Carbonaceous mudstone coarse-grained soil is commonly used as embankment fill in southwestern China. It exhibits significant creep characteristics, and its long-term deformation is closely related to particle breakage. In this paper, large-scale confined creep tests, sieving tests, and single-particle compression tests were carried out, combined with a discrete element model considering stress corrosion effects, to investigate the creep characteristics, particle strength evolution, and mesoscopic breakage mechanisms of carbonaceous mudstone subgrade fill under different gradations. The results show that carbonaceous mudstone exhibits strain-hardening behavior. At low stress levels, the total strain is positively correlated with the gradation fractal dimension, while the opposite is true at high stress levels. The creep critical stress increases with increasing fractal dimension, whereas the creep rate shows the opposite trend. The single-particle crushing strength follows a Weibull distribution, and the characteristic crushing strength decreases with increasing particle size and applied stress, with particle size having a more significant effect. A time factor was introduced into the traditional stress corrosion model to establish a bond strength degradation model that accounts for the effect of loading time. The improved model can more accurately describe the progressive creep failure under high-stress and long-term conditions. Mesoscopic analysis indicates that during creep, the direction of strong contact loads gradually deflects from vertical to more dispersed orientations. Strong contact failure is concentrated on the particle surfaces. For specimens with a smaller fractal dimension, the force chain pattern is a "few but strong" sparse mode, which is sensitive to applied stress and loading time. For specimens with a larger fractal dimension, the force chain pattern transforms into a "many but weak" dense network mode. The research findings can provide a theoretical basis for the long-term stability evaluation of pre-disintegrated carbonaceous mudstone embankments.
Abstract Accurate forecasting of caisson tilt during construction is important for ensuring structural stability and effective construction management. However, reliable prediction remains challenging when only limited early-stage observations are available. In this study, the caisson tilt forecasting problem was formulated as an early-stage limited observation forecasting task using time-series data generated from numerical simulations. A staged construction model was developed using PLAXIS2D, and construction-related uncertainties in deep cement mixing (DCM)-improved ground, riprap, and heaving soil were incorporated through Latin Hypercube Sampling, resulting in a total of 1,000 analysis cases. Stage-wise tilt time series over 28 construction stages were extracted. A masking strategy, in which only the initial observations were provided as input was applied to compare the forecasting performance of a statistical model (ARIMA) and deep learning models (LSTM, TCN, and Transformer). Model performance was evaluated over the forecast horizon using MAE, RMSE, and $${R}^{2}$$ . The deep learning models consistently outperformed ARIMA under all conditions, and the TCN model showed stable forecasting performance even under highly limited early-stage observations. The LSTM model demonstrated improved accuracy as more observations became available.
Landslides pose a recurring threat to human settlements, infrastructure, and ecosystems in the seismically active Chamoli district of Uttarakhand, India. This study presents a comparative landslide susceptibility assessment using three distinct models: Frequency Ratio (FR), Shannon Entropy (SE), and Analytical Hierarchy Process (AHP). Twenty-four geo-environmental and anthropogenic conditioning factors were integrated to develop landslide susceptibility maps (LSMs) tailored to the region’s complex terrain. Multicollinearity analysis was conducted to ensure statistical robustness, and model performance was validated using the Receiver Operating Characteristic (ROC) curve and the Area Under the Curve (AUC) metric. The FR model achieved the highest predictive accuracy (AUC = 0.819), followed by AHP (0.789) and SE (0.594). While FR demonstrated superior data-driven reliability, AHP offered interpretability grounded in expert judgment. Thematic analysis revealed slope, geology, rainfall, proximity to roads, and land use changes as key landslide triggers. Susceptibility zonation showed significant spatial variability across models, with high-risk zones concentrated near road corridors, riverbanks, and deforested slopes. The study underscores the value of methodological triangulation in landslide prediction and recommends FR as a reliable framework for future hazard planning. These findings provide actionable insights for disaster mitigation, infrastructure planning, and sustainable land use management in Himalayan regions.
Abstract In geotechnical engineering, the angle of repose (AoR) is widely adopted as an indicator of soil stability and particle interaction. However, AoR can arise from different combinations of inter-particle sliding and rolling friction, each representing different mechanical characteristics. This study systematically investigates the influence of sliding and rolling friction coefficients on AoR using the Discrete Element Method (DEM). Lifting cylinder tests were simulated while varying both sliding and rolling coefficients from 0.1 to 0.9 to examine their combined effects on AoR formation. Additionally, a stress wave propagation model was developed to simulate laboratory stress wave testing and to explore the relationship between inter-particle friction and wave response characteristics. The results reveal that while sliding and rolling frictions significantly influence stress wave responses, such as stress magnitude, particle velocity, and displacement, the AoR alone does not exhibit a consistent correlation with these dynamic properties. These findings suggest that AoR, although it is useful for static stability, may not be a reliable standalone indicator of dynamic soil behavior.
Abstract The Representative Elementary Volume (REV) is a key concept in the analysis and simulation of jointed rock masses, playing a crucial role in selecting the dimensions of laboratory samples and the domain size of numerical models. However, a precise understanding of the influence of joint geometric and mechanical parameters on the REV remains limited. In this study, the scale dependency of the mechanical properties of jointed rock masses was systematically investigated using three-dimensional numerical modeling. The REV was defined based on the convergence of peak strength and deformation modulus within a 5% threshold. The influence of joint geometric and mechanical parameters on the REV, as well as on the mechanical response at the REV, was analyzed. The results showed that the REV is highly dependent on the loading orientation relative to joint planes, with the smallest values observed when loading is approximately parallel to the joints, and the largest values occurring when the loading direction is favorably oriented for joint sliding and failure. The introduction of a second joint set significantly increased the REV, particularly in modulus-based evaluations. Increasing joint number and spacing enlarged the REV and intensified the reductions in strength and modulus in loading directions that promote joint activation, whereas variations in mechanical properties had a limited influence. It was also found that the deformation modulus is more sensitive to scale than peak strength, requiring larger model sizes to achieve convergence. These findings highlight the importance of joint network geometry in REV estimation and provide practical guidance for selecting representative sample sizes and numerical model dimensions in tunneling, slope stability, and underground excavation projects.
