A method for rolling bearing fault diagnosis based on GSC-MDRNN with multi-dimensional input

Abstract

The traditional fault diagnosis methods for rolling bearings through neural networks mostly use data sources collected by a single sensor and use single-dimensional data input, leading to fault features in bearings not be completely extracted. Moreover, traditional convolution often uses single-size convolution kernels, which are insufficient for fault feature extraction. In response to these problems, the global shortcut connection (GSC)-multichannel deep ResNet network model is proposed. First, a new residual structure, the GSC, is proposed to fuse two-dimensional and one-dimensional signal features. Second, involution is introduced into the field of fault diagnosis to address the problem of insufficient network feature extraction caused by using single-size convolution kernels. In addition, a convolutional block attention module can adaptively assign the weight of each channel feature to achieve adaptive channel fusion. The verification was performed on the four-category and eight-category data sets collected in the laboratory, and the results show that this method has a high fault recognition rate.