On-line optimizing networks for reconfigurable control

An indirect reconfigurable control system is developed, that online identifies the system parameters and redesigns the control laws. This is done for a maneuvering aircraft in a tracking control scenario where a horizontal tail failure is encountered. Critical stability and control derivatives are continuously identified using a constrained least squares approach. Prior information on the parameters is incorporated in the constraints. The identified derivatives are used by a Hopfield network, where the systems dynamics are incorporated using a penalty function. The Hopfield network generates an optimal model following open-loop control law. The optimization is performed every time cycle, thus yielding feedback control. The Hopfield neuromorphic approach, due to its massively parallel architecture, holds great promise for fast computations in analog hardware