Recent performance breakthroughs in artificial intelligence (AI) and machine learning (ML) have led to unprecedented interest among nuclear engineers. Despite the progress, the lack of dedicated benchmark exercises for the application of AI and ML techniques in nuclear engineering analyses limits their applicability and broader usage. In line with the NEA strategic target to contribute to building a solid scientific and technical basis for the development of future generation nuclear systems and deployment of innovations, the Task Force on Artificial Intelligence and Machine Learning for Scientific Computing in Nuclear Engineering was established within the Expert Group on Reactor Systems Multi-Physics (EGMUP) of the Nuclear Science Committee’s Working Party on Scientific Issues and Uncertainty Analysis of Reactor Systems (WPRS). The Task Force will focus on designing benchmark exercises that will target important AI and ML activities, and cover various computational domains of interest, from single physics to multi-scale and multi-physics.
A significant milestone has been reached with the successful launch of the first comprehensive benchmark of AI and ML to predict the Critical Heat Flux (CHF). This CHF corresponds in a boiling system to the limit beyond which wall heat transfer decreases significantly, which is often referred to as critical boiling transition, boiling crisis and (depending on operating conditions) departure from nucleate boiling (DNB), or dryout. In a heat transfer-controlled system, such as a nuclear reactor core, CHF can result in a significant wall temperature increase leading to accelerated wall oxidation, and potentially to fuel rod failure. While constituting an important design limit criterion for the safe operation of reactors, CHF is challenging to predict accurately due to the complexities of the local fluid flow and heat exchange dynamics.
Current CHF models are mainly based on empirical correlations developed and validated for a specific application case domain. Through this benchmark, improvements in the CHF modelling are sought using AI and ML methods directly leveraging a comprehensive experimental database provided by the US Nuclear Regulatory Commission (NRC), forming the cornerstone of this benchmark exercise. The improved modelling can lead to a better understanding of the safety margins and provide new opportunities for design or operational optimisations.
The CHF benchmark phase 1 kick-off meeting on 30 October 2023 gathered 78 participants, representing 48 institutions from 16 countries. This robust engagement underscores the profound interest and commitment within the global scientific community toward integrating AI and ML technologies into nuclear engineering. The ultimate goal of the Task Force is to leverage insights from the benchmarks and distill lessons learnt to provide guidelines for future AI and ML applications in scientific computing in nuclear engineering.
