BLASTNetv3: Multiphysics Simulation Dataset

Mission: BLASTNet is developed to provide researchers in the reacting and non-reacting flow physics communities with high-fidelity simulation datasets in a convenient format for machine learning (ML) applications. This release substantially expands the scale and diversity of the BLASTNet dataset, containing approximately 25 TB of data, 79,475 full-domain samples, and 337 configurations. By providing a large and diverse collection of high-fidelity flow simulations, BLASTNet aims to facilitate the development, evaluation, and fair comparison of ML methods for scientific and engineering applications. Application: The dataset is intended for a wide range of ML applications in fluid dynamics and related fields, including automotive, propulsion, energy, and environmental applications. In particular, the data can support research in areas such as turbulent closure modeling, spatio-temporal modeling, surrogate modeling, super-resolution, reduced-order modeling, and inverse modeling. The scale and diversity of this release also make BLASTNet suitable for investigating data-driven scaling behavior, generalization across flow configurations, and the development of foundation models for fluid physics.