Intelligent control utilizing localized/embedded expert systems in an object-oriented environment
Description
The objective of this research is to evaluate a new approach to intelligent control system architecture utilizing localized intelligence in the Object Oriented design paradigm. Intelligent Control is a fusion of Control Theory with Artificial Intelligence techniques. The desired benefit of such a fusion is a control system based on proven analytical techniques, and also one that can respond to varying control parameters and control objectives in an intelligent manner. Such a system would be able to correctly interpret and respond to many control situations, would be able to intelligently manipulate the physical system to generate additional state information needed to make a control decision, and would be capable of tailored performance characteristics. It should utilize both deterministic methods and heuristic methods To implement these goals, a control system structured as objects or modules that map closely to physical or abstract devices and boundaries naturally encountered in a control loop will be proposed. It will be shown that if properly decomposed, such objects enclose natural boundaries in the problem space, and that knowledge concerning the domain enclosed by these boundaries can be defined and encapsulated within these objects. It will be shown that elements in a typical control system can be constructed by linking such objects in a taxonomy like structure which progresses from very generic and flexible objects to very domain specific objects. Finally a fast goal directed Expert System architecture that has been cast into objects and simplified to operate in these small very specific knowledge domains will be developed A prototype passive vision based weld seam survey control system was implemented in hardware as a platform to validate this proposed control structure. The control loop was constructed from three major taxonomies; Sensors, Sensor Manager, and Controller. Each taxonomy was designed to maximize flexibility and reusability in subsequent variant control loops. Knowledge was collected piece wise from objects contained in each taxonomy. Five intelligent devices (three vision sensors, a sensor manager and a robot controller) utilized local collection of this knowledge and an embedded inference engine object to exercise intelligent control of the robot in unstructured weld seam environments