基於ROS的遠端自動多螺栓 檢測機器人系統開發;The Development of a Remote Automated Multi-Bolt Inspection Robot System Based on ROS

NCUIR > Engineering College > Graduate Institute of Civil Engineering > Electronic Thesis & Dissertation > Item 987654321/94496

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/94496

Title:	基於ROS的遠端自動多螺栓檢測機器人系統開發;The Development of a Remote Automated Multi-Bolt Inspection Robot System Based on ROS
Authors:	陳立;Chen, Li
Contributors:	土木工程學系
Keywords:	螺栓缺陷檢測;物件檢測;微型機器學習;ROS架構;機械手臂;結構健康監測;Bolt defect detection;YOLO;tiny machine learning;ROS;robotic arm;structural health monitoring
Date:	2024-07-29
Issue Date:	2024-10-09 14:48:16 (UTC+8)
Publisher:	國立中央大學
Abstract:	本研究提出了一種基於ROS架構，結合YOLO視覺辨識模型與微型機器學習的遠端自動多螺栓檢測機器人系統，名為RAMBIRobot。該系統旨在解決目前鋼結構螺栓檢測方法中存在的人工高依賴度、效率低和安全性不足等問題。RAMBIRobot系統利用機械手臂結合YOLO視覺辨識技術自動檢測並定位螺栓位置，再通過微型敲擊裝置進行音訊辨識，判斷螺栓的鬆緊程度。該系統採用了高效的YOLOv5s模型，能在各種光照條件下準確識別螺栓位置，並且與深度攝影機結合，獲取螺栓的三維空間座標。音訊辨識部分使用了梅爾頻率倒譜係數（MFCC）進行音訊特徵提取，並利用二維卷積神經網路（2D CNN）進行音訊訊號的分析，從而實現對螺栓鬆緊狀態的精確判斷。為了驗證系統的有效性，本研究進行了多組實驗，包括音訊辨識單元測試、視覺辨識模組測試以及整體系統的整合測試。實驗結果顯示，在不同檢測角度下，音訊辨識單元的總體準確率達到了0.793（90度）、0.711（67.5度）和0.711（45度），其中在90度檢測角度下的鬆動螺栓辨識精確度達到0.902。視覺辨識單元在不同光照條件下的檢測準確率均超過90%，展示了其在多種環境下穩定工作的能力。整體系統的整合測試顯示，RAMBIRobot能夠在遠程操作下高效完成螺栓的定位和檢測，總體檢測精度高達75%。本研究的創新之處在於將ROS架構、YOLO視覺辨識與音訊分析技術相結合，實現了對螺栓的高效、自動化檢測。這不僅提高了檢測精度，還減少了對專業技術人員的依賴，降低了人力成本。同時，系統設計考慮了現場應用的便利性和靈活性，能夠適應不同的檢測場景和環境。未來，該系統還可進一步擴展應用於其他類型的結構健康監測，具有廣泛的發展潛力和應用價值。 ;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.
Appears in Collections:	[Graduate Institute of Civil Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	24	View/Open

社群 sharing

Loading...