應用整體式分類模型於多變量製程變異性異常來源之辨識

Identifying the source of variance shifts in the multivariate process using ensemble classifiers

林志鴻(Chih-Hung Lin)

指導教授 ： 鄭春生

元智大學/工程學院/工業工程與管理學系/碩士(2011年)

摘要

多變量管制圖可以同時監控多個具有相關性之品質特性，針對多變量製程變異性之監控，常用廣義性變異數 |S| 管制圖進行偵測。多變量管制圖之管制外訊號產生可能是由一個或多個變量所導致，然而當管制圖偵測到製程異常發生時，管制圖卻無法進一步地判別是由多變量製程中哪一個品質特性所造成之變異。為了能夠有效地判斷出發生異常之品質特性為何，在過去研究中提出以人工智慧方法進行異常變量之辨識，且僅使用一個分類模型來辨認異常來源，由於單一分類模型受限於相同的訓練樣本以及固定的參數，其辨識績效有限。本研究以整體式分類模型提升其辨識績效。
本研究將辨識管制外之訊號來源視為一個分類問題，並且提出一個整體式支援向量機辨識系統包含偵測和辨識。當多變量管制圖偵測到異常時，將以辨識系統判斷製程變異性之異常來源。一般常用來建立整體式分類器之多樣性策略為操控訓練樣本，在本研究中提出一個創新之資料多樣性方法，我們依據管制圖之不同統計特性所產生之樣本，來建構整體式分類模型，並以特徵值作為支援向量機分類器之輸入向量來提升辨識系統之績效。本研究以分類之正確率作為評估不同方法之指標，其結果顯示整體式辨識系統有助於提升異常來源辨識之績效。

關鍵字

多變量管制圖 ； 變異性變化 ； 整體式學習 ； 支援向量機

並列摘要

Multivariate control chart is used in the situation which simultaneous monitoring of two or more related quality characteristics. The generalized variance, |S|, control chart is usually applied to monitor process variability. Out-of-control signals in multivariate control charts may be caused by one or more variables. Although control chart is efficient in detecting a general multivariate shift in the variance, it fails to determine which variables are responsible for the variance shift. Many research papers address these problems and present various artificial intelligence approaches to identify aberrant variables. In the previous studies, only one classifier is applied in recognizing abnormal sources. However, the single model is limited to the same data or parameters setting and none of them could consistently perform well over all datasets.
In this paper, we formulate the interpretation of out-of-control signal as a classification problem. The proposed system includes a SVM ensemble classifiers and a shift detector. When an out-of-control signal is generated, ensemble classifiers will determine which variable is responsible for the variance shift. Manipulating the training sample is usually used to create the diverse models. In the proposed approach, we base on some different statistical properties to construct ensemble and propose using extracted features as predictors to enhance the performance. The performance of the proposed system is evaluated by computing its classification accuracy. Results from studies indicate that the proposed approach is beneficial for identifying the source of variance changes.

並列關鍵字

multivariate control chart ； variance shift ； ensemble learning ； support vector machine

參考文獻

Breiman, L., “Bagging predictors,” Machine Learning, 24, 123-140 (1996).

Chang, C. C., and Lin, C. J., “LIBSVM: a library for support vector machines,” (2001). Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm

Cheng, C. S., and Tzeng, C. A., “A neural network model for detecting shifts in the process mean and variability,” Journal of Chinese Institute of Industrial Engineers, 11, 67-75 (1994).

Cheng, C. S., and Cheng, H. P., “Identifying the source of variance shifts in the multivariate process using neural networks and support vector machines,” Expert System with Applications, 35, 198-206 (2008).

Cheng, Z. Q., Ma, Y. Z., and Bu, A. J., “Variance shift identification model of bivariate process based on LS-SVM pattern recognizer,” Communication in Statistics: Simulation and Computation, 40, 286-296 (2011).