Apply
the inspection and mapping guidelines in a real bridge use case, ensuring flight safety and data quality.
Test
AI algorithms on real inspection data to verify their capability to detect and classify selected damage classes (i.e., spalling, cracking, corrosion, leaching) with sufficient accuracy.
Evaluate
performance indicators (e.g., duration of survey, risk assessment time, dataset quality).
U-AInSPECT
EUCENTRE is a No-profit foundation (Pavia, Italy) that promotes, and supports training and research in the field of seismic risk mitigation and structural safety. We have an important asset of experimental labs able to reproduce seismic events for testing both structural and non-structural elements. We promote several training programs in collaboration with Pavia’s universities. As a Centre of Competence of the National Civil Protection Department, Eucentre provides emergency support, risk scenarios, and vulnerability assessments. Additionally, Eucentre has an internal drone unit supporting structural and infrastructural assessment and post-emergency rapid mapping. The Foundation actively engages with stakeholders and partners to advance safety engineering, promote a culture of prevention aiming to improve living conditions by enhancing safety and resilience. Overall the U-AInSPECT project is an Enhanced bridge safety and serviceability, crucial for rural areas characterized by a limited redundancy of the local transportation Network. Develop a UAS-based methodology for bridge infrastructure inspection, including the use of innovative procedures based on artificial intelligence algorithms for damage detection. This will result in a comprehensive service value chain, beginning with the infrastructure inspection and concluding with the fast assessment of the bridge attention class, as a basis for supporting proper priority planning for interventions. The U-AInSPECT project applies UAS and AI-based technologies to optimize the inspection of bridge infrastructure in rural areas. In this framework, the related main scenarios investigated are: 1. Structural damage detection 2. 3D bridge reconstruction 3. Evaluation of the bridge attention class (according to the Italian Technical Guideline)
DJI Air 2S
DJI Mini 2
RGB high-resolution cameras with stabilised gimbals and optical zoom, obstacle detection and collision avoidance sensors.
Ground Control Station (GCS) with flight control software for real-time remote flight monitoring and mission planning (i.e., DJI Fly) with the option to extend the inspection monitor (i.e., connecting an external tablet) in order to separate piloting and payload operations, ensuring safer and more efficient inspections.
Deep learning and computer vision algorithms to analyse the collected data (e.g. YOLO).
Web-based GIS application that allows users to fill out the assessment form for the evaluation of the bridge attention class (according to the Italian Technical Guideline). Access to the GIS application requires only a web browser and an internet connection.
Power supply and connectivity, such as portable power banks to sustain field operations and internet connection for support flight operations, access the GIS application and enable data transmission.
SfM photogrammetric software (i.e., Agisoft Metashape) has been used for the generation of 3D point cloud of the bridge together with open source cloud-based software solution (i.e., Potree) for the visualisation and sharing of 3D results. "
Model 1: AI-based damage detection
Purpose: analyses of UAS video frames and/or images to detect and classify selected bridge damage.
Technology: Machine learning algorithms (Deep convolutional neural networks, DCNNs) trained on available datasets integrated with computer vision techniques for accurate detection.
Model 2: 3D reconstruction
Purpose: generate 3D point cloud of the bridge to support structural evaluation.
Technology: Structure from Motion (SfM) commercially available photogrammetric software (i.e., Agisoft Metashape), together with open source cloud-based software solution (i.e., Potree) for the visualisation and sharing of 3D results.
Outcomes
Operative UAS guidelines supporting the optimisation of dataset collection and ensuring survey repeatability (Fast risk assessment, Check lists, Damage assessment, 3D mapping).
Application in a real-world scenario: Reinforced concrete girder bridge inserted on a primary local route in a rural area in Northern Italy, overpassing a river: total longitudinal length: 121m (9 bays with a maximum span length of 17m) superstructure: primary system of 4 longitudinal RC beams and a system of rectangular transverse RC beams with a transversal deck width of about 6.6m.Substructure: RC rectangular piers with a maximum height of about 4.3m.
Application of AI-based algorithm: Deep learning algorithms based on YOLOv8 and YOLOv10 networks trained on two datasets: approximately 5000 real images acquired after recent earthquakes in Italy.
Detection of 4 selected damage classes (most recurrent): concrete cover spalling, cracking, reinforcement corrosion, leaching"
Assessment of the bridge attention class (according to the Italian Guidelines), using UAS-AI based inspection data (e.g. AI-detected damage, 3D reconstruction).
Challenges Faced
Weather sensitivity: changing light conditions occasionally reduced data quality (e.g. potentially influencing image processing accuracy of automated damage detection) and required repeating a few flights.
Environmental conditions: during some flight tests, obstacles such as surrounding vegetation close to the bridge made it difficult to capture complete datasets, especially for point cloud generation and the inspection of hidden areas.
Regulatory constraints: UAS operations needed to be carefully planned in order to avoid or minimize the overflight of uninvolved persons (including the traffic on the bridge). This affected not only the time slot available for inspection but also the choice of the drones in order to comply with European and national UAS regulations.
Technical limitations: minor problems were encountered with the tablet used for flight mission control and data visualization, which occasionally showed delays and overheating under direct sunlight, affecting survey efficiency. Backup devices and shaded working areas have been used to maintain continuity of operations and mitigate this issue.
Skilled human resources: since the service is deployed within a structural safety and inspection framework, the operation mandates the involvement of qualified and certified personnel. Specifically, the system requires a licensed UAV pilot and a payload operator responsible for mission execution and data acquisition, as well as domain experts trained in interpreting the outputs produced by the AI-driven structural damage assessment pipeline. Furthermore, outcomes from AI-based data processing are to be intended as a support for decision making, therefore structural safety assessment must be performed by expert structural engineers."
Impact (Socio-economic & Environmental)
Increased safety and lower inspection costs through drone- and AI-based inspections.
Reduced downtime and better connectivity for rural mobility, agriculture and local economies.
Improved decision-making for infrastructure managers through repeatable and objective assessments.
Lower emissions and environmental impact compared to traditional inspection methods, with less traffic disruption and noise.
Stronger resilience and sustainability of rural infrastructure, supporting long-term rural development."
