A comparative analysis of multi-machine learning algorithms for data-driven RANS turbulence modelling

Abstract

To improve the speed or accuracy of Reynolds-averaged Navier-Stokes in numerical simulations of turbulence, five machine learning algorithms including Random Forests, eXtreme Gradient Boosting, K-Nearest Neighbours, Support Vector Machine and Artificial Neural Network are introduced in this paper. We establish the nonlinear mapping relationship between the average flow field and the steady-state eddy viscosity. The machine learning surrogate models for the Spalart-Allmaras turbulence model are constructed and used to solve the backward facing step problem. It is demonstrated that the machine learning surrogate models constructed by the five machine learning algorithms can effectively express the complex mapping relationship between the average turbulent flow field and the steady-state eddy viscosity. They can not only obtain reasonable simulation accuracy but also significantly accelerate the solution time of the flow field. The algorithms of eXtreme Gradient Boosting, K-Nearest Neighbours and Artificial Neural Network have better performance when considering the calculation accuracy, time cost and memory cost. It shows the great potential of applying machine learning algorithms to RANS turbulence model and also provides a new idea for industrial simulations of turbulent flows.