Automated low-cost framework for crack measurements in RC structures using deep learning approach

This research presents a comprehensive and automated framework for detecting surface cracks and measuring their widths in reinforced concrete (RC) members using a modified YOLO-V11 deep learning (DL) architecture. Basically, manual surface-crack inspection is subjective and labor-intensive, relying heavily on an inspector's skill and conditions on-site. This often leads to inconsistent assessments and longer inspection time. Thus, the proposed approach mitigates these limitations by integrating automated crack detection with direct quantitative crack measurement. The proposed framework presents: (1) a DL crack segmentation model trained on a diverse dataset to enhance generalization in realistic inspection conditions for crack detection and segmentation, (2) a crack width measurement algorithm using patching and stitching method, and (3) a customized image calibration and scaling approach to transfer crack dimensions from pixel-size to real-size using low-cost imaging devices and tools. Finally, the proposed framework was validated using 230 measured crack points collected from both experimental specimens and existing RC structures. The prediction accuracy reached a coefficient of variation of 16.82% and a mean relative error of 12.65%, confirming the reliability of the proposed framework for crack measurements of RC structures.