Speaker
Description
The Reynolds-Averaged Navier-Stokes (RANS) equations, crucial in predicting
turbulent flows through computational fluid dynamics (CFD), involve the decomposition
of flow variables into time-averaged and fluctuating components using Reynolds
decomposition applied to the Navier-Stokes equations. Predicting turbulent stresses
accurately necessitates turbulence models due to flow complexity, which are mathematical
or empirically based. Commonly used models include k − and k − , focused on
turbulent kinetic energy and dissipation rate, and turbulent kinetic energy and specific
dissipation rate, respectively. These models offer strengths and weaknesses, and their
selection is dependent on simulation specifics and result accuracy. The first review section
delves into model variations, discussing closure term functions and constants. Machine
learning (ML) enhances turbulent models by enabling data-driven closures and rapid
precise predictions. The second part explores ML's role in enhancing turbulent models—
predicting quantities, optimizing models, and creating efficient reduced-order models. This
ML integration in modelling holds the potential for improved accuracy, efficiency, and cost
reduction. Challenges and prospects in this field are also addressed in this review.