Early fault detection in railway point machine via adaptive-weight one-class support matrix machine

Railway point machine is a fundamental element of signalling infrastructure and is indispensable for route selection, schedule adherence, and safety. An early fault detection scheme in railway point machine via adaptive-weight one-class support matrix machine (AWOCSMM) and time series monitoring data is proposed herein. As opposed to existing methods, this study focuses on scenarios with early faults and weak symptoms. These challenges are addressed by constructing sensitive features and developing an effective model. Besides, a novel matrix classifier, AWOCSMM, is developed to solve noise sensitivity issue of the conventional one class support matrix machine. Initially, time series monitoring records are acquired and pre-processed. Field expertise is then applied to convert the data into discriminative matrix features. These features support training of the AWOCSMM, which enables effective identification of early faults. The proposed scheme is evaluated on operational datasets of current and power signals collected from two types of turnouts, ZDJ9 and S700K. In the validation experiments, the proposed scheme achieves higher precision, recall, and F1-scores than all baseline methods, and these results confirm its effectiveness.

© 2026 IOP Publishing Ltd. All rights, including for text and data mining, AI training, and similar technologies, are reserved.