Highway slope disasters pose a significant threat to the safe operation of road traffic. Clarifying the fundamental causes of these disasters and developing lightweight monitoring and early warning systems have become urgent priorities to ensure the safety of road networks. Based on a comprehensive review of the characteristics and spatial distribution of highway slope disasters, this study systematically summarized five typical failure modes-collapse, landslide, debris flow, subgrade subsidence, and surface slumping-and analyzed the specific features and failure mechanisms of each type. An in-depth analysis was also conducted on the contributing factors and triggering conditions. Building on these insights, the study proposed a framework for lightweight monitoring and early warning of highway slope disasters. This framework consisted of five main components: on-site inspection, selection of monitoring indicators, optimization of monitoring locations, refinement of warning models, and implementation of preventive measures. The study also outlined the essential conditions required for implementing such lightweight monitoring and early warning. Based on this, the paper focused on highway embankment slopes characterized by “long corridors and strong constraints”. A core framework of “pre-screening before measurement, low-cost sensing, and lightweight modeling” was summarized, and a feasible implementation scheme was proposed. This work serves as a reference for the development and application of lightweight monitoring and early warning technologies for highway subgrade slopes and holds important implications for improving the resilience and safety of road infrastructure.

To address the challenges of complex spatiotemporal evolution, diverse triggering mechanisms, and delayed responses of conventional treatments for highway-associated landslides, this study proposed an integrated full-process “identification-early warning-control” approach for landslide hazard management. In the hazard identification stage, a rapid landslide detection and evolution-tracking method was developed by integrating multi-source data including remote sensing imagery, Interferometric Synthetic Aperture Radar (InSAR) interferometry, Unmanned Aerial Vehicle (UAV) photogrammetry, and Light Detection and Ranging (LiDAR), combined with geophysical prospecting and borehole inclinometers to accurately delineate the geometric characteristics of sliding surfaces. In the monitoring and early warning stage, a multi-parameter monitoring system was established by integrating Global Navigation Satellite System (GNSS), inclinometers, stress meters, microseismic sensors, and meteorological instruments, and a hierarchical early warning model was constructed based on deformation-mechanical control-environmental responses, enabling dynamic monitoring and accurate early warning under complex geological conditions. In the control strategy stage, a phased rapid emergency treating strategy was proposed, integrating structural optimization with components such as micropiles and intelligent anchoring systems, to achieve highly targeted and efficiently deployable countermeasures. The proposed methodology was validated through its application to the Naliang landslide treatment project along the Duba Expressway. The results demonstrate its strong engineering adaptability and high potential for broader implementation in integrated “identification-early warning-control” treatment of landslides along highway.

The dynamic evolution of earthquake-induced high-level rock mass collapses is highly complex, characterized by significant energy release, and poses severe threats to the safety of transportation lifeline projects. To elucidate the mechanism of dynamic instability and the associated hazard effects of high-level rock masses under strong seismic excitation, this study focused on the Lanhuazhai high-level dangerous rock zone, located merely 2.5 km from the epicenter in the Hailuogou Scenic Area during the 2022 Luding earthquake. A dynamic model incorporating real terrain features was established. Using the finite-discrete element method, the catastrophic evolution of the Lanhuazhai rock mass was numerically reproduced, and the failure modes of the highway subgrade under collapse-induced impact loading were identified. In addition, a quantitative assessment of the post-seismic hazard characteristics of the unstable rock mass was conducted, clarifying rockfall trajectories and their implications for highway safety. The results demonstrate that: under the “9·5” Luding earthquake, the Lanhuazhai high-level rock zone was governed by three dominant structural planes. Collapse instability generated a loose accumulation body of 90 000-160 000 m³ aligned along the NW-SE direction, constituting a critical obstacle for emergency highway clearance and post-disaster reconstruction; the catastrophic process of high-level rock collapse can be categorized into four dynamic stages: seismic cracking and sliding initiation, projectile flight, collision-induced fragmentation, and frictional accumulation; the hazard effects of earthquake-induced collapse on the highway subgrade were driven by a dual mechanism, in which seismic shaking induced micro-damage within the subgrade structure, while impact loading from collapsed rock masses triggered compressive-shear failure; two rockfall trajectories observed on January 18 and January 27, 2023, posed direct threats to the highway subgrade. For emergency restoration, a combined strategy of ‘stepped slope excavation and clearance of accumulation body + passive net protection of unstable rock’ is recommended, whereas a tunnel bypass scheme is advised for long-term reconstruction. These findings provide critical scientific insights and practical guidance for disaster prevention and mitigation of transportation infrastructure in mountainous regions affected by strong earthquakes.

Aiming at the problems that there is a lack of a comprehensive assessment framework for landslide risks of mountainous expressways and existing studies only focus on landslide susceptibility, which makes it difficult to fully support risk management and control, this study intends to construct a scientific landslide risk assessment system for mountainous expressways. Based on Geographic Information System (GIS) technology, this study proposes a three-dimensional assessment framework integrating “landslide susceptibility of slope units-vulnerability of expressways-exposure”. Specifically, firstly, combined with the landslide triggering mechanism and engineering geological survey data, 15 landslide influencing factors including elevation, slope gradient and slope position were selected, and a LightGBM landslide susceptibility evaluation model was constructed using the semi-supervised learning method. Secondly, by quantifying the traffic volume after road completion and the economic losses that may be caused by the damage of different structures, 5 vulnerability indicators were determined, including life loss, vehicle loss, direct loss of expressway structures, repair costs and indirect loss caused by traffic interruption. The entropy weight method was used to calculate the weight of each indicator to complete the vulnerability evaluation. Finally, the exposure of the expressway was characterized by the area of the slope unit where the road affected by landslides is located and the average distance from the slope unit to the road. By multiplying susceptibility, vulnerability and exposure, this study realized the quantitative comprehensive assessment of landslide risks of mountainous expressways. This study breaks through the limitation of traditional studies that only focus on landslide susceptibility, and constructs a “susceptibility-vulnerability-exposure” trinity comprehensive assessment framework for landslide risks of mountainous expressways. Moreover, the selection and quantification methods of indicators in each dimension are more in line with practical engineering scenarios. The research results can provide methodological support for the comprehensive assessment of landslide risks, and at the same time offer a scientific basis for the mitigation, control and risk management decision-making of landslide risks of mountainous expressways.

