液晶顯示器為今應用層面最廣、技術最成熟的平面顯示器，近幾年來，TFT-LCD已經是PC產業中的標準產品，不但如此，更進而擴大到家電市場，成為21世紀顯示器的主流產品。因應此一時代趨勢，在業界便開始擴大產能，開發新產品、開發新技術、開發新材料，不斷地精益求精，為了就是要讓平面顯示器的技術能更能超越以往，普及於社會。TFT-LCD屬於精密光電科技製造業，製造的環境均是在無塵室的環境中進行生產製造，因此，對於品質要求的嚴謹，是錙銖必較的。由於產能不斷的擴大，在一連串的生產製造過程中，如何有效的利用經驗及技術，增加產能減少報廢，快速的解決不良缺陷問題，讓受害減至最小，便是一項重要的課題。因此，要如何將這些經驗及技術，化為有效的對策，縮短時間成本，即成為業界目前所急需要面對的問題。 本研究所要探討的主題是如何有效且快速地找出TFT-LCD G4.5代Cell廠不良缺陷的分類，進而轉化為可利用的知識，為異常發生時，提供一個有效正確的處理。本研究擬以倒傳遞類神經網路建構TFT-LCD G4.5代Cell廠不良缺陷與對策方向的關係，並且以統計的主效果圖及交互影響圖，找出倒傳遞類神經網路參數中的學習速率和慣性因子及學習次數的最佳參數組合，使得網路學習能快速趨於穩定，以利研究進行。 TFT-LCD is the widest used flat panel display in the world. During the past few years, Not only TFT-LCD has become the standard products for the PC equipment in the world but has been expanded to the share of consumptions. Because TFT LCDs are the top fashion products in the world，many producers make more efforts to make their product to become more and more better and wide used. TFT-LCD is a very high technical products, so we must keep manufacturing in the fab to keep their quality is best. TFT LCD have a very long manufacturing procedure so it becomes very important issue to reduce waste and defects to make more perfect yield. To make this we have to use our experience and technician to decrease tact time to get biggest profit during manufacturing path. Our research topic is to use Back-propagation network(BPN) to construct TFT-LCD G4.5 cell process defect analysis to make our technician can do best and most fast solving treatment. We use statistic method to find their Main Effects Plot and Interaction Plot to find Back-propagation network(BPN) best relationship of Learning rate and Momentum factor and Train cycle in their parameters to make net learning can fast become stable to get better research result.
