A team of researchers from the U.S. Department of Energy's (DOE) Argonne National Laboratory has developed a new machine-learning algorithm that can rapidly and accurately determine the geometry of gaps in porous materials. The algorithm, called PGNet (Pore Geometry Network), could be used to better understand the structure of liquids and how they flow through porous media, which has applications in a wide range of fields, such as energy, environmental science, and pharmaceutical manufacturing.

"Gap geometry is a fundamental property of porous materials that governs their ability to store and transport fluids," said Argonne scientist Dongxiao Zhang, a co-author of the study. "However, accurately determining gap geometry from experimental data or simulations is a challenging task, especially for complex porous materials."

The researchers developed the PGNet algorithm using a machine-learning technique called convolutional neural networks (CNNs). CNNs are a type of deep learning algorithm that is well-suited for image analysis and recognition tasks. The researchers trained the PGNet algorithm on a large dataset of images of simulated porous materials, and they showed that it could accurately determine the gap geometry of these materials.

The researchers then used the PGNet algorithm to study the structure of liquids in porous materials. They found that the gap geometry of porous materials has a significant impact on the structure of liquids confined within the pores.

This work was funded by the DOE's Office of Basic Energy Sciences. The research team included Dongxiao Zhang, Yuan Cheng, and Yongqiang Cheng of Argonne National Laboratory; and Jialin Li and Ruiqiang Li of the University of Nebraska at Omaha.

The study is published in the journal Nature Communications.