To address the problem of random noise in micro-electro-mechanical system (MEMS)-based landslide deep deformation monitoring signals, a noise reduction method that integrates an improved Honey Badger algorithm (IHBA), variational mode decomposition (VMD), and an improved wavelet threshold is proposed. First, the Tent chaotic mapping, the Whale Optimization algorithm (WOA) spiral search mechanism, and the Levy flight strategy were introduced into the traditional Honey Badger Algorithm (HBA). The IHBA was used to optimize the two key parameters of VMD decomposition, K (number of modes) and α (penalty factor), to obtain an optimal parameter combination. The noisy signal was then decomposed into multiple intrinsic mode functions (IMFs) via VMD. The variance contribution rate of each IMF was calculated to filter out components with low contributions. Subsequently, the remaining components were subdivided into effective, noisy, and noise components based on their correlation coefficients. An improved wavelet threshold function algorithm was designed by incorporating a high-precision fourth-power modulus processing method to overcome the limitation of the constant deviation in traditional threshold functions. The noisy components were denoised using this improved wavelet threshold algorithm. Finally, the denoised and effective components were reconstructed into a denoised signal. Simulation experiments demonstrated that the proposed method achieved the highest SNR (28.7642) and lowest RMSE (0.0101) compared with conventional denoising methods. Furthermore, a comparative analysis of real-world landslide deep deformation monitoring signal denoising revealed that the new method yielded the smallest RMSE and RVR and the highest NM and SNR, with all SERs exceeding 99%. It can effectively reduce noise while preserving the detailed features of the original signals, significantly improving the accuracy and reliability of MEMS-based landslide deep deformation monitoring data, thereby demonstrating promising application prospects.

To enhance the rationality of the seismic stability assessment of soil slopes, this study integrates the pseudo-dynamic method with Spencer's limit equilibrium method to derive a formula for calculating the factor of safety under pseudo-dynamic conditions. The Karhunen-Loève expansion method was employed to generate random fields of the soil parameters, and a Monte Carlo simulation was used to compute the failure probability. A risk assessment model for soil slopes under pseudo-dynamic conditions was established to comprehensively account for the combined effects of seismic dynamic loading and spatial variability of soil parameters on slope stability. By analyzing the peak failure probability under pseudo-dynamic conditions and average failure probability over the seismic period, the influence patterns of seismic motion parameters, slope geometric characteristics, and spatial variability of soil parameters (including autocorrelation distance, coefficient of variation, and correlation coefficient) on failure probability and slope safety factor were revealed. The results indicate that as the horizontal seismic acceleration coefficient increases, the pseudo-dynamic factor of slope safety is reduced nonlinearly, while significantly increasing the peak failure probability within the seismic period. With an increase in seismic wavelength, the potential slip surface was distributed more dispersedly and extended deeper, leading to a higher peak failure probability. The failure probability of the slope was positively correlated with the slope height and slope ratio, and an increase in the spatial variability parameters led to a higher failure probability. Therefore, the spatial variability of soil parameters should not be overlooked in seismic slope risk assessment.

Highway slopes are prone to soil erosion, which poses a significant threat to slope stability and ecological security along road corridors. Vegetation is widely recognized as an effective measure for controlling slope erosion. Most existing studies have focused on qualitative analysis of the influence of vegetation on runoff velocity and erosion processes, while quantitative investigations into flow velocity variations and critical vegetation coverage for soil erosion control on highway-vegetated slopes remain limited. To address this gap, laboratory runoff scouring experiments were conducted under varying vegetation coverage levels, flow discharges, and slope angles. The flow regime characteristics and mean flow velocities were systematically analyzed. A predictive model for the mean flow velocity on vegetated slopes was developed, and a method for calculating the critical vegetation coverage was derived from the model. The results indicate that the surface runoff predominantly exhibits laminar and transitional flow regimes, with all flows operating in the supercritical regime. With increasing vegetation coverage, the Reynolds number increases, whereas the Froude number decreases. By contrast, higher flow discharges and steeper slope angles significantly increase both dimensionless numbers. The mean flow velocity decreases with increasing vegetation coverage, exhibiting a saturation effect at higher coverage levels. Both the flow discharge and slope angle have significant accelerating effects on flow velocity, following exponential growth patterns. Based on the principles of energy conservation and hydraulic theory, a predictive model for the mean flow velocity was established by incorporating modified Manning's roughness and local resistance coefficients under varying environmental conditions. The model was validated against 294 experimental datasets under diverse conditions, achieving high accuracy (R²>0.900) and demonstrating its robust applicability and stability. Based on the experimental results, the critical incipient velocity equation for soil particles and the Soil Conservation Service Curve Number model were integrated to establish a method for determining critical vegetation coverage on highway slopes. The effectiveness and reliability of this method in erosion control were evaluated using the cumulative sediment yield and particle size distribution data from runoff scouring experiments.

To investigate the 3D stability of slopes reinforced by anti-slide piles under steady-state infiltration. The ultimate lateral resistance of anti-slide piles was introduced into the 3D slope model considering the seepage effect. The analytical solution of safety factor was deduced by the upper bound theorem, and the optimal upper limit value was solved via the dichotomy. The result shows that, when the aspect ratio B/H≤4, the 3D weakening effect on the slope is marked, and the safety factor reduces rapidly. Additionally, when the aspect ratio B/H > 4, the safety factor has good convergence performance, while no obvious changes. The 3D weakening effect of slopes mainly depends on the aspect ratio B/H, and is not influenced by the seepage effect. With the increase of infiltration intensity, the safety factor presents remarkable decreasing trend. Meanwhile, the air-entry value and matrix suction of soils are extremely favorable to the stability, but the degree of influence on groundwater level and air-entry value on the stability determined by the infiltration intensity. The effective range of pile spacing is 1 < k_p≤4, and the reinforcement effect is enhanced with the decrease of pile spacing in the effective range. Adversely, with the increase of pile spacing, 3D slopes present a change from deep failure into shallow failure, and the stability of 3D slopes becomes worse. In addition, the optimum pile position of anti-slide piles is n_p≈0.85, and the safety factor can be increased by more than 60%. The results can provide a theoretical guidance for the support design of similar 3D slopes.

