High-Speed Train Brake Pads Condition Monitoring Based on Trade-Off Contrastive Learning Network

Abstract

The braking system of high-speed trains is directly related to the operation safety of the train. The brake pads, which play a crucial role, will inevitably undergo uneven wear in long-term use, posing safety hazards to train braking. As the trains are in normal operating condition for long periods, it is difficult to collect usable uneven wear data, and there is a situation of data imbalance. This article proposes a trade-off contrastive learning network (TCLN), utilizing the differences between data and balancing the weights of different classes, which can realize the condition monitoring under the data imbalance of brake pads. First, data augmentation is employed to provide sufficient and diverse data for contrastive learning, and nonlinear features are extracted by a quadratic convolutional neural network (QCNN). Then, the designed class-weighted method is utilized to improve the characterization ability of the minority class data and realize the equidistant representation of features for each class, which in turn achieves the purpose of paying equal attention to all classes. Finally, the effectiveness of the proposed method is verified using the dataset collected from the scaling experiments, and the results show that the proposed method has higher accuracy and efficiency compared to other methods, which can still accurately identify the brake pad condition when the data are highly imbalanced.