透過您的圖書館登入
IP:18.116.42.208
  • 學位論文

應用決策樹於含檢驗誤差之批量檢驗計劃~C2F6半導體原料為個案分析

Applying Decision Tree to Reduce Costs When Inspection Sampling Errors are Present~ A Case Study of C2F6 Gas in Semi-conductor Manufacturing

指導教授 : 徐旭昇
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


氣體產品是現代工業重要的基礎原料,廣泛應用於化工、石化、機械、電子、食品、核工業等諸多領域。隨著半導體業的蓬勃發展,氣體也為半導體製造不可或缺的流體原料,為半導體製程的重大考量因素。目前供應半導體製程所需氣體均採用全檢方式檢驗氣體中不純物含量。本研究運用決策樹工具搭配貝氏理論作為本研究之研究工具,以最佳經濟成本為考量,在液態氣體產品出貨之前,選擇最好的抽樣計劃,採用貝氏決策方法使得期望損失成本為最小。 研究中以液態氣體產品C2F6 產品為個案,研究方法以決策樹為工具,建立問題決策模式,並以決策樹理論推導出整體決策目標函數,再以C++語言撰寫模式程式。研究結果考量的因素包括:批量大小、單位檢驗成本、賠償損失成本與檢驗誤差等,即可得到一個最佳抽樣數及臨界判斷值之最小期望損失成本為目標之最佳經濟型抽樣檢驗計劃,以此方式將實際之檢驗數據透過自行撰寫之程式來計算最小期望成本,並以此例對模式之相關參數(檢驗成本、期望良率、檢驗誤差、損失成本)做敏感度分析。 本研究搜集個案公司四年資料,使用前三年資料估計模式之驗前分配與檢驗誤差等參數,再與第四年資料做驗証比較。由於現實因素部份資料難以取得,本研究另發展驗証模式以估算所提經濟檢驗模式與實計狀況所產生的成本之差異,結果皆顯示所提模式適用案例公司,能獲得較小期望成本。

並列摘要


Gas products are an important basic material in modern industry and have been widely applied to many industries such as chemical, petrochemical, mechanical, electronic, semi-conductor, nuclear etc. This research studies a quality-and-loss relation problem arising in a semi-conductor manufacturing environment. The upstream supplier provides C2F6 to its downstream manufacturer for thin film and etching. The study develops a Bayesian rectifying inspection sampling model with the objective of minimizing the expected total cost from the producer’s point of view. The factors that influence the total cost include item quality, inspection cost, sampling information, product failure cost, and decision on the remaining items after the sampling. In addition, it is assumed that the inspection is imperfect; that is, both type I and type II errors may occur. A decision tree is used to represent the cost model, and derive the total objective function. The optimal solution is solved by backward dynamic programming. The application of the model is presented through a real world case. To validate the model, we collect four years of data and apply the first three years of data to estimate four model parameters. The estimated parameters are then used to calculate the expected benefit by comparing the results of the fourth year. Additionally, sensitivity analysis is performed to provide reference for the decision-maker regarding the effects of respective model parameters in the problem.

參考文獻


48. 李智明(2004),單期季節性商品之需求預測投資預算-使用貝氏估計法,交大管理學刊,24卷2期,頁37-60。
57. 林章權(2008),逐步加速壽命試驗之貝氏可靠度分析與最佳化設計,中央大學統計學系碩士論文。
51. 黃珮菁(2005),模糊理論在單次抽樣檢驗計畫之應用,成功大學統計學系碩士論文。
54. 洪啟彰(2006),可修復退化系統之貝氏預防保養模式,成功大學工業與資訊管理學系碩士論文。
55. 賴政昌(2007),利用決策樹模式評估 TFT-LCD 產品出貨之最佳處置方案,元智大學工業工程管理研究所碩士論文。

延伸閱讀