成功大學博碩士論文系統

簡易檢索 / 詳目顯示

回結果列表

研究生：	王紀瑞 Wang, Jee-Ray
論文名稱：	類神經網路服役負載估算及記錄器參數驗證與處理 SERVICE LOADS PREDICTION AND RECORDER DATA VALIDATION USING ARTIFICIAL NEURAL NETWORKS
指導教授：	陸鵬舉 Lu, Pong-Jeu
學位類別：	博士 Doctor
系所名稱：	工學院 - 航空太空工程學系 Department of Aeronautics & Astronautics
論文出版年：	2003
畢業學年度：	91
語文別：	中文
論文頁數：	141
中文關鍵詞：	結構疲勞、神經網路
外文關鍵詞：	Neural Network, DNN, SFDIA
相關次數：	點閱：91 下載：3
分享至:	分享至facebook 分享至twitter

查詢本校圖書館目錄查詢臺灣博碩士論文知識加值系統勘誤回報

視情維修(On-condition)的觀念於1970年提出後，已被廣泛的應用於飛機結構的維修上，並且取代原有之定期維修(Inspection/Repair as Necessary，IRAN)方式。飛機結構維修的需求主要源自於疲勞壽命(Fatigue Life)的確保，而疲勞壽命分析的準確與否，與破壞力學(Fracture Mechanics)及服役負載(Service Loads)息息相關，前者於數十年間之發展已趨成熟，而後者則尚有精進空間。
本文利用類神經網路(Artificial Neural Networks)進行服役負載的估算。應用類神經網路為基礎的方法，其主要的優點為：1)不需要以數學模式，來定義服役負載與飛行參數之間的關係；2)在預測服役負載時，可以容忍參數之數據中，包含某種程度的缺陷；3)可以輕易的自動化，不需人力的大量投入。研究中使用迴傳式神經網路(Recurrent Neural Network)模擬飛行參數及結構負載間之動態關係，可準確的估算結構負載，並具備簡單及廣泛應用之特性。此方法與現有參考文獻中之線性回歸(Linear Regression)及類神經網路映射(Artificial Neural Network Mapping)有很大差異。本研究所使用的網路(Network)可學習系統的動態模式，因此可掌握負載隨時間的變化歷程，這對疲勞分析至為重要。
其次，為掌握服役負載的估算品質，除網路的設計之外，最重要的便是輸入飛行參數正確性的判斷。飛行參數來自於服役飛機的飛行資料記錄器(Flight Data Recorder)，紀錄之參數不可避免的將包含各種雜訊或甚至損壞。為確保飛行參數的可用性，我們使用自映射自相關神經網路(Self-mapping Auto-associative Neural Network，AANN Neural Network，DNN)及分散式參數驗證神經網路(Decentralized )來進行感測器失效之檢出、識別、與補正(Sensor Failure Detection Identification and Accommodation，SFDIA)。這個感測器確認系統，為具有雜訊過濾及修正各種偏差功能的ANN。為了證明此一新方法的正確性，我們使用六自由度模擬數據進行測試，結果非常良好。
前述之研究結果顯示，本文所提出的服役負載估算及飛行參數之SFDIA神經網路，能有效的結合於飛機結構疲勞壽命追蹤之流程中，可藉以提昇疲勞壽命追蹤的品質，並提高飛行安全及降低維修成本。

