AN EXPERIENCE WITH THE NEURAL NETWORK FOR AUTO-LANDING SYSTEM OF AN AIRCRAFT Dr. Sreenatha G. Anavatti School of Aerospace, Civil and Mechanical Engineering, University of New South Wales at ADFA, Canberra, Australia Dr. Choi J. Young School of Electrical Engineering and Computer Science, Seoul National University, Seoul, Korea Mr. Francois Pischery Laboratoire d’Automatique, Industrielle Institut National des Sciences Appliquees, de Lyon, Villeurbanne, France Keywords: Auto-landing, Robust Control, Neural Network, Aircraft Dynamics Abstract: Generalization by the Neural Networks is an added advantage that can provide very good robustness and disturbance rejection properties. By providing a sufficient number of training samples (inputs and their corresponding outputs), a network can deal with some inputs it has never seen before. This ability makes them very interesting for control applications because not only they can learn complicated control functions but they are able to respond to changing or unexpected environments. Aircraft landing system provides one such scenario wherein the flight conditions change quite dramatically over the path of descent. The present work discusses the training of a neural network to imitate a robust controller for auto-landing of an aircraft. The comparisons with the robust controller indicate the additional advantages of the neural network 1 INTRODUCTION Auto-landing is a requirement in the modern aircraft due to the necessity for operations under all weather conditions, whether it is civilian aircraft or military aircraft. Considerable efforts have gone in designing suitable control systems for enhancing the auto- landing capability[1,5]. The auto-landing consists of the two phases, the descent phase and the flare. During the descent phase, the glide slope control system guides the aircraft down a pre-determined glide-slope. When the aircraft reaches a pre-selected altitude, the flare control system reduces the rate of descent and causes the aircraft to flare out and touch down with an acceptably low rate of descent. The control system achieves this by the control of the flight path angle γ. It is shown in reference (John H. Blakelock, 1991) that the automatic control of the flight path angle without simultaneous control of the airspeed (either manual or automatic) is practically not possible. The combination of these three systems provides the full longitudinal control of the aircraft. The dynamics of the aircraft is governed by stability derivatives which are functions of flight regime (speed, altitude, density, temperature, etc.). Due to the variations in the flight regime during landing, the dynamics of the aircraft change considerably over the entire flight regime. Hence, a time varying mathematical model is required. Due to the difficulty in handling time-varying differential equations, mathematical models at a number of points in the descent are considered simultaneously. This adds a large amount of uncertainty in modelling employed in the design of flight control systems. In addition, disturbances in terms of gusts and sensor and actuator noise can alter the performance of control system considerably. Hence, there is a necessity for having robust control systems that can handle parameter variations along with good disturbance rejection properties. H-infinity(Ching- Fang Lin, 1995) controller provides one such 393 G. Anavatti S., J. Young C. and Pischery F. (2004). AN EXPERIENCE WITH THE NEURAL NETWORK FOR AUTO-LANDING SYSTEM OF AN AIRCRAFT. In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 393-400 DOI: 10.5220/0002627603930400 Copyright c  SciTePress