GNNePCSAFT Project

Project focused in the use of graph neural networks to estimate the pure-component parameters of the Equation of State ePC-SAFT.

The motivation of this work is to be able to use a robust Equation of State, ePC-SAFT, without prior need of experimental data. Equations of State are important to calculate thermodynamic properties, and are pre-requisite in process simulators.

Currently, the model takes in account only the hard-chain, dispersive and assoc terms of ePC-SAFT. Future work on polar and ionic terms are being studied.

Code is being developed mainly in Pytorch (PYG).

You can find the model deployed at GNNePCSAFT Webapp.

A CLI to use the model can be found at GNNePCSAFT CLI and installed with pipx:

pipx install gnnepcsaftcli

Checkpoints can be found at Hugging Face.

Use cases of this package are demonstrated in Jupyter Notebooks:

• compare.ipynb (Open in Colab): comparison of the performance between two or more trained models
• demo.ipynb (Open in Colab): pt-br demonstration of models capabilities
• training.ipynb (Open in Colab): notebook for model training
• tuning.ipynb (Open in Colab): notebook for hyperparameter tuning

---

Work in progess.