LATTICEPT转发了
Latticept is excited to provide this massive volumetric dataset (1TB) for the #AI training community! I'm hoping the NVIDIA Modulus users of the world (like Michal Taká?, PhD.) can use this dataset for some fully transient 3D training of Fourier neural operators. Maybe Rishi Puri can glue a GNN to it?... Computed in Dotmatics M-Star CFD using the Lattice Boltzmann Method, this is a full volumetric dump of hydrofoil agitator starting from rest at different RPMs. All of it saved in Kitware Inc.'s Paraview format. The dataset is 1.2 million structured cells of a hydrofoil agitator in a pressure vessel starting from rest. It contains vorticity, strain rate vectors, velocity vectors, energy dissipation rate and pressure for 1 second of fluid flow. Timesteps are 5e-4 seconds. Plain LES turbulence is assumed for the subgrid turbulence model via eddy viscosity. RPMs included are from 5 to 60 are included in 5 RPM increments. Download here: https://lnkd.in/eiDzia_f Other finite volume solvers are showing off their AI-capabilities by "solving CFD in seconds". What they're really doing is spending weeks to generate data, then using it to train another model and expect you to be impressed when they click a button. Congrats? I guess? All of this was solved in just hours on a desktop RTX4090. Most of it is limited by hard drive saving -- it took me LONGER TO UPLOAD IT to AWS! DM me here or visit my website to to learn more about M-Star CFD! #ai #cfd #nvidia #modulus #kitware
