Correlation Graph Fourier Spectrum Model and Its Applications in Rolling Bearing Fault Diagnosis

Abstract

Abstract: The transmission system of wind turbines contains complex nonlinear coupled noise and interference, which poses significant challenges to traditional signal decomposition and deconvolution algorithms. This paper proposes a method based on feature dimensionality enhancement and denoising in the graph spectral domain to extract fault features in wind turbine bearings, referred to as the Correlation Graph Fourier Spectrum Model(CGFS Model). First, a spectrum-based Laplacian matrix is constructed to transform the signal from a one-dimensional representation to the graph spectral domain. A transfer frequency-domain feature analysis method is designed, and a linear graph-spectral order separation strategy is developed to overcome the limitations of traditional methods in connecting the time domain and the graph spectral domain. A hybrid correlation entropy is designed to quantify the characteristics present in different graph spectral orders, enabling precise localization of wind turbine bearing faults. Simulation signals and bearing inner/outer-race fault signals validate the effectiveness and practicality of the proposed model.

Conflict of Interest Statement
The authors declare no conflicts of interest.