Underground structures are typically subjected to multidirectional seismic excitation during earthquakes. To investigate the dynamic response of tunnels under such conditions, large-scale shaking-table model tests were conducted. The dynamic responses of the surrounding rock and key locations on the tunnel cross-section were analyzed, and the evolution of the response along the tunnel axis was examined. The results indicate that, regardless of the loading direction, the acceleration response generally increased from unidirectional to bidirectional and then to three-directional loading. Significant differences were observed in the acceleration responses at different locations on the tunnel cross-section. In particular, horizontal seismic excitation perpendicular to the tunnel axis substantially amplifies the acceleration response along the approximately 45°diagonal direction of the cross-section. At the tunnel invert, both the peak acceleration and the energy fractions in the 0–15.625 and 15.625–31.250 Hz frequency bands exhibit pronounced changes at a portal distance of approximately 1.67 tunnel spans. This finding indicates that, for the unsupported tunnel with no excavated upslope at the portal, a distinct seismic amplification zone extends approximately 1.67 tunnel spans from the portal. These findings provide insights for evaluating the seismic response of tunnels and optimizing their seismic design.

The variations in shear force induced by the orebody dip angle significantly affect the bearing capacity of pillars. To investigate the bearing characteristics and failure modes of pillars under different dip angles, a series of “roof-inclined pillar-floor” loading tests were designed and conducted. Acoustic emission (AE) monitoring was employed to capture real-time crack evolution signals, while an adaptive denoising algorithm, based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and permutation entropy (PE), was applied to enhance the signal-to-noise ratio and reliability of the AE signals. By combining AE response characteristics with parameter sensitivity analysis, the mechanism by which the dip angle affects the bearing characteristics and failure modes of the specimens was investigated. The results indicate that the dip angle significantly reduces the bearing capacity of the pillars. As the dip angle increased from 0° to 15°, the average bearing capacities of the three pillar aspect ratios (0.8, 1.0, and 1.5) decreased by 4.28%, 6.49%, and 14.84%, respectively. Furthermore, the higher the aspect ratio, the greater the sensitivity of the bearing capacity to the dip angle. The CEEMDAN-PE adaptive denoising algorithm improved the signal-to-noise ratio by 7.86 dB compared with conventional wavelet denoising, effectively preserving microcrack signals. AF (average frequency)-RA (rise angle) parameter analysis revealed that the proportion of shear failure increased by 4.4%, 3.8%, and 10.4% for the three aspect ratios compared to the 0° dip angle. Both the cumulative ringdown count and cumulative energy increased significantly, with average increases of 145.9% and 90.06%, respectively. These findings reveal that as the dip angle increases, the damage severity of the “roof-inclined pillar-floor” system intensifies, with the failure mode evolving from tension-dominated to combined tensile–shear failure. At 15°, shear failure along a single inclined plane became dominant. Moreover, as the aspect ratio increases, the failure mode becomes increasingly sensitive to the dip angle. These results provide an experimental foundation for the stability assessment of inclined pillars in hard-rock.

To improve the accuracy of lithology identification and uniaxial compressive strength (UCS) prediction based on drilling vibration signals and to elucidate the effects of multidirectional vibration signals and lithological differences, this study investigates coal, sandstone, granite, and marble. Triaxial vibration and acoustic pressure signals generated during core drilling were acquired using a core drilling machine and a dynamic signal acquisition system, from which time- and frequency-domain features were extracted. A backpropagation (BP) neural network was then developed for lithology identification and UCS prediction. The results show that the combined X-, Y-, and Z-axis vibration and acoustic pressure signals provide the best overall predictive performance. Specifically, the maximum accuracy of lithology identification on the test set reaches 93%, while UCS prediction yields a coefficient of determination of 0.76, a root mean square error of 7.61 MPa, and a mean absolute percentage error of 15.56%. Vibration signals perpendicular to the drilling direction provide significantly higher prediction accuracy than those parallel to the drilling direction, while multidirectional signals further improve predictive performance through information complementarity. Among the individual lithologies, granite exhibits the highest UCS prediction accuracy. In addition, the prediction accuracy for soft-hard rock combinations, such as coal-granite and coal-marble, is significantly improved. These findings reveal the relationship between multidirectional signals and lithological combinations, thereby providing technical support for the engineering application of measurement-while-drilling technology.

