Rolling bearings are widely employed in rotating machinery. It’s of great importance to conduct the effective bearing fault diagnosis to guarantee the safety of machines. Vibration and sound are homology signals which contain some complementary information in characterizing the health states of machine. In order to make full use of the information in the vibration and sound signals to achieve higher accuracy of fault diagnosis, this paper proposed a vibration and sound fusion convolution neural network (VS-CNN) model. The proposed model adds a 2-D convolution layer before the classical 1-D CNN to automatically extract complementary features of sound and vibration signals and minimize the loss of information. An experiment on a rolling bearing test rig is carried out to verify the proposed VS-CNN method. Vibration and sound signals are collected synchronously at different working speeds and put into the model directly for training and testing. Results show that the proposed method can achieve high classification accuracy for rolling bearing fault diagnosis under nonstationary conditions. Its ability to deal with signals with strong noise is also verified by adding white Gaussian noise manually to the raw data. Compared with the classical machine learning method, the proposed method shows better diagnosis performance with raw signals as the input.

Rolling bearing,Fault diagnosis,Vibration and sound fusion,CNN