Smart Digital Twin for ZDM-based job-shop scheduling

Abstract

The growing digitization of manufacturing processes is revolutionizing the production job-shop by leading it toward the Smart Manufacturing (SM) paradigm. For a process to be smart, it is necessary to combine a given blend of data technologies, information and knowledge that enable it to perceive its environment and to autonomously perform actions that maximize its success possibilities in its assigned tasks. Of all the different ways leading to this transformation, both the generation of virtual replicas of processes and applying artificial intelligence (AI) techniques provide a wide range of possibilities whose exploration is today a far from negligible sources of opportunities to increase industrial companies' competitiveness. As a complex manufacturing process, production order scheduling in the job-shop is a necessary scenario to act by implementing these technologies. This research work considers an initial conceptual smart digital twin (SDT) framework for scheduling job-shop orders in a zero-defect manufacturing (ZDM) environment. The SDT virtually replicates the job-shop scheduling issue to simulate it and, based on the deep reinforcement learning (DRL) methodology, trains a prescriber agent and a process monitor. This simulation and training setting will facilitate analyses, optimization, defect and failure avoidance and, in short, decision making, to improve job-shop scheduling.