製造業中對於監控製程平均值發生偏移的情況,管制圖扮演著相當重要的角色,而同時監控、管制多個有相關性的品質特性也是現今製程管制中必然之趨勢。然而當多變量管制圖發出製程處於管制外之訊號時,如何有效辨別出是哪一個變量所造成的,是本研究主要的研究課題之一。 本研究主要的目的在於建立一個程序去定義管制外的訊號是由哪一個品質特性所造成的。因此研究中提出一套偵測平均值發生偏移以及分類其異常情形之系統,將管制外信號之解釋定義成一個分類的問題。透過傳統的T-squared管制圖監控製程平均值是否發生偏移,當發生管制外之訊號,再透過類神經網路以及支援向量機定義哪一個品質特性發生平均值偏移的狀況。本實驗使用分解法以及最小距離分類法當作比較之基準,並且考慮不同的共變異數矩陣下各種不同的偏移型態。在此,研究中採用正確分類率作為評估指標。結果顯示,在辨識平均值異常來源時,類神經網路以及支援向量機之效益差異不大,但兩分類器之效益皆優於傳統分解法和最小距離分類法。另外,從研究結果可得知,將原始資料中萃取出的特徵値當作輸入向量,可以有效提升正確分類率。
Control charts are important tools for monitoring process mean shifts in manufacturing industries. There are many situations in which the simultaneous monitoring or control of two or more related quality characteristics is necessary. One difficulty encountered of multivariate control charts is to identify exactly which variable results in an out-of-control signal. The main purpose of this research was to establish a procedure used for determining which variables should be responsible for the signal. In this study, we formulated the interpretation of out-of-control signal as a classification problem. The proposed method includes a shift detector and a classifier. The traditional T-squared chart works as a mean shift detector. When an out-of-control signal is generated, ANN-based and SVM-based classifier will determine which variables should be responsible for the mean shift. The traditional decomposition method and minimum distance classification method were used as baselines. In this research, various shift scenarios expressed in covariance matrices were considered. The performance was evaluated by computing the correct classification rate. Results from our researches indicated that the ANN and SVM have similar classification performance. Both classifiers can perform significantly better than the decomposition method and minimum distance classifier. Furthermore, the results indicated that the classification performance can be further improved by using features extracted from the data.