Published July 29, 2026 | Version v1

Distributed Training of U-Net Architecture for Data-Driven Aerodynamic Flow and Pollutant Dispersion Prediction in Urban Environments - DATASET

  • 1. ROR icon Universitat Jaume I
  • 2. ROR icon Barcelona Supercomputing Center

Description

This repository is intented to share an example of the CFD simulation from the research paper entitled Distributed Training of U-Net Architecture for Data-Driven
Aerodynamic Flow and Pollutant Dispersion Prediction in Urban Environments. In addition the source code for the distributed training of the model is uploaded in a Github repository. All CFD cases cannot be uploaded due to the large amount of storage it would take.

Description

CFD_case01.zip -> a reference case with the raw Velocity (U) and Tracer (T) from OpenFOAM.

cell_centers.zip -> buildings.stl and raw cell centers file (C) from OpenFOAM for geometry depiction.

best_model.pth -> best model already trained for predicting U and T fields.

interpolated_cases.h5 -> a set of cases interpolated to the regular mesh.

mesh_data.h5 -> mesh data from the cases + derived fields.

 

Files

cell_centers.zip

Files (10.2 GB)

Name Size
md5:7dd755b956035d7555690c4b22317eb4
16.9 MB Download
md5:285031bea2beaabfdd7f758f3e48a539
3.1 GB Preview Download
md5:971810cbcb2940e49a551de5cc40d4df
3.5 GB Preview Download
md5:f31c3ebad74affc026016c04069d14b7
525.7 MB Download
md5:462bef32ff8f2b322e07a54c5660b1be
3.1 GB Download