On-condition maintenance demands a constant monitoring of aircraft structures. To assure structural integrity, the fatigue life analysis must be conducted with precise loads assessment provided by prediction methods. Loads prediction used to come from design results which often are crude and cannot reflect the real loading conditions experienced in service flights. To improve the quality of load prediction, the state-of-the-art approaches all use actual service loads that are obtained indirectly from the recorded flight parameters. Regression methods and feedforward artificial neural networks have been adopted to convert the information of flight parameters into service loads at critical areas of airframe. These methods are basically the static fittings of the selected flight parameters with respect to the structural ground testing results. The applicability of these service loads predictions could be inaccurate and misleading. The present research proposed a dynamic loads prediction method to solve the basic difficulties encountered in the current static approaches. In the meantime, the design criteria of developing a minimal human intervened system that can autonomously process large amount service data are included. Artificial neural networks (ANN) are utilized as the ground for establishing the present methodology. For loads prediction, the recurrent neural network (RNN) is adopted to establish the mapping between flight parameters and structural loads. It is shown that complex system dynamics can be learned by RNN. Moreover, the generalization ability of this trained RNN is enormous. For the test flight maneuvers, namely, symmetric wind-up turn and 1-g 360° roll, RNN trained using data obtained form one Mach/altitude block regime can successfully apply to other regimes in the test flight envelope. This demonstrates that for a viable service loads prediction, the loads approximator should possess a temporal causality relationship rather than a static mapping between the flight parameters and the structural loads. Since loads are inferred from the measured flight parameters, hence, the quality of recorded data is also crucial for an accurate loads prediction. Sensor data are constantly complicated by noises, biases and complete failure. The present research proposed a new sensor failure detection, identification, and accommodation (SFDIA) procedure. The neural network-based method completely circumvents the drawbacks of the conventional model-based approaches. Auto-associative neural network (AANN) and decentralized neural network (DNN) are utilized to construct a new SFDIA algorithm for flight parameter validation. Various sensor imperfections including noise, outlier, sensor hard and soft failures can be treated satisfactorily. Moreover, with two threshold values predetermined, the present SFDIA procedure can efficiently and autonomously process sensor data without any human intervention. A newly proposed ANN-based flight service loads prediction methodology is developed. Loads prediction quality can thus be assured by the present integrated sensor validation and dynamic approximation methods. Aircraft fatigue life monitoring and proper structural maintenance actions can be more effectively enforced with the future developments proceeded along this line of thoughts.

