Designing Suction-Inducing Geometries for Flow Control using CFD-Coupled Artificial Neural Networks

Abstract

Suction of the boundary layer is an effective method of flow control, but it often requires too much energy to be worthwhile. Using pressure gradients inherent in the flow to generate the suction eliminates this efficiency issue. Designing geometries to achieve this is difficult, however. In this study we train an artificial neural network to aid the process. A toy problem was investigated of a straight channel divided by a porous medium. The geometry of the upper wall is modified to induce a spanwise 'suction' through the porous medium. The task of the network is to design the geometry of this upper wall so that the desired suction profile is produced. Training data was generated using numerical simulations. The network proved effective at generating geometries that resulted in close approximations to the desired suction profiles. This is a good first step toward fully autogenous suction flow control.