Dual crane stocker scheduling problem using reinforcement learning

Abstract

The stocker system is the most widely used auto material handling system in TFT- LCD fabrication facilities(FABs). The stocker mainly consists of one or two cranes moving along a single track to transport cassettes, containing thin LCD glass substrates between processing machines. Because the stocker system is the primary material handling system in the TFT-LCD FABs, its performance directly affects the overall performance. In this thesis, we investigate the scheduling of a dual stocker system operating with two cranes simultaneously on a single track. We proposed dynamic programming formulation for static case and approximation approach using linear features, furthermore, deep Q network modeling considering non-linearity. In particular, we suggested trace shape of input to improve the performance of a neural network and exploited convolution layer.