| dc.description.abstract | This study presents RAMBIRobot, a remote automated multi-target bolt inspection system for steel structures, integrating the YOLO visual recognition model and tiny machine learning under the ROS framework. RAMBIRobot addresses current challenges in bolt inspection, such as high manual labor dependency, low efficiency, and safety concerns. The system uses a robotic arm with YOLO visual recognition to detect and locate bolts, followed by an audio recognition module with a miniature striking device to determine bolt tightness. The efficient YOLOv5s model accurately identifies bolt positions under various lighting conditions, obtaining 3D spatial coordinates with a depth camera. The audio recognition component employs Mel-Frequency Cepstral Coefficients (MFCC) and a 2D Convolutional Neural Network (2D CNN) for precise bolt tightness determination.
Experiments validated the system′s effectiveness, showing overall audio recognition accuracy of 0.793 (90 degrees), 0.711 (67.5 degrees), and 0.711 (45 degrees), with a loosened bolt detection precision of 0.902 at 90 degrees. The visual recognition module maintained detection accuracy above 90% under various lighting conditions. Integrated system testing indicated RAMBIRobot′s efficiency in remote bolt location and detection, with an overall detection accuracy of 75%.
This study′s innovation lies in integrating the ROS framework, YOLO visual recognition, and audio analysis technologies for efficient, automated bolt detection, reducing reliance on specialized personnel and lowering labor costs. The system′s design ensures ease of application and flexibility in various inspection scenarios. Future expansions could include other types of structural health monitoring, showcasing significant development potential. | en_US |