Abstract: With the advent of the Industry 5.0 paradigm, global smart manufacturing is transitioning from the “efficiency-first” focus of Industry 4.0 to a new stage characterized by human-centricity, sustainability, and resilience. Against this backdrop, this paper first reviews the development trends of artificial intelligence (AI) technologies and analyzes key issues in human-machine collaboration, where humans and smart systems work together to enhance each other’s capabilities. The necessity of such collaboration in smart manufacturing is emphasized. Subsequently, the evolution of multi-agent systems and explores the challenges they face in distributed environments is summarized. It introduces the integration of multi-agent systems with large language models (LLMs), highlighting the breakthroughs this synergy brings to AI. Building on these insights, a cloud-edge-end collaborative wisdom manufacturing architecture based on large reasoning models and multi-agent systems is proposed. By leveraging a coordinated scheduling mechanism between “full-scale models” and “lightweight edge models”, and integrating techniques such as complex network modeling, knowledge distillation, and federated transfer learning, the proposed framework enhances the intelligence, resilience, and interpretability of production scheduling tasks. Experimental results demonstrate that LLMs have the potential to replace traditional deep reinforcement learning models in executing certain high-level scheduling logic in manufacturing contexts. Furthermore, the proposed collaborative architecture offers a more efficient, trustworthy, and resilient foundation for next-generation smart manufacturing operations with multi-agents in the Industry 5.0 era.

Key words: Industry 5.0, large language model, reasoning intelligence, smart manufacturing, multi-agent systems, production scheduling