Bearing fault diagnosis based on TQWT and graph under strong noise

Fault diagnosis technology based on intelligent methods has high requirements for the quantity and quality of training samples. But bearing fault data that can be collected in real life is usually limited and contains a lot of noise which leads to the weak generalization ability of intelligent fault diagnosis methods. Therefore, this paper proposed a new method to surmount the defect which can effectuate bearing fault diagnosis under strong noise. Firstly, Tunable Q-factor Wavelet Transform (TQWT) with sparse penalty optimization is introduced to decompose the vibration signals to obtain sparse wavelet coefficients of each frequency band. Then, the wavelet coefficients of all frequency bands of each sample form a graph data sample. Finally, the diagnosis model built by Graph Neural Network (GNN) can learn features from graph data and realize fault diagnosis. Verified by real data experiments, the proposed method shows a superior diagnostic effect.