Is attention all geosciences need? Advancing quantitative petrography with attention-based deep learning

Computers & Geosciences, Journal Year: 2023, Volume and Issue: 181, P. 105466 - 105466

Published: Sept. 27, 2023

Language: Английский

---

Using Internal Standards in Time-resolved X-ray Micro-computed Tomography to Quantify Grain-scale Developments in Solid State Mineral Reactions

Published: Aug. 14, 2023

Abstract. X-ray computed tomography has established itself as a crucial tool in the analysis of rock materials, providing ability to visualise intricate 3D microstructures and capture quantitative information about internal phenomena such structural damage, mineral reactions, fluid-rock interactions. The efficacy this tool, however, depends significantly on precision image segmentation, process that seen varied results across different methodologies, ranging from simple histogram thresholding more complex machine learning deep strategies. irregularity these segmentation outcomes raises concerns reproducibility results, challenge we aim address work. In our study, employ mass balance metamorphic reaction an standard verify accuracy shed light advantages approaches, particularly their capacity efficiently expansive datasets. Our methodology utilises achieve accurate time-resolved volumetric images gypsum dehydration reaction, traditional techniques have struggled with due poor contrast between reactants products. We utilise 2D U-net architecture for introduce learning-obtained labelled data (specifically, random forest classification) innovative solution limitations training obtained imaging. algorithm developed demonstrated remarkable resilience, consistently segmenting volume phases all experiments. Furthermore, trained neural network exhibits impressively short run times workstation equipped Graphic Processing Unit (GPU). To evaluate workflow, compared theoretical measured molar evolution bassanite during dehydration. errors predicted segmented volumes time-series experiments fell within 2 % confidence intervals curves, affirming methodology. also by proposed method methods found significant improvement volumes. This makes CT suited extracting data, variations growth rate pore size reaction. work, distinctive approach using validate model, demonstrating its potential robust reliable field. measure paves way advanced modelling verification physical properties those involved tectono-metamorphic processes. work underscores promise approaches elevating quality research geosciences.

Language: Английский

---

Application of improved and efficient image repair algorithm in rock damage experimental research

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: June 27, 2024

Abstract In the petroleum and coal industries, digital image technology acoustic emission are employed to study rock properties, but both exhibit flaws during data processing. Digital is vulnerable interference from fractures scaling, leading potential loss of data; while not hindered by these issues, noise destruction can interfere with electrical signals, causing errors. The monitoring errors techniques undermine effectiveness damage analysis. To address this issue, paper focuses on restoration acquired through technology, leveraging deep learning techniques, using soft hard rocks made similar materials as research subjects, an improved Incremental Transformer algorithm repair distorted or missing strain nephograms uniaxial compression experiments. concrete implementation entails a comprehensive training set derived fabricating masks for absent segments, predicting full detail. Additionally, we adopt separable convolutional networks optimize algorithm’s operational efficiency. Based this, analysis conducted repaired nephograms, achieving closer correlation actual physical processes compared conventional techniques. incremental presented in will contribute enhancing efficiency realm damage, saving time money, offering innovative approach traditional

Language: Английский

---

RTG-GNN: A novel rock topology-guided approach for permeability prediction using graph neural networks

Geoenergy Science and Engineering, Journal Year: 2024, Volume and Issue: unknown, P. 213358 - 213358

Published: Sept. 1, 2024

Language: Английский

---

Source localization in subsurface aquifers based on conservation data by learning a Gaussian kernel

Computational Geosciences, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 26, 2024

Language: Английский

---