Rolling bearing faults typically exhibit sideband structures in the frequency domain, yet conventional blind deconvolution methods operate in the time domain and rely on prior fault periods. To reduce this dependence and broaden applicability, this paper presents a frequency-domain blind deconvolution method based on weight feature extraction (WFE-FDBD). The method constructs rectangular pulse weights with a finite bandwidth to mitigate the effects of limited frequency resolution and minor sideband fluctuations. By incorporating the correlated kurtosis index, an inverse filter with adaptive period adjustment is built, enabling automatic selection of the fault period. Operating directly in the frequency domain, the proposed approach effectively enhances fault-related harmonics and suppresses noise interference. Both simulated and experimental signals validate the WFE-FDBD method, demonstrating its capability to diagnose localized defects on inner and outer raceways of rolling bearings.
Wind turbine bearings operate long-term under complex and variable operating conditions, where fault impulse characteristics are easily submerged by strong noise. Traditional graph signal processing-based bearing fault diagnosis methods are limited by fixed graph topology, empirical feature selection and poor noise robustness. This paper proposes an adaptive frequency graph spectrum (AFGS) model for bearing fault diagnosis. The model constructs graph signals in the frequency domain and determines the core analysis interval adaptively via eigenvalue sequences, which eliminates fixed topology constraints. Combined with a fast bisection search framework and a correlation spectral negative entropy (CSNE) index sensitive to periodic fault impulses, the proposed method realizes fully automatic optimal band selection without manual intervention and improves noise resistance. The AFGS method first transforms vibration signals via fast Fourier transform and constructs frequency-domain graph features based on Laplacian matrix decomposition. The optimal fault characteristic band is adaptively determined using the bisection framework and CSNE criterion. Finally, signal reconstruction and envelope spectrum analysis are implemented for fault identification. Simulation results under −3 dB low signal-to-noise ratio show that AFGS can effectively extract the 1st to 8th fault harmonics. Further validation on the measured inner and outer race fault signals of 6205 bearings demonstrates that the proposed method can clearly identify fault characteristic frequencies and their multi-order harmonics. Comparative tests with Fast Kurtogram and Autogram indicate that the two benchmark algorithms only extract limited low-order harmonics under simulated noise and completely fail in practical strong noise environments. Experimental results verify that AFGS outperforms conventional methods in band localization accuracy, noise suppression, and fault feature extraction completeness, providing a reliable solution for the bearing fault diagnosis of rotating machinery under complex working conditions.