Multi Scale Inspection
about.
We present a multi-scale robotic system that finds and precisely measures hairline cracks in concrete. A CNN first detects surface cracks; those regions are back-projected into 3D, then a robotic arm guides a high-resolution laser line scanner to capture crack shape/width. Finally, laser scans are fused with a LiDAR map to build a 3D digital twin of the surrounding structure. The approach is validated in simulation and lab tests and benchmarked against vision-based methods and a transparent crack ruler.
challenge.
Manual inspections are variable and labor-intensive, and full high-resolution coverage of large assets is impractical in time, power, and cost.
Pure vision methods struggle to measure crack shape/depth—photogrammetry/stereo are sensitive to texture, lighting, and calibration.
Inspectors need a way to prioritize ROIs for fine scanning while maintaining a global context of the structure.
results.
End-to-end pipeline: U-Net crack segmentation runs fast enough for real-time use; ROIs are converted from 2D to 3D for targeted robotic scanning.
Micron-level measurement: Experimental protocol produced 12,800 scan slices over 6 cm with ~0.004 mm slice resolution.
Hairline cracks quantified: Measured widths 0.02–0.14 mm at different locations; readings checked against a transparent crack ruler.
Minimum detectable width: Comparative table shows this method achieves < 0.01 mm, outperforming prior laser-based approaches.
Context + detail: Depth cameras alone were not suitable for tiny cracks; LiDAR gives a holistic map while the laser provides precise local geometry—then both are fused.
testimonial.
“This system lets us see the whole structure and the hairline details at once. We flag a suspect region, the robot scans it at ~0.004 mm resolution, and the results snap into our 3D map. It’s faster, more consistent, and far more informative than manual crack checks.”
Ali Ghadimzadeh
Ph.D. Candidate, Mechanical Engineering
