Integrated process-system modelling and control through graph neural network and reinforcement learning

Abstract

Modern manufacturing systems are becoming increasingly complex, dynamic, and connected, and their performance is being affected by not only their constituent processes but also their system-level interactions. This paper presents an integrated modelling method based on a graph neural network (GNN) and multi-agent reinforcement learning (MARL) collaborative control for adjusting individual machining process parameters in response to system- and process-level conditions. The structural and operational dependencies among process machines are captured with a GNN. Iteratively trained with MARL, machines learn to adaptively control local process parameters, e.g., machining speed and depth of cut, while achieving the global goal of improving production yield.