Generative neural network research has surged in the past year, especially bringing it to mainstream popularity with the introduction of openAIs ChatGPT. Another way to utilize generative networks besides NLP, is decreasing computational costs of expensive calculations through estimated data generation, like in the field of computational fluid dynamics.

This thesis proposes to utilize generative AI as a tool to upsample fluid simulations beyond the accuracy of regular interpolation methods by use of super resolution techniques, more specifically Super Resolution Convolutional Neural Networks. Solving computational fluid dynamics problems at a low data density and introducing a coupled CNN to upsample them to higher resolutions, allows for fast FEM computations, which are coupled in real-time through the preCICE library. By use of computational experiments the effectiveness and limitations of the proposed technique are demonstrated.