To fill the theoretical gap in the design guidance for permeable concrete pile composite foundations and improve the consolidation theory of this pile type, this study proposes a composite foundation treatment method based on the combination of piles with different porosities. Based on four typical pile arrangement patterns commonly used in engineering practice, three consolidation models of composite foundations with different porosity-permeable concrete pile combinations were established. An analytical solution for the consolidation process under instantaneous loading was derived by considering pile penetration-induced deformation and bidirectional radial seepage effects. The rationality and applicability of the proposed models were verified using a degenerate solution comparison, numerical simulation, and field monitoring data. A parameter sensitivity analysis was conducted to investigate the influence of key variables on the consolidation behavior of the composite foundation, and the evolution of the pore water pressure within the pile-soil system was analyzed. The results showed that the three proposed models can flexibly adapt to different on-site pile-arrangement conditions, thereby expanding the design approach of composite foundations. Model 2-with low-porosity piles in the center and high-porosity piles on the periphery-achieved the highest drainage consolidation rate and was suitable for rapid consolidation conditions, whereas Model 3-with uniformly low-porosity piles-provided the greatest bearing capacity, making it appropriate for cases with high foundation strength requirements. When treating soft soil foundations, Model 1-with a dense rectangular pile arrangement-offered relatively high drainage efficiency and bearing capacity and better met treatment demands. From an economic perspective, Model 2 was the optimal model. The use of a modified equal-strain assumption enabled a more accurate prediction of the consolidation process and improved design reliability. In addition, clogging significantly affected the drainage performance, whereas the parameters of the disturbed zone had minimal influence on the consolidation rate.

Coarse-grained soil fillings with complex morphologies are prone to breaking under impact loads, which in turn influences the reinforcement efficiency of dynamic compaction. Seldom studies have considered the effect of particle breakage. So it is difficult to truly reveal the impact compaction characteristics of coarse-grained soil under impact load. The 3D blue-light scanning technology was adopted in this paper to realize the digital twin of complex morphologies particles. Combined with the Voronoi tessellation method, a breakable numerical particle library consistent with the real particle morphology was constructed. The Python programming language and the Particle Flow Code in Three Dimension (PFC3D) were integrated to establish a simulation model for the reinforced coarse-grained soil roadbed. A series of dynamic compaction numerical tests were conducted and the cross-scale responses including the crater depth, porosity distribution, displacement field, particle contact relationship, etc. as the number of compaction impacts were comparatively analysed between breakable and unbreakable fillers. It was found that the effect of particle breakage on the reinforcement efficiency of dynamic compaction was significantly stage-dependent during repeated impacts. The compaction process of coarse-grained soil and the reinforcement efficiency of dynamic compaction are weakened by particle breakage. As the number of tampings increased, particle breakage improved the compaction process and the final reinforcement efficiency of dynamic compaction. Finally, the crater depth of the breakable group increased by 11.6%, the porosity decreased by 20.5%, and the coordination number increased by 0.9% compared with the unbreakable group. The reason is that particle breakage consumed part of the tamping energy in the early stage of dynamic compaction. As the number of tampings increased, the amount of particle breakage decreases. The small particles produced by breakage filled the voids between large particles, which promotes particle compaction and thereby enhanced the reinforcement efficiency of dynamic compaction. Ultimately, the effect of dynamic compaction on the breakable particles is better. The research results may be beneficial to revealing the physical and mechanical processes of breakable coarse-grained soil under impact loading and optimizing dynamic compaction technology.

Loess foundation treatment is one of the key issues in geotechnical engineering. Due to its poor engineering characteristics, such as low strength and high compressibility, loess has always been a focus in foundation treatment. This paper investigates the effects of microwave sintering on the temperature field and compressibility characteristics of loess. By using a waveguide fracture antenna for deep uniform heating, the distribution of the temperature field and its evolution over time are systematically analyzed, along with compressibility tests to study the changes in compressibility after microwave sintering. The study shows that the temperature field in the microwave heating process can be divided into three stages: a rapid temperature rise in the first 500 minutes, mainly influenced by the dipole rotation and friction effect of water molecules; a slowed temperature rise from 500 to 1 500 minutes, mainly driven by the response of polar molecules in the soil; and stabilization of the temperature after 1 500 minutes, reaching thermal equilibrium. Spatially, a high-temperature zone is formed within the vertical range of 100 cm and the radial range of 20 cm, showing significant heating effects. In deeper regions and areas far from the hole wall, the temperature gradually decreases. Microwave sintering significantly improves the compressibility of loess, with the largest reduction in compressibility coefficient and compression index near the shallow hole wall, reaching over 90%; in deeper regions and areas further from the hole wall, the reduction is smaller but still ranges from 30% to 50%. Furthermore, the moisture content and density of loess change during sintering, which is closely related to the temperature and compressibility changes. The findings provide theoretical support for the application of microwave strengthening technology in loess foundation treatment.

Most provinces in southern China have warm and humid climates. As a subgrade filling material, high-liquid-limit clay has a low bearing capacity and shear strength; therefore, it is often reinforced during construction. Geogrid reinforcement is a common physical-strengthening method used for subgrades. To accurately characterize the resilient and permanent deformation characteristics of geogrid-reinforced high-liquid-limit clay, a new model was established to predict its resilient modulus and permanent deformation, which considers the relative displacement between the soil and geogrid, lateral constraint of the geogrid-reinforced clay, and residual stress effect induced by the compaction process. Laboratory tests were conducted to evaluate the resilient modulus and permanent deformation of the geogrid-reinforced subgrade soil under various conditions, including different types of geogrids, reinforcement methods, confining pressure levels, and deviatoric stress levels. The influence of these factors on the deformation characteristics of geogrid-reinforced soil was analyzed. The experimental results indicated that geogrid reinforcement enhanced the cohesion and internal friction angle of soil. In particular, the effect of the double-layer reinforcement was superior to that of the single-layer reinforcement, and the effect of the biaxial geogrid reinforcement was better than that of the triaxial geogrids. Moreover, the reinforcement effect of the geogrid was positively correlated with its tensile strength. Geogrid reinforcement enhanced the resilient modulus of the soil, and the rate of increase in the resilient modulus decreased with increasing deviatoric stress levels. Geogrid reinforcement reduces the permanent deformation of soil under cyclic loading. The rate of reduction in permanent deformation increased with increasing deviatoric stress and decreased with increasing confining pressure. The prediction accuracy of the additional equivalent stress-based resilient modulus and permanent deformation models for the geogrid-reinforced subgrade soil were verified experimentally. The new model can accurately predict the resilient modulus and permanent deformation of geogrid-reinforced subgrade soil under different stress levels, load times, geogrid types, and reinforcement modes, providing a useful reference for subgrade design and engineering practice.

