Bearing fault diagnosis algorithm based on multi-source data and auxiliary classifier

In recent years, with the development of sensor technology and deep learning, intelligent bearing diagnosis algorithms using multi-source data have achieved great success. However, when faced with the input of single sensor data, the deep learning model trained with multi-source data cannot work, which affects the flexibility of model application. Meanwhile, the method of only adding a loss function in the final classification layer is insufficient to restrict each branch model, and will also affect the performance of the model. Based on this, this paper proposes a parallel convolutional neural network with auxiliary classifier. The auxiliary classifiers added in each branch can not only give fault diagnosis results when only a single data is used, but also improve the performance of the final multi-data fusion by optimizing the output features of each branch, and broaden the application range of the diagnosis model. The proposed method is tested on the bearing fault dataset, the experimental results show that, after adding multiple losses, the structural form of the auxiliary classifier can effectively improve the fault diagnosis accuracy and robustness of the parallel convolutional neural network. Meanwhile, the auxiliary classifiers of each branch can also obtain good classification performance when facing single sensor data, which better improves the flexibility of the model.