Reinforcement learning-based decision making for sustainable manufacturing operations

Abstract

Abstract. Sustainable manufacturing involves being able to optimize productivity, energy efficiency, and environmental impact simultaneously given dynamic and uncertain operating conditions. The traditional optimization methods are unadaptable and cannot easily reflect the real-time changes in the system. This paper provides a sophisticated reinforcement learning (RL)-based decision-making model of sustainable manufacturing process. The manufacturing system is modelled as a Markov Decision Process (MDP) and a Deep Q-Network (DQN) is used to learn about the optimal control policies by interacting with the environment continuously. Multi-objective reward function is created to include production rate, energy usage, machine usage and minimization of waste. The suggested framework is tested in a virtualized smart factory setting, where the demand is stochastic and machines have variability. Comparative analysis shows that the RL-based approach outperforms the rule-based and heuristic strategies and reports remarkable energy efficiency and operational sustainability. The findings prove RL as a potential solution to adaptive and intelligent manufacturing control.