To consistently identify disturbance-sensitive zones in the surrounding rock of deep roadways under varying confining pressures and disturbance conditions, triaxial rheological disturbance tests were conducted on red sandstone specimens using the RRTS-IV system, complemented by nuclear magnetic resonance (NMR) measurements. A method for delineating disturbance-sensitive zones based on a modified failure proximity index F* was proposed by jointly considering strength attenuation, disturbance strain evolution, and pore-structure changes. The index F* incorporates both the confinement-induced strengthening effect and the disturbance-induced strength attenuation within a unified framework, thereby enabling the proximity to failure under different conditions to be compared on a common dimensionless scale. The results show that: (1) As F* increases, the disturbance strain undergoes a three-stage evolution characterized by an initial decrease, a gradual increase, and a subsequent abrupt increase. These stages correspond to the disturbance-insensitive zone (F*＜0.70), disturbance-sensitive zone (0.70 ≤ F*＜0.80), and disturbance-sensitive failure zone (F*≥0.80), respectively. (2) The post-disturbance failure stress decreases exponentially with the number of disturbances. The fitted disturbance-sensitivity coefficient ? decreases with increasing confining pressure, indicating that confining pressure suppresses disturbance-induced damage. (3) Under different confining pressures and disturbance levels, the minimum and abrupt-change points of the disturbance strain remain stable in F* space and are concentrated near F* ≈ 0.70 and F* ≈ 0.80, respectively. (4) Porosity measurements and NMR images show that fine pores dominate when F*＜0.70; medium-sized pores gradually develop when 0.70≤F*＜0.80; and relatively large pores and interconnected fractures develop rapidly when F*≥0.80, accompanied by an abrupt increase in porosity. This microstructural evolution is consistent with the three-stage evolution of the disturbance strain. The proposed F*-based method integrates the effects of confining pressure, disturbance accumulation, and proximity to failure into a unified dimensionless measure, providing a basis for disturbance-risk assessment and the selection of zone-specific support parameters in deep roadways.

With the accelerated development of urban underground space, the construction of multi-tunnel clusters has become increasingly common. The cumulative ground disturbance induced by sequential excavation poses a critical safety challenge. However, traditional analytical solutions often fail to adequately characterize the time-dependent stress redistribution in multiply connected domains, resulting in insufficiently accurate predictions of the complex group-hole effect. To address this issue, this study proposes a novel analytical method for evaluating the stress and displacement evolution of shallow-buried, triple-parallel tunnels while explicitly accounting for construction sequencing. By integrating the Schwarz alternating iterative method with complex-variable theory, the proposed approach iteratively transfers excavation-induced additional stresses between tunnel boundaries. This strategy transforms the complex triply connected domain into a sequence of solvable singly connected-domain problems, enabling an accurate description of tunnel–tunnel interaction mechanisms. Validation results indicate that the proposed model predicts surface settlement with a relative error of less than 10%. Moreover, its computational efficiency is approximately three times that of the three-dimensional finite element method (FEM), demonstrating its considerable potential for rapid engineering pre-evaluation. The results further reveal several critical relationships involving construction sequencing and spatial parameters: (1) Excavation sequence: Excavating the middle tunnel last effectively mitigates the superposition effect and reduces the maximum surface settlement by approximately 15%; (2) Spacing threshold: Clear tunnel spacing exhibits a nonlinear negative correlation with surface settlement. Specifically, the superposition effect increases by more than 40% when the spacing is less than 2.5 times the tunnel diameter (2.5D); (3) Depth effect: Although increasing the depth of the middle tunnel can significantly reduce the maximum surface settlement by 20%–30%, it also increases the depth of plastic-zone development in the surrounding rock. This study overcomes the limitations of traditional empirical formulas in characterizing complex spatial effects and provides robust theoretical support for optimizing tunnel spacing and selecting construction methods in densely developed urban tunnel clusters.

