Degradation Assessment of Train Axle Bearing based on A Deep Transfer Learning

Train axle bearings which support whole vehicle weight and transmit speed are one kind of key components in railway system. Degradation assessment of train axle bearings is significant for ensuring the safety and reliability of train operation. Deep learning algorithms have been widely applied for machinery degradation assessment using run-to-failure datasets. However, it is hard to collect railway train bearings’ run-to-failure datasets in the real operating condition. In this paper, a deep transfer learning algorithm is proposed to address this problem. In the proposed method, the deep convolution inner-ensemble learning (DCIEL) model is firstly trained by using source domain data and labels. The labelled data in the target domain are then fed into the DCIEL model. The trained model is used to obtain pseudo-labels for unlabeled data in the target domain. Finally, the model for health indicator construction can be obtained by minimizing the loss function. Experiments are conducted to test the proposed method and results verified its effectiveness.