Rolling Bearing Fault Diagnosis Based on Multiscale Block Convolution Neural Network

Abstract

Abstract The working conditions of rolling bearings are always non-stationary, which will degrade the performance of the traditional intelligent fault diagnosis algorithm can achieves better results under constant working conditions. Aiming to the problems mentioned above, a novel fault diagnosis algorithm based on Multiscale Block Convolutional Neural Network (MBCNN) has been proposed in this paper. Compared with other intelligent fault diagnosis algorithms, the proposed algorithm achieves the reuse of effective features while simultaneously extracting global and local fault features end-to-end synchronously. Owing to the original one-dimensional vibration signal can effectively reveal the non-stationarity of bearing fault signal, it is selected as the input of MBCNN and the output of multiple fault categories. Finally, a experiment is conducted to verify the validity of the model, and the advantages of the model are analyzed through the visualization of features. The results show that compared with other methods, this method has higher prediction performance under variable working conditions.