To clarify the factors controlling coal failure modes in the end-face region of longwall panels with large mining heights, this study investigates the failure mechanisms of burst-prone coal specimens with different height-to-diameter ratios (?l = 1, 2, 3). Uniaxial compression tests were combined with in situ X-ray computed tomography (CT) scanning to capture the three-dimensional evolution of fractures during loading. Image processing and three-dimensional reconstruction were used to quantitatively analyze fracture propagation, damage evolution, and the effects of mineral distribution. The results show that specimens with ? l= 1 and 3 exhibit greater dispersion in their stress–strain responses, compressive strengths, and elastic energy indices than those with ?l = 2. The uniaxial compressive strength decreases with increasing ?, and all specimens are classified as highly burst-prone according to their elastic energy indices. The failure modes are strongly dependent on l. Specimens with ? l= 1 exhibit peripheral splitting failure caused by strong end restraint, with limited influence from pre-existing internal damage. Specimens with l = 2 exhibit shear failure, with fracture angles consistent with the Mohr-Coulomb criterion. Specimens with l = 3 exhibit a combined shear–splitting failure mode controlled by the initial damage zone as a result of weakened end restraint, leading to the lowest strength. The influence of mineral distribution on fracture propagation increases with ?. For specimens with l = 2 and 3, the principal fractures preferentially propagate along mineral–matrix interfaces, with their azimuth closely corresponding to the mineral distribution. CT-number-based damage variables at 70% of the peak load increase with ?, whereas the post-peak fracture-volume ratio decreases with ?. The fracture networks of specimens with ? l= 2 exhibit the highest fractal dimension, indicating the greatest spatial complexity. These findings clarify the mechanisms by which the height-to-diameter ratio, mineral distribution, and initial damage control coal failure modes and three-dimensional fracture evolution across macro- and mesoscales, thereby providing theoretical support for rockburst prevention in highly burst-prone coal seams with large mining heights.

Landfill leachate leakage causes pollutant migration, posing a risk of ecological damage to the surrounding environment. To address the key challenge of predicting the contamination range caused by leachate leakage, this study employs mathematical methods, including separation of variables, to derive a concise analytical solution for two-dimensional pollutant migration in a soil medium with infinite horizontal extent. The solution provides a practical tool for preliminary prediction of the contamination range resulting from leachate leakage. The derived analytical solution is then verified by comparison with two existing analytical solutions and the results of numerical simulations. Finally, using toluene as a representative pollutant, the effects of factors such as leakage duration on the contamination range are predicted based on the defined average and maximum contamination distances. The results show that, although increasing the spatial nonuniformity of pollutant concentrations in the leaked leachate reduces the average contamination distance, it increases the maximum contamination distance. Compared with the scenario involving a uniform concentration distribution, the maximum contamination distance can increase by 10.5% under a nonuniform concentration distribution. Both the average and maximum contamination distances increase markedly with leakage duration. When the leakage duration increases from 1 to 3 years, the average contamination distance increases by 133.0%. An increase in the hydraulic conductivity of the soil surrounding the landfill also substantially increases both contamination distances; a tenfold increase in hydraulic conductivity can increase them by more than fivefold. Therefore, related pollution prevention and control measures should account for both the effect of concentration nonuniformity in the leaked leachate on the maximum contamination distance and the influence of pre-existing contamination conditions on the subsequent effectiveness of these measures. Overall, this study contributes to the effective implementation of pollution prevention and control measures for landfill leachate leakage.

Flow-dissolution coupling in soluble-rock dam foundations is a key factor governing the long-term seepage-control performance of high dams. Under the combined effects of tectonic stress and karstification, heterogeneous fracture networks commonly develop within dam foundations, and their evolution directly controls the formation and breakthrough of leakage pathways. Most existing discrete fracture network (DFN) flow-dissolution simulations neglect the influence of in situ stress on the initial distribution of fracture apertures, limiting their ability to realistically capture the evolution of dissolution channels under dam-foundation conditions. In this study, a DFN-based flow-dissolution numerical model is developed by coupling the Bandis fracture deformation model with a two-dimensional depth-averaged flow-dissolution framework. The model simultaneously considers stress-controlled initial fracture apertures, turbulent flow, and nonlinear reaction kinetics. Extensive numerical simulations are conducted to elucidate the evolution of dissolution channels and the associated growth of leakage under stress-controlled aperture conditions. The effects of hydraulic head, impermeable apron width, and grouting-curtain depth on seepage evolution in the dam foundation and the breakthrough time of dissolution channels are quantitatively investigated. On this basis, a scaling law for the breakthrough time of dissolution channels under stress-controlled aperture conditions is proposed. The results indicate that, when stress effects are considered, the mean fracture aperture decreases with depth, causing dissolution channels to preferentially develop in shallow zones or near seepage-control structures, thereby significantly shortening their breakthrough time. Reducing the upstream hydraulic head, increasing the impermeable apron width, and deepening the grouting curtain can effectively delay dissolution-channel breakthrough, with grouting-curtain depth exerting the most dominant influence. This study provides a theoretical basis and methodological support for quantitatively assessing seepage risks and optimizing seepage-control systems in soluble-rock dam foundations.

