Figure - available from: Acta Geotechnica
This content is subject to copyright. Terms and conditions apply.
Loss value over epochs of the training and validation sets using different loss functions: a Dice loss; b weighted cross-entropy. Note that the model is considered best and selected if there is no improvement in the validation loss over the next 25 consecutive epochs
Source publication
![U-Net introduced by Ronneberger et al. [65]. The dimensions (i.e.,...](publication/353400870/figure/fig1/AS:11431281118062728@1675654808132/U-Net-introduced-by-Ronneberger-et-al-65-The-dimensions-ie-width-x-height-of_Q320.jpg)
This paper reports the use of the deep learning-based technique to characterize the particle orientation of clay samples. The U-Net model was applied to perform semantic segmentation for identifying individual kaolinite particles, based on the scanning electron microscopic images taken from clay samples subjected to 1-D consolidation. The measurabl...
Similar publications
Foot ulcer is a common complication of diabetes mellitus; it is associated with substantial morbidity and mortality and remains a major risk factor for lower leg amputation. Extracting accurate morphological features from the foot wounds is crucial for proper treatment. Although visual and manual inspection by medical professionals is the common ap...
Citations
... With the addition of lignin fibers, it can be seen that the percentage of cavernous pore in the expansive soil decreases. The porosity of the soil can be obtained by artificial intelligence to segment SEM images [46], extracting porosity from X-ray images [47] and using ImageJ software [14]. In order to further obtain the effect of lignin fibers on the porosity of expansive soils, SEM images of expansive soils amended with different dosages of lignin Content courtesy of Springer Nature, terms of use apply. ...
With the increasing emphasis on environmental protection and sustainable economic development, recycling industrial by-products for soil improvement has received increasing attention in geotechnical engineering. Lignin fiber, a by-product of the paper industry, has the advantages of good flexibility and dispersion, etc. Given these advantages of lignin fiber, this paper uses lignin fiber to improve expansive soils’ strength and swell-shrink characteristics, then carries out a series of experiments to evaluate the effect of the improvement. Soil specimens with 0%, 1%, 2%, and 4% doped lignocellulosic fibers were prepared indoors, and these specimens were subjected to the unconfined compressive strength test, the consolidated undrained triaxial shear test, the unloaded expansion rate test, and the shrinkage test. The mechanism of lignin fiber’s action in improving expansive soil was revealed by X-ray diffraction test (XRD) and scanning electron microscope test (SEM). XRD and SEM tests have shown that lignin fibers act as a ‘‘bridge’’ lap and fiber web in the soil. The ‘‘bridge’’ lap connects the soil particles and enhances the connecting force between the soil particles. The fiber mesh gives the soil a good stress structure and limits the sliding of the soil particles to a certain extent. The value of unconfined compressive strength of expansive soils is maximum under 2% content of lignin fibers with an increase of 54%. Under 4% content of lignin fibers, the expansion soil had the least unloaded expansion rate, which was reduced by 32.8%. Combining all the test results, it was obtained that expansive soil was best modified with 2% lignin fiber content. In conclusion, using lignin fiber as an additive to modified expansive soil is viable and can lead to resource recycling.
Graphical Abstract
Adequate amount of lignin fiber, in the soil body can play a “bridge” lap and “fiber web” role. The “bridge” lap can connect the soil particles together, enhance the connection between soil particles, so that the soil body to form a stable structure. The fiber web gives the soil body a good stress structure, which can evenly spread the load to other regions of the soil body. The fiber web restricts the sliding of the soil particles to a certain extent, which can enhance the friction of the soil body. However, Excessive lignin fibers tend to collect into clusters within the soil body, separating the soil particles, which tends to form weak surfaces within the soil body, and the presence of weak surfaces affects the overall stability of the soil body.
... Indoor model tests revealed the pattern of pore water pressure dissipation in hydraulic fill soft clay, showing that the change and dissipation of pore water pressure are small and slow in the early consolidation phase, while significant dissipation mainly occurs in the middle of consolidation (Berilgen et al. 2006a, b;Sun et al. 2011;Lei et al. 2014). Some scholars have also attempted to conduct research using the discrete element method (Li et al. 2017Yuan et al. 2019;Chow et al. 2021), yet its application in soft soil environments remains limited. Table 1 shows the mathematical models for the relationship between soft soil compression and permeability proposed by researchers based on consolidation and permeability test results. ...
