以X-bar管制圖為基之模糊診斷系統

Abstract

本論文提出以管制圖為基礎之製程診斷系統，其目的在於輔助線上操作員判讀製程是否處於穩定狀態。若判斷有異常原因存在時，該系統能告知造成製程不穩定的可能異常原因，以協助操作員快速找出此異常原因並消除之，以恢復製程的穩定狀態。 該系統之診斷流程可分為兩個階段：(1) 由管制圖異常症狀之判讀機制判讀管制圖上的點集與各異常症狀的相關程度; (2) 根據此相關程度及異常症狀與異常原因之連結關係，綜合研判製程穩定與否，若不穩定則列出造成不穩定之可能原因。因此，本論文在第一階段以檢定規則判讀法及異常圖形辨識法兩種最常被業界採用的方法為基，提出模糊檢定規則集與灰關聯辨識器兩種判讀機制。在第二階段提出二種連結異常症狀與異常原因的知識擷取方法:(1)最大相似法;(2)模糊決策樹法。 本文最後提出兩套製程診斷系統。第一套係整合模糊檢定規則集與模糊關係矩陣而成，稱為辨識檢定規則為基之診斷系統。第二套則是整合灰關聯辨識器與模糊決策樹而成，稱為辨識異常圖形為基之診斷系統。

In this dissertation, a control chart based diagnosis system has been developed to assist operators in monitoring whether or not a process is under control. When a process is becoming unstable, the system will give an alarm signal and justify the possible causes of instability. By removing the root cause of instability, the process can be improved. The diagnosis procedure in the system includes two phases: (1) to calculate the relationships between observations and predefined unnatural symptoms. (2) According to the relationships mentioned above and the linkage knowledge between unnatural symptoms and assignable causes, the system will infer whether or not the process is keeping in a control state. If not, then the possible causes of instability might be shown. Therefore, this dissertation firstly proposes the fuzzy rules-test method and the grey relation pattern recognizer to define the relationships between observations and predefined unnatural symptoms. Then, two knowledge acquisition methods: (1) maximum similarity method (MSM) and (2) fuzzy decision tree, are proposed for acquiring the linkage knowledge between unnatural symptoms and assignable causes. Finally, two diagnosis systems are illustrated.