在應用單變項失誤偵測技術(如SPC)的過程中，常利用『使用規則』來提升系統狀態的判定效率，但是多變項失誤監視技術(如PCA)之使用規則至今在文獻中尚未見到任何探討。因此在本論文中，我們提出一套系統化的使用規則建構方法。具體而言，我們利用EKF降低估測數據變異度的優點，以及PCA中投影量各自獨立的特性，依警報邏輯的設計技巧[Tsai and Chang, 1996]，發展出可根據線上狀態估測值即時調整使用規則的多變項失誤監視策略，而使系統狀態判斷之期望損失值得以最小化。從利用TE製程模擬程式製造出的數據測試結果中可以發現，本研究所提出的方法不但可以顯著減少傳統監視方法的誤判比例，還可提早在失誤發生初期就偵測出誤源。

In implementing the single-variate process monitoring procedures, such as those adopted for statistical process control (SPC), a set of run rules are often used to facilitate proper assessment of the system states (Bavadas, 1993; Western Electric Company, 1956). However, the run rules of multi-variate monitoring methods, e.g. the principal component analysis (PCA), have never been discussed in the literature. A systematic procedure is thus developed in this work to generate these rules on-line. Specifically, the well-established alarm-logic design techniques (Tsai and Chang, 1996) are adopted to minimize the expected loss of misjudgment in fault detection. Unlike the traditional monitoring approach, the proposed run rules can be adjusted in real time according to state estimates. To enhance alarm accuracy, the EKFs are used to reduce the variation in state estimation. Also, in order to ensure that the alarm variables are mutually uncorrelated, the projections in the principal directions obtained in PCA are adopted for characterizing the system states. The feasibility of the proposed approach has been verified with measurement data produced with the Tennessee-Eastman process simulation program (Downs and Vogel, 1993). The test results show that the multi-variate run rules are not only useful in avoiding spurious alarms, but also effective in detecting incipient abnormal conditions.

[1]Anderson, T. W., An Introduction to Multivariate Statistical Analysis, John Wiley & Sons, Inc.,New York(1984).
[2]Badavas, P. C., Real-Time Statistical Process Control , Prentice-Hall, Inc., Englewood Cliffs, New Jersey(1993).
[3]Downs, J. J., and E. F. Vogel, “A Plant –Wide Industrial Process Control Problem,” Computers Chem. Engng., 17, pp.245(1993).
[4]Grant, E. L., and R. S. Leavenworth, Statistical Quality Control, 6th ed., McGraw-Hill Book Company, Singapore(1988).
[5]George, H. D., Principal Component Analysis, Sage Publications, Inc., USA(1989).
[6]Grewal, M. S. and A. P. Andrews, Kalman Filtering-Theory and Practice, Englewood Cliffs, New Jersey (1993).
[7]Himmelblau, D. M., Fault Detection and Diagnosis in Chemical and Petrochemical Process, New York(1978).
[8]Kohda, T., H. Kumamoto, and K. Inoue, “Optimal Shut Down Logic for Pro- tective Systems,” IEEE Trans. Rellab, R-32, pp.26 (Apr. 1983).
[9]Lees, F. P., Loss Prevention in the Process Industries, Vol. 1, Butterworth, Lon- don, pp.379 (1980).
[10]MacGregor, J. F. and T. Kourti,“Statistical Process Control of Multivariate Process ,” Control Eng. Practice, 3, 3, pp.403,(1995).
[11]Mah, R. S. H., G. M. Stanley, and D. M. Downing, “Reconciliation and Rectification of Process Flow and Inventory Data,” I&EC. Proc. Des. Dev., 15, pp.175 (1976).
[12]Mah, R. S. H., Chemical Process Structu An Engineer’s Manual of Statistical Methods res and Information Flows, Butterworth, Boston (1990).
[13]Nomikos, P., and J. F. MacGregor, “Monitoring Batch Process Using Multiway Principal Component Analysis,” AIChE J., 40, 1361 (1994).
[14]Passano, A. I. et al, ASTM Manual on Quality Control of Material, American Society for Testing Materials, Baltimore(1956).
[15]Raich, A., and A. Cinar, “Diagnosis of Process Disturbances by Statistical Distance in Multivariable Continuous Processes,” AIChE J., 42, 995(1996).
[16]Sharma, S., Applied Multivariate Techniques, John Wiley & Sons, Inc., New York(1996).
[17]Tsai, C. S., and C. T. Chang, “Statistical Operating Strategies for Charging Batch Reactors,” AIChE J., 42, 1304 (1996).
[18]Tsai, C. S., and C. T. Chang, “Optimal Alarm Logic Design for Mass Flow Networks ” AIChE J., 43, 3021 (1997).
[19]Tsai, C. S., and C. T. Chang, “Resilient Alarm Logic Design for Process Networks ,” I&EC Research, 39, 4974 (2000).