The high water content of soft soil leads to complex sedimentation, consolidation, and permeability characteristics, posing challenges to engineering design and construction. Although existing research has made progress in the field of consolidation and settlement characteristics, discussions on the low stress stage are still insufficient, and there is a lack of appropriate mathematical models to describe it. At the same time, the correlation and differences between the degree of consolidation used for settlement calculation and the degree of consolidation used for pore water pressure calculation have not been fully clarified. This study improves the measurement method for permeability coefficients and optimizes consolidation equipment, conducting research on the permeability and consolidation characteristics of soft soil with high water content. The research results show that soft soil with high water content exhibits significant consolidation settlement under low stress, and the incremental settlement decreases with the increase of consolidation stress. The void ratio and compression coefficient undergo drastic changes during consolidation, with a difference of 2 to 4 orders of magnitude, especially significant during the low-pressure stage. This study indicates the existence of a critical stress of about 4 kPa and proposes a segmented method to describe the consolidation and permeability characteristics of soft soil with high water content, establishing an e-lgσ-lgk relationship model, which can effectively reflect the consolidation behavior of super-high water content soft soil. At the same time, the study also finds that in practical engineering applications, the degree of consolidation used for settlement calculation and the degree of consolidation used for pore water pressure calculation should be considered comprehensively to more accurately predict the consolidation process and guide construction.
... A dispersed structure refers to a structure in which the static interaction forces between adjacent particles during sedimentation are repulsive and soil particles are arranged face-to-face (FF). Subsequent researchers conducted a series of tests on flocculated and dispersed samples and indicated that structure affects the mechanical characteristics of clay [18][19][20][21][22][23][24]. Their research results found that the deviatoric stress-axial-strain relationship of dispersed samples exhibited strain softening, whereas flocculated structure samples exhibited strain hardening. ...
This paper conducts triaxial undrained tests on flocculated and dispersed kaolin samples at strain rate range 0.005–1%/min to investigate the effects of structure and strain rate on shear strength. The test results show that the flocculated samples exhibit strain hardening behaviour, while the dispersed samples show strain softening behaviour. The strain rate sensitivity parameter reflects the degree to which shear strength increases with increasing strain rate. The structure affects the strain rate sensitivity parameter, with values of 4.79% and 2.31% for flocculated and dispersed samples, respectively. When the strain rate is 1%/min, due to the low permeability of the dispersed sample, the high strain rate causes a rapid increase in local pore pressure, while the postponed dissipation of excess pore pressure destroys the sample. When studying the influence of clay structure, it is important to use the same strain rate; otherwise, the differences in shear strength may be underestimated.
... Instead of constructing precise mathematical or logical models of the problem, it operates directly on input data to generate outcomes. Additionally, it can automatically distinguish between solid and pore architectures in CT images with varying structures to overcome the issue of unclear soilparticle boundaries (Zhu, J. et al. 2021;Han et al. 2019;Lavrukhin et al. 2023;Chow et al. 2022). ...
... Lavrukhin et al. (2021) employed a hybrid U-net + ResNet-101 model to segment soil CT images and explained the inaccuracies caused by the insufficient representativeness of some soil sample structures. Chow et al. (2022) used the U-Net model to enhance segmentation results and quantify the directional distribution of particles to address the challenge of unclear soil particle boundaries. In addition, other deep-learning models, such as SegNet and DeepLab, have been employed to train datasets for more precise blueprints. ...
Microstructure and pore characteristics of soil determine its physical and mechanical properties such as deformation, strength, and permeability. The accurate characterization of soil microstructure is a crucial prerequisite for understanding soil texture and for the effective characterization of soil properties. This study aimed to evaluate the applicability and limitations of various soil micro-test methods, compare the resolution of different micro-test techniques, and present their results. Several different techniques and methods have been used to analyze soil micropore structures. In terms of micro-visualization, scanning electron microscopy (SEM) and computed tomography (CT) are common imaging methods that can present the microstructure of the soil surface and its interior through optical means. In addition, some methods, such as soil–water retention curve (SWRC), mercury intrusion porosimetry (MIP), gas adsorption (GA), and nuclear magnetic resonance (NMR,) indirectly assess the size-related information of soil pores through the pore characteristics of porous media. The targeted joint application may be selected according to varying objectives—MIP is used to obtain the main structure when studying the overall internal pores, supplemented by CT for three-dimensional remodeling; NMR is used when studying local pore damage to reflect the evolution of pore characteristics related to water storage, supplemented by SEM to support observations of surface or morphological structure damage. Finally, the direction for future development is to process the test results and transform the existing technical equipment.
... The parameters used, such as shape parameter and tortuosity, are determined by the back-calculations of laboratory scale tests. An alternative way to determine these parameters can be microscopic techniques such as AI-based SEM (Chow et al., 2022) and CT scan analysis (Li et al, 2024). More information about back-analysis for the calculation of thermal and hydraulic Water retention curve for all EBS components, fitted by reference porosities (A). ...
