Aircraft structures are prone to vibrations due to their lightweight construction. In most cases, these vibrations are not critical. They decay due to damping effects from aerodynamic forces and damping forces from the structure itself. Varying flight conditions, like airspeed, altitude and temperature, have a large impact on the damping behavior. During flight vibration tests the complex aeroelastic behavior of an aircraft is experimentally identified, to be sure that no critical vibrations can occur within the flight envelope. The DLR Institute of Aeroelasticity has enhanced its flight-testing methods by the use of artificial intelligence to get faster and even more reliable data.
The flutter phenomenon
A well-known vibration phenomenon on aircraft wings is "flutter". In case of aircraft flutter, eigenmodes can couple with each other. This happens when their frequencies shift, in particular also due to aerodynamic influences. Consequently, the damping of one mode becomes zero or negative. Only an initial small disturbance is required to cause instabilities with fast increasing vibration amplitudes. The permanent excitation with external forces is not required to produce large vibrations. This is referred to as "self-excited vibrations".
System identification on ground and in flight
Before first flight of an aircraft prototype the ground vibration test is performed to deliver a structural dynamic identification of the whole aircraft on ground. The results cover modal parameters such as eigenfrequencies, damping ratios and mode shapes in the frequency range of interest. This test explicitly excludes aerodynamic effects to update the structural model for further clearance of the first flight. An example of a ground vibration test is given in the following blog article (ISTAR Ground Vibration Test).
Since the vibration behavior of an aircraft changes with varying flight conditions, the flight envelope is successively expanded during the flight test campaign of an aircraft prototype. This campaign is used to verify the predicted aeroelastic stability in terms of eigenfrequencies and damping ratios with varying flight conditions. Typically, artificial excitation is used with dedicated control surface inputs or other external excitation devices which induce aircraft vibrations in order to analyze the aeroelastic behavior. The DLR Institute of Aeroelasticity is propagating a method that uses the natural turbulence as the only source of excitation during flight.
Output-only modal analysis
The identification of aircraft modal parameters during GVT is a core competence of the Institute of Aeroelasticity. In recent years, the experimental identification methods have been extended in such a way that the modal parameters of an aircraft can also be identified in-flight [1-4] without further artificial excitation. Natural excitation by turbulent flow is sufficient in most cases. However, since the excitation forces cannot be measured, the methods are based solely on the measured response signals of the aircraft ("output-only modal analysis"). Modal analysis is a complex identification procedure and is therefore applied by trained engineers. The underlying mathematical algorithms have been accelerated by efficient programming that modal parameters are available within a few seconds. However, operation by an engineer would not be possible in this short time, so in this work autonomous algorithms take over the engineer's tasks which is described by the three following enhancements for the standard methods.
1. Autonomous analysis using artificial intelligence
The core task of modal analysis is to extract the physically correct solution from many possible mathematical solutions. This is solved manually via a so-called stabilization diagram (Fig. 1a). For the autonomous analysis, multilayer clustering is applied (Fig. 1b). This is a method from the field of unsupervised machine learning.
Figure 1: Autonomous analysis of the stabilization chart
a: Unprocessed stabilization diagram b: Multi-layer clustering c: Cleaned and sorted result
Clustering as well as the identification methods require setting parameters in order to achieve optimal results. In the field of machine learning those parameters are referred to as hyperparameters. In order to avoid time-consuming manual adjustment of the hyperparameters, a system was developed that optimizes them semi-autonomously using Gaussian processes. The intelligent system can thus teach itself an optimal way of analyzing, e.g. for a new prototype [5].
3. Data fusion – Combination of different analysis methods
Since during a flight the boundary conditions are different from those in the laboratory, the uncertainties of the modal identification methods are larger than, for example, in a ground vibration test. In order to reduce these uncertainties, the analysis system has been extended to include various identification methods. Now, different methods (e.g. time-domain methods and frequency-domain methods) with different strengths can be used in parallel. The results of these methods are fused automatically so that the advantages of both approaches can be combined [6].
The developed methods have been successfully demonstrated during a ground vibration test in Brunswick [7], a wind tunnel test campaign in the European transonic wind tunnel (ETW) in Cologne, a UAV flight test campaign in Cochstedt [5, 8] and finally on board the DLR research aircraft ISTAR.
Future research outlook
In the future, further reduction of uncertainty bounds for damping estimates is envisaged to provide real-time methods not only for flight vibration testing but also for further applications where online identification is necessary.
Jelicic, Goran and Schwochow, Jan and Govers, Yves and Böswald, Marc (2016) Automatische Schwingungsüberwachung von aeroelastischen Systemen. VDI-Berichte, 2259, Seiten 211-222. VDI-Verlag. VDI-Fachtagung Schwingungsanalyse und Identifikation, 15.-16. Mrz. 2016, Fulda, Deutschland. ISBN 978-3-18-092259-1. ISSN 0083-5560.
Volkmar, Robin and Soal, Keith Ian and Buchbach, Ralf and Sinske, Julian and Govers, Yves and Böswald, Marc (2022) Semi-autonomous analysis of large aircraft ground vibration tests. International Forum on Aeroelasticity and Structural Dynamics - IFASD, 13.-17. Jun. 2022, Madrid, Spanien.
Soal, Keith Ian and Thiem, Carsten and Meier, Tobias and Volkmar, Robin and Sinske, Julian and Govers, Yves and Böswald, Marc (2022) Embedded flight vibration testing system for online flutter monitoring of UAVs. International Forum on Aeroelasticity and Structural Dynamics - IFASD, 13.-17. Jun. 2022, Madrid, Spanien.
Author
Robin Volkmar, Abteilung Strukturdynamik und Systemidentifikation, DLR-Institut für Aeroelastik
