Differential graph spectral modulation and its application in bearing fault feature identification

Reliable identification of incipient bearing faults remains challenging in industrial scenarios involving complex operating conditions and intense noise interference. To address this issue, a Differential Graph Spectral Modulation (DGSM) method is developed to improve fault feature extraction performance. A dual-order screening model based on the differential order and reconstruction order is designed to dynamically enhance the signal spectrum, thereby improving the signal-to-noise ratio and adaptivity. The introduced differential order decomposes the signal into frequency-domain representations with energy concentrated in different frequency bands. By exploiting the spectral clustering characteristics of the Laplacian matrix, fault-related information is concentrated within a compact range of graph spectral reconstruction orders, enabling targeted reconstruction of fault-sensitive resonance components while suppressing noise and interference. Harmonic spectral kurtosis is designed to quantify feature information in different graph spectral orders, allowing accurate and effective selection of the resonance frequency band while eliminating noise interference. Validation using both simulation and experimental cases demonstrates that the proposed method effectively strengthens fault-related information. Compared with existing techniques, the proposed method achieves better feature enhancement performance and higher diagnostic reliability under complex interference conditions.