Identification and control of aircraft dynamics using radial basis function networks

Abstract

Recently, the eniergeiice of neural networks as a promising tool for approximating complex system input-output mappings has generated a great (leal of interest in the area of inodeling, identification and control of noiiliiiear dynamical systeins. One specific research are8 that woiild tieiiteticlously beiiefit from this approach is the area of identification and control of Iiiglt performance aircraft, especially at high angles of attack. At those flight conditions, the control task becomes est.remclq dificult due to added design complexity and hard nonlinearities characterizing the system. In this paper, we investigate one type of neiiml iietworks, iianiely the Radial Basis Fniictioii (RBF) networks, and apply lliein to the identification and control problems of ari aircraft system. The RBI: network is used as an on-line appiosiniator of the aircraft pitch dynamics, conibinecl with a nonlinear cotit rol law to iiiipiove the closecl-loop system perforinanre. Tlie results are illustrated through simulations using a nonlinear model of the F-18 aircraft pitch dynamics.

Learning control of mobile robots using a multiprocessor system

“…The absolute localization module based on adaptive extended Kalman filter (Jetto, Longhi, & Venturini, 1999;Jetto, Longhi, & Vitali, 1999) guarantees high accuracy with average error in module lower than 1 cm. The environment perception module acquires data on the environment by the same set of external sensors (sonar sensors) and produces a local map of the environment with the detected obstacles (Angeloni et al, 1996). This map is used by the real-time obstacle avoidance module for modifying the planned trajectory to be tracked with a new collision free trajectory.…”

“…The real-time experimental tests of the control algorithms have been performed on the LabMate mobile robot, available at the Robotics Lab of the UniversitaP olitecnica delle Marche and provided with an automatic navigation system (Conte, Longhi, & Zulli, 1995 (Angeloni, Leo, Longhi, & Zulli, 1996;Conte, Longhi, & Zulli, 1996;Jetto, Longhi, & Venturini, 1999) composed of the motion planning module which runs off-line, the localization and control modules, the perception module and the obstacle avoidance module which work on-line with real-time implementations.…”

“…The experimental results using Radial Basis Function Networks (RBFNs) control schemes (D'Amico, Ippoliti, & Longhi, 2001b) have shown that RBFN are a possible alternative to control schemes based on MNN (Corradini et al, 2003). Moreover, these nets are useful for the identification and control of complex non-linear dynamical systems in the presence of uncertainties because the particular structure of RBFN allows their applicability to adaptive and adaptive-learning control schemes (Jagannathan, Lewis, & Pastravanu, 1994;Ahmed-Zaid, Ioannou, Polycarpou, & Youssef, 1993;Polycarpou & Ioannou, 1995;Polycarpou, 1996;Sadegh, 1995;Sanner & Slotine, 1991Sanner & Essex, 1996).…”

<title>Active neurocontrol of large flexible aerospace structures</title>

“…It is usual to use one of two possible forms of equations of motion which can result: either a set of equations with constant coefficients which have been obtained from steady state aerodynamics with the unstaedy aerodynamics being approximated by Küssner & Wagner lift growth functions; or a set of equations with non-constant coefficients which depend upon the use of more exact methods of accounting for the unsteady aerodynamics. The equations of motion employed corresponded to the former set which leads to a formulation of the form: Moq+M2+ M4q+Miq* W+M3* W = c1:*K (1) where M0 represents the matrix of the generalised stiffness of the aircraft structure, M1 represents the matrix of the generalised aerodynamic stiffness, M2 represents the matrix of the generalised inertia of the aircraft structure, M3 represents the matrix of the generalised aerodynamic damping, and M4 represents the matrix of the generalised structural damping. The term F is the forcing function, C1 is a matrix of the coefficients of the generalised forcing function; W(t) is the Wagner lift function and IC(t) is the Küssner function, and * denotes convolution.…”

Identification of a non-linear F/A-18 model by the use of fuzzy logic and neural network methods

“…Radial basis function network were used to map aircraft pitch dynamics. Pitch dynamics was modelled with a non-linear system based on the F-16 A/C non-linear model [1]. Online learning is used with the NN theory.…”