Incipient fault detection of railway point machines using flexible convex hull-based one-class tensor machine under weak-feature condition

Railway point machines are critical devices that control train direction, which directly influences punctuality, operational efficiency, and safety. From a non-destructive testing and evaluation (NDT&E) perspective, this study proposes a data-driven method for railway turnout incipient-fault detection using monitoring-signal images and tensor-based machine learning. Unlike conventional methods, this research specifically addresses scenarios of weak fault characteristics in early-stage railway point machine faults by developing sensitive features and an effective model. Furthermore, the flexible convex hull-based one-class tensor machine (FCH-OCSTM) is proposed, which lowers the computational complexity of conventional one-class support tensor machines. First, time-series monitoring data are collected and preprocessed. Next, image-informed sensitive tensor features are generated using domain knowledge derived from practical detection scenarios. Subsequently, the FCH-OCSTM model is established based on these extracted features to achieve incipient fault detection. The proposed approach is validated on two real-world datasets of operational time-series monitoring data. Experimental results show improvements in recall, precision, and F1 score over the baselines, which confirms the effectiveness of the proposed scheme.