Streamlining Digital Rock Analysis with cloud based workflows

This blog is part 1 of a two-blog series on digital rock and outcrop reconstruction and analysis on Amazon Web Services.

Part 1 will focus on digital rock analysis (DRA) and fluid flow simulation on Amazon Web Services. Part 2 will focus on digital outcrop models for large-scale structural analysis.

The key to accelerating research and production in the energy and mining industries resides in the efficient and generally applicable determination of physical properties of rocks. In this blog, we present recent advancements in the generation of statistical digital twins of reservoir rocks in a workflow that might commonly be applied to any grain-based sample.

The study of flow-and-transport phenomena in sedimentary rocks is important in a range of scientific and engineering applications, including enhanced hydrocarbon recovery, mineral exploration, geothermal energy, groundwater resources, hydrogen storage, CO2 sequestration, mining, and geomechanics. The numerical simulation of various physical and chemical processes in digital rock samples allows for pore-scale analysis and upscaling of rock properties, such as electric resistivity, permeability, and elastic moduli. Moreover, DRA facilitates the nondestructive assessment of different scenarios at in situ and ex situ conditions. For example, CO2 sequestration, or the injection of incondensable gases into geothermal fields, causes rock-matrix dissolution or mineral precipitation, changing the macroscopic properties of rocks, thus requiring detailed preliminary numerical study before applying this technique to real aquifers.

The Australian National Low Emissions Coal Research and Development (ANLEC R&D) used DRA workflow to understand the physics of CO2-brine systems at the pore scale. Their goal was used to assess potential CO2 storage sites within the Surat Basin. By micro-CT, the organization demonstrated site-specific 3D imaging of in situ supercritical CO2 saturation at the pore scale, and it conducted direct 3D pore-scale imaging of supercritical CO2 and brine within Surat Basin core material during CO2 injection at aquifer pressure and temperature conditions. Through this study, ANLEC R&D illustrated that capillary trapping is a significant mechanism for CO2 storage in the Precipice Sandstone and is likely to be stable over timescales of decades to centuries. Additionally, CO2-brine displacement properties are typical of a strongly water-wet system.

X-ray microtomography and high-resolution tomographic images of rock cores make it possible to study flow-and-transport phenomena in detail at the pore scale. However, these high-resolution images require large storage from existing IT infrastructure, stressing data center environments and IT managers. Researchers and experts use DRT to save time and resources that could have otherwise been spent on other methods, such as laboratory core tests especially for unconventional reservoirs. On average, it takes about 4–6 weeks for lab analysis, and results could take up to 5 months when labs are backlogged, leading to slow decision-making. Moreover, modeling of such high-resolution images requires huge amounts of computational power that is often not available in on-premises environments.

Approach

For running DRA on Amazon Web Services, we use Amazon SageMaker —a service to build, train, and deploy machine learning (ML) models for any use case with fully managed infrastructure, tools, and workflows—to launch a Jupyter Notebook instance to install PoreSpy. PoreSpy is an open-source quantitative image analysis tool that we use to create a 3D volume of the image data obtained from the digital rock portals. We extract the pore network from the 3D rock volume using the Snow2 algorithm in PoreSpy, and we use deep learning to predict the diffusional conductance in the extracted pore network. Both the 3D output and the raw images are stored in Amazon Simple Storage Service (Amazon S3), an object storage service offering industry-leading scalability, data availability, security, and performance. Figures 1 and 2 below show an example micro-CT image from a sandstone core section, which is used to create a digital rock for petrophysical analysis.

