多變量管制圖之主要目的是用來偵測製程中是否發生異常訊號。若偵測出製程發生變異,則應儘快解決異常之現象並診斷其原因為何,使製程回歸統計管制狀態內。雖然使用Hotelling T2 管制圖可以監控多個品質特性之異常發生,且擁有良好之績效,然而,Hotelling T2 管制圖卻無法判斷是由製程中哪一個品質特性發生變異。為了有效確認發生異常之品質特性為何,且提高其辨識績效,有研究提出以人工智慧法來判斷是何種品質特性發生異常,且利用單一個或多個分類模型以辨識異常來源。此種方法雖可處理多個品質特性之情形,但其辨識提升之成效不彰且耗費時間。 本研究以決策樹與支援向量機為基礎建構診斷系統。此系統是以Hotelling T2 管制圖進行監控並應用決策樹與支援向量機來辨識。本研究提出以樣本多樣性之方法建構多個分類模型,並以統計特徵值 (平均數與馬氏距離) 作為診斷系統之輸入向量。由研究結果顯示,以整體式支援向量機整合之辨識系統,其辨識績效為最佳。
The main purpose of the multi-variable control chart is used to detect whether the process exists abnormal signals. If the variation is detected in the process, the abnormal signal should be diagnosed and resolved as soon as possible and resume the process back to the in-control statistical state. Although Hotelling chart is efficient in detecting a general multivariate shift in the mean vector, it fails to identify which variables are responsible for the mean shift. In order to effectively identify the occurrence of abnormal quality characteristic and to improve the recognition performance, many research papers address these problems and present various artificial intelligence approaches to identify aberrant variables; and they have been using one or more classification models to identify abnormal source. Although these methods can handle the multiple quality characteristics fault detection, but they remain ineffective and time-consuming. In this study, Decision Tree (DT) and Support Vector Machine (SVM) are the two methods using the ensemble classification model respectively to construct two different diagnostic systems. This study proposes to construct multiple classification model by data diversity method and statistical characteristic value (mean and Mahalanobis distance) as the input vector of the diagnostic system. After comparing against the single classification model, the results show that the recognition performance of the SVM with ensemble is better than the one of the Decision Tree.