Aeroelastic wing flutter testing and analysis

Resumo

The integration of new underwing stores in an aircraft modifies the characteristics of mass distribution (center of gravity) and moment of inertia of the wing. This effect, in addition to to the contribution of aerodynamic loads, causes the vibration modes and frequencies to vary with the dynamic pressure (a function of flight speed and altitude). This strongly non-linear phenomenon implies that, under certain conditions of dynamic pressure, the coupling in frequency (selfsustaining resonance) of two or more modes of vibration initially orthogonal to each other occurs, an aeroelastic phenomenon known as "flutter", which will lead to the loss of the aircraft and the life of the pilot unless the flight conditions change. Thus, the integration of new underwing stores requires carrying out a series of processes that will lead to a new flight envelope, within which it is guaranteed that the aircraft can fly safely. This study requires carrying out theoretical calculations to predict flutter conditions and subsequent validation through flight tests, known as "envelope expansion". Carrying out this task safely requires highly qualified and specialized means and personnel, and this implies extraordinarily high costs, which leads to companies specialized in carrying out these tests to guarding data and results as an industrial secret, and among other things it is very difficult to find validated methods to process flight data and extract vibration parameters at different dynamic pressures. Among the different published methods to identify flutter test flight vibration parameters, the vast majority have been verified only with theoretical models, with the fact that many of them give results that are inconsistent with each other or, when validated with real data, yield inconsistent results. For this reason, the main objective was to develop fast, robust and coherent techniques, capable of returning repetitive and consistent results in real time. The author had access to a flutter flight test database, courtesy of the Spanish Air Force, and has authorization from the Air Force Communications Office to use and publish results derived from his research on those data. This thesis will present a research dedicated to developing two data processing methods for flutter flight tests, in particular on data from a "Sine-Dwell" type excitation, one based on a mathematical model and optimization techniques, and another based on deep learning techniques. The development of both techniques, is based on a first verification of different techniques documented in the bibliography by different authors, as well as on the training of different neural networks; Multilayer perceptrons, deep neural networks and convolutional neural networks. Once a comparison baseline was available, a classical technique was selected (based on a theoretical model and optimization), according to the bibliographic source validated with real data from flutter flight tests, and one of the trained neural networks. Based on the lessons learned, an innovative technique was developed based on the classical model of theoretical model and optimization, verification with synthetic data and comparison of the three previously selected techniques. Finally, the three techniques were validated with real data from flutter flight tests. The results obtained are highly satisfactory, reaching the initially planned objectives. The techniques presented have been verified with synthetic data, compared with previously independently validated bibliographic models and validated in this study with real data. The results are consistent with expectations. The speed of the process allows the analysis of data in real time, increases the situational awareness of the test director and facilitates decision-making to continue or stop the test, in dangerous conditions, with greater safety.