This paper presents modelling of the long-term performance of engineered barrier systems (EBS) in crystalline host rock in terms of coupled thermo-hydro-mechanical (THM) processes in a specific case, considering also the impact of salinity linked with geochemistry. This study has been used as a supporting document for the safety case in the operating licence application for the Olkiluoto spent nuclear fuel repository in Finland. The disposal design chosen is the KBS-3V (Kärnbränslesäkerhet in Swedish, “nuclear fuel safety”; 3, version number; V, vertical) which consists of placing the canisters in vertical deposition holes surrounded by the EBS. The buffer components consist of compacted blocks of Wyoming-type bentonite surrounded by pillow pellets manufactured with the same material. Bulgarian and/or Italian granular filling (GraFi) materials are the backfilling material in the deposition tunnels. The Barcelona basic model (BBM) was considered for modelling the geomechanical behaviour of compacted buffer blocks and GraFi materials filling the deposition tunnels. The Barcelona expansive model (BExM), which consists of a double structure (macro–micro porosity), was considered for the pellets. A laboratory testing campaign (thermal conductivity, water retention curve, oedometer, and infiltration tests) was carried out in order to calibrate the THM model parameters of the corresponding materials. Model-data uncertainties, challenges in 3D THM modelling, and the methodology followed have been provided in terms of modelling capabilities. We implemented 3D THM simulation of an individual deposition hole (canister, notch/chamfer, and buffer materials) drilled into a deposition tunnel (backfill material) in CODE_BRIGHT as a finite element method (FEM) program. This study presents results related to THM performance of the EBS, such as peak temperature, time required to reach full saturation in buffer and backfill, the evolution of dry densities according to permeabilities, the development of swelling pressure in buffer and backfill, and, consequently, deformations in buffer and backfill domains. A sensitivity analysis plan was followed in order to deal with various factors affecting the long-term THM performance of the EBS. In the sensitivity analysis, buffer and backfill design options (different filling material alternatives), geological conditions (saline water, rock permeability, and heterogeneous rock) and numerical simulation options (different numerical model options, issues related to geometry and meshing) were investigated. The performance targets and design specifications set for the buffer and backfill are also discussed. The paper concludes with a summary how the THM design under a certain configuration (geometry, initial conditions, boundary conditions, and buffer and backfill materials) meets the performance targets set for the buffer, backfill, and host rock.
... Machine learning techniques have been widely applied in geotechnical engineering, including the prediction of slope stability, analysis of failure behavior like rock mass squeezing, assessment of fracture toughness, and the modeling of soil p-q curves. Image-based methods have been also employed for tasks such as the segmentation of clay particles and cracks within expansive soils (Chow et al. 2022;Guan and Yang 2023;Guardiani et al. 2022;Hsiao et al. 2022;Hu et al. 2023;Soranzo et al. 2023;Wang et al. 2021;Zhou et al. 2022). Fracture detection and segmentation have been conducted in pavement and structural health monitoring based on photography (Abdel-Qader et al. 2003;Li et al. 2011), as well as in the investigation of concrete autogenous shrinkage and characterization of cement microfractures based on CT images (Mac et al. 2021a, b). ...
Quantitatively assessing the morphology of fractures by hydraulic stimulation is crucial for evaluating the effectiveness of fracturing operations. Herein, a series of laboratory tests are conducted to induce fractures originating from a borehole situated within cylindrical-shaped granite specimens. A total of 13 specimens are pressurized with various viscosities of fracturing fluids and injection rates. For each fractured specimen, 3D X-ray computed tomography imaging is performed. To accurately identify and segment the various morphologies of fractures, a convolutional neural network (CNN) model is employed. A nested U-Net model, based on an encoder–decoder structure, is designed to obtain a more sophisticated segmentation of fractures compared with conventional image processing methods. The results indicate that the complicated fracture patterns, such as the hair-like thin fracture, fracture junction, and complex fracture network, are successfully segmented. For the 3D reconstructed binary fractures, the morphological parameters are quantitatively computed. The increase in fracturing fluid viscosity and injection rates, which in turn led to an increase in breakdown pressure, results in the increase in fracture volume and aperture, whereas the tortuosity and surface roughness tend to decrease. Thus, the CNN-based model enables the accurate segmentation of fractures induced by hydraulic stimulation and helps clarify the relationship between morphological consequences and pressurization conditions.
