Enkelmann, Franz Alexander Richard (2024)
A Hybrid Model for the Estimation of Time-Variant Aircraft Parameters.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00027405
Ph.D. Thesis, Primary publication, Publisher's Version
Text
A_Hybrid_Model_for_the_Estimation_of_Time_Variant_Aircraft_Parameters_Dissertation_Franz_Enkelmann_05_2024.pdf Copyright Information: CC BY-SA 4.0 International - Creative Commons, Attribution ShareAlike. Download (38MB) |
Item Type: | Ph.D. Thesis | ||||
---|---|---|---|---|---|
Type of entry: | Primary publication | ||||
Title: | A Hybrid Model for the Estimation of Time-Variant Aircraft Parameters | ||||
Language: | English | ||||
Referees: | Klingauf, Prof. Dr. Uwe ; Kirchner, Prof. Dr. Eckhard | ||||
Date: | 17 June 2024 | ||||
Place of Publication: | Darmstadt | ||||
Collation: | xxv, 236 Seiten | ||||
Date of oral examination: | 10 January 2024 | ||||
DOI: | 10.26083/tuprints-00027405 | ||||
Abstract: | Knowing the health of the aircraft supports safe, efficient, and sustainable operations. It allows maintenance measures and retrofits of the aircraft in service to be used economically. To meet the safety criteria for flight operations, a reliable assessment is required. On the other hand, the assessment methods must become more precise. This ensures technical modifications that even have a minor impact on the health of the aircraft, such as sharkskin, can be classified as significant. If such modifications are used fleet-wide and in a targeted manner, their potential can better be utilized. To address the demand for increased precision, machine learning methods are used in the context of prognostics and health management. However, the safety criterion in aviation manifests the need for reliable and comprehensive methods, which are experienced in physical model building. Consequently, the combination of both, the hybrid model, found a branch of research to profit from the individual advantages. The hybrid model is considered in this thesis to estimate and predict aircraft system states accurately and physically consistently. The states are set to include time-variant aircraft parameters representing system degradation and recovery within the aircraft life cycle. Their estimation is assumed to be a key element for health assessment. However, present approaches to hybrid model building are shown to be inappropriate in such state estimation tasks. Therefore, a new hybrid model is developed in this thesis. The developed hybrid model is characterized by a parameter scheduling structure and a recursive filter method for system identification. Thus, a data-driven model, an artificial neural network, can extend a physical model. Furthermore, requirements are defined in initialization, generalization, interpretation, and recovery, addressed within the development process. The new hybrid model offers stepwise learning while considering measurement noise and joint state estimation. In this way, the parameters and the weights of the artificial neural network are considered states besides the dynamical states. Consequently, both model parts adapt simultaneously. To handle nonlinearities, a modified unscented Kalman filter is implemented. The use of state constraints further improves physical consistency and filter stability. The developed hybrid model is evaluated using an unmanned aircraft system example. Therefore, a flight test platform is introduced, and a flight simulation environment is developed. A database of flight tests and flight simulations is built, including aircraft modifications that abstract degradation and recovery effects. In four test series, the aircraft is investigated by estimating the system states, including selected parameters, using different physical models and databases. In detail, the unmodified and aerodynamically modified aircraft are considered. In the case of real flight tests, the aircraft’s main wing area is extended, and the virtual aircraft’s parameters are changed in the case of flight simulations. The hybrid model can estimate and predict the aircraft system states. The modified parameters are estimated physically consistently in some of the test series. In conclusion, the developed hybrid model can estimate and predict system degradation and recovery in perspective and meets the defined requirements. However, two dilemmas are recognized, which require further improvement. One concerns the hybrid model structure, and the other one the learning algorithm, the filtering procedure. The open hybrid model structure involves an artificial neural network, whose weighting enables high adaptability to the considered database but decreases the model’s generalizability. Second, the application-specific initialization of the covariance matrices of the filtering procedure allows for high adaptability but negatively affects the filtering quality. As a result, anomalies within the database cannot be properly detected and separated. Finally, the new hybrid model is discussed in the context of artificial and natural intelligence. Descriptions learned via data-driven models are compared to the system knowledge that can be physically experienced. The main contribution is extending a physical model by an artificial neural network, which is recursively and simultaneously adapted. For future work, the expansion of physical knowledge using artificial intelligence is proposed, where any dynamical systems can be considered. |
||||
Alternative Abstract: |
|
||||
Uncontrolled Keywords: | hybrid, physical, data-driven, modelbuilding, artificial intelligence, aircraft, unmanned aircraft system, phm, prognostics, health management, predictive, maintenance, neural network, unscented kalman filter, experimental, flight simulation | ||||
Status: | Publisher's Version | ||||
URN: | urn:nbn:de:tuda-tuprints-274054 | ||||
Classification DDC: | 600 Technology, medicine, applied sciences > 620 Engineering and machine engineering | ||||
Divisions: | 16 Department of Mechanical Engineering > Institute of Flight Systems and Automatic Control (FSR) | ||||
Date Deposited: | 17 Jun 2024 12:04 | ||||
Last Modified: | 18 Jun 2024 06:57 | ||||
URI: | https://tuprints.ulb.tu-darmstadt.de/id/eprint/27405 | ||||
PPN: | 519192397 | ||||
Export: |
View Item |