Özet:
In dynamic and stochastic manufacturing environments, a scheduling system should be able to modify its schedule to adapt to changing situations such as machine breakdowns, due-date changes, new rush orders, canceled orders and so on. In centralized monolithic scheduling systems, dynamic updating may be prohibitive mostly due to computational constraints in generating new schedules, or due to non-existence of effective shop floor data monitoring/data collection systems across the organization. One of the approaches to generate timely scheduling decisions, especially for large scale problems, is to decompose the problem and distribute responsibility of sub-problems to various agents. When agents are organized as loosely coupled autonomous scheduling units, it is easier to manage and control the data with respective local availability constraints, and eventually, faster to respond to system changes. However, distributing the scheduling problem brings in additional issues such as: choice of the decomposition methodology (defines the boundaries of independence for the agents); accuracy and timing of representation of local reality in each agent; scheduling algorithms and frequency of re-scheduling in each agent; and maybe most importantly, handling mechanism for coupling constraints between agents. In this study, we propose a state based modeling approach to represent the workings of a scheduling agent so that effect of choices made in distributing the problem in terms of the overall performance of the scheduling methodology can formally be defined and analyzed. The proposed model will be a framework to properly identify conditions and timing of possible inconsistencies and inaccuracy, and analyze the effectiveness of various distribution mechanisms in terms of scheduling performance.
