A novel method for simultaneous-fault diagnosis based on between-class learning

Abstract Condition monitoring and fault diagnosis are crucial to ensure the safety and efficiency of modern railway systems. The simultaneous fault may lead to catastrophic consequences and can be difficult to accurately detect when components are tightly coupled, which poses particular challenges to the automatic diagnosis. This paper proposes a novel method for simultaneous-fault diagnosis based on the combination of between-class learning and Bayesian deep learning. A modified between-class learning strategy with the multi-label approach is developed for model training. The detection results are obtained through an enhanced estimation method based on Bayesian deep learning, which can capture suspicious samples and identify simultaneous faults. The proposed method can distinguish simultaneous faults from regular faults and identify corresponding fault classes without using simultaneous-fault samples in the training phase. The experiments are conducted for the case of fault detection of high-speed trains, which demonstrates the accuracy and validity of the proposed method.