AI-Powered Digital Twinning and Monitoring of Concrete Dams with Drones and Photogrammetry

Abstract

Concrete dams are critical infrastructure assets that deteriorate over time, making effective monitoring and inspection essential. Traditional manual inspections are often time-consuming, costly, and risky. In contrast, drones provide a faster, safer, and more reliable solution for structural health monitoring. This research develops a digital twin of the dam using aerial imaging and photogrammetry techniques. Creating a highquality 3D model presents challenges due to variations in lighting conditions, camera motion, and shooting angles, which can result in stitching errors, distortions, and blurriness. To address these issues, a preprocessing pipeline is introduced, including image filtering, outlier removal, mesh smoothing, illumination normalization, and local feature enhancement. For damage detection, a human-AI collaboration approach is employed, using an iterative procedure that reduces annotation efforts and improves detection accuracy through human supervision of small data batches. To map detected damage onto the 3D model, a texture updating method is proposed, which avoids the need for model reconstruction by using damage-segmented images or a virtual camera to recapture imagery. Furthermore, a comprehensive dam inspection platform is developed, featuring a user-friendly application that visualizes the 3D model, analyzes and measures various types of damage, integrates GPS coordinates for spatial context, and generates detailed inspection reports. This robust system provides a practical and effective solution for automated dam condition assessment, documentation, and management.