The use of machine learning for filtered statistic turbulent channel flow

The use of machine learning, a recent topic in computer engineering, has become popular in various branches of science for different purposes, including prediction, image processing, classification and clustering. In this work, machine learning is used for the prediction of filtered Reynolds stresses in turbulent channel flow. For this purpose, first turbulent statistics are filtered with a specific filter size to prepare learning data for the neural network. For output, filtered stresses are expected. In this research, machine learning method used instead of direct filtering, as a first step toward subgrid-scale modeling. Several methods exist for machine learning, but linear regression and neural network method is used here. In order to test the accuracy of trained neural network, unfiltered stresses other than those used for the training are used. The results reported a high correlation between the neural network output and the data from numerical simulation.