Data-Driven Thermal Imaging for Real-Time Nuclear Safety Monitoring and Temperature Reconstruction

Real-time monitoring of safety-critical nuclear systems requires accurate detection of thermo-hydraulic behaviors to prevent loss-of-coolant accidents (LOCA). We develop computational methods using dynamic mode decomposition with control (DMDc) and Kalman filtering applied to experimental thermal imaging data from the OPTI-TWIST (Transient Water Irradiation System) prototype at Idaho National Laboratory. Our approach addresses two problems: multi-class classification of thermal profiles to identify power levels using DMDc-based feature extraction, and an inverse problem estimating internal temperatures from external surface measurements through Kalman filtering. The experimental dataset consists of thermal imaging sequences from heating experiments with ground truth from physical thermocouple sensors. We present a DMDc-based classification algorithm mapping thermal features to power categories and a Kalman filter-based state estimation method for inferring interior thermal states from boundary observations. Results demonstrate effective temperature reconstruction validated against real experimental measurements, enabling accurate thermal monitoring for nuclear safety applications.