透過您的圖書館登入
IP:18.218.209.8
  • 學位論文

應用資料探勘技術於自相關製程中即時偵測管制圖異常形狀之研究

Real-time Pattern Recognition of Control Charts Patterns in Autocorrelated Process by a Data Mining Based Approach

指導教授 : 顧瑞祥
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


工業界將統計製程管制(statistical process control, SPC) 方法廣泛的應用於製程監控的領域,主要的目的是在監控製程是否存在可歸屬原因 (assignable causes) 所產生的變異,以便提早採取製程改善之行動,避免增加產品額外的生產成本。而在統計製程管制方法中,管制圖 (control chart) 是最常被應用之重要工具,用來決定系統狀態並偵測製程中可能隨時發生的異常情況,這是屬於一種分類問題。管制圖上出現異常的形狀 (pattern) 常是造成製程失控的特定原因,所以辨識與分析管制圖形狀 (control chart pattern, CCP) 就成了一個重要課題。近年來資料採礦技術的其中一個技術決策樹 (decision tree, DT) 被廣泛的應用在分類問題上,並且有許多研究指出決策樹具有良好的效益,因此本研究擬應用決策樹來做為線上偵測製程之監控系統。傳統上,統計製程管制圖是在製程數據滿足常態分配且獨立性的假設下發展。然而,在實際的工業製程中已採用自動化之生產及檢驗方式,導致製程數據間具有高度之自我相關性,在這種情況下傳統之製程管制法並不適用,將造成錯誤警報增加。因此,本研究所探討的製程是模擬在自我相關製程環境中。本研究主要目的在於將決策樹學習應用於即時辨識自相關製程(autocorrelated processes)中的CCP,並且探討其可行性。模擬的結果顯示以DT為基礎之系統模式的快速學習特性讓使用者可以線上監控製程,且在線上即時情況下能學習新的知識 (knowledge),此特性讓一個CCP辨識系統在面對一個動態製造環境時,更具彈性。

並列摘要


Statistical process control (SPC) is an important method for control process in industry. It can detect assignable cause during the process control which may occur and provide help to improve process and reduce unnecessary product cost. Hence, control chart is an important tool at statistical process control. Control charts can detect abnormal status during the process control which may occur at any time. Essentially, the judgement of the process states can be seen as a classification problem in artificial intelligence. Effectively recognizing control chart patterns (CCPs) is a critical issue in statistical process control, since unnatural CCPs indicate potential quality problems at an early stage, to avoid defects before they are produced. Recently, decision tree (DT) is generally used in classification pattern, and a lot of researches point out that DT have excellent performances. This study examines the feasibility of utilizing a data mining technique DT learning in on-line CCP recognition for process with various levels of autocorrelation. An empirical comparison using simulation indicates that the fast learning of the DT model gives the SPC user the potential for building an automated CCP recognition system that can not only be applied on-line but also be trained in real time. This feature could make the CCP recognition system more adaptable to a dynamic manufacturing scenario.

參考文獻


1. 丁勝興 (2003),資料探勘在用藥輔助決策系統之研究,元智大學資訊管理研究所,碩士論文。
2. 王仁達 (1988),應用類神經網路偵測製程平均值變化:設計策略之研究,元智大學,碩士論文。
3. 吳育儒 (1998),決策樹中移除不相關值問題的研究,淡江大學資訊工程研究所,碩士論文。
4. 吳宗燦 (1997),決策樹漸進學習法中熵數變化之估測研究,中原大學資訊工程研究所,碩士論文。
5. 吳柏林 (1995),時間數列分析導論,華泰書局。

延伸閱讀