Flow Around a Cylinder and Periodic Wake-Impacted Flow for Generative Learning

This data repository contains two test cases of turbulent flow. The first dataset contains 100,000 images, each with a resolution of 1000 × 600 pixels, from a flow around a cylinder at a Reynolds number of 3900. The flow field is characterized by a Kármán vortex street developing in the cylinder’s wake. The vortex street consists of a characteristic coherent vortex system in which the rotational axes of the individual vortices are aligned with the cylinder axis. The second dataset contains 2,250 images, each with a resolution of 1000 × 625 pixels, from a flow around an academic low-pressure turbine blade (T106 LPT) under periodic wake impact. The wakes are artificially generated by an upstream rotating bar grid and convected into the stator passages, where the rotation of the flow within the passages leads to their deformation. The datasets consist of grayscale images generated by post-processing the transient Large Eddy Simulation (LES) velocity-field data using a projection mapping, in the sense that the system remains ergodic on a reduced state space. The data was originally used to train and validate generative models. The results of these studies are presented in the articles [1] Drygala, C., Winhart, B., di Mare, F., & Gottschalk, H. (2022). Generative modeling of turbulence. Physics of Fluids, 34(3), [2] Drygala, C., di Mare, F., & Gottschalk, H. (2023). Generalization capabilities of conditional GAN for turbulent flow under changes of geometry. EUROGEN 2023, [3] Drygala, C.*, Ross, E.*, di Mare, F., & Gottschalk, H. (2024). Comparison of generative learning methods for turbulence modeling. arXiv preprint arXiv:2411.16417, [4] Ross, E., Drygala, C., Schwarz, L., Kaiser, S., di Mare, F., Breiten, T., & Gottschalk, H. (2025). When do World Models Successfully Learn Dynamical Systems?. arXiv preprint arXiv:2507.04898, where a detailed description of the numerical setup can be found in [1].