Using deep learning to capture gravel soil microstructure and hydraulic characteristics
Abstract The special hydraulic properties of gravel soil, attributed to its varying fine particle content, can be effectively analyzed using the Wasserstein Generative Adversarial Networks (WGANs) technique. This approach enables the reconstruction of 3D digital samples of gravel soil, allowing for...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-07-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-04879-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849402567421853696 |
|---|---|
| author | Bin Zhu Yu-Fei Xie Xiang-Gang Hu Dai-Rong Su |
| author_facet | Bin Zhu Yu-Fei Xie Xiang-Gang Hu Dai-Rong Su |
| author_sort | Bin Zhu |
| collection | DOAJ |
| description | Abstract The special hydraulic properties of gravel soil, attributed to its varying fine particle content, can be effectively analyzed using the Wasserstein Generative Adversarial Networks (WGANs) technique. This approach enables the reconstruction of 3D digital samples of gravel soil, allowing for the generation of specific microstructure realizations, including complex pore characteristics. This capability is crucial for gaining insights into the hydraulic behavior of gravel soil. In a specific case involving gravel soil samples from Guilin city, China, three samples with similar structural features were carefully selected for analysis. These samples were scanned using µ-CT to create the training dataset for the reconstruction model. The WGAN with Gradient Penalty technique was then applied to simulate the reconstruction of the digital gravel soil samples. The results demonstrated a high consistency between the reconstructed model of gravel soil realizations and the original samples in terms of porosity, two-point correlation function, linear path function, specific surface, and Euler characteristics number. Furthermore, through the evaluation of permeability, it was shown that the reconstructed realizations effectively captured and represented the actual soil prototype. This allowed for the analysis of seepage characteristics and internal stability within a range of magnitudes up to 10− 2 cm/s. Compared with the machine learning model generated in previous literature, the machine learning model recommended in this paper can capture the hydraulic properties of gravel soil with obvious difference in coarse and fine particle size, which is reflected in the difference of permeability of the two orders of magnitude. |
| format | Article |
| id | doaj-art-9db3653c2e79471e8c798a0adc44c7f6 |
| institution | Kabale University |
| issn | 2045-2322 |
| language | English |
| publishDate | 2025-07-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Reports |
| spelling | doaj-art-9db3653c2e79471e8c798a0adc44c7f62025-08-20T03:37:30ZengNature PortfolioScientific Reports2045-23222025-07-0115111910.1038/s41598-025-04879-4Using deep learning to capture gravel soil microstructure and hydraulic characteristicsBin Zhu0Yu-Fei Xie1Xiang-Gang Hu2Dai-Rong Su3Earth Sciences College, Guilin University of TechnologyEarth Sciences College, Guilin University of TechnologyEarth Sciences College, Guilin University of TechnologyEarth Sciences College, Guilin University of TechnologyAbstract The special hydraulic properties of gravel soil, attributed to its varying fine particle content, can be effectively analyzed using the Wasserstein Generative Adversarial Networks (WGANs) technique. This approach enables the reconstruction of 3D digital samples of gravel soil, allowing for the generation of specific microstructure realizations, including complex pore characteristics. This capability is crucial for gaining insights into the hydraulic behavior of gravel soil. In a specific case involving gravel soil samples from Guilin city, China, three samples with similar structural features were carefully selected for analysis. These samples were scanned using µ-CT to create the training dataset for the reconstruction model. The WGAN with Gradient Penalty technique was then applied to simulate the reconstruction of the digital gravel soil samples. The results demonstrated a high consistency between the reconstructed model of gravel soil realizations and the original samples in terms of porosity, two-point correlation function, linear path function, specific surface, and Euler characteristics number. Furthermore, through the evaluation of permeability, it was shown that the reconstructed realizations effectively captured and represented the actual soil prototype. This allowed for the analysis of seepage characteristics and internal stability within a range of magnitudes up to 10− 2 cm/s. Compared with the machine learning model generated in previous literature, the machine learning model recommended in this paper can capture the hydraulic properties of gravel soil with obvious difference in coarse and fine particle size, which is reflected in the difference of permeability of the two orders of magnitude.https://doi.org/10.1038/s41598-025-04879-4WGANs (Wasserstein generative adversarial networks)Machine learningPorous structurePore-scale CFD simulationHydraulic properties imitation |
| spellingShingle | Bin Zhu Yu-Fei Xie Xiang-Gang Hu Dai-Rong Su Using deep learning to capture gravel soil microstructure and hydraulic characteristics Scientific Reports WGANs (Wasserstein generative adversarial networks) Machine learning Porous structure Pore-scale CFD simulation Hydraulic properties imitation |
| title | Using deep learning to capture gravel soil microstructure and hydraulic characteristics |
| title_full | Using deep learning to capture gravel soil microstructure and hydraulic characteristics |
| title_fullStr | Using deep learning to capture gravel soil microstructure and hydraulic characteristics |
| title_full_unstemmed | Using deep learning to capture gravel soil microstructure and hydraulic characteristics |
| title_short | Using deep learning to capture gravel soil microstructure and hydraulic characteristics |
| title_sort | using deep learning to capture gravel soil microstructure and hydraulic characteristics |
| topic | WGANs (Wasserstein generative adversarial networks) Machine learning Porous structure Pore-scale CFD simulation Hydraulic properties imitation |
| url | https://doi.org/10.1038/s41598-025-04879-4 |
| work_keys_str_mv | AT binzhu usingdeeplearningtocapturegravelsoilmicrostructureandhydrauliccharacteristics AT yufeixie usingdeeplearningtocapturegravelsoilmicrostructureandhydrauliccharacteristics AT xiangganghu usingdeeplearningtocapturegravelsoilmicrostructureandhydrauliccharacteristics AT dairongsu usingdeeplearningtocapturegravelsoilmicrostructureandhydrauliccharacteristics |