Wall Shear Stress Estimation Using Deep Learning with Combined Flow Velocity Distributions of Two Orthogonal Cross Sections in an Intermediate Layer

Accurate evaluation of wall shear stress is important for increasing aircraft speed and improving fuel efficiency. Wall shear stress is estimated from the flow velocity in the viscous sublayer, but it is very difficult to measure this value. Therefore, we tried to estimate it without information on the viscous sublayer with convolutional neural network (CNN) which is a deep learning model. We used velocity and wall shear stress calculated by Large-Eddy Simulation (LES) as trained data. In this study, we studied input data that is effective in improving the accuracy of the model. An approximately 15% improvement in accuracy was achieved by using, in addition to the conventionally used velocity distribution extracted from the xy-plane, the velocity distribution extracted from the xz-plane and the velocity distribution with randomly masked viscous sublayer regions. We believe that these results represent progress in estimating wall shear stress without using information on viscous sublayer.