fast-eng-super-resolution

Fast super-resolution for engineering simulations on 3D unstructured meshes

Introduction

This repository contains the code for the paper "Fast super-resolution analysis of low-pressure duct air flow through adaptive domain decomposition" by Wenzhuo Xu, Akibi Archer, Mike McCarrell, Scott Hesser, Noelia Grande Gutiérrez and Christopher McComb.

Quick Start

Use the following command to configure the environment:

pip install -r requirements.txt

Training and prediction on the framework can be done by running the python script in the project folder. An example using the duct geometry dataset, and neural operator (NO) is shown below:

python run_DS_3D.py --mode=train

After excuting the above command, the model will be trained on the duct dataset. To predict the results, run the following command:

python run_DS_3D.py --mode=predict

All results will be saved to the ./logs folder.

Dataset

The duct dataset in the paper is NOT included in this repository due to copyright reasons. However, the code can be easily adapted to other ANSYS Fluent simulation results. ANSYS mesh files .msh and Fluent output summary binary files are required for the training and prediction. Place the raw data in the ./data/raw folder. Specify raw data file names in the ./dataset/GraphDataset.py file, and the code should be able to process the data automatically.

Configurations

Several other models are also available in the models folder. To use a different model, change the model parameter when submitting the python script. Logging folders can also be specified. For example, to train a neural operator (NO) model, use the following command:

python run_DS_3D.py --mode=train --model=neuralop --exp_name=duct_neuralop

And a similar command to predict the results:

python run_DS_3D.py --mode=predict --model=neuralop --exp_name=duct_neuralop