CHAPTER I
INTRODUCTION	1
1-1 ON-CONDITION STRUCTURAL MAINTENANCE	1
1-2 STRUCTURE SERVICE LOADS	3
1-2-1 Empirical Equations	4
1-2-2 Regression-Based Loads Correlation	4
1-2-3 Neural-Network-Based Loads Prediction	6
1-3 FLIGHT DATA RECORDER	8
1-3-1 F-16 Flight Data Recording System	9
1-3-2 The Quality of Recorded Flight Parameters	10
1-4 LOADS MONITORING PROGRAM AND FATIGUE LOADS SPECTRA	11
1-5 RESEARCH OBJECTIVES	12
1-6 THESIS OUTLINE	13
CHAPTER II
STRUCTURAL FATIGUE LIFE MONITORING	15
2-1 AIRCRAFT STRUCTURAL INTEGRITY PROGRAM	15
2-1-1 Loads/Environment Spectra Survey	16
2-1-2 Fleet Structural Maintenance Plan	17
2-1-3 Individual Airplane Tracking Program	18
2-1-4 Fleet Usage Variability Program	18
2-2 PEAK-VALLEY SEARCH ALGORITHM	18
2-3 AIRFRAME OPERATIONAL LOADS SPECTRA	22
2-3-1 Regression Methods	22
2-3-2 Regression Equation Development	23
2-4 FATIGUE LIFE MONITORING PROCEDURES	24
CHAPTER III
FLIGHT LOADS TEST OF THE INDIGENOUS DEFENSIVE FIGHTER	28
3-1 BACKGROUND	28
3-2 TEST ARTICLE DESCRIPTION	29
3-3 INSTRUMENTATION AND DATA RECORDING	31
3-4 GROUND CALIBRATION TEST	32
3-5 FLIGHT TEST	35
3-6 RESULTS	37
CHAPTER IV
ARTIFICIAL NEURAL NETWORKS FOR NONLINEAR DYNAMIC FIGHTER SERVICE LOADS PREDICTION	38
4-1 BACKGROUND	38
4-2 LOADS-PREDICTING RECURRENT NEURAL NETWORK	39
4-2-1 Recurrent Neural Network Architecture	39
4-2-2 Nonlinear Auto-regression with Exogenous Inputs Model	40
4-2-3 Selections of Input and Output Parameters	44
4-3-4 Transfer Function and Back-propagation Learning	44
4-3 THE DATABASE OF FLIGHT LOADS TEST	46
4-4 RNN LOADS PREDICTION USING FLIGHT TEST DATA	47
4-4-1 Dynamic and Static Service Loads Prediction	47
4-4-2 RNN Model for Symmetric Wind-Up Turn Maneuver	48
4-4-3 RNN Model of 1-g 360° Roll Maneuver	50
4-4-4 Error Survey of Full Flight Envelope	51
4-4-5 RNN Model of Elevated-g Roll Maneuver	52
4-4-6 Discussions	53
CHAPTER V
RECORDER DATA VALIDATION	55
5-1 BACKGROUND	55
5-2 SIMULATION PROGRAM AND SIMULATED PARAMETERS	56
5-3 NOISE FILTERING USING AANN	58
5-3-1 Auto-associative Neural Networks	58
5-3-2 AANN Training	59
5-3-3 Training Sample Generation	60
5-3-4 AANN Training, Testing and Filtering Demonstration	62
5-4 FLIGHT RECORDER DATA VERIFICATION ALGORITHM	63
5-4-1 Sensor Failure Detection, Identification and Accommodation (SFDIA)	63
5-4-2 Self-mapping AANN and Decentralized Neural Network	64
5-4-3 Training and Testing of Self-mapping AANN and DNN	65
5-4-4 Flight Data Verification Algorithm	66
5-5 NUMERICAL DEMONSTRATION OF FLIGHT DATA VERIFICATION ALGORITHM	68
CHAPTER VI
CONCLUSION AND RECOMMENDATIONS	70
REFERENCES	73
APPENDIX A	78



