Bearing fault diagnosis method based on data augmentation and MCNN-LSTM

Due to the extensive and critical use of rolling bearings, real-time state analysis of rolling bearings is particularly important to ensure the normal operation of mechanical equipment. Nowadays, intelligent fault diagnosis methods based on vibration signal analysis are widely used in this field, but there are few studies on fault diagnosis for bearing compound faults. This paper proposes a data augmented bearing fault diagnosis method based on a MCNN-LSTM model for compound faults diagnosis. This study superposes bearing signals of single state in frequency domain to obtain signals of compound faults, to increase the number of training sets and improve the generalization ability of the model. The CWRU public data set with data augmentation were used for experiments. The results showed that the proposed method greatly improved the fault diagnosis accuracy of model for single states and compound faults, which proves the effectiveness of this method. In addition, the noise test results show that this method can also be used in the presence of noise.