This study systematically investigates the compressibility of dredged clay with high water content using Portland cement (PC) and sodium silicate (SS) as composite stabilizers. The effects of SS content (0%, 3%, 6%, 12%) and curing time (3 hours, 7 days, 28 days) on compression behavior of PC-SS-stabilized specimens with the same PC content of 10% were investigated via one-dimensional consolidation tests, with microstructural evolution analyzed using SEM. Results show that PC-SS-stabilized soil exhibits significantly reduced compressibility compared to untreated soil. The e-lgp curve became bilinear, with a yield stress (p_y) as the turning point, compressibility was low before p_y and increased markedly after. Compression indices (C_e, C_c) of stabilized soil were lower than those of untreated soil, initially decreasing and then increasing with SS content, reaching minima at 6%. The swelling index (C_s) was also the smallest at this content. Yield stress increased rapidly, then slightly decreased with SS addition, peaking at 6%. Longer curing time increased yield stress, most notably within 7 days, reduced porosity, and decreased compressibility. After 28 days, compression curves shifted upward. Microstructural analysis indicated that at 6% SS and 10% cement, continuous cementitious products formed, filling pores and cementing particles, thereby enhancing compressive performance.

The studied earth dam was remediated works for seepage and instability problems in the downstream area. The effectiveness of slurry cutoff walls was investigated on seepage behavior. The three-dimensional (3-D) finite element (FEM) analysis at normal high-water level (NHWL) for steady-state flow was established without and with a cutoff wall. Based on the results, the overall effectiveness of the cutoff wall was 85%, which exceeded the predicted design efficiency criterion of 75%. The embankment, residual soil, and phyllite were the three major materials that mainly intercepted by cutoff wall, with decreasing flows of 72-89%, 85-94%, and 55-94% respectively. The phyllite and quarzitic schist below cutoff wall coverage, the flows increased by 179% and 33-61%. Thus, extension of the cutoff lower tip to cover the full depth through phyllite and into the quarzitic schist is highly recommended. The cutoff wall effectiveness reduced potential wet areas on downstream area by 97%, indicating lower risk of seepage erosion. Major reductions in the phreatic line drop were observed of 12-20 m, indicated the hydraulic gradient across the wall thickness of 2.00-3.33 which are less than 4.00 against wall blowout. The piezometric heads of the existing field monitoring are higher than the model without the cutoff wall indicated the possibility of internal erosion. However, this would be clearly solved after cutoff wall construction. The estimated newly installed piezometer heads across the cutoff wall are planned. The model estimates the heads from dam axis to downstream area would lower from 160 to 140-145 m Mean Sea Level (MSL). That would certainly increase the safety on seepage and stability of the downstream slope.

Stone columns are widely used to improve soft ground, and their mechanical behavior is commonly represented by homogenized composite assumptions at the unit-cell level. However, such assumptions are often adopted without sufficient large-scale experimental validation and may not adequately capture the nonuniform load transfer, confinement-dependent friction mobilization, and drainage-sensitive nonlinear response caused by the interaction between a granular column and the surrounding soft clay. To address this issue, a series of large-scale triaxial tests was conducted on stone column-reinforced composite unit cells under static loading. The effects of area replacement ratio, confining pressure, and drainage condition on the stress-strain behavior, peak deviatoric stress, and equivalent shear strength parameters were systematically examined. Based on the test results, a modified Duncan-Chang hyperbolic model was developed, in which the area replacement ratio was explicitly incorporated into the constitutive parameters through experimental calibration. The results show that both the soft soil and the composite unit exhibit strain-hardening behavior, while the reinforced composite develops higher stiffness and strength. The peak deviatoric stress increases approximately linearly with confining pressure and replacement ratio. In addition, the equivalent cohesion changes only slightly after reinforcement, whereas the equivalent friction angle increases significantly, indicating that the strengthening effect is governed mainly by frictional enhancement. The modified model reproduces the measured stress-strain curves with good accuracy, with coefficients of determination generally above 0.90 and relative errors in peak deviatoric stress within about 10% for the tested cases. The proposed framework provides a practical constitutive description for stone column composite unit cells and a useful basis for geotechnical analysis and design.

This study explores the effectiveness of machine learning approaches for estimating mode-I fracture toughness (KIC) of rocks, a crucial factor in both geotechnical engineering and materials science. Six sophisticated machine learning algorithms were constructed and thoroughly assessed using a compilation of 500 experimentally acquired samples. Vital input variables such as grain size, porosity, density, water saturation, and rock classification were gathered from varied geological environments to encompass inherent diversity. The outcomes demonstrated that the Random Forest Regressor (RFR) delivered the highest average forecasting accuracy among the evaluated models. Nonetheless, the Friedman-Nemenyi statistical test showed that models like SVR and ANN exhibited similar performance levels, with no statistically significant differences within the critical difference interval. SHAP analysis additionally improved model transparency by pinpointing porosity and water saturation as the primary factors influencing fracture toughness estimates. The results also underscore the existence of complex nonlinear relationships among input features, demonstrating the capability of machine learning models to capture meaningful physical relationships in heterogeneous rock systems. By merging robust experimental datasets with cutting-edge modeling and interpretability methods, this research advances data-driven techniques in rock mechanics and offers a dependable framework for anticipating fracture behavior in intricate geological materials.

A series of model tests were conducted to investigate how the bearing capacity of adjacent footings on sand varies with interference effects. The spacing ratio between adjacent footings, the embedment depth of the footings, and the relative density of the soil were selected as key parameters, and the interference effects were quantitatively analyzed using an efficiency factor defined as the ratio of the bearing capacity of adjacent footings to that of a single footing. The model ground was prepared using Han River sand with relative densities of 40% and 80%, and the tests were conducted by varying the spacing ratio between adjacent footings from 0 to 3. The results of the model tests revealed that the bearing capacity of adjacent footings on sand was greater than that of a single footing because of interference effects. For all the test results, both the bearing capacity and the efficiency factor reached their maximum values at a spacing ratio of approximately 0.5 and then gradually decreased with increasing spacing ratio. In addition, the bearing capacity of adjacent footings generally increased with increasing footing embedment depth and soil relative density. The trend of the efficiency factor with respect to the spacing ratio between adjacent footings in this study showed overall agreement with previously reported results. The findings of this study are expected to provide fundamental data for establishing rational footing layouts that consider the effects of interference in the design of shallow foundations on sand.

This study proposes a multi-resolution Convolutional Long Short-Term Memory (ConvLSTM) ensemble framework that leverages diverse temporal input resolutions to mitigate error accumulation and improve long-horizon forecasting of retaining-structure behavior during staged excavation. An extensive database of lateral wall displacement responses was generated through PLAXIS2D simulations incorporating five-layered soil stratigraphy, two excavation depths (14 and 20 m), and stochastically varied geotechnical and structural parameters, yielding 2,000 time-series deflection profiles. Three ConvLSTM models trained at different input resolutions were integrated using a fully connected neural network meta-learner to construct the ensemble model. Validation using both numerical results and field measurements demonstrated that the ensemble approach consistently outperformed the standalone ConvLSTM models, particularly in long-term multi-step prediction, exhibiting reduced error propagation and improved generalization. These findings underscore the potential of multi-resolution ensemble strategies that jointly exploit diverse temporal input scales to enhance predictive stability and accuracy in AI-driven geotechnical forecasting.