Online aerodynamic parameter estimation for a fault tolerant flight control system.

摘要

Wichita State University (WSU) and Raytheon Aircraft Company are working toward the development of a flight control system to reduce the workload for a pilot under normal as well as deteriorated flight conditions. An 'easy fly system' for a Bonanza Raytheon NASA test-bed has been used by WSU to develop a neural network-based adaptive flight control system. In this thesis an online technique for aerodynamic parameter estimation is presented, which is developed to improve the adaptation. The neural-based adaptive flight controller uses an artificial neural network for immediate adaptation in dynamic inverse control to compensate for modeling error or control failure. Long-term adaptation to modeling error requires a permanent correction of the aerodynamic parameters used in the inverse controller. This method is designed to update parameters inside the controller and to provide slow and long-term adaptation to compliment the existing immediate adaptation provided by neural networks. The method employs gradient descent optimization, guided by the modeling error for updating each parameter. It also uses the linearized equations of motion where the aerodynamic forces are represented by their coefficients and derivatives. Some convergence enhancement techniques are also used to reduce the time required for parameter identification. (Abstract shortened by UMI.)