Modeling and simulation of multiphase flow in highly fractured porous media with a data-driven multiscale approach

Computational Mechanics, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 10, 2025

Abstract The pseudo-direct numerical simulation (P-DNS) method is a recently developed multiscale strategy designed for high-fidelity computational of complex flow physics. This physics-based data-driven approach involves numerically solving both the fine and global scales. former precomputed into representative volume elements (RVEs), whose homogenized responses serve to train machine learning-based surrogate models. upscaling model feeds scale, which then effectively solved in coarse meshes. In this work, P-DNS applied study multiphase highly fractured porous media. aim overcoming current limitations techniques oil reservoirs due geological heterogeneities. A novel characterization geometry fracture networks proposed. local intrinsic permeability tensor via RVE simulations accounting embedded fractures, thus allowing efficient computation reservoir-scale transport. two-dimensional single-phase two-phase problems on different reservoir scenarios. accuracy predictions assessed relative detailed with fractures very For cases considered, it shown that homogenization technique capable compute accurate rates pressure fields coarser meshes than approach, while achieving speedups solution time about factor 500.

Language: Английский

---

High resolution in non-destructive testing: A review

Journal of Applied Physics, Journal Year: 2022, Volume and Issue: 132(10)

Published: Sept. 9, 2022

Since the beginning of applications non-destructive testing/evaluation (NDT/NDE) techniques, efforts have been made consistently to improve their detection sensitivity and resolution. In present paper, factors governing lateral resolution in three major NDT viz., ultrasonic testing (UT), x-ray radiographic (XRT), eddy current (ECT) are presented. Furthermore, a review recent advances these techniques reach theoretically achievable limit or even surpassing same using alternate approaches is also discussed. For example, UT limited half wavelength by Rayleigh limit; however, subwavelength resolutions achieved through near field methods capturing evanescent field. On other hand, XRT primarily source/focal spot size, which many orders magnitude larger than wavelength. Over years, reduction focal from macro-focus micro-focus now nano-focus has led improvement few nanometers, course, combination with suitable magnification required due detectors pixel size (a μm 10 s μm). Similarly, innovations electromagnetic/magnetic sensors significantly improved ECT. Atomic force microscopy, metamaterials, artificial neural network-based employed for obtaining high-resolution NDE images. At end, authors' perspective toward possible directions

Language: Английский

---

A comparative analysis of super-resolution techniques for enhancing micro-CT images of carbonate rocks

Applied Computing and Geosciences, Journal Year: 2023, Volume and Issue: 20, P. 100143 - 100143

Published: Nov. 14, 2023

High-resolution digital rock micro-CT images captured from a wide field of view are essential for various geosystem engineering and geoscience applications. However, the resolution these is often constrained by capabilities scanners. To overcome this limitation achieve superior image quality, advanced deep learning techniques have been used. This study compares four different super-resolution techniques, including convolutional neural network (SRCNN), efficient sub-pixel networks (ESPCN), enhanced residual (EDRN), generative adversarial (SRGAN) to enhance obtained heterogeneous porous media. Our investigation employs dataset consisting 5,000 acquired highly carbonate rock. The performance each algorithm evaluated based on its accuracy reconstruct pore geometry connectivity, grain-pore edge sharpness, preservation petrophysical properties, such as porosity. findings indicate that EDRN outperforms other in terms peak signal-to-noise ratio (PSNR) structural similarity (SSIM) index, increased nearly 4 dB 17%, respectively, compared bicubic interpolation. Furthermore, SRGAN exhibits technqiues learned perceptual patch (LPIPS) index porosity error. shows 30% reduction LPIPS results provide deeper insights into practical applications domain media characterizations, facilitating selection optimal CNN-based methodologies.

