Example Dataset for GNN-PNM: A Graph Neural Network–Embedded Pore Network Model for Permeability Prediction

This dataset accompanies the project GNN-PNM Permeability, which introduces an end-to-end differentiable framework combining Graph Neural Networks (GNNs) with a pore network solver to predict bulk permeability from 3D digital rock images.

The archive contains:

• data.zip: A collection of graph representations (graph_data.pt) and network parameters (network_params.npz),  derived from synthetic microstructures. These files are used as inputs to the GNN model.

• permeability_values.txt: Ground-truth (GT) permeability values corresponding to each image, used as training targets during model optimization.

This dataset is intended for training and validating the GNN-PNM model. An example is included in the GitHub repository; users should replace it with the full dataset by extracting data.zip into the data/ folder and placing permeability_values.txt accordingly.

GitHub Repository:
🔗 https://github.com/ITLR-DDSim/gnn-pnm-permeability