Reduced order fluid modeling with generative adversarial networks

Surrogate models based on convolutional neural networks (CNNs) for computational fluid dynamics (CFD) simulations are investigated. In particular, the flow field inside two-dimensional channels with a sudden expansion and an obstacle is predicted using image representation of geometry as input. Generative adversarial (GANs) have been shown to excel at such image-to-image translation tasks. This motivates focus this work investigating specific effect training model performance. Numerical results show that overall accuracy GANs generally lower compared identical generator trained directly ground truth L1 data loss. On other hand, GAN predictions often visually more convincing exhibit continuity residual.