An AI-Driven Thermal-Fluid Testbed for Advanced Small Modular Reactors: Integration of Digital Twin and Large Language Models

This paper presents a multipurpose artificial intelligence (AI)-driven thermal-fluid testbed designed to advance Small Modular Reactor (SMR) technologies through the seamless integration of physical experimentation, high-fidelity digital twin, and sophisticated AI frameworks. The platform uniquely combines a versatile three-loop thermal-fluid testbed with a System Analysis Module (SAM)-based digital twin that is accelerated by a Gated Recurrent Unit (GRU) neural network to achieve real-time operational capabilities. The GRU model, trained on a composite dataset of experimental and simulation data, accurately forecasts future system states and corresponding control actions with a temperature prediction RMSE of 4.25 K, enabling long-term transient predictions for operational planning. The digital twin’s performance was validated through comprehensive experimental campaigns, demonstrating its ability to predict complex thermal-fluid dynamics during power transients. Furthermore, an intelligent operator assistance system powered by a large language model (LLM) was developed using context engineering techniques, which synthesizes real-time experimental data, digital twin predictions, and user queries to provide actionable operational guidance in natural language. The LLM implementation demonstrated robust performance in multi-parameter correlation analysis, predictive reasoning, and safety-aware recommendations. This integrated platform establishes a new paradigm for nuclear research infrastructure, demonstrating how the convergence of physical testbeds, AI-accelerated digital twins, and natural language interfaces can accelerate the development and deployment of next-generation intelligent nuclear systems. • Integrated thermal-fluid testbed with AI-driven digital twin and large language model. • GRU-accelerated SAM model enables real-time DT operation. • Long-term transient prediction validated with experimental data. • LLM-based operator assistance via context engineering framework. • Multi-purpose platform for thermal fluid and advanced reactor research.