Multi-agentic production planning utilising simulation and optimisation

While applications of artificial intelligence proliferate, planning complex, volatile production remains a challenge. Hybrid methods that enable large language models to interact with classic models, such as simulation and optimisation, could address this problem. This work develops a workflow-based multi-agent architecture that generates production plans from unstructured user prompts. The system is tested using data from an automotive supplier, with 7 distinct experiments in lot sizing and setup sequencing, using simulated validation. The experiments show the system can reliably create suitable production plans while reducing execution times compared to manual planning. This work presents a step towards hybrid artificial intelligence.