The integration of the shotcrete system with Ultra-High Performance Concrete (UHPC) to reinforce deficient concrete structures has been recognized as having significant potential. However, the execution of manual spraying operations within confined spaces or hazardous environments presents considerable risks, thereby highlighting the pressing necessity for implementing robotic solutions. This research focuses on developing an AI-powered shotcrete spraying robot prototype to address these challenges. Integrating ultra-high performance concrete (UHPC) and visual recognition techniques, the robot is designed to identify and repair concrete defects in real-time automatically. Defect detection is conducted using Google's MobileNetV1, a convolutional neural network (CNN)-based algorithm, which provides results comparable to standard CNNs with lower computational costs. The findings of this study indicate that the robot can precisely identify and reinforce defects using advanced UHPC materials, which offer superior mechanical properties and durability compared to conventional shotcrete. This performance is substantiated by F1-scores exceeding 90% for both training and testing datasets, with the Float32 model outperforming the Int8 model, yielding an impressive 98.0% accuracy compared to 94.2%. These results underline the feasibility and practicality of the proposed AI-powered shotcrete robot concept. This technological advancement could revolutionize concrete defect repair in challenging environments, such as large-span bridges, dams, and high-risk areas, enhancing safety and efficiency. Lastly, this prototype presents significant potential for future automatic concrete defect repair advancements.
Read full abstract