Learning-based adaptive dispatching method for batch processing machines

This study aims to solve the scheduling problem of batch processing machines (BPMs) in semiconductor manufacturing by using a learning-based adaptive dispatching method (LBADM). First, an adaptive ant system algorithm (AAS) is proposed to solve the scheduling problem of BPMs according to their characteristics. Then AAS generates a lot of solutions for the jobs with different distribution of arrival time and due date. These solutions are taken as learning samples. Second, we analyze influencing factors by sample learning method from those solutions. With the help of linear regression, the coefficients of influencing factors can be calculated to build a dynamic dispatching rule adaptive to running environments. Finally, simulation results based on a Minifab model show that the proposed method is better than traditional ways (such as FIFO and EDD with maximum batchsize) with lower makespan and weighted tardiness.