Intelligent Fault Diagnosis Based on Receptive Field of DCNN for Rotary Machine under Variable Conditions

Abstract

The trend of methodology in the field of rotary machine fault diagnosis is getting increasingly intelligent. Deep convolutional neural network (DCNN) has been widely applied to intelligent fault diagnosis. However, few of these works have dealt well with intelligent feature extraction and fault diagnosis of machine by DCNN in variable conditions. In this paper, a deep convolutional neural network based intelligent fault feature extraction and diagnosis method is proposed to address the problem. A specialized DCNN architecture with the corresponding training method and diagnostic scheme are proposed, and the receptive field of DCNN is revealed for fault feature extraction. By combining the receptive field of DCNN with the trait of rotary machine vibration signals, the intelligent rotary machine fault diagnosis methodology is implemented. After that, the experiments on measured gearbox vibration signals are conducted to verify the feasibility of the scheme. The results show that the proposed method is capable of fault feature extraction and diagnosis, and it has strong generalization ability and robustness in variable conditions.