Proceedings of the XMO Industrial Seminar 2026: Excellence in Manufacturing and Operations
Keywords
Large language model; Multi-agent system; Manufacturing decision support; Material–process recommendation; Process parameter optimization
Tracks
DIGITAL MANUFACTURING
DOI
10.5703/1288284318681
Abstract
As product fabrication environments become increasingly diverse, there is a growing need for decision support that connects users’ fabrication intents and requirements to appropriate manufacturing conditions. In this study, we propose an LLM-based multi-agent system for manufacturing decision support that systematically supports material and process recommendation, equipment selection, process parameter optimization, and result prediction. The proposed system utilizes manufacturing literature and technical documents through RAG-based manufacturing knowledge retrieval and excludes low-relevance documents before response generation by verifying the relevance of retrieved documents to the user query. In addition, it explores support relationships among materials, processes, and equipment based on a knowledge graph and derives process parameters that satisfy the target characteristics using optimization metadata and surrogate models for each combination of process, material, and equipment. To address users’ diverse queries and complex requirements for manufacturing decision support, task-specific agents were defined and organized into two teams: a material and process recommendation team and a process optimization team. These teams were then integrated into a hierarchical multi-agent architecture coordinated by a Supervisor agent. The application results showed that the proposed system can dynamically assign appropriate agents according to the purpose and requirements of user queries, and support manufacturing decision-support tasks by linking unstructured knowledge retrieval, structured relationship exploration, process optimization, and result prediction. These results suggest the potential of the proposed system to support information retrieval, condition evaluation, and optimization in complex manufacturing decision-support workflows.
An LLM-based multi-agent system for manufacturing decision support: material–process recommendation and process parameter optimization
As product fabrication environments become increasingly diverse, there is a growing need for decision support that connects users’ fabrication intents and requirements to appropriate manufacturing conditions. In this study, we propose an LLM-based multi-agent system for manufacturing decision support that systematically supports material and process recommendation, equipment selection, process parameter optimization, and result prediction. The proposed system utilizes manufacturing literature and technical documents through RAG-based manufacturing knowledge retrieval and excludes low-relevance documents before response generation by verifying the relevance of retrieved documents to the user query. In addition, it explores support relationships among materials, processes, and equipment based on a knowledge graph and derives process parameters that satisfy the target characteristics using optimization metadata and surrogate models for each combination of process, material, and equipment. To address users’ diverse queries and complex requirements for manufacturing decision support, task-specific agents were defined and organized into two teams: a material and process recommendation team and a process optimization team. These teams were then integrated into a hierarchical multi-agent architecture coordinated by a Supervisor agent. The application results showed that the proposed system can dynamically assign appropriate agents according to the purpose and requirements of user queries, and support manufacturing decision-support tasks by linking unstructured knowledge retrieval, structured relationship exploration, process optimization, and result prediction. These results suggest the potential of the proposed system to support information retrieval, condition evaluation, and optimization in complex manufacturing decision-support workflows.