Bearing Fault Diagnosis Based on Attentional Multi-scale CNN

Abstract

Bearing is an indispensable component of industrial production equipment. The health status of bearing affects the production efficiency of equipment, so it is necessary to detect the health status of bearing in real time. In this paper, a multi-scale feature fusion convolutional neural network with attention mechanism (AMMNet) is proposed for bearing fault diagnosis. Firstly, different scale shallow features of the input signal are extracted by parallel convolutional layers with different kernel sizes. Then, the shallow features are sent to the feature fusion module based on channel attention mechanism. After that, the fused features are fed to the deep feature extractor. Finally, the bearing fault type is identified by the classifier. We introduce a novel dropout mechanism to the input signal to improve the generalization ability of the network. Experiments show that the proposed method has high stability and generalization ability. It can not only achieve high average accuracy in fixed load environment, but also has higher recognition accuracy and better stability than some intelligent algorithms in variable load conditions.