臺灣博碩士論文加值系統

含有記憶功能的PCA(Principal Component Analysis)將被應用於靜態及動態的異常偵測問題。PCA是目前常被用於偵測化工製程方面的一種多變數統計管制方法，但是PCA是基於目前觀測值的一種管制方法，當量測變數有微小的異常變化時，以PCA方式並不容易偵測出異常現象。而其他的多變數管制方法如MEWMA，MCUSUM及MSSUM，因為利用一記憶參數將變數過去資訊加以累積，因此容易偵測出變數的微小異常變化。所以將傳統的PCA方法與上述含有記憶功能的管制方法加以結合，並以此種新的管制方法與傳統PCA方法以數學式加以比較。
對於動態系統將使用類神經網路與含有記憶功能的PCA方法加以結合。因為類神經網路的優點在於使用正常線性或非線性動態系統的數據建立代表此系統線性或非線性動態系統的模式，並以其預測誤差配合新的含有記憶功能的PCA作為動態系統的異常偵測，將優於一般的DPCA(Dynamic Principal Component Analysis)方法。在本文中將以Tennessee Eastman問題及二個實際工廠例子，一為釉料製程工業，一為鋼板熔爐製程，加以驗證結合類神經網路的含有記憶功能的PCA方法對於動態異常偵測的能力。

A general learning methodology based on PCA (Principal component analysis) control charts with memory effect for steady-state and dynamic process monitoring is proposed. PCA is currently a standard multivariate statistical technique. It has been applied to a lot of monitoring problems in chemical processes. However, the PCA model cannot explain the trend relationship among the measured variables with a small or a moderate shift in one or more process variables because it is built only based on the most recent observations. On the other hand, multivariate control charts, like MEWMA, MCUSUM, and SSUM etc, use additional information from the past history of the process, so the models possess the memory effect of the process behavior trend. Naturally, combining the PCA model with the above methods is a logical extension of the standard PCA model. In steady-state process monitoring, the combination method is developed and compared with the traditional PCA in terms of the mathematical definitions. In dynamic process monitoring, a technique combined with the neural network and PCA is proposed. The neural network is used to model the nonlinear dynamic system. The actual behavior of the process to be supervised is compared with that of a nominal fault-free neural network model driven by the same observations. The PCA based control charts evaluate the multivariable residuals driven from the differences between these outputs. This shows how static PCA can be applied on the dynamic system. The complementary of the proposed method not only leads to some cross-fertilization between various techniques but also results in a better model. Finally, the effectiveness of the proposed method for steady-state and dynamic process monitoring is demonstrated through the simulated Tennessee Eastman process problem and real industrial case studies - a melting process in a glaze industry and surface quality in a stainless steel slab to indicate the potential applications.

摘要…………………………………………………………………I
Abstract……………………………………………………………II
目錄…………………………………………………………………III
圖目錄………………………………………………………………V
表目錄………………………………………………………………XI
第一章：1
1-1異常偵測分析 …………………………………………1
1-2研究動機 ………………………………………………3
第二章：靜態統計管制分析(I)- T2為基礎的管制 …………5
2-1T2管制 …………………………………………………6
2-2MEWMA管制……………………………………………7
2-3MSSUM管制 ……………………………………………11
2-4T2、MEWMA及MSSUM管制的比較 …………………13
2-5討論 ……………………………………………………26
第三章：靜態統計管制分析(II)- PCA為基礎的管制方式 …29
3-1PCA管制 ………………………………………………29
3-2MEWMAPCA管制………………………………………34
3-3MSSUMPCA管制 ………………………………………36
3-4PCA、MEWMAPCA及MSSUMPCA管制的比較 ……38
3-5討論 ……………………………………………………57
第四章：動態統計管制分析-NN與PCA為基礎的管制方式……61
4-1關連性分析 ……………………………………………62
4-2DPCA管制 ………………………………………………64
4-3結合類神經網路與多變數管制 ………………………67
4-3-1穩定靜態與穩定動態…………………………… 68
4-3-2類神經網路……………………………………… 69
4-3-3以類神經網路為基礎的多變數管制…………… 71
4-4DPCA與NNMEWMAPCA管制的比較 ………………74
第五章：結論 ……………………………………………………100
附錄(一)MEWMA管制的共變異係數 ……………………………102
附錄(二)MSSUM管制的共變異係數………………………………103
附錄(三) TE製程是否有雜訊產生………………………………106
附錄(四) TE製程是否有異常產生………………………………107
符號說明 …………………………………………………………108
參考文獻 …………………………………………………………110

