By Martina Mandirola 1, Ilaria Enrica Senaldi 1, Chiara Casarotti 1, Igor Lanese 1, Alessio Cantoni 1
1 European Centre for Training and Research in Earthquake Engineering (EUCENTRE foundation), via A. Ferrata 1, 27100, Pavia – Italy
1. Introduction
Rural areas are inherently characterised by limited redundancy in local transportation routes. Therefore, a safe and efficient road network is essential in order to stimulate economic growth, improve living standards and enhance social resilience. The use of unmanned aerial systems (UAS), equipped with appropriate sensors, provides a safe and efficient support to traditional inspection practices (primarily based on visual approaches) of bridge infrastructure, reducing the need for operational personnel to work directly in hazardous environments.
The U-AInSPECT project, developed within the ICAERUS PULL 2[1] funded by the European Union’s Horizon Europe research and innovation programme, aims to define a bridge assessment service in rural areas by integrating drones and artificial intelligence techniques. The proposed operational approach aims to improve inspection efficiency by reducing time, costs and operator risks, while providing decision-making tools for infrastructure maintenance and management.
[1] “Innovations and Capacity Building in Agricultural Environmental and Rural UAV Services”, Grant Agreement No. 101060643.
2. Methodology and Assumptions
The U-AInSPECT project aims to deliver an operational service value chain enabling the transition from data collection during the inspection phase to the preliminary assessment of bridge condition. The project focuses on reinforced concrete girder bridges, as they represent one of the most widespread structural typologies, particularly in Italy. For automatic damage recognition, the machine learning models were trained on the most recurrent damage typologies associated with these structures, namely: concrete cover spalling, cracking, reinforcement corrosion, and leaching.
The operational workflow, outlined in Figure 1, begins with in situ surveys conducted using multirotor drones, which perform visual inspections and photogrammetric mapping for the 3D reconstruction of the bridge. Flight trajectories are carefully planned to optimise data acquisition and facilitate the repeatability and ex-post interpretation of the survey.
The acquired images and videos are then analysed using deep learning algorithms to identify the selected damage classes, followed by human expert validation. In parallel, photogrammetric data are processed to generate a 3D point cloud of the bridge, which is useful for extracting detailed metric and structural information.
All collected datasets are ultimately integrated into a web-GIS platform that guides users in completing the so-called bridge’s “attention class” assessment form, according to the Italian Guidelines issued by the Ministry of Infrastructure and Transport (MIT, 2022). This assessment provides a practical tool for prioritising maintenance interventions and improving risk management.
3. Application and results
The methodology was tested on a real case study: a reinforced concrete bridge located in a rural area of northern Italy. Aerial operations, conducted in visual line-of-sight (VLOS) in compliance with national and European regulations, included missions (mostly carried out in manual control mode) dedicated to 3D mapping and visual damage inspection using two drones of the EUCENTRE fleet (i.e., DJI Mini 2 and DJI Air 2S).
The visual inspection dataset, particularly images and videos, were processed using deep learning algorithms based on YOLOv8 and YOLOv10 networks to detect the selected damage classes (Figure 2). The algorithms were trained on two datasets: the first comprising approximately 5000 real images acquired after recent earthquakes in Italy (IDEA dataset, Senaldi et al., 2025), and the second consisting of semi-synthetic images generated for data augmentation purposes (Dondi et al., 2025).
The flights dedicated to 3D reconstruction collected a dataset of more than 800 images, which were processed using a commercial photogrammetric software based on Structure from Motion (SfM), Agisoft Metashape Pro (Agisoft LLC, 2025), to generate the 3D point cloud of the bridge (Figure 3).
All obtained data were used as input for the assessment of the bridge’s attention class according to the Italian Guidelines (MIT, 2022). Knowing the attention class of a bridge is fundamental to establish inspection and monitoring priorities and to plan maintenance. This classification provides a practical method to estimate potential risk levels based on observable parameters, helping infrastructure managers optimise the allocation of technical and financial resources.
Furthermore, the collected dataset was made available through the ICAERUS Drone Data Analytics Library platform (Zenodo), contributing to the development of drone-based bridge inspections and promoting the sharing of data and methodologies useful for research and innovation in infrastructure monitoring.
4. Conclusions
The U-AInSPECT project confirmed the validity of an integrated approach for the inspection and assessment of bridges in rural areas, based on the combined use of drones, artificial intelligence and 3D reconstruction techniques.
The system was tested on a real case study, demonstrating the operational capability of the proposed workflow and enabling automatic damage detection and bridge attention class evaluation in accordance with national guidelines.
The investigated approach represents valuable support for infrastructure managers, capable of optimising periodic monitoring activities and facilitating post-event inspections in areas that are difficult to access.
Looking ahead, the goal is to extend the methodology to other bridge typologies and to improve the performance of artificial intelligence tools thanks to broader and more diverse datasets.
The integrated adoption of BIM modelling tools, RTK positioning systems and increasingly high-performance drone platforms will make the service progressively more robust, replicable and scalable, supporting large-scale deployment and strengthening the contribution of the U-AInSPECT project to the digitalisation and safety of rural infrastructure.
References
Agisoft LLC. Agisoft Metashape Professional edition Version 2.2.1 (2025). User manual is available on-line: https://www.agisoft.com/downloads/user-manuals/.
Dondi, P., Gullotti, A., Inchingolo, M., Senaldi, I., Casarotti, C., Lombardi, L., & Piastra, M. (2025). Post-earthquake structural damage detection with tunable semi-synthetic image generation. Engineering Applications of Artificial Intelligence, 147, 110302. https://doi.org/10.1016/j.engappai.2025.110302.
MIT Ministry of Infrastructure and Transport, 2022 (in italian): “Linee guida per la classificazione e gestione del rischio, la valutazione della sicurezza ed il monitoraggio dei ponti esistenti”, allegate al parere del Consiglio Superiore dei Lavori Pubblici n.88/2019, espresso in modalità “agile” a distanza dall’Assemblea Generale in data 17.04.2020. Updated versions available online: https://cslp.mit.gov.it/circolari-e-linee-guida/linee-guida-la-classificazione-e-gestione-del-rischio-la-valutazione-della.
Senaldi, I., Casarotti, C., Mandirola, M., & Cantoni, A. (2025). IDEA: Image database for earthquake damage annotation. Data in Brief, 111733. https://doi.org/10.1016/j.dib.2025.111733.
