BlendedNet++: A Large‑Scale Blended Wing Body Aerodynamics Dataset and Benchmark

BlendedNet++ is a large-scale aerodynamic dataset and benchmark focused on Blended Wing Body (BWB) aircraft, developed by the MIT DeCoDE Lab and MIT Lincoln Laboratory. It significantly expands upon the original BlendedNet dataset. Dataset Overview: Geometries: 12,490 unique BWB geometries. Simulation: Steady RANS CFD (Spalart–Allmaras turbulence model) using NASA FUN3D. Data Content: Integrated force and moment coefficients (CL, CD, CM). Dense surface fields for pressure (Cp) and skin friction (Cfx, Cfy, Cfz). Benchmarks: Includes baselines for forward surrogate modeling (PointNet, GraphSAGE, GraphUNet, Transolver, FiLMNet, GNOT) and inverse design (Conditional Diffusion). File Reconstruction Instructions (IMPORTANT): Due to the large size of the dataset (approx 60GB uncompressed), the main archive has been split into 2GB chunks named blendednet++_dataset.tar.gz.part-**. You must download all parts and recombine them before extracting. Linux / macOS: Open a terminal in the directory where you downloaded the files and run: cat blendednet++_dataset.tar.gz.part-* > blendednet++_dataset.tar.gz tar -xzvf blendednet++_dataset.tar.gz Windows (Command Prompt): copy /b blendednet++_dataset.tar.gz.part-* blendednet++_dataset.tar.gz (Then extract the resulting .tar.gz file using 7-Zip or similar).