Research on Fault diagnosis Method of Marine Motor based on WPD-MSCNN

In order to reasonably process the non-stationary vibration data of faulty motor, effectively extract the information of faulty motor, adaptively obtain multi-scale features, avoid the useful information lost by manual feature extraction, and make up for the multi-scale feature extraction capability of CNN, an intelligent fault diagnosis method combining wavelet packet decomposition (WPD) and deep convolutional neural network (CNN) was proposed. Firstly, the vibration signal of the motor is decomposed by WPD, and then the decomposed signal is sent to CNN with a multi-layer structure, so as to extract multi-scale features adaptively, and the extracted features are visualized and analyzed. Finally, the confusion matrix was drawn by the support vector machine (SVM). Based on the motor vibration data of Huazhong University of Science and Technology, the traditional CNN, multi-scale CNN(MSCNN), and WPD-MSCNN are compared. The corresponding average fault identification accuracy is 97.72%, 98.87%, and 99.97%, respectively. Compared with traditional CNN and MSCNN, WPD-MSCNN has better feature extraction ability and better diagnostic performance.