Convolutional Neural Network Based on Spiral Arrangement of Features and Its Application in Bearing Fault Diagnosis

With the coming of artificial intelligence and the era of big data, convolutional neural network (CNN) has become one of the research hotspots in many scientific fields. However, there exist serious edge information loss problems in the information transmission process of CNN. Therefore, in order to suppress the loss of important information during the information transmission process, new methods are proposed which use the singular value decomposition (SVD) algorithm based on the phase space reconstruction to analyze the bearing vibration signal. Singular values are regarded as the features for evaluating the bearing's health condition. Then, the features radiate from the center around to form a spiral matrix as the input of the CNN, which can effectively resist information loss problems during the information delivery process. In order to verify the performance of the proposed methods better than the conventional ones, experiments are carried out. This paper shows that the proposed methods have excellent performances in the field of bearing fault diagnosis.