Time series classification-based anomaly detection for train passage health monitoring

Time series anomaly detection has been widely applied in various fields such as healthcare, industrial equipment monitoring, and transportation infrastructure management, becoming an important means of ensuring system safety. In the field of railway bridge health monitoring, this paper analyzes how time series anomaly detection can be used to analyze vibration and stress data during train passages. Drawing on successful experiences from other fields, a detection model based on time series classification is constructed, characterized by the following three points: (1) Using the STL time series decomposition tool to break down the raw data into trend and seasonal components. (2) Extracting key features from the decomposed components, which are then fed into a hierarchical contrastive learning module. (3) Performing prototype learning co-training on the extracted features simultaneously. The proposed method significantly enhances the performance of time series classification, achieving an accuracy (ACC) of 98.78% on the Bridge dataset and 80.17% on the RoadBank dataset, thereby improving the accuracy of anomaly detection and providing safety and reliability support for the health monitoring of railway bridges. Code available at: https://github.com/G-AILab/civil.git