Nuclear engineers now show an unprecedented level of interest in artificial intelligence (AI) and machine learning (ML) due to recent performance breakthroughs in these fields. Despite the advancements, the applicability and wider usage of AI and ML techniques in nuclear engineering analyses is limited by the absence of specific benchmark exercises. 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 on Artificial Intelligence and Machine Learning for Scientific Computing in Nuclear Engineering was established in accordance with the NEA strategic target to contribute to the establishment of a solid scientific and technical basis for the development of future generation nuclear systems and deployment of innovations.
The Task Force will concentrate on creating benchmark exercises that cover a range of computational domains of interest, from single physics to multi-scale and multi-physics, and that will target significant AI and ML activities.
The successful launch of the first comprehensive AI and ML benchmark to predict the Critical Heat Flux (CHF) marks a significant milestone. This critical boiling point, also known as the boiling crisis, critical boiling transition, departure from nucleate boiling (DNB), or dryout, refers, in a boiling system, to the point beyond which wall heat transfer drastically reduces. CHF can cause a sizable increase in wall temperature in a heat transfer-controlled system, like the core of a nuclear reactor. This accelerated wall oxidation can eventually lead to fuel rod failure. Because of the complexity of the local fluid flow and heat exchange dynamics, CHF is difficult to predict with accuracy, despite being a crucial design limit criterion for the safe operation of reactors.
The majority of the empirical correlations used in current CHF models were created and verified for use in a particular application case domain. The cornerstone of this benchmark exercise is the use of AI and ML techniques to directly leverage a comprehensive experimental database provided by the US Nuclear Regulatory Commission (NRC) in order to seek improvements in the CHF modeling. Better comprehension of the safety margins and fresh chances for operational or design optimizations can result from the enhanced modeling.
On October 30, 2023, 78 people attended the CHF benchmark phase 1 kick-off meeting, representing 48 institutions from 16 different countries. This active participation demonstrates the worldwide scientific community’s strong interest in and dedication to the integration of AI and ML technologies into nuclear engineering. The Task Force’s ultimate objective is to use the benchmarks’ insights and lessons learned to create guidelines for future applications of AI and ML in scientific computing for nuclear engineering.