PSI - Issue 84

Giorgia Ghirelli et al. / Procedia Structural Integrity 84 (2026) 1047–1054

1053

Subsequently, a dynamic time-history analysis is performed by applying the train load as a series of moving vertical forces, following the actual axle configuration and speed recorded during the monitoring campaign. Both the train type and the convoy speed were identified through the analysis of the video frames. The convoy is an ETR-350 train consisting of five coaches, with the two end coaches housing the transformers, inverters, and electric motors. Since it was not possible to determine the number of passengers or their distribution among the carriages, only the weight of the empty train was considered, without accounting for any additional passenger load. This assumption is considered acceptable because the bridge is located on a secondary railway line and the monitoring was conducted outside peak hours, when the expected number of passengers is very low. The passage of a train traveling at 50 km/h – consistent with the actual operational conditions observed during the monitoring campaign – is simulated to reproduce the dynamic excitation on the structure. This loading strategy allows the model to simulate the transient response induced by train passage and to reproduce the deformation patterns observed experimentally. The numerical displacement time histories obtained at locations corresponding to the checkerboard targets are extracted for direct comparison with the vision-based results. This comparison enables the evaluation of the consistency between experimental measurements and numerical predictions and provides insight into the capability of the FE model to capture the key features of the structural response.

(a)

(b)

Fig. 4. Numerical approach: (a) finite element model; (b) comparison between numerical and vision-based displacement time history.

As shown in Fig. 4, the positive and negative peaks, as well as the overall displacement trend, exhibit a close correspondence between the vision-based measurements and the FE predictions. Each peak clearly reflects the passage of an individual train coach. A noticeable deviation arises at the passage of the final coach, when the train exits the bridge. This discrepancy is plausibly associated with the FE model, which includes only two spans rather than the full bridge, thereby not fully capturing the global structural interaction with the remaining spans. 6. Conclusion This study explores the applicability of vision-based monitoring for the dynamic assessment of reinforced concrete railway arch bridge subjected to train loads. By integrating high-resolution video acquisition with an advanced feature tracking strategy, accurate displacement measurements can be achieved using consumer-grade equipment. Among the three tracking approaches evaluated, a hybrid method combining continuous tracking with periodic re-identification provides the best balance between accuracy and computational efficiency. Experimental results are complemented by a finite element model simulating a train passing at 50 km/h over two adjacent spans, capturing span-to-span interactions. The comparison between numerical predictions and vision-based measurements shows strong agreement in both peak responses and overall dynamic trends. Overall, the findings demonstrate that vision-based monitoring offers a non-invasive, cost-effective, and reliable alternative to traditional instrumentation, underlining the potential of this approach for future structural health monitoring applications and for extension to more complex bridge structures and dynamic scenarios.