Automation of AI and Pore Scale Transport Research by Python scripting in GeoDict

The automation of complex computations in porous media research is greatly enhanced by use of the GeoPy module in GeoDict, which enables streamlined scripting for diverse applications. One such application is the computation of the Representative Elementary Volume (REV), a key property in characterizing porous media (Fig 1). The REV for different parameters, such as porosity (using the PoroDict module), permeability (through FlowDict), tortuosity (with PoroDict), and Minkowski functionals (with MatDict) was automated through Python scripting in GeoDict, to analyze multiple high-resolution micro-CT scan datasets (1). On top of prediction of porous media properties on single scans, this type of automation is invaluable to efficiently and automatically process large datasets of thousands structures for generating training data for AI and machine learning tasks (2, 3). By fully automation of tasks, the computational effort is reduced and reproducibility is enhanced.

Extending beyond pore scale property prediction, Python scripting also facilitates more complicated dynamic modeling tasks, such as simulating particle transport and retention at the pore scale (4). By integrating Eulerian–Lagrangian models, transport properties can be dynamically updated to investigate processes like surface deposition and clogging (Fig 1b). The automated workflows also support the generation and validation of datasets for advanced modeling, such as using deep learning networks to predict the effects of particle deposition on porosity and permeability (Fig 1c) (3). Together, these examples highlight how GeoPy in GeoDict enables a seamless transition from static analysis to dynamic modeling, demonstrating its versatility in addressing both fundamental and applied challenges in porous media research.