Integrating Large Language Model Agents with Digital Twins for Industrial Autonomous Systems

Researchers propose a three-layer framework integrating large language models with digital twins and automation systems to enable adaptive industrial autonomous systems. The TPSR model transforms user tasks into executable processes through LLM-based reasoning, demonstrated across five peer-reviewed studies with prototypes showing improved task executability and reduced manual effort.

This research addresses a fundamental challenge in industrial automation: the rigidity of traditional rule-based systems in dynamic manufacturing environments. As factories increasingly adopt cyber-physical systems and digitalization, the inability to autonomously adapt to changing conditions represents a significant operational bottleneck. This dissertation bridges that gap by leveraging LLMs as reasoning engines capable of interpreting ambiguous user intent, planning adaptive workflows, and orchestrating complex automation tasks across heterogeneous systems.

The integration of digital twins—virtual representations of physical assets—with LLM agents creates a feedback loop where models can simulate outcomes before execution, reducing errors and optimizing resource allocation. The TPSR framework systematizing this integration represents meaningful progress toward truly autonomous systems that require less human intervention for reconfiguration and decision-making.

Industrially, this framework has substantial implications for manufacturing efficiency and workforce dynamics. Organizations implementing LLM-enabled automation could reduce downtime during production changes, accelerate deployment of new processes, and lower training requirements for operators. The research demonstrates practical viability through multiple case studies, though acknowledged limitations around computational demands and safety-critical dependence on human oversight remain critical barriers to immediate widespread adoption.

The trajectory suggests growing convergence between AI reasoning capabilities and industrial control systems. Future development will likely focus on addressing computational efficiency, ensuring safety certification for autonomous decision-making, and improving digital twin fidelity. Organizations monitoring this space should track when these frameworks transition from research prototypes to production-ready implementations, as early adopters could gain competitive advantages in manufacturing agility.

• →LLM-based agents integrated with digital twins enable adaptive industrial automation without fixed rule-based constraints.
• →The TPSR model systematically transforms high-level user tasks into executable processes through AI reasoning.
• →Prototypes demonstrate high task executability and command correctness while significantly reducing manual effort.
• →Computational demands and safety-critical limitations currently restrict deployment in mission-critical production environments.
• →Digital twin accuracy directly impacts system performance, making data quality a critical success factor for implementation.