Language: Английский

---

Efficiently reconstructing high-quality details of 3D digital rocks with super-resolution Transformer

Energy, Journal Year: 2024, Volume and Issue: 300, P. 131499 - 131499

Published: May 6, 2024

Language: Английский

---

A machine learning model for textured X-ray scattering and diffraction image denoising

npj Computational Materials, Journal Year: 2023, Volume and Issue: 9(1)

Published: April 10, 2023

Abstract With the advancements in instrumentations of next-generation synchrotron light sources, methodologies for small-angle X-ray scattering (SAXS)/wide-angle diffraction (WAXD) experiments have dramatically evolved. Such developed into dynamic and multiscale situ characterizations, leaving prolonged exposure time as well radiation-induced damage a serious concern. However, reduction on or dose may result noisier images with lower signal-to-noise ratio, requiring powerful denoising mechanisms physical information retrieval. Here, we tackle problem from an algorithmic perspective by proposing small yet effective machine-learning model experimental SAXS/WAXD image denoising, allowing more redundancy reduction. Compared classic models natural scenarios, our provides bespoke solution, demonstrating superior performance highly textured images. The is versatile can be applied to other imaging when data volume complexity concerned.

Language: Английский

---

Upscaling Permeability Using Multiscale X‐Ray‐CT Images With Digital Rock Modeling and Deep Learning Techniques

Water Resources Research, Journal Year: 2023, Volume and Issue: 59(3)

Published: Feb. 9, 2023

Abstract This study presents a workflow to predict the upscaled absolute permeability of rock core direct from CT images whose resolution is not sufficient allow pore‐scale computation. exploits deep learning technique with data raw rocks and their corresponding value obtained by performing flow simulation on high‐resolution images. The map much larger region in predicted trained neural network. Finally, entire calculated Darcy solver, results showed good agreement experiment data. proposed based upscaling method allows estimating large‐scale samples while preserving effects fine‐scale pore structure variations due local heterogeneity.

Language: Английский

---

On the generation of realistic synthetic petrographic datasets using a style-based GAN

Scientific Reports, Journal Year: 2022, Volume and Issue: 12(1)

Published: July 27, 2022

Deep learning architectures have transformed data analytics in geosciences, complementing traditional approaches to geological problems. Although deep applications geosciences show encouraging signs, their potential remains untapped due limited availability and the required in-depth knowledge provide a high-quality labeled dataset. We approached these issues by developing novel style-based generative adversarial network (GAN) model, PetroGAN, create first realistic synthetic petrographic datasets across different rock types. PetroGAN adopts architecture of StyleGAN2 with adaptive discriminator augmentation (ADA) allow robust replication statistical esthetical characteristics improve internal variance data. In this study, training dataset consists > 10,000 thin section images both under plane- cross-polarized lights. Here, using our proposed approach, model reached state-of-the-art Fréchet Inception Distance (FID) score 12.49 for images. further observed that FID values vary lithology type image resolution. The generated were validated through survey where participants various backgrounds level expertise geosciences. established even subject matter expert indistinguishable from real This study highlights GANs are powerful method generating Moreover, they future tool self-labeling, reducing effort producing big, geoscience datasets. Furthermore, shows can be applied other datasets, opening new research horizons application fields particularly presence

Language: Английский

---

Addressing Class Imbalance in Micro-CT Image Segmentation: A Modified U-Net Model with Pixel-Level Class Weighting

Computers & Geosciences, Journal Year: 2025, Volume and Issue: unknown, P. 105853 - 105853

Published: Jan. 1, 2025

Language: Английский

---

Digital rock super-resolution reconstruction with efficient 3D spatial-adaptive feature modulation network

Geoenergy Science and Engineering, Journal Year: 2025, Volume and Issue: unknown, P. 213748 - 213748

Published: Feb. 1, 2025

Language: Английский

---

Super-resolution enhancement of X-ray microscopic images of solder joints

NDT & E International, Journal Year: 2025, Volume and Issue: unknown, P. 103382 - 103382

Published: March 1, 2025

Language: Английский

---