基於人工智慧之車殼瑕疵檢測系統初步研究

以作者查詢圖書館館藏

、以作者查詢臺灣博碩士

、以作者查詢全國書目

、勘誤回報

、線上人數：89

、訪客IP：18.119.163.95

姓名

陳昱昇(Yu-Sheng Chen) 查詢紙本館藏

畢業系所

機械工程學系

論文名稱

基於人工智慧之車殼瑕疵檢測系統初步研究
(Preliminary Study on a Car Body Shell Defect Detection System Based Artificial Intelligence)

相關論文

★ 自動平衡裝置在吊扇上之運用	★ 以USB通訊界面實現X-Y Table之位置控制
★ 液體平衡環在立式轉動機械上之運用	★ 液流阻尼裝置設計與特性之研究
★ 液晶電視喇叭結構共振異音研究	★ 液態自動平衡環之研究
★ 抑制牙叉式機械臂移載時產生振幅之設計	★ 立體拼圖式組合音箱共振雜音消除之設計
★ 電梯纜繩振動抑制設計研究	★ 以機器學習導入電梯生產結果預測之研究
★ 新環保冷媒R454取代R410A冷媒迴轉式單缸壓縮機效能分析與可靠性驗證	★ 高速銑削Al7475-T7351的銑削參數與基因演算法研究
★ 自動化鞋型切削機之設計與實現	★ 以FPGA為基礎之精密位置控制IC
★ CNC三維圓弧插補器	★ PID與模糊控制在營建工程自動化的探討

檔案

[Endnote RIS 格式]

[Bibtex 格式]

[相關文章]

[文章引用]

[完整記錄]

[館藏目錄]

至系統瀏覽論文 (2029-7-1以後開放)

摘要(中)

在當前工業 4.0 與智慧製造的發展趨勢下，實現生產線的自動化和品質
控制是提高效率與產品品質的關鍵。本研究的目標是整合 RFID 技術和深度學習的 AI 瑕疵檢測系統，並在合作汽車製造廠商的生產線上進行實施。RFID 系統架構與實驗建置包括 RFID 標籤和讀取器的配置，以及與 PLC 控制和資料庫管理系統的整合。這個系統實現了車殼在生產過程中的即時監控與管理，大幅提升了物流與生產流程的協同效率。此外，本研究開發了一套 AI 瑕疵檢測系統，專門用於檢測汽車漆面上的點不良和面不良等瑕疵。透過使用高解析度相機來採集圖像、自定義數據集與數據擴增進行Cascade R-CNN 模型訓練，這套系統採用了多階段級聯檢測框架，每個階段逐步優化檢測結果，從而提高瑕疵檢測的準確率和效率。
研究結果顯示，所提出的瑕疵檢測模型在使用少量圖像數據集進行訓練
的情況下，對於瑕疵的精確率與召回率均達到 90%，這表明模型能夠有效
地檢測出汽車漆面上的各種瑕疵，並且在有限數據集的情況下，仍具有較高
的準確率。

摘要(英)

In the context of the current trends in Industry 4.0 and smart manufacturing, achieving production line automation and quality control is crucial for enhancing efficiency and product quality. This study aims to integrate RFID technology with a deep learning-based AI defect detection system and implement it on the production line of a partner automotive manufacturer. The RFID system architecture and experimental setup include the configuration of RFID tags and readers, and integration with PLC control and database management systems. This setup enables real-time monitoring and management of car bodies during the production process, significantly improving the coordination efficiency of
logistics and production workflows. Additionally, this research developed an AI defect detection system specifically designed to detect defects on car paint
surfaces, such as particles and surface. Using high-resolution cameras to capture images, custom datasets, and data augmentation, the Cascade R-CNN model was trained within a multi-stage cascade detection framework, with each stage progressively refining the detection results, thereby enhancing the accuracy and efficiency of defect detection.
The experimental results indicate that the proposed defect detection model, when trained on a dataset with a limited number of images, achieved a precision and recall rate of 90% for defect detection. This demonstrates that the model can effectively identify car defects on car paint surfaces and maintain high accuracy even when trained with a limited dataset.

關鍵字(中)

★ 瑕疵檢測
★ 深度學習

關鍵字(英)

★ RFID
★ SQL
★ PLC

論文目次

目錄
摘要........................................................................................................................i
Abstract .................................................................................................................ii
誌謝......................................................................................................................iii
目錄......................................................................................................................iv
圖目錄..................................................................................................................vi
表目錄..................................................................................................................xi
一、緒論 ......................................................................................................... 1
1-1 研究動機與目的 ................................................................................... 1
1-2 文獻回顧................................................................................................ 3
1-3 論文架構................................................................................................ 6
二、實驗和方法............................................................................................. 7
2-1 智慧工廠.............................................................................................. 7
2-1-1 RFID 系統原理................................................................................ 7
2-1-2 RFID 實驗環境................................................................................ 9
2-1-3 RFID 實驗設備.............................................................................. 10
2-1-4 PC 即時監控與管理介面 .............................................................. 15
2-2 AI 瑕疵檢測系統 .............................................................................. 20
2-2-1 圖像採集系統 ................................................................................ 21
v
2-2-2 自定義數據集 ................................................................................ 23
2-2-3 圖像預處理 .................................................................................... 24
2-2-4 模型架構 ........................................................................................ 26
2-2-5 LOSS............................................................................................... 42
2-2-6 優化器 ............................................................................................ 45
2-3 訓練、驗證、測試............................................................................ 45
2-4 評估指標............................................................................................ 46
三、結果與討論........................................................................................... 47
3-1 不同模型架構的性能比較 ................................................................. 47
3-1-1 Cascade R-CNN ............................................................................. 47
3-1-2 Backbone 的選擇........................................................................... 55
3-1-3 不同的 Anchor 設計...................................................................... 62
3-1-4 優化器選擇 .................................................................................... 69
3-2 模型最佳架構與測試結果 ................................................................. 77
3-3 模型對正常圖像的評估 ..................................................................... 80
四、結論 ....................................................................................................... 83
五、未來展望............................................................................................... 85
六、參考文獻............................................................................................... 86

