- 學位論文
應用維度縮減與資料精簡技術建構TFT-LCD面板零組件良率預測模式
Using Feature Selection and Data Reduction for TFT-LCD Panel Components Yield Prediction
摘要
生產良率是生產事業的主要競爭力，它涵蓋生產力、品質力、速度性、財務性等範圍，是經營重要成敗關健因素之一。目前TFT-LCD面板的關鍵零組件及材料成本占面板總生產成本的75％~80％多仍依賴進口且替代不易，雖然最根本降底成本的主因為突破材料來源以提高材料自主性，但材料製造技術主要來自日本，國內廠商要突破實為不易。因此，唯有從『改善』製造技術著力，透過提昇製程良率來增加獲利，藉由生產流程的調整或機台參數的改善以提昇生產良率增加生產效能，才能在這艱苛的市場狀況下生存。 本研究主要是藉由TFT-LCD面板的關鍵零組件個案公司，現有的實際製程參數設定資料，針對每道製程，分別以不同的屬性選取進行研究變項分析進而找出影響良率關鍵要素進行探討。 首先由專家工程師依經驗法則所挑選的一組屬性集合，再依專家所挑選的屬性個數做為weka.attributeSelection.GainRatioAttributeEval模組挑選屬性數量的依據，最後匯整參考文獻中所說明影響良率的三大因素，以三種維度縮減方式所建構的研究變項資料為基礎。 在所有的研究變項都計算處理完畢之後，分別將三組研究變項，透過資料探勘的分群技術，將現有的資料按不同製程區隔出差異性較小的資料群。再從差異性較小的資料群當中，按20%隨機抽樣進行資料的精簡，最後透過資料探勘的分類技術，再進行影響良率關鍵要素預測的分析。
並列摘要
Production yield is the main cause of the competitiveness of the production, which contains productivity, quality, speed, financial and other range that is important to the success of operating one of the key factors. This research is mainly by TFT-LCD panel key components of the case company, the existing actual process parameter setting data for each channel processes were selected to study the properties of different variables analyzed further to identify key elements affecting yield discussed. The first of thumb by the expert engineers are selected according to a set of attribute set, then by the number of attributes selected by experts as weka.attribute Selection.GainRatioAttributeEval modules based on the number of attributes selected, the final aggregated impact as described in reference yield the three major factors in three dimension reduction methods research variables are constructed based on the data. All of the research variables calculated after dealing with, three groups of variables, respectively, through data mining clustering techniques, the existing information on business segments according to different processes heterosexual smaller data base. Further information from the smaller group differences among a random sample of 20% for data streamlining, the final classification by data mining techniques, and then the key elements affecting yield prediction analysis.