Real-time identification of missile aerodynamics using a linearised Kalman filter aided by an artificial neural network

Abstract

The paper investigates the problem of real-time identification of aerodynamic derivatives in a guided missile application. This application provides a severe test for any parameter estimator, since it has to identify the linearised parameters of a multivariable, nonlinear, time variant, noisy plant, which is initially unstable and then becomes lightly damped. Initially, two radically different approaches are taken by designing both a linearised Kalman filter (LKF) estimator and an artificial neural network (ANN) based estimator. A hybrid estimator is then formed by an LKF, which is aided by the ANN. This produces a new estimator which has superior performance to those from which it is derived. The performance of these estimators is assessed with a nonlinear single plane model against eight types of engagements.