Adaptive narrowband spectral amplitude modulation for rolling bearing fault diagnosis

Abstract The application of robust signal processing techniques to extract clear fault features from bearing vibration signals has always been crucial for rolling bearing fault diagnosis due to various interferences. As a nonlinear filtering method, the recently proposed spectral amplitude modulation (SAM) enables the simultaneous use of both amplitude and phase features embedded in the spectrum for fault feature extraction, and thus exhibits superior feature extraction capability compared to traditional linear filtering methods. However, the performance of the SAM tends to degrade when applied to complex signals because it uses all frequency components of the modified amplitude spectrum for signal reconstruction and is unable to suppress interference components with spectral amplitudes similar or comparable to those of the fault components. To address this problem, an adaptive narrowband spectral amplitude modulation (ANSAM) is proposed in this study. In the proposed approach, a B-spline fitting spectral trend-based adaptive spectral segmentation method is first introduced, which can separate the fault frequency band from the interference frequency bands in the modified amplitude spectrum. Then, a new indicator in the squared envelope spectrum (SES) domain, termed peak-index-constrained localized averaged kurtosis, is designed to select the narrowband signal containing the richest fault information from those generated by spectral segmentation. Finally, the fault characteristic frequency is extracted from the SES of the selected narrowband signal. The effectiveness of the ANSAM is validated through a simulated signal with multiple interference components and three sets of complex experimental bearing vibration data from distinct test benches. The results show that the ANSAM achieves superior performance in fault feature extraction compared to the SAM and fast kurtogram.