Parameter estimation from flight data with process and measurement noise using neural networks

Abstract

Estimating parameters from flight data having both the measurement and process noise poses difficulties with most of the available parameter estimation methods. For such flight data, a recently proposed neural network based parameter estimation method '(called the Delta method) is used for extraction of longitudinal parameters. Results are presented for simulated flight data in turbulent atmosphere, and it is shown that the Delta method estimates are good even from flight data in severe turbulence and also having measurement noise in them. For nonlinear terms present in the aerodynamic model, a frequency-domain application of the Delta method is also illustrated. A comparison of estimates via the Delta method with those via the output error method and time varying filter error approach is presented. Nomenclature