An Overview of Relevant Issues for Aircraft Model Identification


A. Bucharles, C. Cumer, G. Hardier, B. Jacquier, A. Janot, T. Le Moing, C. Seren, C. Toussaint, P. Vacher

PDF icon AL04-02.pdf2.23 MB

A number of activities in aeronautical engineering rely on the availability of models to represent the real behavior of the aircraft. Let us quote, for example, the development of autopilots and synthesis of flight control laws, the study of the handling qualities, the fault monitoring process, the prediction of hazardous behaviors, or the implementation of simulators used to train the pilots and to validate hardware and software systems. The initial modeling derived from CFD, wind tunnel or ground tests is seldom reliable enough with respect to the requirements. Hence, the needed accuracy is finally achieved thanks to suitable identification techniques and to a set of peculiar flight tests. In addition, the complexity of the models has increased in recent years, along with more stringent accuracy requirements to satisfy the raising constraints of the new aeronautical devices which make use of these models; e.g., an increasing number of vibration modes in the low frequency range for flexible aircraft, or a larger complexity and non-linearity of the aerodynamical models in the rigid case. Hence, the variety of problems and models under consideration entails taking an interest in a wide range of identification techniques. These include basic methods, like least-squares or maximum likelihood and their variants, spectral analysis and estimators based on Kalman filtering, as well as more recent approaches like neural-based or subspace methods. Special care is given to the frequency domain formulation of the algorithms, especially in the flexible A/C case. Most of these methods are not directly usable as they are and need to be adapted to the peculiarities of  aeronautics. Accordingly, this paper reviews the various issues related to the identification process when applied to such applications. These steps include data pre-processing, input design, time vs. frequency domain methods, model validation, etc., and are illustrated by industrial problems dealt with by Onera, for rigid as well as for flexible A/C modeling.