ABSTRACT This study proposes a novel sequential classification algorithm based on a one‐class support vector machine (OC‐SVM) for detecting crossing‐gate rod breakage using time‐series data from the railway telemeter system operated by the Shikoku Railway Company (JR Shikoku), Japan. The algorithm is specifically designed for real‐time anomaly detection by incorporating sequential score monitoring and a threshold‐based alert mechanism. The proposed method uses OC‐SVM to classify the operational state of the rod without requiring failure labels. Sequential classification enables continuous inference, and alerts are triggered when the moving average of the classification‐difference score exceeds a dynamically updated threshold. The threshold is computed online through recursive updates of the mean and standard deviation. The method was evaluated using six rod breakage cases at five locations. Among them, four cases were successfully detected, whereas the two undetected cases were likely due to discrepancies between the actual breakage time and the reported alert time—highlighting a limitation in the evaluation framework. These results demonstrate that the proposed algorithm is suitable for real‐time deployment and contributes to the advancement of condition‐based maintenance (CBM) in railway infrastructure.
Kashiwao et al. (Thu,) studied this question.