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

NCUIR > Engineering College > Graduate Institute of Mechanical Engineering > Electronic Thesis & Dissertation > Item 987654321/95927

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95927

Title:	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
Authors:	陳昱昇;Chen, Yu-Sheng
Contributors:	機械工程學系
Keywords:	瑕疵檢測;深度學習;RFID;SQL;PLC
Date:	2024-08-02
Issue Date:	2024-10-09 17:24:37 (UTC+8)
Publisher:	國立中央大學
Abstract:	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。;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.
Appears in Collections:	[Graduate Institute of Mechanical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	17	View/Open

社群 sharing

Loading...