Discussing how ML can be applied to CFD
CFD, being a field that has been around for decades, has matured to be very useful to companies in various domains and has also been implemented at scale using cloud providers. Recent advances in ML have been applied to CFD, and in this section, we will provide readers with pointers to articles written about this domain.
Overall, we see deep learning techniques being applied in two primary ways:
- Using deep learning to map inputs to outputs. We explored the flow over an airfoil in this chapter and visualized these results. If we had enough input variation and saved the outputs as images, we could use autoencoders or Generative Adversarial Networks (GANs) to generate these images. As an example, the following paper uses GANs to predict flows over airfoils using sparse data: https://www.sciencedirect.com/science/article/pii/S1000936121000728. As we can see in Figure 11.22, the flow fields predicted by CFD and the GAN are visually very...
