Artificial intelligence-driven intelligent concrete crack detection and real-time width estimation for smart structural health monitoring systems: A review

It is partly the concrete infrastructure degradation (mainly crack formation and crack propagation) that is one of the most significant challenges facing the contemporary civil engineering. The traditional inspection procedures, which are based on a regular manual survey, non-destructive testing (NDT) guidelines, and an elementary image processing pipeline, have proven their constant incapacity to achieve sensitivity, scalability, and real-time reactivity. This combination of deep learning, computer vision, and embedded edge computing has also been the driving force behind a paradigmatic shift to the creation of intelligent, autonomous structural health monitoring (SHM) systems with the capability to operate in continuous, high-fidelity crack detection and quantitative width measurements. The present paper is a critical, interdisciplinary survey of the current state-of-the-art in AI-driven concrete crack detection that includes the following architectures, object detectors, semantic segmenters and detectors, and emergent paradigms of vision transformer (ViT) and vision-language paradigms (CLIP). Specific analytical focus is given to real-time crack width estimation techniques - such as pixel-to-millimetre calibration, stereo vision triangulation, laser profilometry, and Structure-from-Motion (SfM) 3D reconstruction techniques and their corresponding error metrics and deployment limits. The review also discusses how these AI components are integrated into smart SHM systems, which include edge AI systems, IoT data pipelines, cloud-edge hybrid systems, drone-based inspection systems, and digital twin systems. The critical research gaps that have been observed are lack of standardised benchmark datasets, lack of rigorously validated real time deployment studies, lack of cross-domain model generalisation, and unsolved hardware-software co-design issues.