A Rolling Bearing Fault Detection Method Based on Heterogeneous Fusion Deep Neural Network

To research the problems of the rolling bearing fault diagnosis under variable loads, a dual-input model of rolling bearing fault detection based on heterogeneous fusion deep network was proposed, which was trained based the data set after incremental processing by the sliding window frame. The accuracy of the model was 96.05% in the test set. Compared with single convolution branch network and non-convolution branch network, the proposed method not only ensured the excellent accuracy in test set, but also enhanced the stability of training process. In addition, the test accuracy of the model under different loads was about 96%. The proposed model achieved a high fault recognition rate under variable load as well as satisfactory load adaptability.