傳統的連接器產業中，使用人力將歪斜的導體校正到正確位置是一項極為重要的製程，對整體良率具有關鍵影響。然而，隨著AI伺服器和航太設備等高階產品的出現，這些產品對精度的要求變得格外嚴苛。傳統的人工理線不僅無法提升產能以應對AI技術帶來的需求增長，還難以維持穩定的良率。因此，如何運用AI 技術來替代人力，成為本研究的核心焦點。 過往的連接器相關研究多採用物件偵測技術來識別導體的歪斜情況，但物件偵測在精確定位導體位置上存在不足。此外，傳統的自動光學檢測在辨識顏色相近的物體時效果不佳。針對這些問題，本研究提出結合物件偵測與圖像分割的混合模型，以提升導體抓取的準確性和穩定性，並改善導體與PCB 板上錫膏顏色相近的抓取問題。實驗結果顯示，本研究所提出的混合模型在mIoU 上達到約96%，其推論時時完全符合合產線的實際需求。經由實際產線測試，本研究的方法也證實具備真實應用的可行性。;In the traditional connector industry, manually aligning crooked conductors to their proper positions is a crucial step that significantly affects overall yield. However, with the rise of high-end products like AI servers and aerospace equipment, the precision requirements for these products have become extremely demanding. Traditional manual wiring not only fails to boost production capacity to meet the increasing demands brought by AI technology but also struggles to maintain consistent quality. Therefore, finding a way to use AI to replace manual labor has become the main focus of this research. Previous studies in the connector field have mostly used object detection techniques to identify crooked conductors. However, object detection alone isn’t accurate enough for pinpointing the exact locations of the conductors. Additionally, traditional Automated Optical Inspection (AOI) systems have difficulty distinguishing objects that are similar in color. To tackle these challenges, this study introduces a hybrid model that combines object detection with image segmentation. This approach improves the accuracy and reliability of grabbing the conductors and solves the problem of picking up conductors and solder paste on the PCB board that look alike in color. Our experimental results show that the proposed hybrid model achieves a mean Intersection over Union (mIoU) of approximately 0.96, and its inference time meets the real-world demands of production lines. Tests conducted on actual production lines also confirm that our method is practical and effective for real-life applications.
