Fault Identification of Rotating Machinery Based on Dynamic Feature Reconstruction Signal Graph

• Published in: IEEE/ASME transactions on mechatronics 2024-06, Vol.29 (3), p.2056-2066
• Main Authors: He, Wenbin, Mao, Jianxu, Li, Zhe, Wang, Yaonan, Fang, Qiu, Wu, Haotian
• Format: Article
• Language: English
• Subjects: 2-D-convolutional neural network (2-D-CNN); Artificial neural networks; Coefficients; Convolutional neural networks; Decomposition; Discrete wavelet transforms; dynamic feature selection (DFSL); Energy distribution; Fault detection; Fault diagnosis; fault identification; Feature extraction; feature signal graph; Graphs; Machinery; Rotating machinery; Signal reconstruction; Signal to noise ratio; wavelet packet; Wavelet packets; Wavelet transforms
• ISSN: 1083-4435; 1941-014X
• DOI: 10.1109/TMECH.2023.3318373

To improve the performance in identifying the faults under strong noise for rotating machinery, this article presents a dynamic feature reconstruction signal graph method, which plays a key role in the proposed end-to-end fault diagnosis model. Specifically, the original mechanical signal is first decomposed by wavelet packet decomposition (WPD) to obtain multiple subbands including the coefficient matrices. Then, with the originally defined two feature extraction factors maximum distribution difference (MDD) and discrete distribution difference (DDD), a dynamic feature selection method based on the L2 energy norm (DFSL) is proposed, which can dynamically select the feature coefficient matrix of WPD based on the difference in the distribution of norm energy, and enable each subsignal to take adaptive signal reconstruction. Next, the coefficient matrices of the optimal feature subbands are reconstructed and reorganized to obtain the feature signal graphs. Finally, deep features are extracted from the feature signal graphs by 2-D-convolutional neural network (2-D-CNN). Experimental results verify that this method achieves superior performances than the existing methods under different noise intensities, based on the data of both the public platform of a bearing and our laboratory platform of robotic grinding.