Figure 1. Example of a micro-CT image with the selected area chosen for reconstruction (Doddington Sandstone)

Figure 2. 3D view of reconstructed digital rock from micro-CT images (showing only voxelized pore spaces)

To determine flow properties such as permeability from this generated pore network, we perform single- and two-phase flow simulations using the open-source product PoreFoam. Using Amazon Web Services ParallelCluster , an open-source cluster management tool, we built a cluster using PoreFoam running on highly parallel nodes. Compute, storage, and network requirements are defined in Amazon Web Services ParallelCluster configuration files and automatically populated by user input. After benchmarking, the c5.12xlarge instance type (48 vCPU, 96 GB RAM), which is designed to run compute- and memory-intensive workloads, is the best choice for compute in this use case. This configuration can run multiple simulations in parallel without waiting in queues, leading to faster results compared to the on-premises configuration. As shown in figure 3, we used Amazon Web Services ParallelCluster to launch a cluster of instances from Amazon Elastic Compute Cloud (Amazon EC2), which offers secure and resizable compute capacity for virtually any workload. With Amazon Web Services ParallelCluster, you create a simple text configuration file through a user-friendly interface to model your cluster’s resources, which are used to automate the provisioning of compute, storage, and networking. Our cluster was deployed in the US-East-1 region within 20 minutes.

Figure 3. Amazon Web Services ParallelCluster architecture

Numerical simulation

Porosity values are determined based on the analysis of the images. As Bijeljic et al. explained in their 2013 article , porosity is calculated by dividing the number of pore voxels (Npvox) by the total number of voxels (Nvox). Voxels that lack connections to the inlet or outlet through the pore space are not considered in the flow calculations.

To calculate the flow, we use a standard finite volume method, which is implemented in OpenFoam. The software directly simulates incompressible, steady, viscous flow through the pore-space images by solving the volume conservation equation and the Navier-Stokes equations. Normalized flow fields are determined where the ratios of the magnitude of U at the voxel centers divided by the average flow speed Uav are represented as streamlines. Red and green indicate high values, while blue indicates low values.

Figure 4a. Normalized pressure fields with a unit pressure difference across the digital rock sample

Figure 4b. Velocity flow field, which characterizes the nature of fluid flow

The area of investigation, voxel sizes, and computed porosity and permeability are shown in table 1 below.

Computation time and convergence

The simulation results were analyzed to determine the optimal convergence time for estimating rock properties. Simulation data was collected at 0.1, 0.2, 0.5, 1, and 2 seconds for the rock sample. Analysis of the rock property values demonstrates that the porosity and permeability calculations stabilized after 0.2 seconds, as shown in figure 5. This observation implies that extended simulations may not be needed in this case, because convergence was achieved early on. However, while shorter convergence times are desirable for computational costs, longer simulation times may be necessary to accurately capture more complex flow phenomena, especially in multiphase flow scenarios. Our workflow provides a streamlined solution to address these extended simulations that require large computational overhead. Figure 5 demonstrates how we can save costs by avoiding long-running simulations once we see good convergence.

Figure 5. Simulation times for permeability and porosity calculations

Conclusion

DRA can help operators to better understand the properties of reservoirs—including porosity, fluid saturations, and permeability—which can lead to increased efficiency and accuracy in reservoir characterization. This information can be used to optimize drilling and production strategies.

As earlier stated, DRA facilitates the nondestructive assessment of different scenarios at in situ and ex situ conditions, which is not possible with existing laboratory experimental methods. Using DRA, customers can run multiple simulations and visualize pore and grain space without altering the original rock structure and composition.

Running DRA on Amazon Web Services provides the following benefits:

1. Scalability: Amazon Web Services provides on-demand access to a wide range of compute instances with varying CPU, GPU, and memory configurations, helping users to easily scale their simulation workloads up or down as needed.
2. Cost savings: Users pay for only the resources that they use and do not need to invest in expensive hardware or worry about maintenance and upgrades, which are typical for this kind of simulation.
3. Speed and performance: Amazon Web Services provides access to high-performance instances and storage, facilitating faster and more efficient analysis of large digital rock datasets.
4. Flexibility: Amazon Web Services provides a wide range of features and services, including the ability to store and analyze large amounts of digital rock data, collaborate with other users, and integrate with other tools and platforms.

Furthermore, by having the flexibility to run hundreds of simulations, customers can accurately study the behavior of fluid in the reservoir, which can lead to the development of more effective enhanced oil-recovery techniques and CO2 underground storage efforts.