Neural control of a sea skimming missile

Jones, Llŷr Campbell (1996) Neural control of a sea skimming missile. University of Southampton, Doctoral Thesis.

Abstract

The use of a neural network to control the lateral acceleration of a missile is a novel concept. In this thesis the purpose of a missile automatic Flight Control System (AFCS) is described first, followed by a review of missile guidance and control techniques and a detailed definition of the role and function of a missile autopilot.

A clear derivation of the mathematical model of the missile airframe dynamics used throughout the research is given. A baseline discrete optimal Linear Quadratic Regulator (LQR) controller is presented and explained in some detail. The results obtained from a digital simulation of this controller follow. These results and many others from similar simulations are used to provide the necessary training and recall data for the design of the neural controller.

A comprehensive discussion of a number of different types of neural networks and neural controller implementation methods is presented, together with information about the training performance of various learning algorithms.

The neural autopilot controller is formulated as a three layer, non-linear, discrete time-series neural network whose parameters are adjusted under a supervised learning environment by using the Levenberg-Marquardt Optimisation algorithm. As opposed to using heuristic reasoning for selecting the network input and output parameters, a novel technique utilising a priori knowledge of the baseline controller is used. The current fin position command is modelled as a function of the past value of commanded fin position, and, current and past values of lateral acceleration command, achieved fin position, measured lateral acceleration and body rate. Lateral acceleration time histories for two flight speeds for a range of incidences are used to illustrate performance. The neural controller is validated using a test lateral acceleration command sequence which differs significantly from that used for generating the network training data. The effectiveness of the proposed neural controller design can be judged by these digital simulation response results. The dynamic performance of the neural controller is compared with that of the baseline LQR controller. Further simulation results in the presence of actuator and sensor noises shed light on the ability of the neural controller to serve as a practical tool for flight control system design.

This record has no associated files available for download.

Contributors

Download statistics