The synergistic evolution of the “stress field-fracture field-seepage field” in the surrounding rock during deep mining, combined with the significant alteration of the rock mass?s physical and mechanical properties by groundwater, makes it difficult to reconcile the scientific validity and practical applicability of traditional acoustic monitoring and early-warning criteria, thereby significantly increasing the difficulty of accurately predicting dynamic disasters such as rockbursts. To elucidate the damage evolution and acoustic response characteristics of hydraulically weakened sandstone and establish reliable detection methods, this study proposes a multidimensional discrimination model based on acoustic features. Four groups of sandstone specimens with different water contents—dry, naturally moist, weakly saturated, and fully saturated—were prepared. Synchronous mechanical-acoustic tests were conducted throughout uniaxial compression using an MTS electro-hydraulic servo system and a P-wave velocity monitoring platform. Ultrasonic P-wave velocity, waveform skewness, and kurtosis were extracted as key acoustic parameters to characterize their qualitative relationships with damage evolution. A random-forest-based classification model incorporating SHAP(SHapley Additive exPlanations)-weighted feature matching was then developed. The results indicate that water significantly modulates the evolution of P-wave velocity, waveform skewness, and kurtosis across different stress stages. Under fully saturated conditions, P-wave velocity does not decrease substantially during the yield stage; instead, it remains high or even increases slightly, producing a misleading acoustic signature. This finding highlights the limitations of relying solely on velocity reduction to assess rock instability. In contrast, waveform skewness and kurtosis exhibit complex but distinct variations associated with microcrack initiation and propagation. SHAP-based interpretability analysis clarifies the respective contributions of acoustic features, including P-wave velocity, skewness, and kurtosis, to damage-state identification under different saturation conditions. An interpretable random forest discrimination model integrating multidimensional acoustic features was established, achieving an overall accuracy of 71.6% in identifying the water-saturation state and stress stage of the rock. This study elucidates the acoustic characterization mechanisms of nonlinear rock damage under hydraulic weakening and proposes an integrated identification approach that combines machine-learning interpretability with rock-physics mechanisms. The findings provide an experimental foundation and theoretical support for stability assessment and intelligent early warning of instability in deeply buried, water-saturated rock masses.

To elucidate the shear mechanical behavior of rock structural surfaces under different shear boundary conditions and the underlying mechanisms governing size effects, a multiscale rock structural-plane model with identical roughness morphology was developed using the particle flow method. A systematic comparative analysis was conducted of the shear strength and deformation characteristics, crack evolution, and stress-field distribution of specimens with different length-to-height ratios under conventional unilateral shear loading and fully surrounding shear-loading conditions. The results indicate that, under unilateral shear loading, both the pre-peak shear stiffness and shear strength decrease continuously with increasing structural-plane size. Shear deformation and crack evolution along the structural surface exhibit pronounced size effects, which primarily arise from the nonuniform transfer of shear deformation and stress along the surface. In contrast, under fully surrounding shear loading, the structural surface undergoes synchronized shear deformation as an integrated whole. Consequently, the shear stiffness, peak strength, and crack density are relatively insensitive to changes in specimen dimensions, resulting in substantially weaker size effects. Microscopic analysis further reveals that fully surrounding shear loading effectively alleviates localized stress concentrations and promotes the uniform propagation of cracks along the structural surface. Shear failure occurs predominantly through an overall wear mechanism. These findings clarify the mechanisms by which shear-loading conditions regulate the mechanical behavior and size effects of rock structural surfaces, while also highlighting the potential limitations of conventional direct shear tests for characterizing shear properties under seismic loading or large-scale sliding conditions. The results provide a new theoretical basis for the rational determination of shear-strength parameters for structural surfaces and the stability assessment of rock masses in seismic engineering.

