Deep Learning-based Damage Detection for Structural Health Monitoring of Civil Infrastructure

Ever-increasing concerns in the structural security of modern urban environments reveal the need for intelligent systems able to autonomously and quickly detect damage in civil infrastructure. Manual or non-automatic inspections are time-consuming, onerous, and inclined to subjectivity, while traditional AI-based approaches meet challenges in generality, comprehensibility, and consistent performance in diverse conditions. This paper presents an Artificial Intelligence pipeline integrating the Segment Anything Model and a classic ResNet model classification for smart surface damage identification in civil infrastructure. The designed framework exhibits flexibility, interpretability and computational efficiency, which makes it particularly suited for resource-limited systems and structural health monitoring tasks. Preliminary experiment results on both synthetic and real-world datasets show how robust segmenta-tion before classification improves damage detection accuracy, especially when dealing with complex textures and hard-to-spot defect boundaries. The proposed framework also opens new perspectives for integrating deep learning-based visual inspection within digital twin environments, supporting predictive maintenance strategies and continuous monitoring of structural health conditions.