RockFlow: Fast Generation of Synthetic Source Rock Images Using Generative Flow Models

Abstract

Image-based evaluation methods are a valuable tool for source rock characterization. The time and resources needed to obtain images has spurred development of machine-learning generative models to create synthetic images of pore structure and rock fabric from limited image data. While generative models have shown success, existing methods for generating 3D volumes from 2D training images are restricted to binary images and grayscale volume generation requires 3D training data. Shale characterization relies on 2D imaging techniques such as scanning electron microscopy (SEM), and grayscale values carry important information about porosity, kerogen content, and mineral composition of the shale. Here, we introduce RockFlow, a method based on generative flow models that creates grayscale volumes from 2D training data. We apply RockFlow to baseline binary micro-CT image volumes and compare performance to a previously proposed model. We also show the extension of our model to 2D grayscale data by generating grayscale image volumes from 2D SEM and dual modality nanoscale shale images. The results show that our method underestimates the porosity and surface area on the binary baseline datasets but is able to generate realistic grayscale image volumes for shales. With improved binary data preprocessing, we believe that our model is capable of generating synthetic porous media volumes for a very broad class of rocks from shale to carbonates to sandstone.

Minireview on Porous Media and Microstructure Reconstruction Using Machine Learning Techniques: Recent Advances and Outlook

“…This is because the fully convolutional architecture of the DCGAN models allows them to generate high-resolution images or volumes for large inputs. In contrast, the RockFlow model has fixed dimensions in the training image plane . In Figure , the ground-truth and DCGAN values for Minkowski functionals were evaluated for 200 3 voxel volumes of Berea sandstone samples, while RockFlow results are only for 128 3 volumes.…”

“…A few studies have used flow-based generative models to reconstruct 3D images from 2D slices based on latent space interpolation. 53,54 RockFlow was the first flow-based generative model proposed by Anderson et al 53 to reconstruct 3D images of porous media. Kamrava and Mirzaee 54 used a similar approach to generate new realizations of a complex microstructures captured by X-ray computed microtomography while additional data are produced stochastically.…”

“…In contrast, the RockFlow model has fixed dimensions in the training image plane. 53 In Figure 13, the ground-truth and DCGAN values for Minkowski functionals were evaluated for 200 3 voxel volumes of Berea sandstone samples, while RockFlow results are only for 128 3 volumes. As can be seen, the RockFlow has underestimated porosity while overestimating the surface area and closely matching the Euler characteristic.…”

“…is Kronecker delta function which is equal to 1 when its argument is zero and zero for all nonzero arguments. 13 Different types of reservoir rocks Fu et al 14 Fontainebleau sandstone Super-resolution Wang et al 16 Rock samples Kamrava et al 17 Shale samples Bai et al 23 Rock core images 2D-to-3D MRF-based reconstruction Fu et al 47 Collection of isotropic and anisotropic samples including polymer nanocomposites, Leopard sandstone, synthetic silica, and porous electrodes Exemplar-based reconstruction Bostanabad et al 49 Steel, tempered martensite, and aluminum alloy 3D-to-3D MRF-based reconstruction Fu et al 14 Fontainebleau sandstone Super-resolution Wang et al 36 Sandstone Unsupervised Learning 2D-to-2D VAE Laloy et al 25 Binary channelized aquifer Deep Belief Network Cang et al 15 Ti-6Al-4 V alloy, Pb63-Sn37 alloy, Fontainebleau sandstone, and spherical colloids GAN/Conditional GAN Feng et al 27 Binary sandstone images Feng et al 28 LSM-YSZ cathode material Kernel-based reconstruction Chan et al 35 Binary channelized image of subsurface reservoirs 2D-to-3D Latent space interpolation You et al 2 Estaillades carbonate rock sample Anderson et al 53 Shale images Kamrava and Mirzaee 54…”

2D-to-3D image translation of complex nanoporous volumes using generative networks

“…In a multimodal imaging workflow such as that shown in Fig. 1, a sample is imaged using two or more imaging modalities at the same resolution, a model is trained to translate between modalities, and the synthesized images used to estimate petrophysical properties of the sample [20][21][22] . Multimodal image prediction or enhancement is common in medical imaging [23][24][25] , but this methodology has yet to gain prominence in source rock characterization.…”