applied to enhance the periodic impact features within the fault signal, improving its recognizability. To address the signal length reduction issue inherent in the MOMEDA process, a waveform extension strategy is introduced to compensate for the missing signal, ensuring signal integrity. Simulation and experimental results demonstrate that the proposed method exhibits robust noise resistance and can effectively extract early fault features of rolling bearings under strong noise conditions, validating its accuracy and effectiveness.