Review of the development of turbulence models for low Prandtl number fluid flows and heat transfer based on machine learning techniques

Abstract

Nuclear energy is a clean and sustainable power source with minimal environmental impact. Its effectiveness relies on advanced cooling systems that use low Prandtl number ($Pr$) fluids, such as liquid metals, to manage extreme heat in liquid-metal cooled fast reactors (LMFRs). However, predicting the turbulent heat transfer and fluid flow behaviour of these fluids is challenging for traditional Reynolds-averaged Navier–Stokes (RANS) models due to their non-equilibrium turbulence and complex flow patterns. While large-eddy simulation (LES) and direct numerical simulation (DNS) methods are sufficiently accurate, they remain impractical for engineering applications due to their high computational cost. Machine learning (ML) offers innovative solutions to improve turbulence modelling, enabling more accurate predictions and efficient designs. This review highlights the need to advance ML-assisted turbulence modelling for low-$Pr$ fluids by reviewing traditional models, ML-augmented approaches and their real-world applications. It also addresses current limitations such as limited data, model reliability and generalization issues, while outlining opportunities to develop smarter, safer and more efficient nuclear thermal hydraulics cooling systems for the nuclear energy sector and other cutting-edge technologies.