Design and evaluation of a learning augmented longitudinal flight control system

A summary of a program to design, develop, and evaluate a hybrid learning augmented longitudinal flight control system for a high performance aircraft is presented. In addition to the learning augmented (hybrid) controller, two baseline controllers (a linear compensator and an adaptive controller) were also developed for purposes of comparison. The three share the same basic structure: all are model-reference controllers, the hybrid system is a learning augmented version of the adaptive controller, which is an adaptive version of the linear compensator. Each was designed to serve as a longitudinal control augmentation system to improve the basic handling qualities of a three-degree-of-freedom, nonlinear aircraft model. The hybrid controller was trained by exposing it to a number of transient responses of this aircraft. The transients were initiated by applying different stick inputs to the aircraft at randomly selected trim conditions throughout a subsonic flight envelope. The desired pitch rate responses corresponding to the same stick inputs and flight conditions were generated by a simple linearized reference model, scheduled on altitude and Mach number. Performance statistics were obtained for all three controllers by analyzing the pitch rate tracking error experienced over a variety of different flight conditions