RL-Based Scheduling Tested on Industry-Inspired Coating Process Using Digital Model Playground

A new paper on arXiv (2608.14122) highlights how reinforcement learning (RL) can be applied to production scheduling in coating processes, addressing real-world problems often missed in simpler studies. The researchers use the open-source Digital Model Playground (DMPG), a framework for discrete event simulation, to train RL agents on the coating procedure. They compare two popular algorithms—Deep Q-Networks and Proximal Policy Optimization—against conventional dispatching methods. The model accounts for factors like setup times that vary by sequence, equipment failures, random disruptions, and fluctuating usage rates, reflecting genuine manufacturing challenges. The authors aim to connect academic findings with industrial practices, suggesting that RL-based scheduling can be a practical alternative to traditional techniques.