整合光源控制與深度學習辨識之平放膠體散料夾取系統開發;Development of a flat lying plastic micro connector component picking system integrating light control and deep learning recognition

NCU Institutional Repository > 工學院 > 光機電工程研究所 > 博碩士論文 > Item 987654321/92336

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/92336

题名:	整合光源控制與深度學習辨識之平放膠體散料夾取系統開發;Development of a flat lying plastic micro connector component picking system integrating light control and deep learning recognition
作者:	陳尚宏;Chen, Shang-Hong
贡献者:	光機電工程研究所
关键词:	微小電子零件辨識;光源參數調整;YOLOv4;基因演算法;機器視覺;六軸機械手臂;ROS;Recognition of Tiny Electronic Components;Light source parameter tuning;YOLOv4;Genetic algorithm;Machine vision;Six-axis robot arm;Robot Operating System
日期:	2023-10-24
上传时间:	2024-09-19 15:46:46 (UTC+8)
出版者:	國立中央大學
摘要:	少量多樣的電子連接器產品製造和裝配需要大量的人力資源，但廠商面臨嚴重的缺工問題。因此，本研究開發一套取放料系統，使用六軸機械手臂進行膠體零件散料夾取，此系統結合機械手臂、光源控制、機器視覺以及深度學習等技術，分別使用兩種不同作業系統來實現系統的開發，使用 Windows 作業系統來完成環狀光源與控制器的整合，且基於 Linux Ubuntu18.04 版本的作業系統環境，透過機器人作業系統(Robot Operating System,ROS)開發軟體系統，經由 ROS 分散式的架構將實驗用電腦、六軸機械手臂、工業相機及自適應夾爪與深度學習、機器視覺進行軟硬體的整合，以達到散料辨識夾取系統的開發。本論文首先透過基因演算法於調整光源參數，進而增強 YOLOv4 模型辨識膠體正反面的效果，並有效地解決了由於環境因素導致辨識率降低的問題，接著使用六軸機械手臂，將散料膠體依照其正反面準確的夾取至指定的治具上，目前此系統的成功率為94%，成果顯示本論文能夠成功克服原先合作廠商須採用在少量多樣或試產時尚未投入成本建置振動盤進行料件取料問題。;Manufacturing and assembling a variety of electronic connector products in small quantities require a significant amount of human resources, yet manufacturers are grappling with severe labor shortages. Hence, this research presents the development of an electronic connector materials handling system that employs a six-axis robotic arm for the precise retrieval of adhesive components. This study utilized two different operating systems for system development. Windows operating system was employed for integrating the light source and controller, while the Linux Ubuntu 18.04 version operating system environment was used to develop the software system through the Robot Operating System (ROS). The computer, six axis robot arm, industrial camera adaptive gripper, and deep learning and machine vision technologies were integrated by ROS distributed architecture, ultimately achieving the development of a system for picking micro connector component. This research employs genetic algorithm to optimize the parameters of the light source, followed by utilizing the YOLOv4 deep learning model to identify the front sides or back sides of the plastic component. Then use six-axis robotic arm to handle these materials with precision, placing them onto designated fixtures. Currently, this system boasts a success rate of 94%. The results highlight the successful resolution of the challenge faced by collaborating manufacturers, eliminating the need to invest in vibration feeders for scenarios involving small scale, diverse, or trial production.
显示于类别:	[光機電工程研究所 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	3	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....