This Is AuburnElectronic Theses and Dissertations

Show simple item record

MACHINE LEARNING BASED MINERAL CHARACTERIZATION FROM PORE TO CORE SCALE AND IMPLICATIONS FOR GEOCHEMICAL REACTIVITY


Metadata FieldValueLanguage
dc.contributor.advisorBeckingham, Lauren
dc.contributor.authorAsadi, Parisa
dc.date.accessioned2022-11-29T14:44:57Z
dc.date.available2022-11-29T14:44:57Z
dc.date.issued2022-11-29
dc.identifier.urihttps://etd.auburn.edu//handle/10415/8467
dc.description.abstractGeological formations have great potential for large-scale carbon sequestration to reduce the net rate of increase in atmospheric CO2. In these systems, CO2 is injected into formations and mineralized through geochemical reactions. Subsurface CO2 sequestration systems include reservoirs which store the injected CO2 and impermeable cap-rocks (mostly shales) which seal the reservoir. Impermeable cap-rocks are a necessary component of subsurface CO2 sequestration systems to prevent fluid migration and leakage. The presence and evolution of fractures in CO2 sequestration systems can not only pose a risk to system integrity but also threaten overlying groundwater resources with acidification and contamination via trace element mobilization. This process is complex and not well understood. Reactive transport simulations can be used to simulate the evolution of subsurface CO2 sequestration systems including mineral reactions and porosity evolution. Micro-scale imaging of geological samples is a powerful technique to obtain the necessary parameters for reactive transport simulations. However, images are typically manually processed by domain experts, which is time-consuming, labor-intensive, and subjective. This study is divided into three main sections. The first two sections of this study propose machine learning and deep learning approaches to facilitate the automatic segmentation of minerals while the third section deals with reactive transport simulation. In the first section, the performance of several filtering techniques with three machine learning methods and a deep learning method was assessed for reliable feature extraction and pixel-level phase segmentation of X-ray CT images. K-means clustering, Random Forest, and Feed Forward Artificial Neural Network, as well as the modified U-Net model, were applied to the extracted input features. The results showed that the U-Net model with the linear combination of focal and dice loss performed best with an accuracy of 0.91 and 0.93 for Mancos and Marcellus shales, respectively. In general, it was found that considering more features provided more promising and reliable segmentation results that are valuable for analyzing the composition of dense samples, such as shales, which are significant unconventional reservoirs in oil recovery and caprocks for CO2 storage systems. The second study proposed an intelligent framework that not only evaluates the accuracy of prediction for each pixel but also investigates the accuracy of predicted neighboring pixels. Random Forest and U-Net machine learning were used as primary model architectures for mineral characterization and surface area analysis of six sandstone samples. Various input variable sets including filter extracted features, scanning electron microscopy (SEM) backscatter electron (BSE) images and SEM-energy dispersive x-ray spectroscopy images (EDS) images were considered. A new methodology was proposed to distinguish the more susceptible places to dissolution on the surface of a given mineral using a ranked mineral dissolution risk assessment map. The results showed both methods had an acceptable performance, especially with extracted features as input to the models. However, the U-Net model outperformed the Random Forest in all samples. In addition to high accuracy in both models, the proposed methodology was shown to reliably identify the locations susceptible to dissolution indicated via proposed risk assessment maps. The intelligent segmentation and surface area analysis framework is a promising tool for accelerating the processing of SEM data and reactivity assessment of samples. The last section of this study aimed to understand the impact of variations in mineralogy of the fracture surface and surrounding matrix on simulated mineral reactions and reaction rates between minerals and CO2 saturated brine. The porosity, mineral composition, and ion concentration evolutions near fracture surfaces in the context of geologic CO2 storage, over various mineral distributions on fracture surfaces measured from image analyses of SEM-BSE images and XRD information, were assessed and compared. Numerical simulations considered reactions with CO2 acidified brine at short-term scales and long-term which are pertinent for understanding reactions for the typical laboratory experiments and field, respectively.en_US
dc.rightsEMBARGO_GLOBALen_US
dc.subjectCivil and Environmental Engineeringen_US
dc.titleMACHINE LEARNING BASED MINERAL CHARACTERIZATION FROM PORE TO CORE SCALE AND IMPLICATIONS FOR GEOCHEMICAL REACTIVITYen_US
dc.typePhD Dissertationen_US
dc.embargo.lengthMONTHS_WITHHELD:12en_US
dc.embargo.statusEMBARGOEDen_US
dc.embargo.enddate2023-11-29en_US
dc.creator.orcidhttps://orcid.org/0000-0001-9988-0308en_US

Files in this item

Show simple item record