Box, G. and Luceno, A., Statistical Control by Monitoring and Feedback Adjustment, John Wiley & Sons, Inc (1997)
Cybenko, G., “Approximation by Superositions of aSigmodial Function,” Mathematics of Control,Signal and Systems, 2, 303 (1989)
Downs, J. J. and Vogel, E. F., “ A Plant-wide Industrial Process Control Problem,” Compters chem. Engng., 17, 245 (1993)
Fuchs, C. and Benjamini, Y., “Multivariate Profile Charts for Statistical Process Control,” Technometrics., 36, 2 (1994)
Gan F. F., “Joint Monitoring Of Process Mean And Variance,” Nonlinear Analysis, Theory, methods & Application., 30, 7 (1997)
Hormik, K., “Approximation Capabilities of Multilayer Feedforward Network,” Neural Network, 4, 251 (1991)
Hotelling, H. “Multivariate Quality Control,” Techniques of Statistical Analysis, Eisenhart, Hastay, and Wallis, Eds., McGraw-Hill, New York, 1947
Jackson, J. E. and Mudholkar G. S. Technometrics, 21 (1979)
Jackson, J. E., “Principal Components and Factor Analysis: Part I-Principal Components,” Journal of Quality Technology, 12 (1980)
Jackson, J. E., A User’s Guide to Principal Components, John Wiley & Sons, New York, NY. (1991)
Jackson, J. E., A user’s Guide to Principal Components, Wiley, New York (1992)
Jobson, J. D. Applied Multivariate Data Analysis Volum I:Regression and Experimental Design, New York (1911)
Ku, W.; Storer, R. H. and Georgakis, C., “Disturbance detection and isolation by dynamic principal component analysis,” Chemometrics and Intelligent Laboratory Systems, 30 (1995)
Lippmann R. P., “An Introduction to Computing with Neural Nets,” IEEE, April (1987)
Lowry, C. A.; Woodall, W. H.; Champ, C. W. and Rigdon, S. E., “A Multivariate Exponentially weighted Moving Average Control Chart,” Technometrics, 34 ,1 (1992)
Lowry, C. A. and Montgomery, D. C., “A review of multivariate control charts,” IIE Transaction., 27 (1995)
Mason, R. L.; Tracy, N. and Young, J. C., “A Practical Approach for Interpreting Multivariate T2 Control Chart Signals,” Journal of Quality Technology, 29, 4 (1997)
Prabhu, S. S. and Runger, G. C., “Designing a Multivariate EWMA Control Chart,” Journal of Quality Technology, 29, 1 (1997)
Runger, G. C. and Prabhu, S., “A Markov Chain Model for the Multivariate exponentially Weighted Moving Averages Control Chart,” Journal of the American Statistical Association, 91, 436 (1996)
Smits, J. R. M.; Melssen, W. J.; Buydens, L. M. C. and Kateman G., “Using Artificial Neural Networks for Solving Chemical Problems,” Chemometrics and Intelligent Laboratory Systems, 22 (1994)
Sullivan, J. H. and Woodall, W. H., “A Comparison of Mu;tivariate Control Charts for Individual Observations,” Journal of Quality Technology, 28, 4 (1996)
Taylor, J. G., Neural Network and Their Applications, John Wiley & Sons Ltd. (1996)
Tracy, N. D.; Young, J. C. and Mason, R. L., “Multivariate Control Charts for Individual Observations,” Journal of Quality Technology, 24, 2 (1992)
Unger, L. H.; Powell, B. A. and Kamens, S. N., “Adaptive Networks for Fault Diagnosis and Process Control,” Computers chem. Engng, 14, 4/5 (1990)
Wold, S., “Exponentially Weighted Moving Principal components analysis and projectipn to latent structures,” Chemometrics and Intelligent Laboratory Systems, 23 (1994)
Wachs, A. and Lewin, D. R., “Improved PCA Methods for Process Disturbance and Failure Identification,” AIChE Journal, 48, 8 (1999)