The rapid deterioration of building facades presents substantial safety hazards in urban environments, necessitating advanced, automated inspection solutions. While computer vision (CV) and deep learning (DL) techniques have shown promise for defect analysis, critical gaps remain in achieving real-time, quantitative, and generalizable damage assessment suitable for robotic deployment. Current methods often lack precise metric quantification, struggle with diverse material appearances, and are computationally intensive for on-site processing. To address these limitations, this paper introduces a fully automated, end-to-end inspection framework integrating a wall-climbing robot, a real-time vision-based analysis system, and a digital twin management platform. The primary contributions are threefold: (1) a novel, fully integrated robotic framework for autonomous navigation, multi-sensor data collection, and real-time analysis; (2) a lightweight, synthetic data-augmented DL model for real-time defect segmentation and metric quantification, achieving a mean Average Precision (mAP) of 0.775 for segmentation, an average defect length error of 1.140 cm, and an average center position error of 0.826 cm; (3) a cloud-based digital twin platform enabling quantitative defect visualization, spatiotemporal traceability, and data-driven project management, with the on-site inspection cycle demonstrating a responsive latency of 2.8–4.8 s. Validated through laboratory tests and real building projects, the framework demonstrates significant improvements in inspection efficiency, quantitative accuracy, and decision support over conventional methods.
Song et al. (Wed,) studied this question.