Abstract: The development of Industry 5.0 presents higher requirements for the informatization, digitization, and intelligence of the manufacturing industry. To address the important challenges of the lack of traditional workshop manufacturing capacity organizational paradigm and intelligent scheduling technology, a digital workshop manufacturing capacity modeling and adaptive scheduling technology in the human-cyber-physical interconnected environment is proposed to achieve high-fidelity modeling of complex workshop manufacturing capacity and efficient scheduling of production resources. In order to effectively manage the production elements in the digital workshop, a digital workshop manufacturing capacity modeling technology that integrates the human-cyber-physical system is proposed. In addition, a deep reinforcement learning-driven adaptive scheduling algorithm (DRL-AS) is devised for the digital workshop, which models the flexible job shop scheduling problem in the form of heterogeneous directed acyclic graphs. Considering the complex coupling relationship between operations and machines, a multi-factor representation method based on hierarchical self-attention mechanism is designed to extract global features of the environmental state and assist the agent in making high-quality decisions. Proximal policy optimization (PPO) algorithm is used to train the proposed adaptive scheduling technology. Experimental results show that the scheduling performance and generalization performance of the proposed method are significantly better than those of the comparison algorithms.

Key words: human-cyber-physical system, flexible job-shop scheduling, deep reinforcement learning, manufacturing capability modeling, digital twin