參考文獻

[1] Z. Ren, F. Fang, N. Yan, and Y. Wu, “State of the art in defect detection based on machine vision,” International Journal of Precision Engineering and ManufacturingGreen Technology, vol. 9, no. 2, pp. 661–691, 2022.
[2] J. Zhang, M. Kikuta, C. Zhang, A. Ito, K. Anmi, Y.Matsui, and T. Akashi, “Defect inspection of coated automobile roofs using a single camera,” IEEJ Transactions on Electrical and Electronic Engineering 15(4), 616-625, 2020.
[3] F. Chang, M. Liu, M. Dong, and Y. Duan, “A mobile vision inspection system for tiny defect detection on smooth car-body surfaces based on deep ensemble learning,” Measurement Science and Technology 30(12), 2019.
[4] Z. Jiang, X. Hu, and S. Wang. “Image Classification of Car Paint Defect Detection Based on Convolutional Neural Network,” Journal of Physics Conference Series 2456(1):012037, 2023.
[5] L. Luo, C. Deng, Z. Wu, S. Wang, and T. Ye, “Automobile airbag defect detection algorithm based on improved Faster RCNN,” 2021 International Conference on Computer Engineering and Artificial Intelligence (ICCEAI), Shanghai, China, pp. 281-285, 2021.
[6] S. Houming, W. Zhen, F. Peng, Z. Chun, X. Tao, D. Qin, X. Jiajun, and L. Jinjuan, “A Detection Method of Self Explosion Defect of Transmission Line Insulator Based on Cascade R-CNN,” 2022 5th International Conference on Electronics and Electrical Engineering 87 Technology (EEET), Beijing, China, pp. 71-75, 2022.
[7] Y. Cao, Y. Wang, H. Feng, and T. Wang, “Method for detecting surface defects of ceramic tile based on improved Cascade RCNN,” 2022 4th International Conference on Frontiers Technology of Information and Computer (ICFTIC), Qingdao, China, pp. 41-45, 2022.
[8] C. -H. Li, K. -W. Lao, and K. -W. Tam, “A Flooding Warning System based on RFID Tag Array for Energy Facility,” 2018 IEEE International Conference on RFID Technology & Application (RFID-TA), Macau, Macao, pp. 1-4, 2018.
[9] G. Kasanga, and J. Phiri, “Passenger and Luggage Tracking System Using Sensor Networks for Public Transport,” The International Conference in Information and Communications Technologies, 2019.
[10] T. Karygiannis, B. Eydt, G. Barber, L. Bunn, T. Phillips, “Guidelines for Securing Radio Frequency Identification (RFID) Systems,” National Institute of Standards and Technology, 2007.
[11] R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation,”2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, pp. 580-587, 2014.
[12] R. Girshick, “Fast R-CNN,” 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, pp. 1440-1448, 2015.
[13] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards RealTime Object Detection with Region Proposal Networks,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 2017.
88
[14] K. He, G. Gkioxari, P. Dollár, and R. Girshick, “Mask R-CNN,” 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, pp. 2980-2988, 2017.
[15] Z. Cai, and N. Vasconcelos, “Cascade R-CNN: Delving Into High Quality Object Detection,” 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 6154-6162, 2018.
[16] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, pp. 770-778, 2016.
[17] S. Xie, R. Girshick, P. Dollár, Z. Tu, and K. He, “Aggregated Residual Transformations for Deep Neural Networks,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, pp. 5987-5995, 2017.
[18] T. -Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature Pyramid Networks for Object Detection,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, pp. 936-944, 2017.
[19] L. Peng, H. Wang, C. Zhou, F. Hu, X. Tian, and Z. Hongtai, “Research on Intelligent Detection and Segmentation of Rock Joints Based on Deep Learning,” Advances in Civil Engineering 2024, 8810092, 13 pages, 2024.
[20] S.J. Pan, and Q. Yang, “A Survey on Transfer Learning,” IEEE Transactions on Knowledge and Data Engineering, Vol. 22, No. 10, pp. 1345-1359, 2010.
[21] A. Kolesnikov, L. Beyer, X. Zhai, J. Puigcerver, J. Yung, S. Gelly, and N. 89 Houlsby, “Big Transfer (BiT): General Visual Representation Learning,” arXiv, 1912.11370, 2020

指導教授

董必正(Pi-Cheng Tung)

審核日期

2024-8-2

推文