... Recent advances in data science, such as machine learning (ML), and in particular, deep learning (DL) models like ANNs, support vector machines (SVMs), Gaussian process regression (GPR), recurrent neural networks (RNNs), and long-short term memory (LSTM) neural networks, have assisted researchers and engineers in utilizing data from measurements in various geotechnical applications (He et al. 2020;Kang et al. 2021;Liu et al. 2015;Makasis, Narsilio, and Bidarmaghz 2018;Pooya Nejad and Jaksa 2017;Tophel et al. 2022;Zhang, Yin, and Jin 2021). To the author's knowledge, no study has been conducted using ML to model the long-term behaviour of a single grain. ...
In geotechnical engineering, the time-dependent behaviour or ageing behaviour is vital for applications such as earthwork compaction and liquefaction potential assessment. This study introduces a novel test apparatus to understand micromechanical factors and deformations at grain contacts. Using a non-contact Digital Image Correlation (DIC) technique, deformations were measured with a 10 μϵ spatial resolution. This enabled quantification of grain creep and contact maturing deformations, surpassing previous experimental methods. To model this complex behaviour, Machine Learning (ML) models, including an artificial neural network (ANN) and long-short term memory neural network (LSTM), were used, achieving a 1-2% error rate with experimental results. The integration of ML offers a promising tool for predicting long-term grain strains, enhancing the assessment of structures' serviceability with the studied materials.
... Various ML models, including Support Vector Regressions (SVRs), GBR, and ANN have been widely used in civil and geotechnical engineering applications [66][67][68][69][70][71][72]. This section describes the three different ML models employed in this paper, viz. ...
Stress wave velocities, i.e. shear and compression wave velocities and hence small strain stiffness, are important parameters required for dynamic analysis and designing various geotechnical structures. Measuring wave velocities has been challenging, let alone developing a model for their prediction. There exist few models in the literature; however, their use is limited and can be used mostly in specific cases, for example, isotropic conditions. This paper uses the dataset developed in the author's previous study and aims to develop a model where the wave velocities can be predicted accurately. The dataset contains wave velocities evolution of three different materials: Toyoura sand, river sand and silica sand at three different initial relative densities sheared by increasing the major principal stress. The three different materials allowed the study of different particle characteristics, including sphericity, convexity, aspect ratio and roughness, in predicting and modelling the wave velocities. This paper utilizes three different machine learning (ML) models viz. artificial neural network (ANN), support vector regression (SVR) and gradient boosting regression (GBR) to model the observed behaviour of wave velocities evolution. All three models were equally efficient in modelling the evolution of both shear (Vs) and compression wave (Vp) velocities. The predicted Vs and Vp using the three different machine learning models are also compared with a literature model available for isotropic stress conditions and showed that the ML models developed in this study are superior by incorporating the effect of grain characteristics which is missing from the literature model. The relationship between the prediction of the wave velocities and input parameters is highlighted using feature importance and Pearson's coefficient. At the end of this study, ANN is used to provide a set of nonlinear equations in the matrix form for Vs and Vp, which can be used by anyone and does not require any prior knowledge of ML or coding.
... Recent developments in data science, such as ML and especially deep learning (DL) models including ANNs, support vector machines (SVMs), and Gaussian process regression (GPR), have helped research scientists and engineers use the data available from measurements in various geotechnical applications [16,19,32,35,44,73,75]. For instance, data-driven models for capturing the complex behaviour of soil compaction in estimating the material properties for quality assurance (QA) and quality control (QC) purposes have recently been considered and integrated with IC [17]. ...
This paper introduces a Theory-Guided Machine Learning (TGML) framework, which combines a theoretical model (TM) and a machine learning (ML) algorithm to predict compaction density under cyclic loading. Several 1-D tests were conducted on uniformly-graded fine sand compacted at varying moisture contents (w), stress levels (σ_v) and loading frequencies (f), simulating the field compaction of materials using a vibratory roller. The laboratory compaction data were first analysed using a revised TM and an artificial neural network (ANN) and their performance was measured using mean absolute error (MAE). Next, the data were analysed using the TGML framework, which involves three different techniques. TGML1 increased the ML’s ability to extrapolate (MAE improved from 2.2×10-3 to 1.2×10-3); TGML2 ensured ML and TM complemented each other to model observations better (MAE improved from 2.3×10-3 to 7.9×10-4); and TGML3 assisted in regularizing the ML with an additional loss function which ensured the model followed the mechanistic understandings of the underlying physics (MAE improved from 9.2×10-3 to 2.7×10-3). Considering TGML3 during modelling is essential when dealing with noisy field datasets, and this is the highlight of this paper. TGML frameworks showed less error and lower model uncertainty, estimated using the novel Monte Carlo (MC) dropout technique. Furthermore, the developed TGML framework was used to demonstrate a termination criterion, i.e., the number of cycles of roller movement required to achieve the desired degree of compaction. Finally, an approach is proposed by which a simplified TM and ML model can estimate field compaction behaviour during roller movement.
