透過您的圖書館登入
IP:3.140.242.165
  • 學位論文

結合資料前處理技術、支援向量機(迴歸)與演化式演算法於 TFT-LCD 製程之分析與應用

A hybrid model of data pre-process techniques, support vector machine (regression), and meta-heuristics in analyzing TFT-LCD manufacturing processes

指導教授 : 白炳豐
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


在台灣的兩兆雙星政策中,TFT-LCD 產業列為其中一項主要重點產業,且在國際市場上與韓國、日本三分天下;在TFT-LCD的製程共需經過約一百道的程序,為相當複雜的製造產業,因此在如何提昇產業競爭力上為相當重要的課題,也衍生出相當多的議題值得探討。 研究中先將焦點集中於異常產品發生的主因判斷上。部份異常現象的發生,目前仍然無法得知真正主要原因為何,因此在異常偵測中,透過資料探勘技術的支援向量機、倒傳遞類神經網路及線性鑑別三項技術,分析製程資料以預期能夠找出根本原因,以期望能夠突破製程上的瓶頸。 而在時間序列的量測資料中,目前只能夠以經驗法則判斷量測結果將超出規格,但無法準確地預估發生時間點,故在研究中以支援向量迴歸、倒傳遞類神經網路兩項技術分析資料,以求能夠正確預估量測結果超出規格的時間點。 研究中除比較兩個案例各三項技術的優劣之外,同時更重要的是期待能夠利用研究結果實際應用在TFT-LCD產業中,以達到提昇製程良率、減少成本支出、縮短產品產出時間的目標,並增加企業在產業中的競爭力。

並列摘要


With the establishment of the new government policy, Taiwan has listed TFT-LCD manufacturing as one of its major industry, and subsequently creates high competition within the international market against Korea and Japan. The processes of TFT-LCD are complicated as it has to go through about 100 different stages of processes. Thus, the topic of 'How to increase better the ability of production competitiveness' is vital to further discuss and explore for its other certain relevant issues. Firstly, we start by verifying the main reason why an abnormal product occurred. If we still unable to detect the cause, and then apply 3 data mining technologies, which are SVM, BPNN and LDA, to investigate the anomaly from processing data. Hence, we expect to figure out the obstacle to the problem and resolve it. Secondly, the estimation timing of the measurement data which out of the specification is still according to engineers’ experience for now. However, this estimation isn’t accurate exactly in terms of the exact timing point. Since the measurement data is one kind of time series data, we apply other 3 data mining technologies, which are SVR, and BPNN to estimate the time of measurement results point beyond the specifications. Apart from comparing the 3 technologies between the 2 cases, the research provide important information and results which allow for maximum utilization of the TFT-LCD manufacturing, and thus increasing efficiency and productivity, and reducing raw cost. Subsequently, the competition of this TFT-LCD Manufacturing within the market will be raised.

並列關鍵字

TFT-LCD Manufacturing PCA SVM BPNN LDA PSO SVR

參考文獻


1-1.經濟工業局重大政策, 挑戰 2008 兩兆雙星計劃, http://www.moeaidb.gov.tw/external/ctlr?PRO=policy.PolicyList&cate=226
1-2.工研院 IBK-ITIS 計畫 Jun. 2004
1-3.DisplaySearch 新聞發表,DisplaySearch Monthly TFT-LCD Shipment Database http://www.displaysearch.com.tw/press_releases/20081222.aspx, Nov. 2008
1-4.王信陽,“次世代 TFT 生產線的對應與挑戰”,光連雙月刊,Sep. 2002
1-5.吳玉祥,“TFT--LCD 顯示原理 - 顯示原理製程技術與設備” ,光機電奈米量測系統科技整合研討會,機械系暨機電光研究所,中華技術學院

延伸閱讀