Post-earthquake investigations indicate that back-to-back reinforced soil walls exhibit good seismic performance, and the reinforcement arrangements have an important influence on the dynamic response. An experimental study of shaking table tests on back-to-back reinforced soil walls with different reinforcement arrangements was designed according to the similitude relationships. The reduced-scale models with modular block facings were excited using a series of horizontal sinusoidal input motions with increasing acceleration, to investigate the influence of reinforcement arrangements on the dynamic response and deformation modes of back-to-back reinforced soil walls. The fundamental frequency of back-to-back reinforced soil walls with connected reinforcements is higher than that of back-to-back reinforced soil walls with overlapped reinforcements. For the back-to-back reinforced soil walls with overlapped reinforcements and connected reinforcements, the acceleration amplification factors on both sides and middle zone increase significantly with increasing elevation and show significant amplification effect. For the back-to-back reinforced soil walls with overlapped reinforcements, the input motion has an important influence on the acceleration amplification effect, and the acceleration amplification factors of the middle zone are larger than those of the two sides. For the back-to-back reinforced soil walls with connected reinforcements, the input motion has a minor influence on the acceleration amplification effect, and the amplification factors of the two sides and the middle zone are almost the same. The displacements of back-to-back reinforced soil walls with different reinforcement arrangements increase along the elevation, and the facing displacements increase significantly for larger input motions. The back-to-back reinforced soil walls with connected reinforcement have smaller facing displacements than those of the back-to-back reinforced soil walls with overlapped reinforcements. For higher input motions, the deformation modes of back-to-back MSE walls with overlapped and connected reinforcement are rotation and tilting, respectively. The results provide important insights for seismic design guidelines.

For large-scale deep miscellaneous fill soil sites, which typically exhibit unfavorable engineering properties, pre-reinforcement treatment is an essential prerequisite prior to construction. The vibratory probe compaction(VPC) method is a novel technology for pre-reinforcement of miscellaneous fill soils. Given that the existing research on the mechanism of the VPC method for treating fine-grained soil is not applicable to miscellaneous fill soil. This study employed model tests as the research method. Through vibration tests on different soil layers during the construction process, the soil vibration response was investigated, the system resonance frequency was determined, and the transmission and attenuation laws of vibration energy were studied. By monitoring soil settlement, the influence of vibration frequency on settlement evolution was examined. Particle analysis tests were conducted to explore the reasons for stiffness changes. Deep soil pressure monitoring during construction was performed to analyze the influence of frequency on soil stress variations. The results indicate that the resonance frequency of the miscellaneous fill system is in the range of 16-18 Hz. During the high-frequency probe penetration phase, surface vibration is concentrated in the vertical direction, while deep soil vibration is concentrated horizontally. During the low-frequency retention phase, surface vibration is primarily horizontal, while deep soil vibration is primarily vertical, exhibiting significant near-resonance amplification. As depth increases, vibration energy decays faster and the reinforcement zone decreases. Soil settlement mainly occurs during the high-frequency penetration and low-frequency retention phases, with the latter being predominant. Performing vibration retention close to the deep soil resonance frequency can expand the settlement-affected zone, shorten the consolidation time for miscellaneous fill foundations, and achieve a settlement deformation zone reaching 46% of the strain deformation zone. This provides a reference for determining the influence range of vibratory probe compaction in miscellaneous fills. VPC induces particle rearrangement at the surface, causing stiffness changes and distinct layering, which gradually weakens with depth. Low-frequency vibration retention significantly increases horizontal stress in deep soil, with the maximum increment observed at 18 Hz. The horizontal stress increment at the 400 mm depth is higher than at 200 mm and 600 mm depths. The findings of this research enrich and advance the theoretical understanding of the VPC mechanism for reinforcing miscellaneous fill foundations.

In order to clarify the dynamic response, damage mechanism and failure mode of highway subgrade under earthquake, and to comprehensively evaluate the seismic performance of highway subgrade, a shaking table test scaled-down model of fill and fill-and-cut subgrades in Luding was established, and the changing laws of acceleration and displacement response were analyzed. A finite element nonlinear dynamic time history analysis model was established, revealing the connection between energy dissipation and damage mechanism of subgrades, and clarifying the damage classification based on permanent settlement. A framework for evaluating the seismic resilience of subgrades was proposed, and the feasibility of the framework was demonstrated by examples. The results show that the dynamic response of fill-and-cut subgrade is significantly larger than that of fill subgrade, and the two present different failure modes. The internal shear damage occurs due to the insufficient strength of the fill subgrade, which is manifested as the overall slip and settlement. Plastic damage occurs at the interface of the fill-and-cut subgrade due to stiffness difference and stress concentration, and the “push-pull” effect of the fill zone dominates the evolution of seismic damage in chronological order, which triggers the expansion of plastic strain within the subgrade and eventually leads to overall instability. The permanent settlement and hysteretic energy dissipation under seismic action have strong correlation, and the damage evaluation system based on permanent settlement can be effectively applied to assess the seismic damage of subgrades. When the peak ground acceleration (PGA) is less than 0.20g, the subgrade is slightly damaged, the structural function remains intact, and the subgrade is still in the high resilience range. When the PGA increases to 0.40g, the post-earthquake function of the subgrade decreases significantly, and the post-earthquake function of fill-and-cut subgrade decreases to 0.24, and the resilience index decreases to 0.83. The research results can optimize the seismic design and repair strategy, which is of great significance in ensuring the normal operation of the highway system and reducing the socio-economic cost.

