本研究之主旨在於運用倒傳遞類神經網路（Back Propagation Neural Network, BPNN）發展出具有發動機氣路部件多重故障（Multiple Faults）診斷能力的故障診斷系統，以改善目前大多數類神經診斷系統皆偏重於單一故障診斷的現象。本研究使用普惠（Pratt & Whitney）公司PW4000-94’型發動機之影響係數矩陣（Influence Coefficient Matrix）產生網路訓練與測試所須之故障樣本。結果顯示，BP網路對於發動機多重故障具有相當優良的診斷能力。然而有些故障所對應的量測參數值明顯較其他故障小，導致這些故障在與其他故障一同發生時，所對應的網路輸入值與其他故障單獨發生下的差異十分微小，造成網路因無法分辨出其間的差異性而產生誤判，因此在研究時必須經過篩選並排除此類故障。而在使用具有4個輸入值的網路架構時，會出現部分故障之間量測參數向量過於接近的現象，也就是相同量測參數對應不同故障案例的矛盾情況，因此本研究建議以具有8個輸入值的網路架構進行發動機多重故障診斷，以確保診斷之正確性。

This research presents the utilization of a backpropagation neural network (BNN) as a fault diagnosis system for detecting multiple faults based on the measurements of gas-path variables of an engine. The influence coefficient matrix of Pratt & Whitney PW4000-94” engine was employed to generate the fault patterns for training and testing a multi-layered neural network. In each generated pattern, due to multiple simultaneous faults, some of the faults may have dominant effects on the values of input variables; that is, unless the faults are present with comparable severity which generate measurement deltas with same order of magnitude, the network may classify the minor faults as less significant noise to the major faults. Thus, those unidentifiable fault patterns were deleted from the training process to avoid incorrect classifications. Computer simulations were conducted to experiment two network structures, one with four input variables and the other one with eight input variables. Because some of the generated fault patterns with four input variables may contain contradictions in the input-output mapping relationship—similar input deltas map onto different output fault types, the network structure with eight input variables was adopted as the diagnosis system and recommended for multiple faults detection and isolation. The results of computer simulations have validated the effectiveness of the proposed diagnosis system for isolating multiple faults of engine gas-path with satisfactory accuracy.

[1] 童遷祥, 翁森棋, 張立德,“飛機引擎維修專題研究, ”財團法人工業技術研究院, 民國83年5月.
[2] Smetana, F. O.,“Turbojet Engine Gas Path Analysis A Review,”AGARD Conference Proceedings, No. 165, 4-5 April, 1974.
[3] 陳大光, 張赫然,“用氣動熱力參數對發動機故障進行診斷與定位的實用方法,”北京航空航天大學碩士論文第二章, 1989年.
[4] Urban, L. A.,“Gas Path Analysis Applied to Turbine Engine Condition Monitoring,”AIAA 72-1082, 1972.
[5] Urban, L. A., and Volponi, Allan J.,Mathematical Methods of Relative Engine Performance Diagnostics,”Journal of Aerospace, Vol. 101, Technical Paper 922048, 1992, pp. 2025-2049.
[6] Doel, D. L., "TEMPER-A Gas Path Analysis Tool for Commercial Jet Engines," ASME Journal of Engineering for Gas Turbines and Power, Vol.116, No.1,1994, pp. 82-89.
[7] 葉怡成, 郭耀煌, 專家系統方法應用與實作, 全欣資訊圖書股份有限公司, 民國八十二年, pp.5-6.
[8] 沈建明,“航空燃氣輪發動機診斷系統研究－專家系統的應用與監視參數的選擇,”北京航空航天大學碩士論文第二章, 1991年.
[9] Haykin, S.,“Neural Networks, A Comprehensive Foundation,”Macmillan College Publishing Co. Inc., 1994, New York.
[10] Fausett, L., "Fundamentals of Neural Networks: Architectures, Algorithms and Applications," Prentice Hall, Englewood Cliffs, New Jersey, ISBN 0-13-334186-0, 1994.
[11] 徐自珍,“航空發動機人工神經網路故障診斷法,”國立成功大學航空太空工程研究所碩士論文, 民國86年.
[12] P. J. Lu, T. C. Hsu, J. Zhang, M.C. Zhang,“An Evaluation of Engine Fault Diagnostics Using Artificial Neural Networks,”ASME Journal of Engineering for Gas Turbines and Power, Vol. 123, No.2, April, pp.340-346.
[13] Zedda, M., and Singh, R., "Fault Diagnosis of a Turbine Engine Using Neural Networks: a Quantitative Approach,". AIAA-98-3602, 1998.
[14] Eustace, R., and Merrington, G., "Fault Diagnosis of Fleet Engines Using Neural Networks," ISABE 95-7085, 1995, pp.926-936.
[15] Cifald, M. L., and Chokani, N., "Engine Monitoring Using Neural Networks," AIAA-98-3548, 1998.
[16] Werbos, P., “Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences,” Ph.D. thesis, Harvard, Cambridge, MA, August 1974.
[17] Parker, D.B.,“Learning Logic,”Technical Report TR-47, Center for Computational Research in Economics and Management Science, MIT, Cambridge, MA, April 1985.
[18] McClelland, J. and Rumelhart, D.,“Learning Internal Representations by Error Propagation,”Parallel Distributed Processing, Vol. 1 and 2, MIT Press, Cambridge, MA, 1986.
[19] Fortuna, L., Graziani, S., Lo Presti, M. and Muscato, G.,“Improving Back-Propagation Learning Using Auxiliary Neural Networks,”International Joint Conference on Control, Vol. 55, No. 4, 1992, pp. 793-807.
[20] Naoki KAMIYAMA, Nobukazu IIJIMA, Akira TAGUCHI,“Tuning of Learning Rate and Momentum on Back Propagation,” Neural Networks, international Joint Conference on Neural Networks, 1991.
[21] Hsin H. C., Li C. C., Mingui Sun, and Robert J. Sclabassi,“An Adaptive Training Algorithm for Back Propagation Neural Networks,” IEEE Transactions on System, Man, and Cybernetics, Vol. 25, No. 3, March 1995, pp.512-514.
[22] Yu, X. H., Chen G. A., Cheng S. X.,“Dynamic Learning Rate Optimization of the Back Propagation Algorithm,”IEEE Transactions on Neural Networks, Vol. 6, No. 3, May 1995, pp.669-677.
[23] Magouals, G. D., Vrahatis, M. N. and Androulakis, G. S.,“Effective Backpropagation Training With Variable Stepsize,” Neural Networks, Vol. 10, No. 1, 1997, pp.69-82.
[24] 羅華強, 類神經網路–Matlab的應用, 清蔚科技股份有限公司, 2001年, pp.5-23 ~ 5-26.
[25] B. D. MacIsaac,“Engine Performance and Health Monitoring Models Using Steady State and Transient Prediction Methods,” AGARD Lecture Series 183, May 1992, pp.9-1~9-21.