A learning reactive scheduler using CBR/L

Abstract

Machine learning techniques can increase the power/expertise and improve the problem-solving efficiency of artificial intelligence applications for complex problems such as reactive scheduling in manufacturing operations management. Reactivity requires adaptively changing behaviour of a performance system that can best be supported by learning from experience. Reactive scheduling is considered in this paper with two options for reactive repair/proactive schedule adjustment to compensate for/prevent the effects of disturbances during the execution time of a current schedule. With a supervisory control perspective to shop floor operations, reactive scheduling of the two options above is embodied as one major intelligent supervisory function in a supervisory reactive scheduler SUPREACT and complemented by a cognitive learning approach to improve search space/control pattern-matching knowledge for guiding iterative schedule repair actions. In a brief overview, some learning approaches that comprise past experience for future re-use in reactive scheduling are referred to. The paper then presents a combined rule- and case-based reasoning/learning approach to opportunistic reactive scheduling based on a blackboard framework of the system's Expert Supervisor Unit, which also supports human integration into supervisory control of executed processes. Inherent learning ability of the case-based component allows to capture new schedule repair/search control knowledge, including also human preferences, and thus improve the system's reactive/proactive schedule repair efficiency in response to unexpected events/performance deterioration trends during the execution of predictive schedules in manufacturing shop floors.