{"id":11271,"date":"2024-01-12T09:48:02","date_gmt":"2024-01-12T08:48:02","guid":{"rendered":"https:\/\/www.iese.fraunhofer.de\/blog\/?p=11271"},"modified":"2024-01-22T10:59:20","modified_gmt":"2024-01-22T09:59:20","slug":"process-planning","status":"publish","type":"post","link":"https:\/\/www.iese.fraunhofer.de\/blog\/process-planning\/","title":{"rendered":"Process Planning and Continuous Scheduling (Part 2)"},"content":{"rendered":"

In our last Fraunhofer IESE blog post, we introduced a holistic process planning and scheduling design called RL design, which addresses individualized production with small lot sizes<\/a>. However, this design cannot deal with scheduling problems in the case of large job quantities. In this post, we introduce the second RL design: process planning with subsequent scheduling. This design aims to address the case where jobs arrive continuously at random times, with all of them sharing several common product types.<\/p>\n

This design separates the process planning module and the scheduling modules into two separate DQN architectures. The DQN for process planning is started first. It has the same construction as the first RL design except that all services have a duration of a single time step and the agents only have to learn to synthesize process plans instead of deriving schedules. This reduces the computation time for process planning dramatically. The process plans are stored in one\u00a0 place and are later accessed by the scheduling module. The continuous scheduling module is responsible for deriving near-optimal schedules based on the process plans.<\/p>\n