Recent advancements in machine learning (ML) and artificial intelligence (AI) have generated significant interest among nuclear engineers. However, the adoption and effectiveness of AI and ML techniques in nuclear architectural analyses are hindered by the lack of standardized benchmark exercises. To address this issue, the Task Force on Artificial Intelligence and Machine Learning for Scientific Computing in Nuclear Engineering was formed under the Expert Group on Reactor Systems Multi-Physics (EGMUP) of the Nuclear Science Committee’s Working Party on Scientific Issues and Uncertainty Analysis of Rear Systems (WPRS). This initiative aligns with the NEA’s strategic goal of establishing a robust scientific and technical foundation for the development of next-generation nuclear systems and the integration of innovations. The Task Force’s primary focus is on developing standard exercises that target key AI and ML applications across various mathematical domains, ranging from single physics to multi-scale and multi-physics scenarios.
A significant milestone was achieved with the successful introduction of the first comprehensive benchmark for AI and ML in predicting the Critical Heat Flux (CHF). In a boiling system, CHF signifies the point at which wall heat transfer significantly diminishes, also known as the critical boiling transition, boiling crisis, departure from nucleate boiling (DNB), or dryout, depending on the operational conditions. CHF can lead to a substantial increase in wall heat, triggering accelerated wall oxidation and, in certain instances, fuel rod failure in heat transfer-controlled systems like nuclear reactor cores. Predicting CHF accurately is challenging due to the intricate dynamics of local smooth movement and temperature exchange, despite its critical importance as a design limit for reactor safety.
Current CHF models primarily rely on empirical correlations tailored to specific software case domains. The benchmark, supported by the US Nuclear Regulatory Commission (NRC), offers an extensive experimental database that directly contributes to enhancing CHF modeling. This improved modeling can provide a deeper understanding of safety margins and unveil new possibilities for design enhancements or operational optimizations.
The CHF standard phase 1 kick-off meeting on October 30, 2023, saw active participation from 78 attendees representing 48 institutions across 16 countries. This global engagement underscores the scientific community’s keen interest and commitment to integrating AI and ML systems into nuclear engineering. The Task Force aims to leverage insights from metrics and distill lessons learned to provide recommendations for future AI and ML applications in computational medicine within nuclear engineering.
