Active Flutter Suppression for a Three-Surface Transport Aircraft by Recurrent Neural Networks

This paper presents an effective approach for the design of a flutter-suppression system by means of recurrent neural networks. This system is used to move flutter instabilities outside the flight envelope of an unconventional three-surface transport aircraft. The design process requires a comprehensive aircraft model in which flight mechanics, structural dynamics, unsteady aerodynamics, and control-surface actuators are represented in state-space form, according to the modern aeroelastic approach. The implemented regulator is based on two recurrent neural networks: one is trained to identify the system dynamics and the other acts as a controller using an indirect inversion of the identified model. Keeping the training of both recurrent networks online leads to an adaptive control system. Extensive numerical tests are used to tune the neural network design parameters and to show how the neural controller increases system damping, widening the flutter-free flight envelope by more than 15% of the uncontrolled flutter velocity.

[1]  W. Rodden,et al.  A doublet-lattice method for calculating lift distributions on oscillating surfaces in subsonic flows. , 1969 .

[2]  W. P. Rodden,et al.  Equations of motion of a quasisteady flight vehicle utilizing restrained static aeroelastic characteristics , 1985 .

[3]  Paolo Mantegazza,et al.  Single Finite States Modeling of Aerodynamic Forces Related to Structural Motions and Gusts , 1999 .

[4]  Bernard Etkin,et al.  Dynamics of Atmospheric Flight , 1972 .

[5]  L. Morino,et al.  Matrix fraction approach for finite-state aerodynamic modeling , 1995 .

[6]  Franco Bernelli-Zazzera,et al.  Adaptive Control of Space Structures via Recurrent Neural Networks , 1999 .

[7]  Gregory J. Hancock.  An introduction to the flight dynamics of rigid aeroplanes , 1995 .

[8]  Ronald J. Williams,et al.  A Learning Algorithm for Continually Running Fully Recurrent Neural Networks , 1989, Neural Computation.

[9]  Tommy W. S. Chow,et al.  A recurrent neural-network-based real-time learning control strategy applying to nonlinear systems with unknown dynamics , 1998, IEEE Trans. Ind. Electron..

[10]  D. Sorensen,et al.  Approximation of large-scale dynamical systems: an overview , 2004 .

[11]  K. L. Roger,et al.  Airplane Math Modeling Methods for Active Control Design , 1977 .

[12]  Jiri Cecrdle,et al.  Experimental investigations of a vibration suppression system for a three surface aeroelastic model , 2005 .

[13]  Paolo Mantegazza,et al.  Active Flutter Suppression Using Recurrent Neural Networks , 2000 .

[14]  Liang Jin,et al.  Adaptive control of discrete-time nonlinear systems using recurrent neural networks , 1994 .

[15]  Sergio Ricci,et al.  Control of an all-movable foreplane for a three surfaces aircraft wind tunnel model , 2006 .

[16]  Kumpati S. Narendra,et al.  Control of nonlinear dynamical systems using neural networks: controllability and stabilization , 1993, IEEE Trans. Neural Networks.

[17]  K S Narendra,et al.  Control of nonlinear dynamical systems using neural networks. II. Observability, identification, and control , 1996, IEEE Trans. Neural Networks.