Real-Time Unsupervised Machine Learning Vibration-Based Monitoring of Railway Bridges

The rapid evolution of measurement technologies has facilitated the deployment of distributed sensor configurations capable of continuously observing the dynamic behaviour of civil infra-structure. Although such systems enhance the capability to identify structural deterioration at multiple spatial scales, they simultaneously produce extensive and complex datasets that chal-lenge conventional analysis techniques. This paper presents an automated methodology for real-time interpretation of high-volume response data collected from distributed sensor sys-tems. The proposed system unifies modern data-driven strategies to detect anomalies directly from vibration measurements through unsupervised machine learning. A newly developed mathematical learning scheme enables continuous tracking of system response, allowing ab-normal patterns associated with damage to be identified without reliance on labelled training data. When statistically significant deviations from the learned baseline behaviour are observed, the framework issues an immediate warning signal. The performance of the proposed methodology is examined using numerical simulations of simply supported beam structure under healthy and damaged conditions. The results confirm the reliability and stability of the approach and demonstrate its suitability for real-time structural condition assessment.