Fault Diagnosis Method for Wind Turbine Equipment Based on Quaternion Graph Frequency Domain Transform in High-Dimensional Sensing Data

Abstract

Wind turbine units are subjected to fluctuating loads, intense noise, and complex operating environments over long periods, making it easy for the faint fault signatures of components such as gearboxes, bearings, and drive trains to be masked. A fault diagnosis method employing quaternion graph frequency-domain transformation is developed to enhance multi-channel information fusion and facilitate the adaptive selection of frequency bands for frequencydomain reconstruction. First, this study formulates the relevant sensor signals as quaternion signals and applies the quaternion Fourier transform to generate a fused spectrum. Subsequently, the spectral amplitudes are used to construct a graph model, and their corresponding graph-frequency representation is then derived via graph Fourier analysis. Based on the graph frequency-domain representation, the proposed method introduces Correlation Spectral Negentropy (CSNE) to determine the optimal reconstruction order and adaptively extract informative fault features through sliding frequency-band reconstruction. The method then integrates time-domain, spectral, and envelope-spectrum analyses , enabling the detection of fault characteristic frequencies and their harmonics.