The first International Rock Imaging Summit (IRIS) was held as a virtual event in November 2020. Digital rock imaging involves 3D computer tomography scans of cores, plugs or cuttings that are used in sedimentological analysis and to support petroleum engineering studies. Image segmentation allows for the identification of grains and pores. Other techniques such scanning electron microscopy are used to identify individual mineral grains. These measurements are also used as input for machine learning-based techniques to automate interpretation.
Carlos Alberto Santos Molina of Repsol Technology Lab showed how reservoir properties are obtained from different types of rock sample. Digital rock physics can be performed on ‘cheap and ubiquitous’ cuttings, providing less spatially-biased sampling across the whole reservoir than a single core. Multi-physics imaging across various measurement types is analyzed with PCA* and clustering. The ‘cheap and powerful’ approach has some drawbacks, notably cuttings quality. ‘Rock flour won’t work!’
* Principle component analysis
Leonardo Salazar presented Thermo Fisher Scientific’s (TFS) new ‘MAPS’ mineralogy software that provides ‘multi-scale and multi-modal’ data insights. New ‘Mixel’ smart X-EDS* aids mineral identification. MAPS produces detailed petrology statistics, elemental assay and particle size. Combined with the Apero scanning electron microscope, the solution is claimed to be the ‘platform of the future’ for automated mineralogy.
* X-ray energy dispersive spectrum analysis.
Andrew Fogden (Wintershall-DEA) provided a comprehensive review of various enhanced oil recovery (EOR) applications. These combine the results of digital core analysis with lab analysis of wettability and other characteristics required for two-phase flow studies. Fogden showed how EOR assessments have been made for different reservoirs: low salinity tertiary waterfloods, carbonates and gas condensate in sandstone. In the latter, digital analysis of fragmentary cuttings and sidewall cores were upscaled for use in the reservoir fluid flow model to evaluate the risk of condensate banking. The outcome was a go-ahead decision on the field’s development.
Thermo Fisher Scientific’s Gwenolé Tallec showed how dual energy computed tomography (DECT) can provide interactive visualization of an entire well. DECT analysis is provided in TFS’ PerGeos digital rock analysis toolset. Voxel by voxel density and atomic number measures provide a rock-typed image of a complete core. This can be used for plug-site selection for SCAL*. A whole core CT-derived ‘heterogeneity log’ can be correlated with LWD measurements. Other functionality includes automatic heterogeneity, dip and strike logs. A supervised machine learning approach is used for facies analysis. The approach has been presented in SPE Paper SPE-197628-MS, ‘Improved reservoir characterization through rapid visualization and analysis of multiscale image data using a digital core analysis ecosystem’, co-authored with Saudi Arabia’s KAUST R&D organization.
* Special core analysis.
Christian Hinz of Math2Market showed the impact of different segmentation methods on digital rock analysis results, including M2M’s trainable, deep learning-based segmentation GeoDICT. DRP is used to predict electrical, flow and mechanical properties. Good image segmentation is the key to accuracy. GeoDict 2021 offers a variety of segmentation methods including deep learning, along with modules for saturation and flow.
Federico Gamba (also with Thermo Fisher Scientific) provided more examples of PerGeos’ supervised ML, applied to facies detection over the whole well. PerGeos works across cores, cuttings, thin sections and slabs. These ‘ground truth’ data sources are merged into a database and used to train a rock model and automate facies analysis. Predicted facies compare well with lab measurements. PerGeos is a complete Python deep learning environment built on standard Python packages such as NumPy, Scikit-Learn and TensorFlow. More on artificial intelligence tools for Amira-Avizo Software and PerGeos Software.
Jun Luo (iRock Technologies) presented on image segmentation with a
U-Net* deep neural network. U-Net image segmentation separates
constituent parts such as matrix, pores and pyrite across a range of CT
scans of different lithologies. With sufficient training datasets,
U-Net compares well with human-interpreted results. iRock Technologies
uses industry-standard analytical tools and its own ‘RockDNA’ pore
network modeling technology. More from IRT.
* A convolutional neural network originally developed for medical images.
Using data from the University of Texas at Austin’s Digital Rocks
Portal, Matthew Andrew (Zeiss) showed
how multivariate statistical regression and reinforcement learning can
be used to predict permeability from core images. The approach has been
published in the open access E3S Web of Conferences.
Mohamed Soufiane Jouini of Khalifa University has used a 3D printer to ‘validate results’ obtained from a machine learning-based analysis of CT core scans. A 3D-Systems’ ProJet MJP 3600 with a 30 micron resolution printed virtual copies of the cores from the scanned/segmented data. Porosity measurements on the printed cores compared well with the real thing. Permeability less so.
We asked TFS for more on the MAPS/PerGeos distinction and learned
that ‘Maps software is embedded with many TFS microscopes and is used
mainly for 2D data acquisition. For more sophisticated 3D and image
processing, PerGeos adds 2D/3D multi-scale multi-modality image
analysis. More from Thermo Fisher Scientific.
The IRIS virtual event was hosted by UAE-based EXPROBIZ FZE located in Ajman, U.A.E. Rock Imaging 2021 is scheduled to be held in November.
© Oil IT Journal - all rights reserved.