A Generic Architecture for Multiple Agent Process Control

This cooperative research effort combined expertise in artificial intelligence and constraint-based reasoning from the Department of Computer Science at NCSU with expertise in Process Control from the Department of Electrical Engineering at Duke. The inter-disciplinary efforts of the researchers were focused through the construction of an in-depth control application of a large-scale hydraulic distribution system. Project activities consisted of three parts. The first was to study the nature of the knowledge necessary for multiple agent process control, including issues deriving from uncertainty in dynamic systems and issues concerned with achieving task coordination between multiple agents including humans in the control loop. The second was to investigate the utility of fuzzy constraint networks for representing and reasoning with the types of knowledge identified in the first part. The third was to integrate the results from the first two parts by applying the knowledge gained to develop a prototype fuzzy constraint network-based process controller for a multi-agent application area.