Pushing the Boundaries of Porous Material Characterization: Universal Stochastic Data Fusion With Deep Learning

Microscopic imaging plays a crucial role in revealing the intricate microstructures of porous materials, enabling detailed investigations into their physical properties and behavior. However, no single imaging modality satisfies the diverse requirements for comprehensive microstructural characterization across multiple scales, limiting our ability to thoroughly analyze and model these porous materials. To overcome this limitation, multiscale and multimodal imaging approaches are increasingly employed. However, effectively integrating heterogeneous data sets into high-fidelity digital representations of porous materials remains a significant challenge. In this study, we propose a deep learning-based framework for multiscale and multimodal data fusion, leveraging advanced generative artificial intelligence to overcome two persistent hurdles: (a) seamless integration of unpaired imaging data sets from different modalities and resolutions, and (b) robust one-to-many mappings that preserve the inherent uncertainty and diversity of the combined data. By applying this framework to a porous media imaging data set, we demonstrate its ability to enhance the characterization of heterogeneous materials and uncover new insights into pore-scale physical processes. This versatile and scalable approach holds broad applicability across disciplines such as geoscience and materials science, paving the way for more comprehensive multiscale porous material analysis and modeling.