 

LIST OF TABLES

Table 2-1 Major tasks of the aircraft structural integrity program.	82
Table 2-2 Terms relevant to the recognition of peaks.	83
Table 2-3 Parameters and functions used in linear regression analysis.	84
Table 4-1 The selection criteria of flight parameters.	85
Table 4-2 Parameter histories of wind-up turn (flight condition 5).	86
Table 4-3 Parameter histories of 1g-360° roll (flight condition 5).	87
Table 5-1 Major design data of CH-1 trainer.	88
Table 5-2 Sensor noise intensities of flight parameters.	89
Table 5-3 Noise-filtering effectiveness (ηfilter)	90
Table 5-4 Decentralized neural networks definitions.	91





 

LIST OF FIGURES

Fig. 1-1 The relationship among flight parameters, strain gauge response and structural loads.	92
Fig. 1-2 Artifical neural network method to improve the prediction	93
of service loads and validate the recorder data.	93
Fig. 2-1 Current fighter aircraft structural fatigue life tracking procedures.	94
Fig. 2-2 Hypothetical NZ data.	95
Fig. 2-3 Wing root predicted vs. actual bending moment.	96
Fig. 2-4 Operational usage variation versus crack growth.	97
Fig. 3-1 Static test and flight loads test flow chart.	98
Fig. 3-2 Roll command time history in automatic roll function.	99
Fig. 3-3 IDF flight loads measurement.	100
Fig. 3-4 Flight loads test data collection procedures.	101
Fig. 3-5 Graphic CRT real time display.	102
Fig. 3-6 IDF ground calibration test setup.	103
Fig. 3-7 Flight loads test sequence.	104
Fig. 3-8 Wing and horizontal tail (HT) bending moment in symmetric maneuvers.	105
Fig. 4-2 Recurrent neural network with hidden neurons.	106
Fig. 4-3 Linear and tangential sigmoid transfer function.	107
Fig. 4-4 The basic architecture of NARX model.	108
Fig. 4-5 Mach/Altitude block definition.	109
Fig. 4-6 Parameter time histories of wind-up turn.	110
Fig. 4-7 Major parameters time history of 1-g 360o roll.	111
Fig. 4-8 The influences of the number of unit in tapped-delay-line memory to the RNN prediction results of wind-up turn.	112
Fig. 4-9 The influences of units used in tapped-delay-line memory (1-g 360 deg. Roll).	113
Fig. 4-10 RNN network architecture.	114
Fig. 4-12a FNN predicted vs. measured loads of flight condition 5.	116
Fig. 4-12b RNN predicted vs. measured loads of flight condition 5.	116
Fig. 4-13 The RNN learning convergent curve of flight condition 5.	117
Fig. 4-15a FNN predicted vs. measured loads of flight condition 5.	119
Fig. 4-15b RNN predicted vs. measured loads of flight condition 5.	119
Fig. 4-16 Error survey results of testing flights.	120
Fig. 4-17 The comparison between the measured and predicted data at conditions 4, 9, 14, and 17.	121
Fig. 4-18a RNN training results of flight condition 5 for elevated-g roll.	122
Fig. 4-18b RNN testing results of flight condition 4 for elevated-g roll.	123
Fig. 4-19 Error survey of testing conditions for symmetric wind-up turn using RNN approximator of elevated-g roll.	124
Fig. 4-20 Insufficient data points appearing in the symmetric portion of the elevated-g roll of flight condition 5.	125
Fig. 5-1 Three views drawing of CH-1 trainer.	126
Fig. 5-3 Body-axis system.	127
Fig. 5-4 The CH-1 simulation results of symmetrical push-over maneuver.	128
Fig. 5-4 The CH-1 simulation results of symmetrical push-over maneuver. (cont.).	129
Fig. 5-5 Auto-associative neural network architecture.	130
Fig. 5-6 The convergence history of the AANN training.	131
Fig. 5-7 Comparison of noise-free, noisy, and AANN filtered data (K=0.04).	132
Fig. 5-8 Comparison of noise-free, noisy, and AANN filtered data (K=0.08).	133
Fig. 5-9 Four sensor fault types: Category I: hard failure (large sudden jump, complete loss), Category II: soft failure (constant bias, drifting).	134
Fig. 5-10 The recorder data verification algorithm.	135
Fig. 5-11a Self-mapping AANN results (alpha parameter contains 20% bias).	136
Fig. 5-11b DNN reconstruction and the resulting AANN error (alpha parameter contains 20% bias).	137
Fig. 5-12a Input/output comparison of self-mapping AANN (pitch rate parameter contains 20% bias).	138
Fig. 5-12b DNN reconstruction and the resulting AANN error (pitch rate parameter contains 20% bias).	139
                                    

