Convolutional Neural Networks for Implementing Opposition Control in Turbulent Channel Flows

Abstract

Opposition control is an innovative technique that has shown promise for reducing turbulence intensity and wall shear stress in turbulent channel flow. The objective of the present work is to develop a Multi Input Multi Output Convolutional Neural Network (CNN) for accurately estimating and controlling the flow in a turbulent channel at a bulk Reynolds number of 3000. We focus on using all three variables, namely, wall pressure, spanwise and streamwise shear-stresses, as inputs to the CNN architecture to predict the wall-normal velocity at different heights of the detection plane (z+ = 10, 15, 20, 25, and 30). The dataset for training the CNN is generated from Direct Numerical Simulation (DNS) of a turbulent channel flow to extract spatial wall information. The correlation coefficients (ρw) between the actual and predicted wall-normal velocities are found to be very high, with values of 0.99, 0.97, 0.94, 0.86, and 0.85 at z+ = 10, 15, 20, 25, and 30, respectively. We also calculated the R2 score, which confirmed the high accuracy of the MIMO-CNN model in predicting wall-normal velocity fluctuations. This indicates the effectiveness of the proposed MIMO-CNN architecture in accurately estimating the flow field in a turbulent channel.