This is the repository for the FlowBench. It contains the link to the full dataset and the code used for training the SciML operators. Additionally, we include a collection of scripts for preparing data into machine learning format and downsampling data into lower resolution.
FlowBench is an extensive flow dataset which contains over 10,000 data samples of a fully resolved numerical simulation for modeling transport phenomena in complex geometries. FlowBench will facilitate the evaluation of the interplay between complex geometry, coupled flow phenomena, and data sufficiency on the performance of current and future neural PDE solvers.
We include workflows to train three types of neural operators: Fourier Neural Operators (FNO), Convolutional Neural Operators (CNO), and Deep Operator Networks (DeepONets). The implementation of the three networks are included here.
FlowBench offers over 10,000 solutions for flow around complex geometries in both 2D and 3D. Simulation include fluid flow and thermal flow scenarios, and our solutions are available as either single snapshots or time sequences, addressing both steady-state and time-dependent scenarios. The dataset is provided at three different resolutions and includes essential features like a geometry mask and a signed distance field. It is specifically designed to support the development of next-generation scientific machine learning (SciML) neural PDE solvers, particularly those tackling complex geometries and multiphysics phenomena.
We have a project website which highlights FlowBench main results. Our website gives an overview of our dataset, geometries, solver, and our research team that worked on this project.