Analysis of air-jet vortex-generator controlled SWBLI using deep encoder-decoder convolutional network

Shock-wave/boundary-layer interactions are commonly occurring complex phenomenon in many aerospace applications with several detrimental effects. A common flow control technique to alleviate its adverse effects and to mitigate separation is the application of air-jet vortex-generators (AJVGs). AJVGs induce counter-rotating vortex pair downstream of jet injection and enhance the momentum redistribution in the boundary layer. Consequently, the flow is more resistant to separation and the extent of shock-induced separation decreases. Such flows are complex in nature and is highly space and time dependent. Hence, we use scientific machine learning to reveal the physics behind the control effectiveness of AJVGs. Once trained, artificial neural networks such as deep encoder-decoder convolutional network allows to extract high quality data at low cost and high accuracy depending on the quality of the network.

Firstly, we build the architecture and validate the network using a 2D test function. We evaluate the network by performing a parametric study before applying it on both unforced and forced AJVG-control cases.