Ground collapse is a common geological hazard worldwide, with pipeline leakage and rainfall infiltration serving as two primary triggers. However, the mechanisms governing soil erosion and cavity development caused by cyclic pipeline leakage, as well as the subsequent rainfall-induced collapse, remain poorly understood. This study investigated the complete evolution of this hazard through a series of model tests, encompassing soil erosion induced by cyclic pipeline leakage, cavity development and stabilization, and eventual rainfall-induced collapse. Key physical and mechanical parameters were monitored throughout the tests to characterize the response at each stage of hazard progression, with particular emphasis on the influence of rainfall intensity on cavity collapse patterns. The results revealed that cavity development under cyclic leakage proceeded through three distinct stages: rapid expansion dominated by seepage forces, gradual development governed by the combined effects of seepage and capillarity, and eventual stabilization controlled by capillary forces. No significant surface displacement was observed during cavity evolution. Depending on rainfall intensity, cavity collapse patterns were classified into three types: washout-dominated, rainfall-infiltration-dominated, and groundwater-seepage-dominated collapse. In all cases, collapse occurred abruptly, without apparent surface deformation before failure. Furthermore, the mechanism governing cavity stability was elucidated at both the macroscopic and microscopic scales based on limit equilibrium and capillary theories. At the macroscopic scale, capillary action generated apparent cohesion, thereby significantly enhancing the shear strength of the soil. At the microscopic scale, interparticle capillary forces exceeded the self-weight of individual particles, enabling them to adhere to one another and preventing detachment. This study provides experimental evidence and theoretical insights that advance the understanding of cavity formation induced by pipeline leakage and subsequent collapse triggered by rainfall infiltration.

Rock brittleness is a key mechanical parameter that affects the failure mode of rock mass structures under complex hydraulic action. To quantify the evolution law of rock brittleness under cyclic water intrusion, this study establishes an improved energy calculation model and, based on energy evolution theory, constructs brittleness indices for the pre-peak, post-peak, and the entire deformation processes. A comprehensive brittleness index (BI) is proposed, and its reliability is validated through uniaxial compression tests on four typical rocks (diorite, marble, sandstone, and mudstone). To explore the applicability of the brittleness index, triaxial compression tests under different confining pressures were conducted on yellow sandstone subjected to different cyclic water intrusions. The research results indicate that: (1) Comparing the comprehensive brittleness index (BI) with existing brittleness indices reveals that the brittleness prediction results for typical rocks show a consistent trend, indicating that this index can accurately reflect rock brittleness; (2) Cyclic water intrusion leads to progressive deterioration in the mechanical properties of sandstone, with the peak stress decreasing by 5.1%-12.0% and the residual strength decreasing by 15.6%-25.0%, and the damage evolution exhibits an S-shaped three-stage pattern; (3) Cyclic water intrusion significantly reduces sandstone brittleness, whereas confining pressure demonstrates an obvious dual-stage regulatory effect on sandstone brittleness, enhancing brittleness under low confining pressure conditions (0-10 MPa) and reducing brittleness under higher confining pressure conditions (10-15 MPa). The results of this study are of great significance for understanding the progressive failure mechanisms of water-involved rock engineering, and can provide a theoretical basis and potential applications for the stability assessment of structures such as reservoir bank slopes and underground caverns.

Although biogrouting has been preliminarily applied to rock-mass reinforcement and sealing, little research has investigated its diffusion mechanism and sealing performance in inclined rock joints. To address this gap, rough rock-joint specimens with different inclination angles were fabricated using 3D-printing technology, and laboratory MICP-based biogrouting experiments were conducted. The diffusion of calcium carbonate and the corresponding variations in transmissivity were monitored in gently inclined rock joints during the biogrouting process. The results indicate that the joint inclination angle is a key factor governing calcium carbonate diffusion. Most of the grouting-induced diffusion occurred during the first grouting cycle, and the evolution of the diffusion area exhibited three distinct stages with increasing grouting time: an initial linear-diffusion stage, a nonlinear accelerated-diffusion stage, and a dense stabilization stage associated with calcium carbonate filling. This evolution was quantitatively described by an empirical exponential-decay function, with a correlation coefficient of approximately 0.9. Moreover, the gently inclined rock-joint specimens with different inclination angles were completely sealed after 2–5 grouting cycles, and their transmissivity decreased by more than 99% on average, demonstrating the significant effectiveness of biogrouting in sealing inclined rock joints. However, owing to the influence of joint inclination, the distribution of the biogrouting products within some specimens was nonuniform, which may compromise the long-term sealing performance. These findings provide technical support for the practical application of microbial grouting in the reinforcement and sealing of jointed rock masses.

