A Learning Enhanced Flight Control System for High Performance Aircraft
Abstract
Numerous approaches to flight control system design have been proposed in an attempt to govern the complex behavior of high performance aircraft. Gain scheduled linear control and adaptive control have traditionally been the most widely used methodologies, but they are not without their limitations. Gain scheduling requires large amounts of a priori design information and costly manual tuning in conjunction with flight tests, while still lacking an ability to accommodate unmodeled dynamics and model uncertainty beyond a limited amount of robustness that can be incorporated into the design. Adaptive control is suitable for nonlinear systems with unmodeled dynamics, but has deficiencies in accounting for quasi-static state dependencies. Moreover, inherent time delays in adaptive control make it difficult to match the performance of a well-designed gain scheduled controller. An alternative approach that is able to compensate for the inadequacies experienced with traditional control techniques and to automate the tuning process is desired. Recent Teaming techniques have demonstrated an ability to synthesize multivariable mappings and are thus able to learn a functional approximation of the initially unknown state dependent dynamic behavior of the vehicle. By combining a learning component with an adaptive controller, a new hybrid control system that is able to adapt to unmodeled dynamics and novel situations, as well as to learn to anticipate quasi-static state dependencies is formed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1992
- Accession Number
- ADA252520
Entities
People
- Noel F. Nistler
Organizations
- Air Force Institute of Technology