A resonance demodulation method based on amplitude Z-score of envelope spectrum for weak bearing fault detection

Resonance demodulation has been extensively employed for fault diagnosis in rolling element bearings due to its effectiveness in extracting fault features. Achieving high diagnostic accuracy largely depends on a precise identification of the optimal resonant frequency band for demodulation. However, accuracy and robustness of existing methods still require further improvement. To address this issue, this study proposes a resonance demodulation algorithm, termed Z-score-gram (Zsgram), which aims to enhance fault feature extraction under low signal-to-noise conditions. The proposed method constructs a hierarchical representation of a signal by summing minimum-scale components through a level-wise decomposition structure, significantly improving computational efficiency by reducing required number of Fourier and inverse Fourier transforms. To incorporate physical fault characteristics into evaluation processes of frequency bands, the algorithm introduces an amplitude-based Z-scores at fault-related frequencies within envelope spectra. This integration improves both sensitivity and robustness of fault-component identification. Experiments on simulated data and public datasets demonstrate that the Zsgram outperforms benchmark methods including Fast-Kurtogram, Autogram, and improved envelope spectrum via candidate fault frequency optimization-gram, achieving superior accuracy and stability in detecting weak fault components. As a result, the proposed method facilitates early and accurate extraction of fault features, improves potential defects observable, and consequently enhances both timeliness of predictive maintenance and intrinsic operational safety of bearings.