Optimization of turbulence models for supercritical fluids in vertical pipes based on Machine learning

• Developing supercritical fluid turbulence model based on machine learning method is studied. • The eddy viscosity is predicted by machine learning algorithm and used to solve the RANS equations. • The predicted results are analyzed in depth using priori and posteriori analysis methods. In this paper, We study a turbulence model based on a machine learning algorithm to predict the mechanism of turbulent heat transfer in supercritical fluids. Firstly, the machine learning algorithm is used to train the unclosed terms in the equations of the RANS model, and then the results of the turbulence model obtained from the machine learning model are coupled to the RANS solver, and the required velocity and temperature distributions are obtained by iterative calculations. The calculated turbulent stress, turbulent heat flux, velocity field and temperature field are analyzed in depth using priori and posteriori analysis methods for the cases with and without heat transfer deterioration, respectively. The results surface that the phenomenon of heat transfer deterioration in supercritical fluids can be predicted readily using machine learning methods. Among the machine learning algorithms studied, the development of turbulence models for supercritical fluids based on the ML-SGB algorithm has better efficiency and accuracy. Compared to the training of the constant physical fluid turbulence model, the supercritical fluid thermophysical properties have a drastic change when approaching the pseudo-critical point, and it is necessary to include S ij , W ij feature vectors in the input features.