本論文將影像識別實際應用於工廠端在生產製造過程中的高速 線品質監控，透過該機器學習後的影像識別可以快速的判斷出 PCB、焊錫、芯線端的外觀差異，並且依照當前芯線的位置，與焊接 區的 PAD 位置做識別，記錄下每根芯線與 PAD 的偏移狀況。 此項研究利用基恩斯 keyence VHX-X1 的顯微鏡，拍照記錄生產 後高速線產品的影像，透過機器學習的方式，自動判斷出高速線之 芯線端的偏移況狀，以符合國際標準 IPC-A-610 電子組件可接受性 的標準。傳統工廠端在生產高速線時只能運用人工的方式，並且搭 配抽檢檢查產品狀況，將高速線放在顯微鏡底下，人員判讀外觀， 當發現芯線偏移出 PAD 區時，再進行手動量測點選位置，確認產品 是否有超出規範的距離，且一個高速線產品通常都會有十幾個以上 的焊接點位， 傳統人工檢查費時又費力。 傳統 AOI 檢測，需要有標 準片當作範本，但是焊接後的成品每個外觀都不盡相同，故也很難 用 AOI 當作檢測方式。 舉例來說，焊接後的 10 個產品外觀，人員用 CCD 檢查需耗時共 300 秒，其中不包含異常芯線偏移的量測處理時間，如其中有芯線 偏移人員需要點對點人工量測，花費的時間往往是 10 分鐘起跳，但 如用訓練完成後的程式運行檢查，只要大約 60 秒的時間即可判別芯線偏移 Pad 的狀況，並且還會把偏移量紀錄下來。 本論文將利用顯微鏡拍一張照片後，交由 AI 運算分析，找出每 一根個別的芯線實際位置，辨識成功率高達 97.29%， 將芯線成功識 別之後分析並記錄偏移狀況後生成 CSV 檔案，方便人員可以直接應 用檔案做後期的資料分析、數據統計，更可以提供 SECS/GEM 系統做 上傳資料，提供了工業 4.0 更便捷的選擇。;This paper applies image recognition to real-world monitoring in high-speed production lines in factory manufacturing processes. Through machine-learning-based image recognition， differences in the appearance of PCBs， solder， and the wire can be rapidly identified. Additionally， based on the current position of each wire， the system recognizes and compares it with the PAD position in the soldering area， recording the offset condition between each wire and its corresponding PAD. This study uses a Keyence VHX-X1 microscope to capture and record images of high-speed cable products after production. Through machine learning techniques， the system automatically determines the offset conditions of the wire ends of high-speed cables， ensuring compliance with the international standard IPC-A-610: Acceptability of Electronic Assemblies. In traditional factory production of highspeed cables， inspection relies primarily on manual methods combined with sampling inspections. The high-speed cable is placed under a microscope， and operators visually evaluate its appearance. When a wire is found to be offset outside the PAD area， manual point selection and measurement are performed to confirm whether the product exceeds the specified tolerance. Moreover， a single highspeed cable product typically contains more than a dozen soldering points， making traditional manual inspection both time-consuming and labor-intensive. Traditional AOI (Automated Optical Inspection) requires a standard reference sample as a template; however， the appearance of each finished product after soldering varies， making it difficult to use AOI as an effective inspection method. In this paper， a single image captured by a microscope is processed through AI-based computational analysis to identify the actual position of each individual wire. After analyzing and recording the offset conditions， the system generates a CSV file， enabling personnel to directly use the data for subsequent analysis and statistical processing. Furthermore， the data can be uploaded to a SECS/GEM system， providing a more convenient solution in support of Industry 4.0.
