Accelerated porous materials design and development using statistics and machine learning

Mass transport properties of porous materials are crucial for many applications. Macroscopic performance is determined by microscopic structure. Hence, optimizing microstructures is key in e.g. hygiene products, pharmaceutics, and batteries.

In this project, we will combine stochastic modeling, spatial statistics, simulations, and machine learning to understand porous materials and their mass transport properties.

The project builds upon earlier and ongoing projects in which material models and prediction models have been developed.

In short, we will generate large sets of realistic materials structure models, both isotropic and anisotropic, and compute their mass transport properties as well as geometrical features useful for characterization, develop new models for mass transport prediction using multiple kinds of statistical models and neural networks, exploring sparsity and separability to obtain ´explainable´ models, explore materials using generative machine learning architectures and optimize structures with respect to geometrical features and mass transport.

The purpose is to develop a toolchain for faster and better tailor-made porous materials design by advanced methods in statistics and machine learning (ML). The project will mainly fund two postdocs.

The proposed research will facilitate tailor-making materials for specific purposes, lead to new methods and models in statistics and ML, and give new insights at the intersection of materials science and ML.