MOCVD機台流量控制器之異常製程判斷與老化現象探討

Abstract

流量控制器(MFC)為MOCVD系統之零件，本論文的目標是找出MFC零件異常模式與老化現象。但MFC資料沒有異常製程的判斷標準，於是本篇論文利用統計假設檢定給出異常製程的判斷標準，以及使用羅吉斯廻歸描述此老化現象。統計假設檢定的想法不只能用在參數檢定，也能適當的用在品質管制上，透過給定合理的誤判機率，並設立管制界線當作正常製程的標準，進而判斷不正常的製程，設法從中找出異常問題。而廣義線性模型的羅吉斯廻歸，則能透過建立正常/不正常值與時間的關係，進得到老化機率(或說是風險)的估計，以提供給工程師做為停機更換零件的依據。論文最後亦模擬四種情況，說明這樣的作法會遇到哪些可能的問題。

MFC is one part of the MOCVD system. The goal of this thesis is to figure out the normal/abnormal manufacturing process and aging process of MFC data. However, there is no standard way to identify the abnormal process of MTC data. This thesis uses statistical hypothesis testing to identify abnormal manufacturing processes, and also uses the logistic regression to describe the aging process. The idea of the hypothesis testing not only uses on testing parameters but also performs well on the quality control. By giving an reasonable significant level α, we can construct control limits as what the normal manufacturing process is like, and monitor the process through these limits to figure out something different inside. Logistic regression, a generalized linear model, can build up the relationship between binary response abnormal/normal and times to obtain the aging probability (or risk), which can help engineers decide when to cut down the manufacturing process and change the MFC. At the end, we simulate four conditions to see what we might meet in the future.