Bearing fault diagnosis of two-dimensional improved Att-CNN2D neural network based on Attention mechanism

Effective fault diagnosis of rolling bearings is essential to the reliability and safety of modern industry. Although traditional intelligent fault diagnosis technologies such as support vector machines, extreme learning machines and artificial neural networks can achieve satisfactory accuracy, they still rely heavily on expert knowledge and manual intervention in the process of feature extraction and selection. This paper proposes a new method for fault diagnosis of rolling bearings based on CNN deep learning based on attention mechanism. First, in order to make full use of the convolutional network to obtain a larger receptive field, a one-dimensional time series is transformed into a two-dimensional matrix as the input of the CNN network by mapping. The neural network is used to extract the advanced features of the input signal. The high-level features of the output are scored, and the results are finally output through a fully connected classifier. Finally, the model was verified by an experiment. The results show that the improved Att-CNN2D network with the attention mechanism has greatly improved the model generalization ability and obtained higher accuracy than the traditional CNN network.