PSI - Issue 84

Giuseppe Santarsiero et al. / Procedia Structural Integrity 84 (2026) 313–320

319

Third, the number of defect classes actually represented in the dataset was lower than the code catalogue, indicating the need for progressive dataset expansion to achieve comprehensive coverage. Anyway, the YOLO-based defect detection provided the same risk evaluation for the four bridge spans analyzed in this paper, which only requires the detection of the highest severity defect. The presence of several false negatives (FN, defect present but not detected, see Fig. 2) suggests that future developments shall include the expansion of the dataset to additional bridges, refinement of segmentation-based approaches, improved handling of panoramic images, and progressive integration of AI outputs into inspection management systems. An effective and timely management of inspection results achieved through this supporting tool can enhance prompt repair interventions that could be able to improve the bridges’ durability and road network resilience (Santarsiero and Picciano 2023). 5. Conclusions The proposed framework combines deep learning–based damage detection on RC/PRC structural elements with a code-oriented approach aimed at producing structured, inspection-ready outputs in bridge risk assessments. A key contribution of the work is the adoption of a unified defect taxonomy shared across different structural components. This harmonization enabled consistent image annotation and model training, allowing the Computer Vision system to recognize similar degradation mechanisms across beams, slabs, piers, columns, and abutments. The case study on eight PRC girder bridges confirmed that, even when applied to real inspection images with heterogeneous quality, the YOLO-based model was able to correctly identify corrosion-related phenomena and, in most cases, the highest-severity defects governing the Class of Attention as structural risk estimates. The results highlight the potential of AI tools to reduce subjectivity and support inspectors in processing large image datasets, particularly in the most time-consuming phases of defect identification and form compilation. At the same time, limitations related to panoramic views, image quality, and dataset completeness were observed, indicating the need for further developments. Future work will focus on dataset expansion, improved handling of panoramic images, and tighter integration of AI outputs into bridge management systems. Overall, the proposed approach represents a practical step toward scalable, code-compliant AI-assisted bridge inspection, where artificial intelligence acts as a support tool for engineering judgement rather than a replacement. Acknowledgements This research was funded by the High Council of Public Works (CSLLPP) and was carried out as part of the activities envisaged by the agreement between CSLLPP and the ReLUIS Consortium implementing Ministerial Decree 578/2020 and Ministerial Decree 204/2022. The contents of this paper represent the authors’ ideas and do not necessarily correspond to the official opinion and policies of CSLLPP. References Dogan, G., Arslan, M.H., Ilki, A., 2023. Detection of damages caused by earthquake and reinforcement corrosion in RC buildings with deep transfer learning. Engineering Structures 115629. Iraniparast, M., Ranjbar, S., Rahai, M., & Moghadas Nejad, F. (2023). Surface concrete cracks detection and segmentation using transfer learning and multi-resolution image processing. Structures, 54, 386-398. https://doi.org/10.1016/j.istruc.2023.05.062 Karimi, N., Mishra, M., & Lourenço, P. B. (2024): Automated Surface Crack Detection in Historical Constructions with Various Materials Using Deep Learning-Based YOLO Network. International Journal of Architectural Heritage, 1–17. https://doi.org/10.1080/15583058.2024.2376177 Liu, J., Sun, H., Liu, H., Yue, Q., Xu, Z., Jia, Y., Wang, S., 2025. Recognition and quantification of apparent damage to concrete structures based on computer vision. Measurement 115635. Ministero delle Infrastrutture e dei Trasporti (MIT), 2020. Linee guida per la classificazione e gestione del rischio, la valutazione della sicurezza ed il monitoraggio dei ponti esistenti (Decreto n. 578 del 17 dicembre 2020). Ministero delle Infrastrutture e dei Trasporti (MIT), 2022. Linee guida per la classificazione e gestione del rischio, la valutazione della sicurezza ed il monitoraggio dei ponti esistenti (Decreto n. 204 del 1 luglio 2022). Redmon, J., Divvala, S., Girshick, R., Farhadi, A., 2016. You only look once: Unified, real-time object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 779–788.