Sustainable Manufacturing Through Digital Twin and Reinforcement Learning

Abstract

Smart manufacturing is on the cusp of a significant transformation as it integrates virtual systems with their real-world counterparts, primarily through the use of digital twins. When combined with deep reinforcement learning, the predictive capability of DTs is sharpened using real-world data, offering valuable insights throughout an entity's life cycle, from inception to retirement. DRL provides a resilient framework for making decisions in unpredictable and ever-changing environments. As agents continually interact with these environments, their decision-making strategies, guided by rewards, are refined. The foundation of this learning lies in the Markov decision process, which steers the efficacy of DRL. This methodology has demonstrated its effectiveness in challenges like scheduling and robot control. This chapter explores the benefits, frameworks, data flow, and pipelines of implementing DRL in smart manufacturing, particularly in resource scheduling. It provides a comparative analysis of existing research and DRL's performance against traditional heuristics.