An end-to-end approach for bearing fault diagnosis based on a deep convolution neural network

Abstract

Traditional methods for bearing fault diagnosis mostly utilized a shallow model like support vector machine (SVM) that required professional machinery skills and much of knowledge. Deep models like deep belief network (DBN) had shown its advantage in fault feature extraction without prior knowledge. In this paper, an end-to-end approach based on deep convolution neural network (DCNN) is presented. The approach embodying the idea of end to end diagnosis has only one simple and elegant convolution neural network and don’t need any exquisite hierarchical structure that was used in the traditional methods. The samples of time-domain signals are inputted into the proposed model without any frequency transformation, and the approach can diagnosis bearing fault types and fault sizes simultaneously as output. Experimental researches had shown that the approach has the advantages such as a simple structure, less iteration and real-time, while its accuracy on the diagnosis of fault types and fault sizes can still be guaranteed.