Authors: V. Govindarajan, A. K. Chauhan, J. Gandhi, M. Rajendrakumar, R. S. Kumar, K. Natesan
Year: 2025
Industry: Nuclear & SMR
Keywords: Use FSE, Steady state, Transient, Dynamic, SFR, Comparison
ABSTRACT
This paper presents the development of machine learning (ML) based thermal-hydraulic models for Sodium-Cooled Fast Reactor (SFR) steam generators, encompassing both steady-state and transient analysis. A comprehensive dataset was systematically generated using Flownex SE simulations. For steady-state analysis, a Multi-Layer Perceptron Regressor (MLPRegressor) was trained to predict key parameters, including sodium and water outlet temperatures and heat transfer rates. The model achieved good accuracy with an MAE of 0.5433, while providing a good computational speed-up compared to Flownex simulations. For transient analysis, Long Short-Term Memory (LSTM) networks were implemented to capture temporal dependencies in dynamic thermal-hydraulic behavior. The LSTM model demonstrated good performance with an MAE value of 3.5 for steam temperature and achieved a good speed-up for transient simulations. These results demonstrate the significant potential of ML-based approaches for real-time thermal-hydraulic analysis in nuclear reactor applications.