Time-Domain System Identification for Long-EZ Fixed-Wing Aircraft Based on Flight Test Data

Abstract

This paper provides a system identification procedure on a fixed-wing aircraft Long-EZ, which is the critical stage of providing accurate aircraft models for autopilot design in the future. Flight test has been carried out by ITPS Canada Ltd. Extensive real flight test data have been utilized for the identification and verification of a linear transfer function model, a nonlinear neural network model, and a block-oriented model consisting of linear and nonlinear elements. Linear transfer function structure has been determined with physical dynamics, and the model parameters have also been identified. Nonlinearity of aircraft dynamics has been identified using a multilayer perceptron (MLP) neural network structure. Flight data has also been utilized to train this MLP structure. Performance comparison results have demonstrated different predicting capabilities of the developed linear, nonlinear, and block-oriented models. The developed block-oriented model shows its predicting capability in a more satisfactory manner.