To investigate the effects of rotation of the principal stress direction on the morphology and size of the plastic zone in surrounding rock, a conformal mapping method based on complex variable theory was adopted. The principal stress rotation angle ? was introduced into the remote stress field, and analytical solutions for the circumferential stress at the boundaries of four roadway sections, namely circular, elliptical, rectangular, and straight wall semicircular arch sections, were derived under different lateral pressure coefficients ? and principal stress rotation angles ?. The circumferential stress distribution in the surrounding rock of different roadway sections was then investigated. Based on the Mohr-Coulomb strength criterion, an implicit equation for the boundary of the plastic zone in surrounding rock was established. The evolution of the morphology and size of the plastic zone for the four roadway sections under different combinations of ? and ? was analyzed, and the theoretical results were verified by numerical simulation. The results show that: (1) ? governs the distribution pattern of circumferential stress and the locations of stress concentration at the roadway boundary, whereas ? causes directional rotation of the stress field and breaks the symmetry of the stress distribution. Noncircular sections with geometric corners or abrupt changes in curvature are more sensitive to changes in the principal stress direction and are more prone to local stress concentration. (2) The evolution of the plastic zone in surrounding rock can be classified into six typical morphologies. ? controls the basic propagation pattern of the plastic zone, whereas ? determines its deflection direction. Their combined effects determine the final propagation morphology, extent, and orientation of the plastic zone. (3) The sensitivity of the maximum plastic zone radius Rpmax to changes in the principal stress direction is significantly affected by the roadway section shape. Circular and elliptical sections show relatively low sensitivity, whereas rectangular and straight wall semicircular arch sections exhibit a strong dependence on the principal stress rotation angle and are also strongly affected by ?. Based on these evolution characteristics of the plastic zone, a targeted advance support scheme integrating coordinated long and short bolts and cables, steel mesh and shotcrete, and shallow grouting reinforcement was proposed. The 11302 return airway of Hongqingliang Coal Mine was used as an engineering case to verify the reliability of the proposed support scheme.

To address the low accuracy of tunnel water-inrush prediction under complex geological conditions and the difficulty in capturing the nonlinear relationships among disaster-inducing factors, an intelligent prediction framework driven by ensemble learning is proposed. Based on a review of 249 documented water-inrush disaster cases in China and abroad, 22 disaster-inducing factors were selected. The synthetic minority over-sampling technique (SMOTE) was employed to overcome the class imbalance of the minority-class data, and principal component analysis (PCA) combined with K-means clustering was used to establish a hazard assessment system. A stacked generalization (SG) strategy was then adopted to integrate a random forest (RF) model and a backpropagation neural network (BPNN) to construct the prediction model, and the analytic hierarchy process (AHP) was further introduced to identify the key disaster-inducing factors. The results show that: (1) By exploiting the complementary error characteristics between the base learners, the SG model achieves a coefficient of determination (R2) of 0.98, a mean absolute error (MAE) of 0.68 and a mean square error (MSE) of 0.87, and its overall prediction performance and generalization ability are superior to those of a single base learner. (2) For major water-inrush cases with water-inflow volumes exceeding 300 m3/(d•m), the relative prediction error of the SG model remains stably within 10%, which effectively reduces the prediction fluctuations of a single base learner in the high-flow range. (3) Based on the initial weights obtained from the RF model and the correction made by engineering judgment, the AHP identifies the hydraulic conductivity, adverse geological conditions, geological structure, negative landform and groundwater occurrence conditions as the five core disaster-inducing factors, with a cumulative weight of more than 48.56%. These findings provide a relatively reliable technical approach for water-inrush disaster prevention and early warning in underground engineering.

