This study introduces a multi-scale robotic inspection framework designed to detect and quantify hairline cracks in concrete through the integration of computer vision, laser scanning, and robotic control. The workflow begins with a convolutional neural network (CNN) that identifies surface cracks from visual data. These detected regions are then projected into 3D space, where a UR3 robotic arm autonomously guides a high-resolution laser line scanner to capture fine geometric details such as crack width and depth.

The system further fuses laser data with LiDAR-based surface mapping to reconstruct a precise 3D digital twin of the damaged area. The approach is validated through both simulation and laboratory experiments, achieving sub-millimetre accuracy and high repeatability. This multi-scale integration bridges global surface mapping with local defect quantification, providing a scalable methodology for robotic NDE of concrete structures.

A conference paper presenting this work has been published, and an extended journal version is currently being prepared, highlighting improvements in 3D fusion accuracy and adaptive robotic trajectory control.