Adaptive retuning of feedforward controller - Application to the airbrake compensation of an aircraft

This paper deals with the adaptive retuning of a feedforward controller, under the normalized lattice form, for a parameter-varying closed-loop system. The objective is to tune the controller in real-time during specific flights so that it does not need the adaptive part in nominal operation. The method was developed to help aeronautical design engineers to retune specific feedforward control laws at the early stage of the design process, i.e. when aircraft models are not fully reliable. The idea is to give more weight to the system itself in the control laws tuning process. Using a design method based on inverse simulation, we use a combination of adaptive filtering, local learning and optimal control techniques to achieve a real-time tuning. The method consists in estimating the system's inverse response in real-time using feedback control and in locally retuning the control law, whose parameters are interpolated using neural networks. The method is tested on the airbrake compensation of a civilian aircraft.

[1]  P. Regalia.  Adaptive IIR Filtering in Signal Processing and Control , 1994 .

[2]  Daniel Alazard,et al.  REAL-TIME AUTOMATIC TUNING OF AN AIRCRAFT SIDESLIP ESTIMATOR , 2007 .

[3]  Ph. Mouyon,et al.  An autotuned deconvolution algorithm with applications to aircraft landing tests matching and autopilot retuning , 2002, Proceedings of the 41st IEEE Conference on Decision and Control, 2002..

[4]  Roderick Murray-Smith,et al.  Local model networks and local learning , 1994 .

[5]  B. Widrow,et al.  Adaptive inverse control , 1987, Proceedings of 8th IEEE International Symposium on Intelligent Control.

[6]  Shaohua Tan,et al.  Gain scheduling control of nonlinear plant using RBF neural network , 1996, Proceedings of the 1996 IEEE International Symposium on Intelligent Control.

[7]  Léon Bottou,et al.  Local Learning Algorithms , 1992, Neural Computation.