The evolution of plastic strain in subgrade clay subjected to long-term freeze-thaw cycles and repeated traffic loading is critical for accurately evaluating the service life and settlement deformation of subgrades in seasonally frozen regions. In this study, a series of static and dynamic triaxial tests were conducted to investigate the effects of the number of freeze-thaw cycles and confining pressure on the strength parameters of clay. The results reveal that cumulative freeze-thaw damage significantly degrades the mechanical properties of the soil. Furthermore, the evolution of plastic strain across different deformation states was quantitatively analyzed. Based on the generalized Kelvin model, a fractional-order damage model was developed to describe plastic strain behavior in the stable, critical, and failure states. In this model, the classical Newtonian dashpot was replaced with an Abel dashpot, and a nonstationary viscoplastic element was introduced to capture the temporal and nonlinear evolution of plastic strain. The test results indicate that the peak strength, cohesion, and internal friction angle of subgrade clay decrease with an increasing number of freeze-thaw cycles, although the rate of decrease gradually diminishes. The development of plastic strain exhibits stress dependence and freeze-thaw sensitivity, indicating that deviatoric stress and freeze-thaw damage are key factors inducing transitions among the stable, critical, and failure states. The proposed fractional-order damage model fits well with the experimental plastic strain curves for all three states. Moreover, an increase in the fractional order increases the proportion of the accelerated strain stage, whereas a higher acceleration index leads to a faster plastic strain rate in this stage. These findings provide theoretical support and modeling guidance for predicting long-term deformation and performing numerical analyses of plastic strain in subgrade soils in seasonally frozen areas.

The long-term stability of mudstone coarse-grained soil subgrades is closely related to their strength. Dynamic-static triaxial tests were conducted on mudstone coarse-grained soils with varying degrees of erosion to accurately predict their strength under the effects of penetrating erosion and cyclic loading. A peak strength database containing 300 datasets was established. Based on this database, six machine-learning methods-random forest (RF), gated recurrent unit (GRU), k-nearest neighbors (KNN), backpropagation neural network (BPNN), extreme gradient boosting (XGBoost), and Gaussian process regression (GPR)-were employed to predict the peak strength. The models were comprehensively evaluated using metrics, such as determination coefficient R², root mean square error (RMSE), mean absolute error (MAE), 95% uncertainty interval (U₉₅), and weighted mean differences (WMD). The results indicate that six factors-penetrating intensity (p₀), loading frequency (f), dynamic confining pressure (σ_3c), dynamic deviatoric stress (σ_d), loading cycle count (N), and static confining pressure (σ_3s)-are the primary contributors to the peak strength (σ₁-σ₃)_max. Among the methods tested, the XGBoost model demonstrates the highest prediction accuracy and lowest uncertainty and is recommended as the optimal method. Sensitivity analysis reveals that penetrating intensity (p₀) makes the greatest contribution to (σ₁-σ₃)max (with a weight of 30.28%), followed by loading cycle count (N) (27.81%) and static confining pressure (σ_3s) (21.91%). Monotonicity analysis shows that with increasing p₀, σ_d, f, and N, (σ₁-σ₃)max generally decreases. Notably, when p₀ ranges from 0 to 40 kPa, the decrease in (σ1_-σ₃)_max is less significant. Thus, coarse particles dominate the mechanical behavior, and the influence of erosion on strength is minimal during this range.

In order to explore the response mechanism of the subgrade to the collapsible deformation of the foundation in the collapsible loess area, the deformation modes of the collapsible loess foundation at the bottom of the subgrade under different collapsible paths were studied by means of numerical simulation, and these research results were applied to the large-scale settlement model test. The research findings show that: The foundation deformation mode of the subgrade base under the condition of foundation subsidence is relatively complex, covering a variety of distribution forms such as linear, quadratic curve, trapezoidal, and “S” -shaped curve; for the subgrade structure with the same thickness under the linear, curved and trapezoidal settlement modes, the stress and deformation of the subgrade body show an increasing trend; before the stable soil arch is formed in the subgrade body, the tensile area and the core area inside it will change in distribution form and shift in position with the conversion of the settlement mode, and the stress level of the subgrade body increases with the increase in the thickness of the embankment; the deformation level on the top surface of the subgrade under the three settlement modes shows an increasing trend, and when the thickness of the lower embankment is 0, 2, and 4 m respectively, the maximum deformation amounts of each group are 10.32, 12.52 mm, and 2.25 mm; when the thickness of the lower embankment of the subgrade body reaches 4 meters, a stable soil arch is formed inside the subgrade body. Under the boundary disturbance of different settlement modes, stress release occurs in the subgrade body between the bottom core area and the arch rib area due to the supporting effect, and the stress level drops by nearly 60%, and the deformation level decreases by more than 80% compared with the situation where the soil arch has not been formed. The study on the response mechanism of subgrade bodies under the condition of foundation collapsibility is of great significance for the safe operation of subgrade engineering throughout its life-cycle.

Intense rainfall can induce strong seepage and suffusion within road subgrade soils, meanwhile dynamic traffic loads on road surfaces further exacerbate the development of subgrade suffusion. Such interactions have frequently caused urban road collapse. To address this issue, a coupled computational fluid dynamics-discrete element method (CFD-DEM) model was developed to simulate the initiation and progression of suffusion. A stepwise dynamic loading algorithm with fixed time-step was implemented to realize dynamic loading within the suffusion model. Furthermore, a custom-designed experimental platform was established, enabling the first visualization of suffusion under arbitrary waveform dynamic loading. The influence of dynamic loading on the internal stability of soil and suffusion progression was systematically investigated. The results indicate that dynamic loading disturbs the soil skeleton as well as generates localized high pore pressures, which significantly increase the drag force acting on the particles, driving them toward critical motion thresholds. The internal stability of soil degrades significantly under dynamic loads, with a noticeable reduction in the critical hydraulic head required to initiate suffusion. As the frequency of cyclic loading increases, disturbances to the soil structure and fluctuations in local hydraulic gradients intensify, leading to an exponential decline in the critical hydraulic gradient for initiating suffusion. The proposed CFD-DEM framework provides an effective analytical tool for simulating subgrade suffusion under dynamic loads. The identified suffusion mechanisms offer valuable insights for evaluating, predicting, and mitigating subgrade failures associated with suffusion.

