Autonomous Concrete Crack Monitoring Using a Mobile Robot with a 2-DoF Manipulator and Stereo Vision Sensors

Area of Science:

• Robotics and Automation
• Structural Health Monitoring
• Computer Vision

Background:

• Maintaining structural integrity in concrete structures necessitates effective crack monitoring.
• Existing methods may lack efficiency or comprehensive mapping capabilities over large areas.

Purpose of the Study:

• To propose and validate a mobile ground robot system for autonomous crack monitoring and mapping in concrete structures.
• To enhance crack detection and measurement accuracy using advanced robotics and deep learning.

Main Methods:

• A mobile robot equipped with a 2-Degrees-of-Freedom (2-DoF) manipulator, stereo vision sensors, and a manual rotation plate was utilized.
• Deep learning algorithms, including YOLO (You Only Look Once) v6-s for detection and SFNet (Semantic Flow Network) for segmentation, were employed.
• Synthetic image generation, preprocessing, and median absolute deviation filtering of point clouds were applied for enhanced crack analysis.

Main Results:

• The system achieved autonomous crack detection, segmentation, and dimension calculation.
• Crack propagation direction was predicted, enabling precise robotic manipulation.
• Combined 3D point clouds from multiple frames allowed for total crack length and width calculation with a maximum relative error of 1%.

Conclusions:

• The proposed robot system demonstrates a viable and accurate solution for autonomous crack monitoring and mapping.
• The integration of robotics, deep learning, and stereo vision significantly improves the efficiency and precision of structural health monitoring.
• The method shows potential for practical application in infrastructure maintenance and safety assessment.