Research on Bearing Fault Diagnosis Based on DS Evidence Theory and Convolutional Neural Network

In order to solve the problem that the accuracy in bearing fault diagnosis based on single sensor is not high, convolutional neural network and DS evidence theory are combined to fuse the information of multiple sensors. The sensor types include vibration acceleration at the bearing and drive motor stator current. The neural network is trained separately for each sensor data and features are extracted adaptively. The model output of the neural network is mapped into a probability distribution by softmax method, and this probability distribution is used as the basic probability assignment in DS evidence theory, and the final decision fusion is carried out by the Dempster-Shafer evidence fusion. The algorithm is validated through experimental data, and the results show that the method provides better diagnostic results than any single sensor, and the accuracy of normal working condition, inner ring fault and outer ring fault identification reached 99.5%, 96.5% and 99.1% respectively