Self-Optimising Manufacturing Network: an Online Optimization Problem Decided by Reinforcement Learning

Networked personalized production necessitates resilient optimization approaches in the face of uncertainty to proactively react to unforeseen scenarios and enable early action to address sudden shifts in consumer and supplier connections. This work presents a comprehensive framework for decision-making within a local unit of a self-optimizing manufacturing network, explicitly focusing on exploring the interplay between optimization, simulation, and reinforcement learning. We propose an initial validation scenario in which the agent needs to accept production demands considering priorities concerning three performance criteria (economic, sustainability, and variability), given the current load on the system that can impact delays and, consequently, financial losses. Following the hyper-parameter setup experiment, the computational results strongly support validating the manufacturing unit simulator and the Proximal Policy Optimization-based decision agent.