DETECTING FLAT ROOF DEFECTS WITH MACHINE LEARNING AND DEEP LEARNING TECHNIQUES

Abstract

Deep learning has emerged as a transformative approach for detecting structural damage and deformations, particularly for flat roofs and large-scale infrastructure. This article synthesizes recent progress in applying convolutional neural networks (CNNs), segmentation models, object detectors (YOLO, Faster R-CNN), and autoencoders for unsupervised anomaly detection. Drones (UAVs), thermal imaging, and vibration sensing all contribute critical data. By training on images or signals indicative of healthy vs. damaged conditions, deep models can locate cracks, spalling, missing fasteners, or stiffness loss at high speed and with impressive accuracy - often above 85%. A review of more than 300 publications indicates that remote inspection with AI can drastically reduce manual labor and improve the consistency of damage identification, even in hazardous or inaccessible areas. A summary table compares deep learning effectiveness across beams, walls, decks, roofs, and other structural compo-nents. Real-world deployments on bridges, high-rise facades, and post-disaster zones confirm that deep learning, coupled with UAV-based inspections, can accelerate maintenance work-flows, detect subtle defects, and reduce safety risks. Ongoing challenges include data scarcity for rare failure modes, generalizing models to new environments, and the integration of physics-based reasoning. Recommendations for future research involve fusing multispectral data, automating calibration of deep models, and embedding AI in digital twins for continuous structural health monitoring.