Characterizing heterogeneous rocks in 3D with a multimodal deep learning approach – Implications for transport simulations

Investigating the heterogeneous transport properties of rock is vital for accurate assessment of radionuclide migration, which is essential for the safety assessment of a nuclear waste disposal facility. Previous studies have combined x-ray computed tomography (XCT) with other methods to obtain three-dimensional (3D) mineral and porosity maps, but such approaches are time consuming and somewhat dependent on the operator. To address these limitations, we have developed a deep learning-based method that combines XCT with fast and modern characterization techniques such as scanning micro x-ray fluorescence (μXRF) and carbon 14 polymethylmethacrylate (PMMA) autoradiography. This innovative approach produces 3D mineral and porosity maps with minimal operator dependency and manual work. The results obtained from our analysis of various rock samples demonstrate the method’s suitability for transport simulation studies in various geological settings.