▣ Title : Closing the Simulation-to-Operation Loop: Differentiable Physics and Agentic AI for Advanced Nuclear Systems

▣ Speaker : Prof. Yang Liu(Texas A&M University)

▣ Date & Time : 2026. 10. 01(Thu) 16:30~17:30

▣ Venue : Research Bldg I, #310

▣ Language : Korean

▣ Abstract :

Advanced nuclear systems require simulation capabilities that connect physical models with measurements, inference, and operational decisions. This seminar presents a closed-loop research framework built on differentiable physics, gradient-enabled inference and control, and agentic workflow orchestration. Differentiable thermal-hydraulic models preserve governing equations and numerical structure while exposing sensitivities for calibration and optimization, with reference-code comparisons against MOOSE and SAM. The discussion progresses from inverse parameter estimation and state estimation to differentiable predictive control, distinguishing demonstrated surrogate-trained controllers from ongoing data-assimilation and solver-based training efforts. Agentic AI then provides an orchestration layer for domain knowledge, engineering documents, specialized simulation tools, and human review. AutoSAM illustrates document-grounded model generation, execution, and error diagnosis. A thermal-fluid laboratory testbed extends these ideas to continuous monitoring, locally deployed language models, and asynchronous forecasting under explicit operating rules. The seminar concludes by identifying the remaining integration and validation requirements for combining these capabilities into governed closed-loop digital engineering for advanced nuclear systems.