IN-FLIGHT ADAPTIVE CONTROL USING NEURAL NETWORKS

REAL-TIME FLIGHT CONTROL IS OFTEN IMPLEMENTED USING CONTROL SYSTEMS AND PRE-PROGRAMMED CONTROLLER SOFTWARE TAILORED TO A SINGLE APPLICATION. HNC, INC. PROPOSES TO BUILD AN ADAPTIVE NEURAL NETWORK CONTROLLER WHICH DOES NOT NEED TO BE PROGRAMMED BUT RATHER CREATES AN INTERNAL REPRESENTATION OF THE CONTROL LAW AS A RESPONSE TO THE ACTION OF THE SYSTEM BEING CONTROLLED OR A SIMULATION THEREOF. HNC, WITH THE COOPERATION OF UNITED TECHNOLOGIES CORPORATION, PROPOSES TO DEMONSTRATE THE OPERATION OF AN ADAPTIVE NEURAL NETWORK FLIGHT CONTROLLER. PHASE I WILL CONVER THE EXTENSION OF HNC/UTC'S CURRENT NEURAL NETWORK CONTROLLER TO INCLUDE IN-FLIGHT RE-CONFIGURATION, AS WELL AS A DEFINITION OF THE HARDWARE REQUIREMENTS AND CORRESPONDING ARCHITECTURE FOR A REAL-TIME SYSTEM. PHASE II WOULD INVOLVE DEMONSTRATING THE CONTROLLER IN REAL-TIME FOR AN AIRFRAME TO BE SELECTED BY AFWAL AND DESIGNING A DEPLOYABLE CONTROLLER. A THIRD PHASE OF THE PROGRAM WOULD BE DIRECTED TOWARD THE PRODUCTION AND FLIGHT TEST OF THE PROTOTYPE NEURAL NETWORK CONTROLLER.