The red clay, fly ash and guar gum has been demonstrated to improve the mechanical properties and reduce the permeability of coarse-grained filler in carbonaceous mudstone. The compression experiments of improved carbonaceous mudstone (ICM) under hydro-mechanical coupling conduction were conducted. A self-made hydro-mechanical coupling loading device was used to determine the stress-strain and permeability coefficient by establishing distinct initial fractal dimensions and axial water pressures. Concurrently, an unconfined compressive strength test was conducted to analysis its compressive properties. In conclusion, revealing the improvement mechanism from a microscopic point of view by SEM and grey correlation analysis was employed to comprehensively evaluate the factors influencing the mechanical and permeability properties of ICM. The results indicate that: Compared with the untreated carbonaceous mudstone, the mechanical properties of the ICM were significantly enhanced, and the fractal dimension and axial water pressure were negatively correlated with the mechanical properties. The increase in fractal dimension and axial water pressure is associated with a decline in the mechanical properties and an enhancement in their compressibility. The reduction of interparticle pores within the ICM lead to a decrease in the permeability of free water. The ICM permeability coefficient significantly decreased. The permeability coefficient decrease with the increased axial stress, and the fractal dimension is positively correlated with the permeability coefficient. The bonding and friction effects of the improver aggregates make the microstructure of the soil sample more compact and stable. The ICM unconfined compressive strength increased significantly by 345.45%, and the ICM compressive capacity was significantly enhanced. The fractal dimension and modifier exerted a substantial influence on the mechanical and permeability properties of carbonaceous mudstone, with the grey correlation coefficients of 0.73 and 0.65, respectively. The roadbed filling should maintain the fractal dimension within 1.78-2.60 to ensure continuous gradation of the filler. Meanwhile, the design of the drainage and waterproofing systems should be further optimized, and improvers with strong cementing capability and effective filling performance should be selected.

To study the micro- and meso-pore evolution and strength deterioration characteristics of recycled red brick (RB) and old concrete (OC) particles from construction and demolition waste and natural aggregate (NA) particles under freeze-thaw cycles, nuclear magnetic resonance scanning technology was used to obtain the T₂ curve distribution of aggregates with different freeze-thaw cycles. The T₂ curves were quantitatively characterized based on the multi-fractal theory. The laboratory uniaxial compression tests were used to test the compressive strength of three different types of aggregates. The correlation mechanisms between multi-fractal parameters and compressive strength of three different types of aggregates under different freeze-thaw cycles were then analyzed. The research results show that the porosities of NA, RB, and OC particles increased by 0.1%-0.3%, 1%-3.5%, and 1%-3% after 15 freeze-thaw cycles, respectively; the proportion of micro-pores of NA particles increased by 3%-6%, while the proportion of macro-pores decreased by 1%-5%; the proportion of macro-pores of RB particles increased by 4%-8%, while the proportions of meso- and micro-pores change less; the proportion of meso-pores of OC particles increased by 2%-6%, while the proportion of micro-pores decreased by 3%-8%. The compressive strength of NA, RB, and OC particles decreased by 7%, 72%, and 31%, respectively; and the compressive strength of NA particles was significantly correlated with the proportions of macro- and micro-pores, the compressive strength of OC particles was highly correlated with the proportions of macro- and meso-pores, and the compressive strength of RB particles was insignificantly correlated with the proportion of internal pores. The pore size distribution in the three different types of aggregate particles showed significant multi-fractal characteristics. In the low probability measurement area, the multi-fractal dimension values descend in the following order: RB>NA>OC; and in the high probability measurement area, the multi-fractal dimension values descend in the following order: OC>NA>RB. The Pearson's correlation coefficient values show that there exists significant correlations between the multi-fractal indices and porosities of NA and RB particles, while such correlations are insignificant for OC particles.

Centrifuge model tests and finite element analysis were carried out to study the deformation development characteristics and failure process of subgrade during earthquakes under the influence of rainfall infiltration, where the dynamic response characteristics of the subgrade under seismic action after rainfall, the development laws of displacement and strain, as well as the characteristics of instability and failure were investigated. Research results show that the previous rainfall causes the development of seepage and water migration within the subgrade, resulting in the significant dynamic response and further development of displacement of the subgrade. After the rainfall ends, the seepage inside the subgrade continues to develop towards the toe of the slope, during which the saturation of the soil at different positions changes. The deformation law and sliding characteristics of the subgrade induced by earthquakes are affected by the water migration caused by the previous rainfall. When seepage causes the soil near the toe of the subgrade to become saturated due to infiltration, the deformation and damage of the subgrade under the action of earthquakes are the most obvious. When the total amount and duration of prior rainfall are the same, as the rainfall intensity decreases, the infiltration rate of rainwater in the early stage of rainfall is higher, the development of seepage inside the subgrade accelerates, the migration of internal moisture is faster, the soil suction rapidly decreases and the soil shifts from the unsaturated state to the saturated state, causing the effective stress to continuously reduce and eventually leading to the larger scale of subgrade deformation caused by earthquakes as well as the wider range of sliding surfaces upon the occurrence of failure. Among the three cases featuring prior rainfall with decreasing, constant, and increasing intensity, under the same seismic loading, the cumulative deformation of the subgrade is the most significant when the prior rainfall has the decreasing intensity, followed by the constant rainfall intensity cases, and the cumulative deformation of the subgrade is the smallest when the prior rainfall has the increasing intensity.