[1] Broek, David, The Practical Use of Fracture Mechanics, Kluwer Academic Publisher, Boston, 1988.
[2] Bartelds, G.,“Aircraft Structural Health Monitoring, Prospects for Smart Solutions from a European Viewpoint,”Journal of Intelligent Material Systems and Structures, Vol. 9, No. 11, 1998, pp. 906-910.
[3] Bach, R. E. and Wingrove, R. C.,“Application of State Estimation in Aircraft Flight Data Analysis,”AIAA Paper 83-2087, Gatlinburg, 1983.
[4] Chang, Y. B., Lin, T. Y., Wang, J. R., and Chen, C. C.,“Flight Loads Test of the Indigenous Defensive Fighter,”Journal of Aircraft, Vol. 31, No. 4, July-Aug. 1994.
[5] Cook, A. B., Fuller, C. R., O’Brien, W. F., and Cabell, R. H.,“Artificial Neural Networks for Predicting Nonlinear Dynamic Helicopter Loads,”AIAA Journal, Vol. 32, No. 5, May 1994.
[6] Dorr, R., Kratz, F., Ragot, J., Loisy, F., and Germain, J. L.,“Detection, Isolation, and Identification of Sensor Faults in Nuclear Power Plants,”IEEE Transactions On Control Systems Technology, Vol. 5, No. 1, Jan. 1997.
[7] Defiore, T. and Barnes, T.,“The FAA Operational Loads Monitoring Program,”The 1999 USAF Aircraft Structural Integrity Program Conference, 30 November – 2 December 1999.
[8] Faller, W. E., Schreck, S. J., and Luttges, M. W.,“Real-Time Prediction and Control of Three-Dimensional Unsteady Separated Flow Fields Using Neural Networks,”AIAA Paper 94-0532, 1994.
[9] Faller, W. E. and Schreck, S. j.,“Neural Networks: Application and Opportunities in Aeronautics,”Progress of Aerospace Science, Vol. 32, pp. 433-456, 1996.
[10] Faller, W. E. and Schreck, S. j.,“Unsteady Fluid Mechanics Applications of Neural Networks,” AIAA Paper 95-0529, January 9-12, Reno, 1995.
[11] Faller, W. E., Smith, W. E., and Huang, T. T.,“Applied Dynamic System Modeling: Six Degree-of-Freedom Simulation of Forced Unsteady Maneuvers Using Recursive Neural Networks,”AIAA Paper 97-0336, January 6-10, 1997
[12] Faller, W. E., Schreck, S. j., and Luttges, M. W.,“Real-Time Prediction and Control of Three-Dimensional Unsteady Separated Flow Fields Using Neural Networks,”AIAA Paper 94-0532, January 10-13, 1994.
[13]Fausett, L.,“Fundamentals of Neural Networks: Architectures, Algorithms and Applications,” Prentice Hall, Englewood Cliffs, New Jersey, ISBN 0-13-334186-0, 1994.
[14] Fujimoto, W.T.,“Regression-Based Loads Correlation: A Tool for Improving the Accuracy of Damage Tolerance Analyses,”The 1999 USAF Aircraft Structural Integrity Program Conference, 30 November – 2 December 1999.
[15] Greenman, R. M., Stepniewski, S. W., Jorgensen, C. C., and Roth K. R.,“Designing Compact Feedforward Neural Models with Small Training Data Sets,”Journal of Aircraft, Vol. 39, No. 3, May-June 2002.
[16] Haykin, S., Neural Networks, A Comprehensive Foundation, Macmillian College Publishing Co. Inc., New York, 1994.
[17] Hagan, M. T., Demuth, H. B., and Beale, M., Neural Network Design, An International Thomson Publishing Company, 1996.
[18] Hoffman, M. E.,“A Neural Network Prototype for Predicting F-14B Strains at B.L. 10 Longeron,”NAWCAD WAR-92042-60, 1992.
[19] Hoffman, M. E.,“Neural Network Prototype for Individual Aircraft Tracking,”Neural Air Warfare Center, Code 6042, pp. 802-819, 1992.
[20] Houston, M. I. And Matthews, A. J.,“The Royal Australian Air Force P-3C Flight Loads Test Program,”The 1999 USAF Aircraft Structural Integrity Program Conference, 30 November – 2 December 1999.
[21] Ikbal, M. S., Misra, H., and Yegnanarayana, B.,“Analysis of Auto-associative Mapping Neural Networks,”IEEE, 0-7803-5529-6/99, 1999.
