A Multi-Domain Feature Fusion Method for Wind Turbine Bearing Fault Diagnosis

Abstract

The vibration signals of bearings in wind turbines contain crucial information for fault diagnosis. To more effectively extract the features of wind turbine bearing vibration signals for this purpose, this study proposes a bearing fault diagnosis method that leverages multi-domain feature fusion for feature extraction and deep modeling. To extract and fuse the features of time-domain signals, signal processing techniques such as Variational Mode Decomposition (VMD) and Fast Fourier Transform (FFT) are employed. The fused features are subsequently selected using Random Forest (RF) methods, and a Deep Belief Network (DBN) model is built for fault classification. The efficacy of the proposed method is evaluated using two different datasets, with results indicating a fault classification accuracy of more than 97.3%. This validates the method's effectiveness and generalization in diagnosing wind turbine bearing faults.