To elucidate the evolution patterns of bike-sharing system demand and clarify the influence of various factors on demand, this study proposed a hierarchical spatio-temporal forecasting framework integrating feature decoupling and deep learning at both cluster and station levels. Firstly, an improved LP algorithm based on station interconnectivity and similarity was employed to construct a station clustering model. Interconnectivity was characterised by spatial distance and travel volume, while similarity was weighted by POI similarity and historical travel volume similarity. The borrowing-return imbalance index was introduced to evaluate clustering effectiveness, with station clustering performed to minimise this imbalance index. Subsequently, cluster-level and station-level demand prediction models based on the TFT model were established respectively, integrating cluster-level prediction results into the station-level prediction process. Finally, the SHAP method was employed to analyse the influence mechanisms of various factors on both cluster-level and station-level bike-sharing demand. The study shows that: during cluster-level demand prediction, hourly features exert the most significant influence on cluster-level demand, manifesting as night-time suppression and daytime promotion. Meteorological factors exhibit bidirectional moderation, promoting demand under conditions of moderate temperature, higher atmospheric pressure, and lower wind speed and humidity. At the station level, cluster demand emerges as the core determinant. Incorporating cluster-level results elevates the station-level prediction R² from 0.767 9 to 0.850 4, while the MAE decreased from 1.215 2 to 0.975 5, representing an error reduction of approximately 19.73%. Meteorological factors exhibits a certain degree of independent influence at the station level; demand may still be stimulated under certain low-temperature, high-humidity, or high-wind-speed conditions. This indicates that station-level demand not only depends on the overall fluctuations of cluster demand but is also influenced by the surrounding environment, demonstrating a degree of independence. This study constructs a bike-sharing demand prediction method based on a hierarchical spatio-temporal framework, providing decision support for identifying demand drivers and dynamic scheduling in bike-sharing systems.

To address the limitations of conventional stop identification methods, including their heavy reliance on nonstandardized POI (Points of Interest) classifications and dwell time, as well as their poor transferability, a novel semantic stop identification method was proposed in this study based on trajectory geometric features. By leveraging the inherent information embedded in trajectory data, the method systematically quantifies the overlapping patterns of subtrajectories near stop points while incorporating standardized POI data related to temporary stop points, thereby enabling effective differentiation between semantic stop points and temporary stop points. Furthermore, enterprise-level order data were employed to annotate trip purposes for the identified semantic stops and construct evaluation metrics, significantly mitigating subjectivity and complexity in the validation process. Finally, focusing on heavy-duty trucks, specifically container trailers, an empirical study was conducted on the Yangtze River Delta business area based on the trajectory and order data of container trailers provided by the transportation management system of a container transportation company. The results demonstrate a 96.5% accuracy rate for business address matching and an 85.5% validity rate for semantic stop identification, representing significant improvements over existing methods. These findings establish a robust foundation for large-scale multi-scenario applications. The results reflect the spatial distribution patterns and regional connectivity of port city freight distribution activities, along with behavioral heterogeneity across different heavy truck types, which can provide support for urban managers to formulate more comprehensive and reasonable freight policies and provide a basis for freight enterprises to manage fleets and expand value-added services.

In mixed-traffic environments, the interaction between cyclists and vehicles is critical for vehicle safety and has become a key factor in traffic safety. To enable advanced driver assistance systems (ADAS) to effectively identify such interactions, a method for interaction behavior recognition based on graph representation learning was proposed. This method integrated the skeletal information of cyclists and key information of non-motor vehicles, modeled the spatiotemporal characteristics of cyclists, and incorporated key information from vehicles to extract the interaction features between cyclists and vehicles. Different labeling methods were proposed for behaviors with varying complexities. For simple and direct basic interaction behaviors, manual annotation was used to generate behavior labels to ensure data quality. For more complex and variable interaction behaviors, a graph kernel-based clustering algorithm was employed to automatically generate behavior labels, addressing behaviors that were difficult to classify manually. A cyclist-vehicle interaction behavior graph model dataset was constructed based on the cyclist spatiotemporal graph model and behavior labels, and the recognition of interaction behaviors was accomplished using a graph kernel-based classification method. Real-vehicle data collection experiments were conducted to validate the effectiveness of this method. In the experiments, the system collected a large amount of actual cyclist-vehicle interaction behavior data, and behavior recognition was performed using the proposed method. The experimental results demonstrate that this method can identify various interaction behaviors between cyclists and vehicles with an accuracy of 99.65%. This high accuracy indicates that the method has significant practical value for improving the safety performance of intelligent driving systems, providing theoretical and practical support for further advancements in intelligent onboard systems.

In terms of the traffic safety issues caused by the speed limit change in work zone areas, this research aims to apply an in-vehicle audio warning to ensure traffic safety through improving driver's response behavior. Two-way four-lane expressway was selected as the test scenario, in which the design speed was 120 km·h^-1 and the speed limit of the work zone area was 80 km·h^-1. Meanwhile, the research considered sunny and foggy weather as two sight conditions, and the visibility under the two conditions was 300 m and 100 m respectively. Based on the driving simulation platform, 23 males (12 professional drivers vs. 11 normal drivers) and 21 females (11 professional drivers vs.10 normal drivers) were recruited to participate and complete the tests. Through analyzing the starting point of deceleration, the average deceleration, the standard deviation of speed, the percentage of speeding and the percentage of uncomfortable deceleration duration, it was found that: compared with normal drivers, the response behavior of professional drivers was more closely related to high risk, since they tended to take larger deceleration and higher percentage of uncomfortable deceleration duration. Although the start point of deceleration would be delayed in foggy condition, the percentage of speeding is lower than that in sunny condition because driver tended to take risk compensation behavior such as reducing speed and keeping a lower standard deviation of speed. The audio warning could advance the start point of deceleration and provide drivers with sufficient preparation time, thereby reducing the percentage of speeding and the uncomfortable deceleration duration in the response process. Moreover, the audio warning could make the distribution of the start point of deceleration more concentrated in the information release location, which is conducive to the traffic management department to carry out accident prevention works. The research can shed some lights on the management to the speed limit change in work zone areas.

