Reconstruction and Synthesis of Source Rock Images at the Pore Scale

Image-based characterization of rock fabric is critical for understanding recovery mechanisms in shale formations due to the significant multiscale nature of shale source rocks. Nanoscale imaging is particularly important for characterizing pore-scale structure of shales. Nanoimaging techniques, however, have a tradeoff between high-resolution/high-contrast sample-destructive imaging modalities and low-contrast/low-resolution sample-preserving modalities. Furthermore, acquisition of nanoscale images is often time-consuming, expensive, and requires signficant levels of expertise, resulting in small image datasets that do not allow for accurate quantification of petrophysical or morphological properties. In this work, we introduce methods for overcoming these challenges in image-based characterization of the fabric of shale source rocks using deep learning models. We present a multimodal/multiscale imaging and characterization workflow for enhancing non-destructive microscopy images of shale. We develop training methods for predicting 3D image volumes from 2D training data and simulate flow through the predicted shale volumes. We then present a novel method for synthesizing porous media images using generative flow models. We apply this method to several datasets, including grayscale and multimodal 3D image volume generation from 2D training images. Results from this work show that the proposed image reconstruction and generation approaches produce realistic pore-scale 3D volumes of shale source rocks even when only 2D image data is available. The models proposed here enable new capabilities for non-destructive imaging of source rocks and we hope will improve our ability to characterize pore-scale properties and phenomena in shales using image data.