Datasets and results for neural approximations based on stress potentials

1. Overview

This is the dataset that we use for testing and training the models as desribed in the article:  
  An approach to encode divergence-free stress fields in neural approximations based on stress potentials

2. Repository Structure

data/
├── datasets/          
│   ├── grains_10_res_128_samples_5000/         
│   └── grains_48_res_128_samples_8/     
├── results/        

3. Directory Details

datasets/

Contains processed material parameter and stress tensor data in .npy format for use in neural operator training and evaluation.

Each dataset folder includes:

• Material parameters: E.npy (Young's modulus), v.npy (Poisson's ratio)
• Stress tensor components: P11.npy, P22.npy, P23.npy,P32.npy,P33.npy (stress tensor)
• Metadata: input_param_data.json.npy

results/

Contains output quantities.

The trained models:

• *PeFNO.eqx 
• *PgFNO.eqx 
• *PiFNO.eqx 

The loss histories and losses on the test set:

• *best_model_losses.npy 
• *test_losses.json 

The results of the experiments: 

• *hyper_param.json 
• *resultsGridSearch.pkl 
• *resultsSensAnaCoefLoss.pkl