Unified Control Concept for Advanced Flight Systems

The position control of aircraft is inherently a two-point boundary-value problem, and aircraft rate and attitude control are desirably treated as such. The aircraft maneuvers from an initial state vector at a given initial time to a final state vector at a given or optimiized final time. The final state is achieved in a manner that minimizes cost functions such as the time integral of induced drag. This is a classical problem in variational calculus, but until recently there has been no practical means for initializing the co-state differential equations of the calculus. This proposal presents a new method for this initialization that uses artificial neural networks trained off-line on simulated optimum solutions. The method is realizable in airborne systems without large computational capabilities. Phase I work will remove the remaining technical obstactles and demonstrate the method via simulations of an advanced aircraft controlled in complex maneuvering. A control synthesis tool that can be used by the entire flight control community will be fully developed in Phase II. The proposed method is called a "unified control concept" because it provides solutions for nonlinear and time varying flight conditions, employs neural network based adaptive control, and provides analytically derived designs for rate, attitude, and position control laws.