[22] Johnson, M. R. and Denegri Jr., C. M.,“Comparison of Static and Dynamic Neural Networks for Limit Cycle Oscillation Prediction,”Journal of Aircraft, Vol. 40, No. 1, January-February 2003, pp. 194-203.
[23] Kim, H. C., Kim, D., and Bang, S. Y.,“A Principal Component Analysis Mixture Model with an Efficient Model Selection Method,”IEEE, 0-7803-7044-9/01, 2001.
[24] Kim, D. and Marcinniak, M.,“Prediction of Vertical Tail Maneuver Loads Using Backpropagation Neural Networks,”Journal of Aircraft, Vol. 37, No. 3, May-June 2000.
[25] Kim, D. and Marcinniak, M.,“Horizontal Tail Maneuver Load Prediction Using Backpropagation Neural Networks, ”Journal of Aircraft, Vol. 39, No. 2, March-April 2002.
[26] Lomax, T. L., Structure Loads Analysis for Commercial Transport Aircraft: Theory and Practice, AIAA Education Series, 1996.
[27] Lippmann, R. P.,“An Introduction to Computing with Neural Nets,”IEEE ASSP Magazine, pp. 4-22, April 1987.
[28] Laban, M.,“On-Line Aircraft Aerodynamic Model Identification,”Proefschrift Technische Universiteit Delft.-Met Lit. Opg., 1994.
[29] Moller, J. C., Litt, J. S., and Guo, T. H.,“Neural-Network-Based Sensor Validation for Turboshaft Engines,”AIAA Paper 98-3605, 1998.
[30] Maine, R. E. and Iliff, K. W.,“Application of Parameter Estimation to Aircraft Stability and Control,”NASA RP-1168, 1986.
[31] MIL-STD-1530A, Aircraft Structural Integrity Program, Airplane Requirements, 11 December 1975.
[32] MIL-A-008866A, Military Specification – Airplane Strength and Rigidity Reliability Requirements, Repeated Loads and Fatigue, 31 March 1971.
[33] MIL-A-83444 (USAF), Airplane Damage Tolerance Requirements, July 1974.
[34] MIL-STD-1553, Multiplex Applications Handbook, 1975.
[35] Neppach, D., and Casdorph, V. A.,“Sensor Failure Detection, Identification and Accommodation in a System Without Sensor Redundancy,”AIAA Paper 95-0011, 1995.
[36] Napolitano, M. R., Neppach, C., Casdorph, V., and Naylor, S.,“Sensor Failure Detection, Identification, and Accommodation Using On-Line Learning Neural Architectures,”AIAA Paper 94-3598, 1994.
[37] Napolitano, M. R., Neppach, C., Casdorph, V., Naylor, S., Innocenti, M., and Silvestri, G., “Neural-Network-Based Scheme for Sensor Failure Detection, Identification, and Accommodation,”Journal of Guidance, Control, and Dynamics, Vol. 18, No. 6, pp. 1280-1286, Nov.-Dec. 1995.
[38] Napolitano, M. R., Windon, D., Casanova, J., and Innocenti, M.,“A Comparison between Kalman Filter and Neural Network Approaches for Sensor Validation,”AIAA Paper 96-3894, 1996.
[39] Napolitano, M. R., An, Y.-H., Seanor, B., Pispitsos, S., and Martinelli, D.,“Application of A Neural Sensor Validation Scheme to Actual Boeing 737 Flight Data,”AIAA Paper 99-4236, 1999.
[40] Narendra, K. S. and Parthasarathy, K.,“Identification and Control of Dynamical Systems Using Neural Networks,”IEEE Transaction on Neural Networks, Vol. 1, No. 1, pp. 1-27, March 1990.
[41] Pidaparti, R. M., Jayanti, S., and Palakal, M. J.,“Residual Strength and Corrosion Prediction of Aging Aircraft Panels: A Neural Network Study,”Journal of Aircraft, Vol. 39, No. 1, 2002, pp. 175-180.
[42] Pidaparti, R. M., Jayanti, S., Palakal, M. J. and Mukhopadhyay, S.,“Structural Integrity Redesign Through Neural-Network Inverse Mapping,”AIAA Journal, Vol. 41, No. 1, January 2003, pp. 119-124.
[43] Pidaparti, R. M., Jayanti, S., and Palakal, M. J.,“Residual Strength and Corrosion Rate Predictions of Aging Aircraft Panels: Neural Network Study,”Journal of Aircraft, Vol. 39, No. 1, January-February 2002.

校內：2004-08-25公開
校外：2004-08-25公開

簡易檢索 / 詳目顯示

相關論文