A Fault Detection Method for Railway Turnout with Convex Hull-based One-Class Tensor Machine

Railway turnout is critical equipment for changing trains' directions, which directly impacts the safety and efficiency of operation. This study presents a novel abnormal detection method for railway turnouts with convex hull-based one-class tensor machine (CH-OCSTM) and monitoring signal images. As opposed to existing methods, it fully preserves the spatial structure and profile information in both data processing and model-building processes. Besides, a novel tensor-form classifier called CH-OCSTM is developed to improve the one-class support tensor machine (OCSTM)'s limitations in high computing complexity. First, the one dimensional original time-series signals are converted into two-dimensional images by data preprocessing and 2D representation. Next, the feature tensor is calculated by the CANDECOMP/PARAFAC decomposition method with the curve image data. Then, the CH-OCSTM model is built with the extracted feature tensor to implement the abnormal detection. The performance of the proposed method is evaluated and tested on two real-world operational current and power datasets. Experimental results show that the proposed method performs better than other existing approaches in accuracy and recall.