LLM agents + digital twins: A new framework for autonomous industrial systems

Yuchen Xia’s doctoral dissertation at the University of Stuttgart introduces a groundbreaking three-layer framework that integrates large language model (LLM) agents with digital twins to create autonomous industrial systems. The framework addresses the rigidity of traditional rule-based automation by embedding LLM-based reasoning for adaptive, goal-oriented behavior. The core innovation is the Task-Process-Service-Resource (TPSR) model, which translates high-level user tasks into executable processes across dynamic environments and heterogeneous components. Four distinct LLM roles are defined: process orchestration (planning task sequences), service matching (linking tasks to available automation services), digital resource generation (creating or updating digital twin models), and agent-as-a-service (providing LLM capabilities on demand).

The framework was developed and validated through five peer-reviewed studies using design science research methodology. Case studies and prototypes demonstrate adaptive task planning, event-driven control, simulation-based parameterization, and automatic digital model generation—all while significantly reducing manual effort. Results show high task executability, command correctness, and content-generation accuracy. However, the system depends heavily on accurate digital representations of physical assets, requires substantial computational resources for LLM inference, and still needs human intervention in safety-critical situations. Despite these limitations, this work provides a systematic blueprint for industrial systems that can autonomously interpret, plan, and execute user tasks, marking a significant step toward truly adaptive manufacturing environments.