Generalized difference mode decomposition for adaptively extracting fault components of rotating machinery under non-stationary conditions

Abstract

Compared with existing adaptive signal decomposition methods, difference mode decomposition can effectively extract repeated transient signal components caused by rotating machinery faults. It assumes that signal components are separable in the frequency domain while there is a frequency aliasing phenomenon among signal components under non-stationary conditions in practice. Therefore, adaptive extraction of fault components from original vibration signals under non-stationary conditions is still a challenging topic in signal decomposition fields. In this paper, a novel adaptive signal decomposition method called generalized difference mode decomposition (GDMD) is proposed as an extended version of difference mode decomposition. Firstly, the proposed GDMD method transforms non-stationary signals in the time domain into stationary signals in the angular domain by using the resampling technique. Secondly, a physically explainable optimal difference spectrum is obtained to make a distinction between health signals and fault signals based on convex optimization. Besides, positive and negative thresholds of the optimal difference spectrum are automatically determined through change-point analysis. Finally, signal components are reconstructed by utilizing the inverse fast Fourier transform according to the two thresholds. The experimental research on the inner and outer race faults of rolling bearings has demonstrated the effectiveness of the proposed GDMD method in feature extraction and fault diagnosis under non-stationary conditions.