摘要
航空发动机一直被喻为飞机的心脏,其性能衰退情况直接影响着飞机的飞行安全和航空公司的经济效益。
航空发动机的性能衰退主要表现在其性能衰退参数呈品质下降趋势,而性能衰退参数属于时间序列参数,必须利用能够处理时间序列的方法来处理和预测航空发动机的性能衰退情况。DEGT(Delta Exhaust Gas Temperature, DEGT)是发动机重要性能衰退参数之一,本文以DEGT为例,对航空发动机性能衰退预测技术展开研究。
针对航空发动机性能衰退参数的时序特点,避开繁琐的、实际操作困难的数学建模和无法反映参数时间累积效应的传统人工神经网络预测方法,提出一种基于过程神经网络的性能衰退预测方法,并将前馈过程神经网络、双并联过程神经网络、小波过程神经网络分别应用于航空发动机性能衰退预测中,对各种模型的预测结果进行比较,分析影响过程神经网络泛化能力的多种因素。
在此基础上,为提高预测精度和克服单一过程神经网络的预测缺陷,提出基于过程神经网络集成的性能衰退预测方法,介绍过程神经网络集成的基本概念和基本理论,分析个体网络输出结果合成阶段的几种方法,并对各种方法的优缺点进行比较。
为提高过程神经网络集成的泛化能力即优化过程神经网络集成模型,对影响过程神经网络集成模型泛化能力的诸多因素进行分析研究。
在上述理论研究的基础上,开发“基于过程神经网络的航空发动机健康状态预测系统”,并集成到“基于Web的航空发动机健康状态监测和维修数据管理系统”中,为实现航空发动机性能衰退预测的实时化、自动化和智能化提供支持。
Aero-engines has been described as the heart, and its performance decline condition directly affects the flight safety of the aircraft and the cost of the airline.
Aero-engines performance decline mainly reflected in the quality of their performance decline parameters showing a declining trend, and the performance decline parameters are time series parameters. Therefore, when we treatment and predict the condition of the aero-engines performance decline, we must use the method which is able to handle the time-series approach. DEGT (Delta Exhaust Gas Temperature, DEGT) is one of the important parameters of the engine performance decline. In this paper, taking DEGT as an example, the predicting technique of aero-engines performance decline is researched.
Because of the timing characteristics of the aero-engines performance decline parameters, this paper avoids the mathematical modeling which tedious and practice difficulty and the method of the traditional artificial neural network forecast which does not reflect the time parameters of the cumulative effect, proposes a method of the performance decline prediction based on process neural network. Then applying respectively the feedforward process neural network, two parallel process neural network, wavelet process neural networks to the predicting of the aero-engines performance decline. In this way, compares the predicted results, analyzes factors affecting the generalization ability of the process neural network.
On this basis, this paper efforts to improve the forecasting accuracy and to overcome the prediction defect of the single process neural network forecasting. First, proposes the prediction method of the performance decline based on the model of process neural network ensemble forecasting. Second, describes the concepts and basic theory of the process neural network ensemble. Third, analyzes the synthesis of the output stage of the network in several ways and compares the advantages and disadvantages of each method at the same time. Forth, analyzes the many factors which impact the generalization ability of the process neural network model.
For optimizing the process neural network ensemble model, this paper also analyzes many factors affecting the generalization ability of the process neural network.
A software system was developed based on the theory study above, named aero-engine health condition prediction system based on process neural network. The system is used in Air China now, and has been integrated into the "Web-based aero-engine health monitoring and maintenance data management system". The system will support to realize the independence, real-time, automation and intelligence of the aero-engine performance decline prediction.
引文
1富涛,许春生.在翼航空发动机剩余寿命预测.中国民航飞行学院学报. 2006, 17(3):18~21
2周茂军.考虑性能衰退的航空发动机总体性能裕度设计研究.航空动力学报. 2008, 23(10): 1868~1874
3 Y L Lü, R L Lang, H Lu, Z Z Tan. Prediction of aeroenging’s performance parameter combining RBFPN and FAR. Beijing Hangkong Hangtian Daxue Xuebao. 2010, 36(2): 131~134
4吴学辉,李志刚,陶增元.参数小偏差对某高推重比航空发动机性能的影响分析[J].航空动力学报. 2005, 20(6): 1028~1031
5何新贵,许少华.过程神经元网络. 1版.科学出版社. 2007: 3~4 40~44
6耿宏,揭俊等. ACARS报文参数的辨识.航空电子技术. 2006, 37(4):6~11
7北京飞行维修工程有限公司.发动机集群的科学管理.民航发动机可靠性研究课题总结. 2000, 7:121~130
8 Y. X. Song. K. X. Zhang. Y. S. Shi. Research on aeroengine performance parameters forecast based on mulitiple linear regression forecasting method. Hangkong Dongli Xuebao. 24(2): 427~431
9张海军.民航发动机性能评估方法与视情维修决策模型研究.南京航天航空大学博士学位论文. 2007, 2~38
10钟诗胜,栾圣罡.面向航空发动机全寿命周期管理的航线数据处理系统.计算机集成制造系统. 2006, 12(8): 1273~1278
