Reinforcement learning for sustainability enhancement of production lines

The importance of sustainability in industry is dramatically rising in recent years. Controlling machine states to achieve the best trade-off between production rate and energy demand is an effective method for improving the energy efficiency of production systems. This technique is referred to as energy-efficient control (EEC) and it triggers machines in a standby state with low power requests. Reinforcement Learning (RL) algorithms can be used to successfully control production systems without the requirement of prior knowledge about system parameters. Due to the difficulty in acquiring comprehensive information about system dynamics in real-world scenarios, this is considered an important factor. The goal of this work is to create a novel RL-based model to apply EEC to multi-stage production lines with parallel machine workstations without relying on full knowledge of the system dynamics. Numerical results confirm model benefits when applied to a real line from the automotive sector. Further experiments confirm the effectiveness and generality of the approach.