Dynamic Scheduling of the Dual Stocker System using Reinforcement Learning

Abstract

The stocker system is the most widely used material handling system in LCD and flat panel fabrication facilities (FABs). The stocker mainly consists of one or two cranes moving along a single track to transport lots, or cassettes, containing 10 to 30 thin glass substrates between processing machines. Because the stocker system is the primary material handling system in the FABs, its performance directly affects the overall performance. In this study, we investigate the scheduling of a dual stocker system operating with two cranes simultaneously on a single track and propose a learning-based scheduling algorithm for the system. We report some of the results of our long-term efforts to dynamically optimize the dual-crane stocker. We fisrt show the modeling and algorithm to minimize the make-span of the jobs. We incorporate the model to dynamically allocate jobs. In particular, we use a reinforcement learning method in the scheduling algorithm. The model is validated in an extensive simulation study based on actual data.