Applying Decision Tree-Based Ensemble Classifiers for Diagnosing Mean Shift Signals in Multivariate Control Chart

應用整體式決策樹分類模型於多變量管制圖平均數偏移之診斷

鄭春生(Chuen-Sheng Cheng) ； 李虹葶(Hung-Ting Lee)

《品質學報》 21卷2期 (2014/04) Pp. 91-103

https://doi.org/10.6220/joq.2014.21(2).02

摘要

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，則應立即處理異常之訊號並診斷其發生異常原因為何，使製程回復至穩定之管制狀態內。Hotelling T^2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 T^2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，本研究以決策樹為基礎建構診斷系統。此系統是以Hotelling T^2管制圖進行監控並應用決策樹整體式分類模型進行辨識。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，以整體式決策樹整合之辨識系統，其辨識績效為最佳。

關鍵字

整體式學習 ； 決策樹 ； 平均數偏移 ； 多變量統計製程管制 ； 統計製程管制

並列摘要

The Hotelling T^2 control chart is an important tool for monitoring process shift in multivariate statistical process control (MSPC). Detecting and diagnosing out-of-control variables are required tasks when a multivariate control chart signals. This paper presents a decision tree-based ensemble model to address diagnosing issue in multivariate process control. The commonly used ensemble methods, including bagging and AdaBoost are considered in this paper. To improve the classification performance, we propose using a set of features extracting from process data. Results from comparative studies indicate that these features with certain ensemble classifiers can significantly improve classification performance. The proposed approach contributes to process monitoring and identifyingmean shift sources in MSPC, which can assist engineers to effectively identifyresponsible variables and accelerate improvement action generation.

並列關鍵字

ensemble learning ； decision tree ； mean shift ； multivariate statistical process control (MSPC) ； statistical process control (SPC)

參考文獻

Alfaro, E.,Alfaro, J. L.,Gámez, M.,García, N.(2009).A boosting approach for understanding out-of-control signals in multivariate control charts.International Journal of Production Research.47(24),6821-6831.

Aparisi, F.,Avendaño, G.,Sanz, J.(2006).Techniques to interpret T2 control chart signals.IIE Transactions.38(8),647-657.

Aparisi, F.,Avendaño, G.,Sanz, J.(2007).Neural networks to identify the out-of-control process variables when a MEWMA chart is employed.Proceedings of the The 16th IASTED International Conference on Applied Simulation and Modelling.(Proceedings of the The 16th IASTED International Conference on Applied Simulation and Modelling).

Aparisi, F.,Carrión, A.(2010).Artificial neural networks for identifying the signals of multivariate EWMA control chart.Proceedings of 2010 10th International Conference on Intelligent Systems Design and Applications.(Proceedings of 2010 10th International Conference on Intelligent Systems Design and Applications).

Breiman, L.,Friedman, J. H.,Olshen, R. A.,Stone, C. J.(1984).Classification and Regressing Tree.Pacific Grove, CA.:Wadsworth.