Bearing fault diagnosis with varying conditions using angular domain resampling technology, SDP and DCNN

Abstract

Due to a difficulty in dealing with the nonlinear vibration signals collected from varying speeds by conventional signal process method, it was worse to develop fault diagnosis in this case, so that this paper proposed a bearing intelligent fault diagnosis method under varying conditions by angular domain resampling, Symmetrized Dot Pattern (SDP) and deep convolutional neural networks (DCNN). To further increase the proportion of fault information in signals, CEEMDAN was applied in denoising and reconstructing the angular domain vibration signals, and the denoised and reconstructed signals were visualized by SDP, where the SDP parameters were determined by Pearson correlation coefficient. Moreover, the SDP patterns were formed by mapping the different reconstructed fault signals into a polar coordinate system, and the formed SDP patterns were classified by DCNN so as to realize fault diagnosis. Finally, the effectiveness of the proposed bearing intelligent fault diagnosis method was verified by a bearing vibration experiment, and the superiority was compared with six diagnosis methods. Results demonstrate that the proposed method can diagnose bearing fault effectively under variable conditions, and both the training and testing samples have also achieved a good performance with diagnosis accuracy of 97.14% and 96.00%, and the training samples and testing samples for the outer ring pitting is 100%. The poorest diagnosis accuracies for training samples and testing samples appear in outer ring crack with diagnostic accuracy of 94% and 92%, indicating that the proposed method is one of the most superior diagnosis methods under the same test conditions.