臺灣博碩士論文加值系統

智慧製造之關鍵中須以數位器材取代以人員視覺辨識及操持工作的過程，主要是須依直覺與經驗做出視覺辨識與決策，包含認知物件類型及方位，目標在於提供下一製程步驟之正確定位。類此人為操作經常因為疲勞與危險之本質，造成管理不易及時程難精算等問題，成為產能掌控的主要障礙。本研究以深度相機拍攝工作區域之點雲畫面，即時辨識工作物件之方位與距離，令後續操作可用於自動連貫進行。實作上將以深度學習方法，預先取得3D工件已知方位之多視角點雲，做為深度學習程式訓練之基礎。訓練完成後將可即時辨識出工件之即時方位與距離。結果顯示所得資料可立即提供如機器手臂等數控工具做即時或離線動作規劃，提升工作效率並大為降低對人員之依賴。

One of the key tasks to achieve intelligent manufacturing is to replace the process of human visual recognition and handling with equivalent digitalized operations. These processes mainly require human visual identification and judgements based on intuition and experience, and the goal is to instantly provides precise positioning for the following operation setups. However, due to the nature of fatigue and danger in tasks, these manual operations are prone to unexpected stops and waits, which compromised the production capacity and thus making the overall management unpredictable. In this study, the depth camera is utilized to capture the point cloud images in the working area, and the precise orientation and position info of the work parts are recognized and extracted in real time, so that the following operations may carry out automatically and coherently by the digital controlled machinery. In practice, a deep learning procedure is employed, where point cloud sample images of the work parts from multiple view angles are pre-obtained, and calculated to get the orientation data of the parts, then the data are used as the basic input for the deep learning program training. After completion, the orientation and distance of the work part can be acquired immediately. The results show that the obtained position can be directly applied to the manipulating tool, such as robots or machine tools, for motion planning, consequently reduce the need of human efforts.

摘要...i
Abstract...ii
誌謝...iii
目錄...iv
表目錄...vi
圖目錄...vii
第一章 緒論...1
1.1 研究背景...1
1.2 研究動機...1
1.3 研究目的...2
1.4 研究限制...2
1.5 章節描述...3
第二章 文獻探討...4
2.1 3D方位識別...4
2.2 深度學習...6
2.3 動作規劃...6
2.4 智慧製造...7
2.5 文獻總結...8
第三章 研究方法...10
3.1 研究架構...10
3.2 研究環境架設...11
3.3 深度相機取得點雲...15
3.5 深度學習訓練...17
3.6 推算工件方位...19
第四章 實作部分...22
4.1 實作設備介紹...22
4.2 實驗點雲資料獲取...22
4.3 點雲資料預處理...23
4.4 深度學習訓練...24
4.5 推算工件方位...36
4.6 實作探討...37
第五章 結論...38
5.1 結論...38
5.2 未來研究...38
參考文獻...39
Extended Abstract...41

Mendoza, J. B. L.、彭彥嘉，2019，「用於軌跡追蹤的移動機械手臂機器人運動規劃」，智慧機器人技術專輯，436期，頁39-45
Ayoobi, H., Kasahi, H., Cao, M., Verbrugge, R., & Verhiji, B., 2022, Local-HDP Interactive open-ended 3D object category recognition in real-time robotic scenarios, Robotics and Autonomous Systems, Vol. 147, pp. 103911.
Bao, Y., Zhang, X., Wang, C., & Ming, X., 2024, Further expansion from smart manufacturing system (SMS) to social smart manufacturing system (SSMS) based on industrial internet, Computers & Industrial Engineering, Vol. 191, pp. 110119.
Floris, A., Frittoli, L., Carrera, D., & Boracchi, G., 2024, Composite convolution A flexible operator for deep learning on 3D point clouds, Pattern Recognition, Vol. 153, pp. 110557.
Fang, T., & Ding, Y., 2022, A sampling-based motion planning method for active visual measurement with an industrial robot, Robotics and Computer-Integrated Manufacturing, Vol. 76, pp. 102322.
Guo, W., Huang, X., Qi, B., Ren, X., Chen, H., & Chen, X., 2024, A novel versatile method for generating machining path directly from point cloud, Journal of Manufacturing Processes, Vol. 120, pp. 15-27.
Huang, T., Yuan, C., Li, P., Xu, S., & Kong, Q., 2023, Spatial displacement tracking of vibrating structure using multiple feature, Engineering Applications of Artificial Intelligence, Vol. 126, pp. 106889.
Liu, T., Kang, H., & Chen, C., 2023, ORB-Livox: A real-time dynamic system for fruit detection and localization, Computers and Electronics in Agriculture, Vol. 209, pp. 107834.
Kler, R., Ashish, Nimmagadda, P., Navarajan, J., Chauhan, D., Babu, G. R., 2024, Recognition and implementation of the smart manufacturing systems in industrial sectors for evolving industry 4.0, Measurement: Sensors, Vol. 31, pp. 100987.
Ma, L., Zhu, J., Yang, L., Liu, H., Fan, Y., & Yang, S., 2024, A point cloud optimization method for accurate line-scan 3-D measurement in motion with dynamic disturbance, Measurement, Vol. 231, pp. 114669.
Rogage, K., & Doukari, O., 2024, 3D object recognition using deep learning for automatically generating semantic BIM data, Automation in Construction, Vol. 162, pp. 105366
Sorathiya, P. C., Singh, S. A., & Desai, K. A., 2023, Mobile-Based Augmented Reality (AR) Module for Guided Operations of CNC Surface Roughness Machine, Manufacturing Letters, Vol.35, pp. 1255-1263.
Yu, T., & Chang, Q., 2022, User-guided motion planning with reinforcement learning for human-robot collaboration in smart manufacturing, Expert Systems With Applications, Vol. 209, pp. 118291.
Zhou, Z., Li, L., Fürsterling, A., Durocher, H. J., Mouridsen, J., & Zhang, X., 2022, Learning-based object detection and localization for a mobile robot manipulator in SME production, Robotics and Computer-Integrated Manufacturing, Vol. 73, pp. 102229.
Zhang, H., Wang, C., Tian, S., Lu, B., Zhang, L., Ning, X., & Bai, X., 2023, Deep learning-based 3D point cloud classification: A systematic survey and outlook, Displays, Vol. 79, pp. 102456.