To address the low accuracy of tunnel water-inrush prediction under complex geological conditions and the difficulty in capturing the nonlinear relationships among disaster-inducing factors, an intelligent prediction framework driven by ensemble learning is proposed. Based on a review of 249 documented water-inrush disaster cases in China and abroad, 22 disaster-inducing factors were selected. The synthetic minority over-sampling technique (SMOTE) was employed to overcome the class imbalance of the minority-class data, and principal component analysis (PCA) combined with K-means clustering was used to establish a hazard assessment system. A stacked generalization (SG) strategy was then adopted to integrate a random forest (RF) model and a backpropagation neural network (BPNN) to construct the prediction model, and the analytic hierarchy process (AHP) was further introduced to identify the key disaster-inducing factors. The results show that: (1) By exploiting the complementary error characteristics between the base learners, the SG model achieves a coefficient of determination (R2) of 0.98, a mean absolute error (MAE) of 0.68 and a mean square error (MSE) of 0.87, and its overall prediction performance and generalization ability are superior to those of a single base learner. (2) For major water-inrush cases with water-inflow volumes exceeding 300 m3/(d•m), the relative prediction error of the SG model remains stably within 10%, which effectively reduces the prediction fluctuations of a single base learner in the high-flow range. (3) Based on the initial weights obtained from the RF model and the correction made by engineering judgment, the AHP identifies the hydraulic conductivity, adverse geological conditions, geological structure, negative landform and groundwater occurrence conditions as the five core disaster-inducing factors, with a cumulative weight of more than 48.56%. These findings provide a relatively reliable technical approach for water-inrush disaster prevention and early warning in underground engineering.

The stress distribution between coarse and fine fractions is nonuniform in internally unstable soils, particularly in gap-graded and widely graded soils. Therefore, accurately estimating the stress reduction coefficient is essential for predicting the critical hydraulic gradient at which internal erosion initiates. For internally unstable coarse-grained soils with gap-graded and widely graded particle-size distributions, the erodible fine particles within the pores of the coarse-grain skeleton were initially assumed to carry no effective stress. Based on this assumption, a theoretical equation was derived to express the stress reduction coefficient as a function of porosity and erodible fine-particle content. In addition, constant-head upward seepage tests were conducted on coarse-grained soils with different gradations. The proposed equation was validated against the experimental results, including previously published data. The results demonstrate that the proposed equation provides more accurate predictions of the critical hydraulic gradient than previously published equations. It is applicable to gap-graded and widely graded coarse-grained soils with erodible fine-particle contents of 6%–20% and porosities of 0.25–0.55. Finally, the proposed equation was used to improve a previously published model describing the evolution of the critical hydraulic gradient under loading. The improved model effectively describes the evolution of both the critical hydraulic gradient and the average vertical effective stress under static and cyclic loading conditions, as demonstrated by previously published test data for coarse-grained soils. The proposed equation provides a theoretical basis for improving the reliability of seepage-erosion assessment in gap-graded and widely graded coarse-grained soils.

To elucidate the long-term dynamic deformation characteristics of subgrade filler under typical prestressed subgrade stress paths and facilitate reliable prediction, cyclic triaxial tests with stepwise confining pressure and variable-amplitude dynamic loading were conducted on silty filler collected from the roadbed layer of the Shuohuang Railway. The effects of the initial confining pressure, confining pressure increment, and dynamic stress amplitude on cumulative plastic strain and its cycle-based growth rate were systematically investigated, and a corresponding prediction model was established. The results show that increasing either the confining pressure increment or the initial confining pressure significantly suppresses the development of cumulative plastic strain, delays specimen failure, and promotes the transformation of potentially failure-prone specimens into critical- or stable-type specimens. As the initial dynamic stress amplitude increases, cumulative plastic strain increases, whereas the number of loading cycles to failure decreases significantly. During the nonaccelerated failure stage, the cycle-based growth rate of cumulative plastic strain exhibits a stepwise, sawtooth-like logarithmic linear decay with increasing loading cycles. The dynamic stress determines the overall magnitude of the growth rate, whereas the confining pressure increment primarily regulates and suppresses its development. The ratio of the confining pressure increment to the dynamic stress increment is a key parameter governing the evolution of cumulative plastic strain and the transformation of deformation patterns. Based on fractional calculus theory, a loading-cycle hardening concept is introduced to characterize the effects of previous loading history. In combination with a critical dynamic stress triggering mechanism, a fractional-order cumulative plastic strain prediction model is established to uniformly describe the development of stable-, critical-, and failure-type deformation. After parameter identification using nonlinear least squares, the coefficient of determination (R²) for cumulative plastic strain prediction under all test conditions in no less than 0.965. The cycle-based growth rate predicted by differentiating the model agrees well with the experimental results for representative cases, further confirming the models rationality and reliability. These findings provide a theoretical basis for controlling the long-term dynamic deformation of prestressed subgrades and evaluating their engineering applicability.