Regarding the motion performance requirements of the new highway slope skeleton construction equipment when completing trajectory-based slotting and sliding mode construction tasks on inclined slopes, Firstly, through the analysis of the structure and working principle of the equipment, it was proposed to study the motion performance of the equipment with the adaptability of slope and slope length. Based on the characteristics of construction tasks, adaptability indicators and motion parameters were selected to establish an equipment adaptability evaluation system. Secondly, kinematic analysis and simulation of the equipment were conducted with adaptability evaluation index as the goal. The modified D-H (Denavit-Hartenberg) parameter method was used to establish a kinematic model of a 7-degree-of-freedom robotic arm, and the forward kinematic equations were obtained and solved. The working space of the end effector in the initial and reference fixed states was obtained through Monte Carlo method simulation. ADAMS (Automatic Dynamic Analysis of Mechanical Systems) was employed to establish a virtual prototype model of the equipment, simulating the overall motion state of the machine under typical construction conditions, and measuring the motion parameters of the end working mechanism considering the constraints of the construction slope and various components. Finally, a prototype construction test was conducted to complete the skeleton construction under specified slope conditions and verify the equipment adaptability. Results indicate that the maximum theoretical inclination angle of the main arm is -75.85°, and the theoretical range of longitudinal movement of the end effector along the slope surface is 9 990 mm. Under component and slope constraints, the longitudinal movement range of the end effector is 8 424 mm, the maximum depth into the slope surface is 290 mm, and the maximum turning radius is 2 375 mm. Research shows that the equipment has good adaptability to slope gradient, slope length, skeleton size, and shape, with motion performance meeting the requirements of complex construction tasks. This study provides a new perspective for kinematic research of equipment with multi-directional and large-scale movement characteristics in slope environments. The findings will help promote engineering applications of highway slope skeleton construction equipment and similar machinery.

Sampling-based rapidly exploring random tree (RRT) algorithms encounter sampling redundancy, low planning efficiency, and suboptimal path quality when performing motion planning for intelligent vehicles in complex obstacle environments. To address these challenges, this study proposes the DFNN-RRT, an optimal motion-planning method that integrates deep learning with the RRT algorithm. The framework employs an automatic point-cloud encoder to extract topological features and encode preprocessed high-dimensional data from complex obstacle scenes by embedding environmental point-cloud data into obstacle-space representations. By fusing these obstacle spatial features with the vehicle's initial and target poses and state-space variables, a spatiotemporal joint input representation was constructed to drive the planning network's training. During forward propagation, the network generates goal-oriented, informed sampling points through progressive optimization. A time-sensitive cost function was designed with path-point intervals as core parameters to quantify dynamic path-growth costs, and we applied a weighted mean squared error loss function between the posterior predicted states and target states to enable efficient model convergence via gradient optimization. Experimental results demonstrate that the probability of finding the global optimal solution peaks when the loss-rate threshold is set to 0.3; at 5 Hz planning frequency, the system minimizes time, and sampling scale shows a significant positive correlation with computational load. Under identical environmental configurations, the DFNN-RRT maintains the RRT's probabilistic completeness while optimizing key performance metrics: computational time, path length, curvature, and sampling density improve by 38.5%, 7.8%, 42.85%, and 52.5%, respectively, compared with the baseline RRT. A comparative analysis of biased RRT and bidirectional RRT demonstrated improvements of 27.1% and 12.2% in temporal efficiency, respectively. Simulations and real-world vehicle tests validate the capability of the method to generate smooth trajectories consistently in complex obstacle environments, confirming its temporal efficiency and reliability.

Trajectory prediction evaluates potential risks posed by surrounding traffic and provides safety guarantees for decision-making and planning in autonomous driving. To overcome the limitations of traditional trajectory prediction models that rely on manually calibrated spatial interaction features, this study proposes a vehicle trajectory prediction model based on a pattern-matching attention mechanism (PMA-LSTM). The proposed approach begins by preprocessing historical vehicle data, organizing it into training batches, and constructing an interaction index table for surrounding vehicles. Temporal features are then extracted using a Long Short-Term Memory (LSTM) network and enriched with kinematic features to enhance prediction accuracy. Predefined patterns with trainable parameters are employed to learn autonomously the potential impact of each pattern on an ego vehicle. A radial basis function is utilized to match attention across specific patterns for varying numbers of surrounding vehicles, whereas deep spatial interaction features are extracted using a convolutional neural network. Finally, a recurrent decoder module generates multi-modal future trajectories and constructs a spatiotemporal risk potential field as a safety constraint in the planning module to support optimal trajectory decision-making. The proposed algorithm was validated on real-world datasets, including the NGSIM and a collected simulation dataset. Subsequently, it was tested for real-time performance and case analysis using the CARLA-SUMO co-simulation platform constructed with a real-world base map. Experimental results demonstrate that PMA-LSTM outperforms competing models in negative log-likelihood error, root mean square error, and final displacement error, achieving reductions of 11.4%, 9.2%, and 14.1%, respectively, over a 5-second prediction horizon compared to CS-LSTM(M). Furthermore, the multimodal output exhibited a 16.06% reduction in negative log-likelihood error compared with the unimodal output. The proposed trajectory prediction algorithm demonstrates superior prediction accuracy, more precise intention recognition, and better alignment with real-world planning requirements, thereby providing reliable and safe support for optimal trajectory planning in autonomous driving.

Path planning is a core technology in intelligent vehicles, placing higher demands on the driving safety and stability of intelligent vehicles in obstacle avoidance scenarios. A dual-layer path planning method that informs tire-road friction limits is proposed for obstacle avoidance scenarios. An improved rapidly-exploring random tree (IRRT) algorithm was developed for upper-layer preplanning. To balance the search efficiency and path accuracy, the expansion step size of the random tree was adapted to the longitudinal velocity of the vehicle. Constraints based on the tire-road friction limits were then applied to the expansion angles of sampling points. Random sampling was performed within these constraints. Moreover, an artificial potential field method was combined to construct attractive potential fields at the sampling and target points, as well as a repulsive potential field at the road boundary. In this manner, the position of the random sampling point was modified. To further improve the smoothness of the path and ensure vehicle stability, nonlinear model predictive control (NMPC) was employed to optimize the preplanned path within the prediction horizon in lower-layer replanning. A vehicle plane motion model and tire-road friction coefficient constraints were established to generate a replanned path. Finally, the proposed path-planning algorithm was validated in both static and dynamic obstacle avoidance scenarios using CarSim/Simulink co-simulation and hardware-in-the-loop test platforms. The results indicate that the IRRT algorithm in the preplanning stage offers advantages over the RRT^* algorithm in terms of search efficiency, path feasibility, and vehicle stability. Compared with single-layer IRRT and traditional NMPC methods, the IRRT-NMPC dual-layer path planning method further ensures that the vehicle remains within tire-road friction limits and enhances its stability and safety while meeting real-time requirements.