Flow imaging as an alternative to non-intrusive measurements and surrogate models through vision transformers and convolutional neural networks

In this work, we propose a framework whereby flow imaging data is leveraged to extract relevant information from flowfield visualizations. To end, vision transformer (ViT) model developed predict the unsteady pressure distribution over an airfoil under dynamic stall images of flowfield. The network capable identifying features present in and associate them response. Results demonstrate that effective interpolating extrapolating between regimes for different motions, meaning ViT-based models may offer promising alternative sensors experimental campaigns building robust surrogate complex flows. addition, uniquely treat image semantic segmentation as image-to-image translation task infers labels structures input supervised way. Given velocity field, resulting convolutional neural (CNN) generates synthetic any corresponding fluid property interest. particular, convert field into order subsequently estimate manner.