一般銲錫品質的檢測,常需要特殊光源,例如LED、結構光源,或儀器諸例如X光、超音波、多角度取像。這些設備由於較昂貴,且在檢測能力上並無法有效檢測出所有的焊錫瑕疵,導致業者接受度不高。 對此,本研究提出以機器視覺為基礎的架構下,利用簡單的光學設備擷取待測影像之後,透過未焊錫空板上的銅箔區域影像,執行最小濾波器的運算,便可分割出所有待測影像上的焊錫點。接著量化焊錫點區域的特徵,將特徵分成二值化影像為基和灰階影像為基兩方面,之後透過所建立的分類樹,便可分類焊錫。依業者要求,本研究所定義的瑕疵類別有破洞、未焊、短路三類,加上正常焊錫點。 經過實驗結果發現,本研究利用組間距離和盒方圖所決定的分類樹之分類正確率可以達到97.26%;而利用Clark和Pregibon(1992)改良CART後的方法,正常焊錫雖被完全正確分類,但卻會將短路全部誤判成正常焊錫點。
To inspect the quality of solder joints defects needs some special illumination arrangement, such as LED、structural light, or some special instrument, such as X-ray、ultrasonic images. Because these equipments are expensive and can’t inspect solder joints defects effectively, the application for these illumination techniques is limited. The focus of this study is to provide a solder joints inspection framework based on machine vision. After capturing images which need inspection by a CCD camera, it utilizes the image of copper region on the PCB bare board to apply a minimum-filter, and it can segment all solder joints regions. The selected features of solder joints are calculated for the region. Then the features are separated into two parts: part one is based on the binary images、part two is based on gray-value images. To classify solder joints is using classification tree. In this study, it defines three types of solder joints defects: Open、No solder、Short and regular solder joints. The experiment results showed that using classification tree determined by the distance between groups and box plots, the classification correctness reached 97.2%. And utilizing the method proposed by Clark and Pregibon(1992), the classified correctness of regular solder joints is 100%, but it would classify Short to regular solder joints.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。