An inference structure for the control and scheduling of manufacturing systems

Abstract Control and scheduling problems in manufacturing systems have long been an intriguing subject to operations researchers and industrial practitioners. In the past few years, Artificial Intelligence (AI) methods for control and scheduling of manufacturing systems have gained considerable attention. A fundamental advantage of AI methods, in contrast to algorithmic approaches, is in the ability to take a “reformulated approach” to scheduling problems. However, most of the existing paradigms of AI methods (e.g. expert system shells) restrict the user to a particular inference strategy and representation method. In many cases, these paradigms are not flexible nor powerful enough to take the full advantage of AI methods. In this paper, an inference structure is described which allows the designer of an AI-based scheduling system to take advantage of (1) a data-driven strategy to move forward from existing knowledge to new conclusions; (2) a goal-driven strategy to prove or disprove a goal for hypothesis by examining its supporting evidence; (3) a simulation interface to conduct “what-if” analysis of the future states of the modeled system, and (4) a direct LISP interface in the form of “rule-programs,” which allows various algorithmic procedure, search strategies, and problem-specific inferencing methods to be adopted easily.