Generative adversarial networks (GANs) are among the most promising methodologies, offering more accurate estimations and better perceptual quality. But can they be employed to reconstruct an entire three-dimensional field?
With 3D-GANS, a generative AI methodology, we have demonstrated for the first time that it is possible to estimate 3D channel flow from wall measurements using a single network. This approach maintains the advantages of analogous planar estimators, such as the capability to extract non-linear patterns from data, while also overcoming certain limitations. It can estimate a 3D region in a single training-estimation step and requires relatively fewer computational resources. Most importantly, turbulent coherent structures can be easily identified, providing valuable information about their wall footprint.
Read the full open access publication here. The data sets and codes are also openly accessible.
This work is a collaboration with Ricardo Vinuesa of KTH Royal Institute of Technology
