Abstract:  Aiming at the difficulty of feature extraction and feature selection of traditional rolling bearing fault recognition algorithms, a rolling bearing fault diagnosis method based on deep dictionary learning (DDL) was proposed. Firstly, the rolling bearing fault vibration data under different working conditions were collected using sensor and the sparsity constraint was implemented by DDL to learn the typical structural features in the fault data layer by layer. Secondly, drawing on the idea of “layer by layer feature extraction” of deep learning method, the deep fault dictionary was constructed according to the fault sample's structure. And the fault samples were fed into the deep fault dictionary to determine the fault category according to the reconstruction error of the samples. Finally, the effectiveness of the DDL model was tested on the rolling bearing test bench. The results of the research indicate that the fault recognition rate of rolling bearing of the proposed deep dictionary learning method reaches 99.28% and the training time is only 765s, which have great advantages in fault recognition accuracy and training time comparing with other deep learning methods such as convolutional neural network and recurrent neural network. The deep dictionary learning method employs the sparse constraint driving dictionary to automatically extract the fault features in the vibration signal samples, while the deep dictionary structure makes the extracted fault features have better hierarchical and physical meaning, which in line with people's intuitive understanding of the fault.

Key words:  rolling bearing; fault recognition; deep dictionary learning (DDL); sparse representation

YU A-dong. Fault recognition of rolling bearing with deep dictionary learning［J］.Journal of Mechanical & Electrical Engineering, 2022,39(2):231-237.