在電子製造業中,所生產的電子產品上幾乎都會用幾顆LED來表示其狀態等資訊,這些LED種類可能有單一顏色與多色,但隨著產品種類繁多與複雜,LED在生產製造時,經常發生極性反、故障、缺件、或錯件等情形,必需於生產時檢測其功能是否正常以確保產品品質。在業界對於LED的檢測通常是由操作人員目視檢驗,來確認其是否有不良現象;由於是透過人工檢查,所以會有下列問題: (1)因為人為疏失而導致漏測。 (2)無法自動化測試。 (3)人為判定標準不一致。 所以若能提供一具有自動檢測與正確判斷率的系統,對於提升品質與生產效率皆有莫大的幫助。 本研究提出以網路攝影機來擷取電子產品之LED影像,並應用倒傳遞網路模型來分析與檢測LED顯示是否正常。其主要特點為採用低成本的USB網路攝影機作為LED影像擷取裝置,並透過倒傳遞網路模型先將標準的LED影像輸入當訓練樣本,訓練完成的網路模型,就可用來自動檢測待測LED是否正常。當然網路輸入層的特徵定義與整個LED影像訓練樣本篩選方法是關係檢驗結果準確的重要因素,也是本研究所要探討的重點。另外在使用USB網路攝影機的架設與影像參數調整也是檢測成敗的關鍵,本文中亦會特別說明。 本系統經過實際導入產線檢測電子產品之LED,根據實驗之數據顯示其可提供非常穩定且正確之判斷,證明本系統可完全取代測試人員之目視檢驗,進而達成電子產品LED自動化檢測之目的。
Light Emitting Diode (LED) is used generally on electronic products to determine the product status and information by electronics manufacturing. Some LEDs are single color and some are multiple colors. Along with the complexity of producing, component missing, misalignment, wrong polarity, and solder bridge are possible happen on LED. The examination of LED on electronic products is necessary to ensure the quality of products. By human inspection, the testing of LED may cause some problems as following, 1.Human errors 2.Unable to test automatically 3.Inconsistent judgment criterion Developing a fully automatic test system will be a great benefit to increase the quality of products and producing efficiency. In our study, we use web camera to fetch the image of LEDs of product, and apply Back-Propagation Network algorithm to analyze and inspect the correction of LED. The adoption of webcam to get the LED image is economic. A standard LED image is used as a training sample data in BPN. A well-trained network model can be used to determine the accuracy of LED. The feature of input layer of neural-network and the selection of LED image for training purpose are emphasized in this thesis. On the other hand, the set-up of webcam and the tuning of image quality parameters will affect the accuracy of LED test and they will be discussed in this paper as well. Our fully automatic test system is employed in the assembly line practically and the experimental results are stable and correct. Human inspection can be totally substituted by our system and the goal of fully automatic test system of LED is achieved.