A multi-scale robotic approach for precise crack measurement in concrete structures

Abstract

This paper introduces a multi-scale robotic approach for measuring the width, length, shape, and profile of hairline cracks in concrete structures. The approach uses a convolutional neural network to identify potential surface cracks, and then robotically navigates a high-resolution laser scanner to measure the detailed shape of the detected cracks. Finally, 3D point cloud registration techniques fuse the laser scans with LiDAR-based scan of the surrounding environment. The proposed method is validated with computer simulations and physical experiments on a concrete specimen. The results are compared against the state-of-the-art, vision-based methods as well as readings of a transparent crack width ruler. The comparison demonstrates the superiority and effectiveness of the proposed multi-scale robotic approach in measuring hairline cracks providing vital data for assessing the conditions of civil infrastructures.