Structural health monitoring of an arch bridge with artificial intelligence-based anomaly detection implemented in a real-time monitoring system architecture

The Etxebarri arch bridge is a reinforced and prestressed concrete bridge approximately 88 m long, with an arch spanning more than 56 m and extended at each end by a 15.50 m span to cross the Nervión River between the municipalities of Etxebarri and Bilbao (Spain). During the second half of 2025, a monitoring system was installed on this bridge to continuously monitor its structure and improve safety during the railway renovation works that began that year. These works are carried out at night, while during the day the bridge returns to operation. This paper explains the real-time monitoring system implemented, which uses the 4G network and Message Queuing Telemetry Transport (MQTT) communication to send high-sampling data. Accelerations and inclinations are measured on the bridge deck using both high-end wired sensors and, as a supplement, IoT devices. Then, an initial warning system based on the vertical deck acceleration limits of the Eurocode has been established. For this research, the obtained eigenfrequencies are first contrasted with the results given by finite element models of the bridge, and these are compared with information from the opening load test in 2003. Finally, an anomaly detection method based on an unsupervised neural network has been functionally included in the monitoring system. This approach considers the full acceleration time series, treating them as response signatures in the time domain, yielding promising results. This technique, together with the modular real-time monitoring system implemented, constitutes a solution that could be easily adapted to other sensors and structures.