LLM-Driven Multi-Agent Control for Skill-Based Smart Manufacturing
LLM-based agents were developed to manage high-mix manufacturing by generating production sequences and handling unexpected faults in automated systems. The study compared three agent structures—orchestrator, peer-to-peer, and monolithic—using an MCP tool server with OPC UA skills. Results showed that both monolithic and peer-to-peer architectures achieved the highest solve rate (93%), while the orchestrator uniquely solved a complex conveyor belt fault by rerouting plates.