A Multi-dilated Fusion Convolutional Neural Network for Fault Diagnosis of Rolling Bearings

Abstract

It is of great significance to implement real-time and effective condition monitoring and fault diagnosis for rolling bearings. Traditional model-based approaches rely heavily on previous experience and expert knowledge, which hinders their practicality in the industrial field. To cope with this problem, we develop a novel CNN mode called a multi-dilated fusion convolutional neural network (MF-CNN) for bearing fault diagnosis in this study. First of all, a CNN model with 1-D convolutional operations is developed to learn features directly from vibration signals. Then, a multi-dilated fusion module (MDFM) is developed to guide the CNN model to extract features from vibration signals at multiple levels. MDFM adopts dilated convolutions with different dilation rates to enlarge the receptive field. Finally, the MF-CNN architecture is built based on the above improvements. Some experiments are carried out to verify the effectiveness of the proposed MF-CNN. Experimental results suggest that MF-CNN outperforms some state-of-the-art approaches.