AN ARTIFICIAL NEURAL NETWORK-BASED CLASSIFIER ENSEMBLE APPROACH FOR ON-LINE RECOGNITION OF CONCURRENT CONTROL CHART PATTERNS

應用以類神經網路為基底的集成式分類技術於線上即時辨識混合型管制圖形狀

顧瑞祥(Ruey-Shiang Guh) ； 李孟樺(Meng-Hua Li)

《品質學報》 25卷1期 (2018/02) Pp. 1-28

https://doi.org/10.6220/joq.201802_25(1).0001

摘要

Effectively recognizing control chart patterns (CCPs) is an important issue in statistical process control. Machine learning techniques (such as artificial neural works and decision tree learning) have been successfully applied into the field of CCP recognition. Classifier ensembles (i.e., combining of multiple classifiers) have been proven to be a method superior to single classifiers in many pattern recognition problems. This study applies classifier ensemble approaches to develop an effective artificial neural work-based model for on-line recognition of concurrent CCPs, in which more than one pattern exists simultaneously. The concurrent CCPs were relatively difficult to be accurately recognized due to the pattern interaction and resultant complexity. Three traditional classifier ensemble methods, including Bagging, Adaptive Boosting (AdaBoost), and Arcing, are employed and enhanced in this study using a diversity measure among the component classifiers. Numerical comparisons obtained using extensive simulation show that both recognition accuracy and recognition speed can be greatly improved by using the proposed classifier ensemble model. The simulation results also indicate that the enhanced classifier ensemble methods proposed in this study perform better than the traditional classifier ensemble methods. These results demonstrate that the classifier ensemble methods can be particularly useful in the field of complex CCP recognition.

關鍵字

statistical process control ； concurrent control chart patterns ； pattern recognition ； classifier ensemble ； artificial neural networks

並列摘要

管制圖上所出現的異常形狀(pattern)常和某些造成製程失控(out-of-control)的特定原因(assignable cause)有關，因此及時且正確地辨識管制圖形狀(control chart pattern, CCP)是統計製程管制(statistical process control)的一個重要課題。近年來，有很多學者積極且成功地應用各種機器學習(machine learning)技術於辨識CCP的領域中，而整合多個單一分類器(single classifier)於一個分類任務的集成式分類器(classifier ensemble)也已在許多領域中被證實其績效優於單一分類器，實務中常見的混合型管制圖形狀(concurrent CCP)，由於同時包含了兩種或兩種以上的CCP，單一分類器常因分類偏差或變異(classification bias/variance)問題，很難對複雜的混合型CCP具有令人滿意的線上即時辨識績效。本研究應用以類神經網路為基底的集成式分類技術構建一個能有效線上即時辨識混合型CCP的模式。本研究運用三個傳統的集成式分類技術，包含掛袋法(Bagging)、適應推進法(AdaBoost)及簡明推進法(Arcing)，並使用一個衡量基底分類器(component classifier)分類能力之離散程度的多樣性指標(diversity measure)增進傳統集成式分類技術的分類正確率。模擬數據顯示本研究提出的以集成式分類技術為基的混合型CCP線上即時辨識模式，其辨識績效，包含辨識精度（正確率）及辨識速度（平均連串長度，average run length），均遠優於文獻中以單一分類器為基的混合型CCP辨識模式，模擬結果同時指出本研究提出的新型集成式分類技術，其績效也可優於傳統的集成式分類技術。本研究的結果證明集成式分類方法可有效地增進以機器學習技術為基的CCP辨識模式之辨識精度及速度。

並列關鍵字

統計製程管制 ； 混合型管制圖形狀 ； 形狀辨識 ； 集成式分類 ； 類神經網路

參考文獻

Tumer, K. and Ghosh, J., 1996, Error correlation and error reduction in ensemble classifiers, Connection Science, 8(3-4), 385-404.

Al-Assaf, Y., 2004, Multi-resolution wavelets analysis approach for the recognition of concurrent control chart patterns, Quality Engineering, 17(1), 11-21.

Brazdil, P., Gama, J., and Henery, B., 1994, Characterizing the applicability of classification algorithms using meta-level learning, Machine Learning: ECML-94, Lecture Notes in Computer Science, vol. 784, edited by Bergadano, F. and de Raedt, L., Springer, Berlin, Germany, 83-102.

Breiman, L., 1996a, Bagging predictors, Machine Learning, 24(2), 123-140.

Breiman, L., 1996b, Bias, variance, and arcing classifiers, Technical report 460, Statistics Department, University of California, Berkeley, CA.