Adaptive output feedback control based on neural networks: Application to flexible aircraft control

One of the major challenges in aeronautical flexible structures control is the uncertain or the non stationary feature of the systems. Transport aircrafts are of unceasingly growing size but are made from increasingly light materials so that their motion dynamics present some flexible low frequency modes coupled to rigid modes. For reasons that range from fuel transfer to random flying conditions, the parameters of these planes may be subject to significative variations during a flight. A single control law that would be robust to so large levels of uncertainties is likely to be limited in performance. For that reason, we follow in this work an adaptive control approach. Given an existing closed-loop system where a basic controller controls the rigid body modes, the problem of interest consists in designing an adaptive controller that could deal with the flexible modes of the system in such a way that the performance of the first controller is not deteriorated even in the presence of parameter variations. To this purpose, we follow a similar strategy as in Hovakimyan (2002) where a reference model adaptive control method has been proposed. The basic model of the rigid modes is regarded as a reference model and a neural network based learning algorithm is used to compensate online for the effects of unmodelled dynamics and parameter variations. We then successfully apply this control policy to the control of an Airbus aircraft. This is a very high dimensional dynamical model (about 200 states) whose direct control is obviously hard. However, by applying the aforementioned adaptive control technique to it, some promising simulation results can be achieved.