A Parallel Feature Fusion Network for Simultaneous Bearing Fault Detection and Remaining Useful Life Prediction

In prognostics and health management, bearing fault diagnosis and remaining useful life (RUL) prediction are two major research fields but invariably considered independently. However, when the remaining useful life of a bearing is insufficient, it must have failed in some way. Given the characteristics and connections of the two tasks, a parallel feature fusion - convolutional neural network and long short-term memory (PFF-CNNLSTM) network was proposed in this paper. The input of this network is two parallel channels while the output is the fault types and normalized remaining useful life. And the effectiveness of the PFF-CNNLSTM model was verified by the comparison with BP neural network and CNN-LSTM on the IMS dataset.