Data-Driven Modeling and Sparse Sensing for Nuclear Energy Systems

Abstract

In nuclear energy systems, the integration of computational and interpretable data-driven models is essential for ensuring the safe and efficient operation of reactors and subsystems. These models leverage real-time sensor data to reconstruct critical variables such as temperature, pressure, and velocity, enabling precise monitoring, control, and optimization. By minimizing the need for costly physical experiments, computational models allow virtual testing of designs and operational strategies while addressing uncertainties and enhancing subsystem performance. Coupled with data-driven methodologies, they form the backbone of modern nuclear engineering, facilitating real-time decision-making and ensuring robust, reliable, and efficient system operations. Accurate reconstruction and monitoring in nuclear systems face challenges due to sensor noise and the impracticality of deploying extensive sensor arrays in harsh environments. This work addresses these limitations by introducing a data-driven framework for spatially constrained sensor placement, designed to minimize reconstruction errors and quantify noise-induced uncertainties. Using a greedy optimization algorithm, this approach ensures physically feasible configurations while maintaining high-fidelity reconstructions. These advancements are applied to critical scenarios, including fuel irradiation experiments, nuclear fuel test rod models, and steam generator subsystems, laying the groundwork for creating reliable and interpretable models for nuclear power plants (NPPs). This work further explores the effects of power transients on reactor core coolant dynamics, developing frameworks that adapt sensor placement and sampling intervals to changing operational states. The integration of classification tools like Linear Discriminant Analysis (LDA) with DMDc enables precise identification of transient regimes, ensuring robust monitoring and control. These contributions extend beyond individual subsystems, enabling the development of comprehensive digital twins for entire NPPs. By addressing noise, spatial constraints, power perturbations and transient conditions, this work establishes a foundation for advanced monitoring and control in nuclear systems. The proposed frameworks enable proactive decision-making, enhance safety, and improve operational efficiency, contributing to a new standard for intelligent, data-driven engineering solutions in the nuclear industry.