Reynolds Averaged Modelling with Convolutional Neural Networks

The jet is a common flow, which can be observed from the nozzle for an aeration installed in a dam. Such a jet flow is one of the complicated problems since it has various sizes of eddies having different momentums. So, it is necessary to reconstruct the jet flow by reflecting the jet flow characteristics such as Reynolds stress, turbulent kinetic energy to understand and also expand its applications.

The previous studies conducting the Reynolds-averaged Navier–Stokes (RANS) model proved that the realizable k-epsilon is more efficient to do research the turbulent mechanism than other models. However, the Reynolds averaged model is hard to reflect all flow properties such as the fluctuations of velocity or pressure and the model need to be improved by including more detailed turbulent processes. As a recommendation, we use a new and suitable technique, a Convolutional Neural Network (CNN), to combine RANS and LES, which algorithm has been used to find the linear and nonlinear relationship.

This study is aimed to pursue two objectives; we try to find out the relationship between LES and RANS models and then, some turbulent processes from the LES are combined with RANS model using CNN. Finally, we compare and evaluate both results of this model and results used just RANS or LES models.