Intelligent Fault Diagnosis with Multi-scale Convolutional Dense Network

Abstract

Abstract With the continuous development of artificial intelligence technology, intelligent fault diagnosis approaches have been successfully developed and achieved promising performance in recent years. However, in the existing methods, the time domain characteristics of the signal are first ignored in the process of network construction, and at the same time, it is less considered in the aspects of multi-scale feature extraction and feature fusion. In order to solve the above problems, a multi-scale convolutional dense network (MCDN) was established. Specifically, the proposed framework mainly includes three parts, among which the multi-scale feature pre-extraction module is used to extract multi-scale features, the dense connection module is used to achieve effective feature fusion, and the classification module realizes the recognition of different failure modes. To verify the performance of MCDN for fault diagnosis, rolling bearing data sets gathered from Xi’an Jiao Tong University (XJTU) are employed and analyzed. The analysis result confirms that the proposed method can achieve superior performance compared with other latest methods under varying degrees of noise.