To address engineering hazards such as strength degradation and reinforcement failure commonly observed in geocell-reinforced silty-clay subgrades in high-temperature regions, this study investigates the effects of temperature rise on the mechanical properties and reinforcement performance of geocell-reinforced silty clay. Silty clay and three types of geocells—high-density polyethylene (HDPE), polyethylene terephthalate (PET), and polypropylene (PP)—were used as test materials. Geocell mechanical tests and temperature-controlled consolidated-undrained triaxial tests were conducted at 20 ℃, 50 ℃ and 70 ℃. A triaxial shear constitutive model for silty clay was established and calibrated using Duncan-Chang parameters. The mechanical behavior of geocell-reinforced soil at different temperatures was further analyzed using the reinforced-soil constitutive model. The results indicate that: (1) the strip stiffness and joint strength of all three geocells are negatively correlated with temperature, with HDPE geocells exhibiting the greatest temperature sensitivity. The joint strengths of HDPE and PET geocells are governed by peel strength, whereas that of PP geocells is controlled by pull-out strength; (2) increasing temperature reduces the shear strength of silty clay, increases pore-water pressure, decreases the initial tangent modulus K and failure ratio Rf, and increases the modulus exponent n in the Duncan-Chang model; (3) the confining pressure induced by the geocells significantly enhances the strength of silty clay. At 20 ℃, PET geocells provide the most effective reinforcement, substantially outperforming HDPE and PP geocells; and (4) the mechanical behavior of geocell-reinforced soil composites at different temperatures is governed by the thermomechanical coupling between the geocell and the soil. Increasing temperature markedly reduces reinforcement efficiency, with the greatest reduction observed for HDPE geocells (71.43%–74.42%), followed by PP geocells (20.07%–31.91%), whereas PET geocells are the least affected (4.97%–18.71%). This behavior is attributed to the intensified movement of the linear saturated hydrocarbon chains in HDPE at elevated temperatures, which promotes material softening. These findings provide guidance for the material selection and structural design of geocell-reinforced subgrades and slopes in high-temperature regions and can effectively improve the long-term service stability of engineering structures.

Long-term slow cyclic loading progressively alters the strength and deformation characteristics of rockfill materials. A series of single-particle crushing tests was conducted to investigate the effects of initial particle size, number of loading cycles, and cyclic loading amplitude on the crushing strength and breakage modes of individual particles. In addition, discrete element simulations were performed to reproduce the complete loading history of a single particle, including cyclic loading-unloading and subsequent reloading to failure. Based on the Paris-Erdogan law, a bond-strength degradation factor was developed and incorporated into the particle breakage criterion, providing particle-scale insight into the underlying mechanism of rockfill strength degradation under slow cyclic loading. The results indicate that: (1) Cyclic loading induces cumulative damage, and the crushing strength of individual particles, which follows a Weibull distribution, decreases with increasing cyclic amplitude and number of cycles; (2) Cyclic loading markedly alters the breakage mode, with particles tending to fragment into a greater number of smaller pieces as the cyclic amplitude and number of cycles increase, while the failure pattern gradually shifts from explosive fragmentation to splitting and, ultimately, crushing; (3) Low cyclic stress levels may not produce pronounced macroscopic breakage, but they promote fatigue-crack growth within particles. The proposed bond-strength degradation factor quantitatively captures the coupled effects of stress level and cycle number on fatigue-crack propagation; and (4) the numerical results agree well with the experimental trends, validating the proposed degradation factor and breakage criterion. They further demonstrate that the fundamental mechanism underlying single-particle strength degradation under slow cyclic loading is the fatigue-driven propagation and progressive coalescence of internal cracks.