NEURAL NETWORKS, YOULA PARAMETERIZATION AND RECONFIGURABLE CONTROL SYSTEM

RESEARCHERS ARE INVESTIGATING A NEW CONTROL ARCHITECTURE IN WHICH A NEURAL NETWORK IS USED TO SET THE COEFFICIENTS OF THE YOULA PARAMETER FOR THE CONTROLLER. THE "DESIGN BY LEARNING" BENEFITS ASSOCIATED WITH NEURAL CONTROL ARE REALIZED BY THE APPROACH WHILE SIMULTANEOUSLY GUARANTEEING THAT THE RESULTANT SYSTEM IS STABLE. THE ARCHITECTURE YIELDS A RECONFIGURABLE SYSTEM IN WHICH THE NEURAL NETWORK AUTOMATICALLY UPDATES THE YOULA PARAMETER WHENEVER A NEW REFERENCE INPUT OR PERFORMANCE CRITERIA IS SPECIFIED. SPECIFIC RESEARCH OBJECTIVES INCLUDE THE DEVELOPMENT OF NEURAL NETWORK ARCHITECTUERS AND TRAINING METHODS APPLICABLE TO THE SYSTEM; FORMULATION OF METHODS FOR UPDATING THE NETWORK TRAINING EACH TIME THE SYSTEM IS OPERATED; IMPLEMENTATION OF THE CONTROLLER AS AN AUTO-RECONFIGURABLE SYSTEM; AND APPLICATION OF THE NEURAL CONTROLLER TO FLIGHT CONTROL SYSTEMS AND ROBOTICS.