資料探勘 (data mining) 是利用自動或半自動的技術,由資料庫中萃取出有效的、事前未知的,以及潛在有用的資訊,以作為兩種用處,一是了解資料特徵與關係可以提供決策制定與問題解決的依據;二是所探勘出來的資訊可以幫助進行分類與預測。 本研究希望以偏光板在面板端的客訴異常,利用資料探勘技術中的決策樹分析,比較Answer tree軟體的四種演算方式CHAID、Exhaustive CHAID、CART與QUEST,利用其錯誤率進行演算法選定,再從其中進行異常真因與現象的法則歸納,並透過失效模式與效應分析 (FMEA),在嚴重度、發生率、探測度、RPN的級別設定上進行檢討分析,可使得分析知識更有效率與正確性,進而預先防堵,避免無謂之異常造成雙方的品質成本。 供應商品質管理 (supplier quality management, SQM) 可以藉由此套品質異常診斷系統快速分類,進行機器設備與參數設定最佳化與特殊檢驗手法的搭配,並同時水平展開到所有供應商管理系統,降低異常重複發生機率,以期達到SQM的源頭管理、事半功倍。
Data mining uses automatic or semi-automatic technology to extract information which is efficient and predictable. There are two potential utilities of the information, one is to understand the meaning of relationship with data and decision making from problem solving and the other is to help categorize the information. This study investigates the polarizer complaint with decision tree. Compare the risks of CHAID, Exhaustive CHAID, CART and QUEST and analyze through FMEA. Investigate each level of severity (S), occurrence (O), detection (D) and Risk Priority Numbers (RPN) ratings to enhance the efficiency and accuracy of the data. Meanwhile, supplier quality management (SQM) can be predicable and classifiable through this quality system with equipment, optimum parameter setup and specific inspection methodology.