Automated Diagnostic System for Bridges through Artificial Intelligence and Image Processing: Case Study in Lima, Peru

This research addresses the gap in comprehensive automated bridge damage diagnosis systems by developing and implementing a system based on Artificial Intelligence (AI) and image processing to diagnose damage to bridges in Lima, Peru. Using the YOLOv7, YOLOv8, and YOLOv10 models, a comparative analysis was performed regarding accuracy, sensitivity, and mAP. The results show that YOLOv8 achieved a mAP50 of 47% and a mAP50-95 of 32%, significantly outperforming YOLOv10 (21% and 9%, respectively) and YOLOv7 (9% and 3.8%). In addition, YOLOv8 obtained a precision of 60% and a recall of 50%, positioning itself as the most effective model for detecting cracks, corrosion, and concrete spalling. The CRISP-DM methodology was selected for the development process, from collecting a robust dataset of 7,934 images to implementing a web application that automates the diagnosis. The system generates detailed reports and specific recommendations, optimizing efficiency and reducing inspection times by up to 40%. The field validation included 202 images collected from Lima bridges, demonstrating the applicability and reliability of the system in real scenarios. This solution, in addition to improving the safety and sus tainability of infrastructures, represents a significant advance in the automation of structural inspections, promoting the adoption of innovative technologies in civil engineering.