11胡金海,谢寿生等.基于支持向量机方法的发动机性能趋势预测.推进技术. 2005, 26(3): 260~264
12陈立波,宋兰琪,宋科,张占纲.基于支持向量机的航空发动机磨损趋势预测.润滑与密封. 2008, 33(5): 84~87
13陈果.用结构自适应神经网络预测航空发动机性能趋势.航空学报. 2007, 28(3): 535~538
14 S. R. Zhao. X. H. Huang. Fault diagnosis for aeroengine gas path components based on neural network multisensor data fusion. Hangkong Dongli Xuebao. 2008, 23(1): 163~168
15 J. L. Qu. C. S. Tang. Integrated diagnosis of aeroengines’gas path faults using artificial neural network. Hangkong Dongli Xuebao. 2008, 23(11): 2124~2127
16 X. G He, J. Z. Liang. Process Neural Network. World Computer Congress 2000, Proceedings of Congerence on Intelligent Information Processing. Beijing: Publishing House of Electronics Industry, 2000, 143~146
17钟诗胜,栾圣罡,丁刚.基于过程神经网络与气动热力参数的航空发动机状态监视.南京理工大学学报. 2006, 30(5): 533~536
18金星姬,贾炜炜.人工神经网络研究概述.林业科技情报. 2008, 40(1): 65-71
19何新贵,梁久祯,许少华.过程神经元网络的训练及其应用.中国工程科学. 2001, 3(3): 31~35
20何新贵,梁久祯.过程神经网络的若干理论问题.中国工程科学. 2000(2): 40~44
21 L G Valiant. Theory of the learnable. Communications of the ACM. 1984, 27(11): 1134~1142
22 L K Hansen, P Salamon. Neural network ensembles. IEEE Transactions on Pattern Analysis and Machine Intelligence. 1990, 12(10): 993~1001
23田雨波.混合神经网络技术. 1版.科学出版社, 2009: 200~210
24 R E Schapire. The strength of weak leamability. Machine Learning. 1990, 5(2): 197~227
25 M P Perrone, L N Coopler. When networks disagree: ensemble method for neural networks. Artificial Neural Networks for Speech and Vision, Chapman & Hall, New York. 1993, 126~143
26 A Krogh, J Vedelsby. Neural network ensembles, cross validation, and active learning. Advances in Neural Information Processing Systems. 1995, 7: 231~238
27 L Breiman. Bagging predictors. Machine Learning. 1996, 24(2): 123~140
28 P Sollich, A Krogh. Learning with ensembles: how over-fitting can be useful. Advance in Neural Information Processing Systems. 1996, 8: 190~196
29 Y Freund, R E Schapire. A decision-theoretic generalization of on-line learing and an application to boosting. Computer and System Sciences. 1997, 55(1): 119~139
30周志华.神经计算中若干问题的研究.南京大学博士学位论文. 2000: 47~70
31 Y Liu, X Yao. Ensemble Learning via Negative Correlation. Neural Networks.1999, 12(10): 1399~1404
32沈掌权.神经网络集成技术及其在土壤学中应用的研究.浙江大学博士学位论文. 2005, 58~59
33周志华.机器学习及其面临的挑战.技术报告.计算机科学面临的挑战高层研讨会.厦门大学.2003
34许少华,何新贵.基于函数正交基展开的过程神经网络学习算法.计算机学报. 2004, 27(5):645-650
35 S. G. Luan, G. Ding, S. S. Zhong. Aeroengine lubricating oil metal elementsconcentration prediction based on double parallel process neural network. Run Hua Yu Mi Feng/Lubrication Engineering. 2006, (5): 32~37
36 S. S. Zhong, Y. Li. Condition monitoring of aeroenging based on wavelet process neural networks. Hangkong xuebao. 2007, 28(1): 68~71
37施彦,韩力群,廉小亲.神经网络设计方法与实例分析. 1版.北京邮电大学出版社. 2009: 1~3 7~13
38 Y Freund. Boosting a weak algorithm by majority. Information and Computation. 1995, 121(2): 256~285
39 W Yates, D Partridge. Use of methodological diversity to improve neural network generalization. Neural Computing and Applications. 1996, 4(2): 114~128
40 D Opitz, J Shavlik. Actively searching for an effective neural network ensemble. Connection Science. 1996, 8(34): 337~353
41 A Lazarevic, Z Obradoric. Effective pruning of neural network classifier ensembles. International Joint Conference on Neural Networks, Washington. DC. United states, 2001. Institute of Electrical and Electronics Engineers Inc, 796~801
42 Q Fu, S X Hu, S Y Zhao. Clustering-based selective neural network ensembles. Journal of Zhejiang University Science, 2005, 6 (5): 387~392
43李凯,黄厚宽.一种基于聚类技术的选择性神经网络集成方法.计算机研究与发展. 2005, 42(4): 594~598
44 Z H Zhou, J X Wu, W Tang, et al. Combining regression estimators: GA-based selective neural network ensemble. International Journal of Computational Intelligence and Applications. 2001, 1(4): 341~356
45乐晓蓉.神经网络集成算法设计及分析.扬州大学硕士学位论文. 2007: 50~55
46张全平,吴耿锋.基于人工免疫网络的神经网络集成方法.计算机工程. 2008, 23(34): 199 ~201
47 C E Brodley, T Lane. Creating and Exploiting Coverage and Diversity, Work Notes AAAI-96 Workshop Integrating Multiple Learned Models. 1996, 8~14
48 D. Jimenez. N. Walsh. Dynamically weighted ensemble neural networks for classification. Proceedings of the 1998 IEEE International Joint Conference on Neural Networks. Part 1 (of 3), Anchorage. AK. USA, 1998. IEEE. Piscataway, NJ. United States, 753~756
