應用貝氏網路於工業製程之診斷與預測

Abstract

工業製程中常常有上百個步驟，每個步驟都會包含相當多的測量項目，所以我們會得到相當龐大的原始數據。而這些數據中包含著一些可能導致製程異常的原因或造成產品異常的因素，因此我們選擇使用聯合樹模型的方法來建立模型，提供工程師或分析師來分析根據聯合樹診斷出來的異常狀態。 本論文以晶圓製程為例，將資料先經過一些處理建立聯合樹模型和應用聯合樹，接著就影響應用的因素作討論，討論樣本數量的影響和變數數量的影響並測試其計算時間跟計算的極限，接著討論被解釋變數分類的問題，最後驗證模型是否正確。 同樣的方法可應用在很多領域上，如類似製程的自動化科技上、大量病歷資料的醫學上、大量數據的氣象上…等等各種需要處理大量資料的領域，都可以應用此方法來進行診斷和預測的工作。

There are hundreds of steps in the process of manufacturing operation. Every step contains lots of measurements. As a result a tremendous amount of data is available. These data maybe contain reasons that case abnormal states of manufacturing process. We use Junction Tree Algorithm to establish Junction Tree Models, in order to provide engineers or analysts to diagnose abnormal states. Take manufacture of silicon wavers for example. After establishing the junction tree model, we use the model to find abnormal states. Then we will discuss factors such as number of cases and variables and the limit of calculation time and the quantization of yield variable. At last we will verify the junction tree model. This method also can be applied to many areas that need to handle a tremendous amount of data, for example in the industrial, medical or meteorological area etc.. We can use this method to diagnose and prognose results after the junction tree is built.