D3514_README File For 'Research Data for PhD Thesis Predicting Aerodynamics with Hierarchical Geometric Deep Learning: The GCN-MM-AE Framework'

Dataset DOI: 10.5258/SOTON/D3514

Date that the file was created: May 2025

-------------------
GENERAL INFORMATION
-------------------

ReadMe Author: David Massegur Sampietro, University of Southampton, ORCID ID 0000-0001-6586-5097

This dataset supports the thesis entitled 'Predicting Aerodynamics with Hierarchical Geometric Deep Learning: The GCN-MM-AE Framework'
AWARDED BY: Univeristy of Southampton
DATE OF AWARD: June 2025

DESCRIPTION OF THE DATA:
This dataset contains all the results obtained for the various results chapters of the PhD thesis, particularly Chapters 4 to 8. 
These results were then used to generate the various results figures illustrated in the thesis. The multiple results files are classified according to the various folders. 
As reported in the thesis, results were obtained eithere with SU2 CFD solver for the reference (ground-truth) and the implemented python/pytorch library developed during the research.
Feel free to contact the author if further information is required.

This dataset contains the four different folders corresponding the results chapters to facilitate consultation:

Chapter_04: 
Mesh results in 'vtu' format for GCN-MM-AE model.
Includes ML predictions (variables ending with _pred), CFD ground truth (variables ending with _gt) and the error (variables ending with _pred_vs_gt). Variable name roots correspond to the same as used in the thesis (Cp, Cfx...).
Readable in Pyvista (python package) and Tecplot.

Chapter_05: 
Results for quasi-steady and recurrent model predictions for validation signal 8 reported in the chapter. 
Excel files consist of the comparison and error reports for each integrated force quantity. Each row correpond to a time step.
The 'npz' files contain the model predictions (Pressure_pred_dyn), the predicted integrated forces (Aeroforces_pred_dyn) and the reference integrated forces for comparions (Aeroforces). Rows are sorted by time step.
The respective files with 'headers' in the name are text files stating the (self-explanatory) name of each variable contained in the respective npz file.
Readable in numpy or excel.

Chapter_06: 
Results for quasi-steady and recurrent model predictions for validation case 111 reported in the chapter. 
Excel files consist of the comparison and error reports for each integrated force quantity. Each row correpond to a time step.
The 'npz' files contain the model predictions (Pressure_pred_dyn), the predicted integrated forces (Aeroforces_pred_dyn) and the reference integrated forces for comparions (Aeroforces). Rows are sorted by time step.
The respective files with 'headers' in the name are text files stating the (self-explanatory) name of each variable contained in the respective npz file.
Readable in numpy or excel.

Chapter_07: 
Mesh results in 'vtu' format for the gradient-loss model (approach 2) for the reduced dataset scenario (scenario 2).
Includes ML predictions (variables ending with _pred), CFD ground truth (variables ending with _gt) and the error (variables ending with _pred_vs_gt). Variable name roots correspond to the same as used in the thesis (Cp, Cfx...).
Readable in Pyvista (python package) and Tecplot.

Chapter_08: 
Mesh results in 'vtu' format for the GE_EDGCN-MM-AE model.
Includes ML predictions (variables ending with _pred), CFD ground truth (variables ending with _gt) and the error (variables ending with _pred_vs_gt). Variable name roots correspond to the same as used in the thesis (Cp, Cfx...).
Readable in Pyvista (python package) and Tecplot.

Date of data collection: March 2021 - September 2024

--------------------------
SHARING/ACCESS INFORMATION
-------------------------- 

Licence:
Restricted

Recommended citation for the data:
Massegur, D. (2025) "Data for thesis Predicting Aerodynamics with Hierarchical Geometric Deep Learning: The GCN-MM-AE Framework" (10.5258/SOTON/D3514) [Dataset], Pure, University of Southampton, DOI: https://doi.org/10.5258/SOTON/D3514 

This dataset supports the thesis:
D. Massegur (2025) "Predicting Aerodynamics with Hierarchical Geometric Deep Learning: The GCN-MM-AE Framework", University of Southampton, Faculty of Engineering and Physical Sciences, Aeronautical and Astronautical Department, PhD Thesis, 375 p., DOI: https://doi.org/10.5258/SOTON/T0080