Login | Register

Dynamic recurrent neural networks for stable adaptive control of wing rock motion

Title:

Dynamic recurrent neural networks for stable adaptive control of wing rock motion

Kooi, Steven Boon-Lam (1999) Dynamic recurrent neural networks for stable adaptive control of wing rock motion. PhD thesis, Concordia University.

[thumbnail of NQ47693.pdf]
Preview
Text (application/pdf)
NQ47693.pdf
6MB

Abstract

Wing rock is a self-sustaining limit cycle oscillation (LCO) which occurs as the result of nonlinear coupling between the dynamic response of the aircraft and the unsteady aerodynamic forces. In this thesis, dynamic recurrent RBF (Radial Basis Function) network control methodology is proposed to control the wing rock motion. The concept based on the properties of the Presiach hysteresis model is used in the design of dynamic neural networks. The structure and memory mechanism in the Preisach model is analogous to the parallel connectivity and memory formation in the RBF neural networks. The proposed dynamic recurrent neural network has a feature for adding or pruning the neurons in the hidden layer according to the growth criteria based on the properties of ensemble average memory formation of the Preisach model. The recurrent feature of the RBF network deals with the dynamic nonlinearities and endowed temporal memories of the hysteresis model. The control of wing rock is a tracking problem, the trajectory starts from non-zero initial conditions and it tends to zero as time goes to infinity. In the proposed neural control structure, the recurrent dynamic RBF network performs identification process in order to approximate the unknown non-linearities of the physical system based on the input-output data obtained from the wing rock phenomenon. The design of the RBF networks together with the network controllers are carried out in discrete time domain. The recurrent RBF networks employ two separate adaptation schemes where the RBF's centre and width are adjusted by the Extended Kalman Filter in order to give a minimum networks size, while the outer networks layer weights are updated using the algorithm derived from Lyapunov stability analysis for the stable closed loop control. The issue of the robustness of the recurrent RBF networks is also addressed. The effectiveness of the proposed dynamic recurrent neural control methodology is demonstrated through simulations to suppress the wing rock motion in AFTI/F-16 testbed aircraft having the delta wing configuration. The potential implementation as well as the practicality of the control methodology are also discussed

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Mechanical and Industrial Engineering
Item Type:Thesis (PhD)
Authors:Kooi, Steven Boon-Lam
Pagination:xx, 229 leaves : ill. ; 29 cm.
Institution:Concordia University
Degree Name:Ph. D.
Program:Mechanical and Industrial Engineering
Date:1999
Thesis Supervisor(s):Neemah, R
Identification Number:TJ 217 K66 1999
ID Code:919
Deposited By: Concordia University Library
Deposited On:27 Aug 2009 17:15
Last Modified:13 Jul 2020 19:48
Related URLs:
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top