最新刊期
- 摘要：Objective The rapid urbanization in China has generated over 3.5 billion tons of construction waste annually and enormous industrial solid wastes (fly ash, slag, red mud), with cumulative stockpiles exceeding 20 billion tons, severely constraining green development. Simultaneously, the cement industry contributes ~8% of global anthropogenic CO₂ emissions. Geopolymer recycled aggregate concrete (GRAC), which replaces ordinary Portland cement (OPC) with alkali‑activated solid‑waste precursors and natural aggregates with recycled concrete aggregates (RCA), offers a triple solution: waste disposal, low‑carbon building material production, and CO₂ mitigation. This review systematically synthesizes recent advances in GRAC, focusing on three interconnected themes – cementitious chemistry and interfacial science, performance optimization, and low‑carbon technologies – to identify scientific bottlenecks and guide future engineering implementation.Progress Si/Al regulation and gel chemistry. The Si/Al molar ratio was established as the most critical parameter governing geopolymerization kinetics, gel nanostructure, and mechanical strength. Optimal Si/Al ratios generally fell within 1.5–2.5, though the exact value shifted with precursor reactivity, activator type, and admixtures. Low‑calcium systems produced N–A–S–H gels, while high‑calcium slag‑based systems formed C–(A)–S–H gels; the interplay between Si/Al, Ca content, and precursor activity determined the dominant gel type and final performance.Conclusions and Prospects This review confirmed that GRAC possessed superior interfacial density and significantly lower GWP than conventional RAC, yet several bottlenecks hindered large‑scale application: (i) multi‑solid‑waste synergy required precise regulation and standardization via rapid characterization–reactivity prediction–mix adjustment closed loops; (ii) long‑term durability prediction under coupled environmental actions (freeze–thaw, wet–dry, chloride, carbonation) demanded in‑situ X‑CT and nanoindentation to model ITZ evolution; (iii) ML integration with materials genomics needed open, multi‑scale, standardized databases that bridged macro‑ and microstructural features; (iv) hybrid fiber systems beyond binary blends, including natural plant fibers (jute, sisal, basalt), warranted systematic investigation for interfacial compatibility and long‑term performance; (v) accelerated carbonation technology required industrial scale‑up (batch uniformity, flue gas adaptability) and unified carbon accounting for carbon trading; (vi) structural‑component‑level studies (beams, columns, slabs, joints) were essential to establish design codes and standards. Overall, GRAC represented a paradigm‑shifting material that converged waste recycling, low‑carbon manufacturing, and intelligent design, providing a viable pathway toward sustainable infrastructure. This review offered a critical and forward‑looking synthesis to catalyze both fundamental research and engineering adoption in the global pursuit of carbon neutrality.关键词：Geopolymer recycled aggregate concrete;interface transition zone;Alkali activation;Fiber reinforcement;machine learning;Life cycle assessment;CO2 mineralization1|0|0更新时间：2026-07-17
- 摘要：Significance During dynamic polymer processing operations such as extrusion, melt mixing, injection molding, fiber spinning, and ball milling, polymer chains are subjected to complex mechanical fields involving shear, extension, and compression. These mechanical forces induce chain stretching, conformational evolution, and, under sufficiently high stress, covalent bond scission, giving rise to a series of mechanochemical processes. Since early studies demonstrated that mechanical processing causes irreversible reductions in polymer molecular weight, processing mechanochemistry has attracted sustained attention because of its critical role in determining processing stability and polymer material performance. Force-induced bond scission has traditionally been regarded as the primary cause of polymer degradation during processing. However, the reactive intermediates generated by bond cleavage, such as polymer-centered radicals, also provide unique opportunities for grafting, crosslinking, functionalization, and controlled degradation. Therefore, elucidating the mechanochemical mechanisms operating during polymer processing is important not only for preventing undesirable degradation but also for transforming mechanochemistry into a versatile strategy for polymer modification, sustainable recycling, and value-added upcycling. Nevertheless, these mechanisms remain insufficiently understood because practical polymer processing involves the simultaneous coupling of complex flow fields, thermal effects, oxidation, and heterogeneous processing environments. This review comprehensively summarizes recent progress in elucidating the mechanisms of force-induced chain scission under processing conditions through investigations of real processing operations, equivalent mechanochemical platforms, and molecular dynamics simulations. It further reviews processing-enabled polymer functionalization and emerging mechanochemical strategies for polymer degradation, recycling, and upcycling. Finally, the remaining challenges and future opportunities for integrating mechanochemistry with polymer processing are discussed to facilitate the rational design, advanced manufacturing, and sustainable utilization of next-generation polymeric materials.Progress Recent studies have significantly advanced the understanding of polymer mechanochemistry during processing from both mechanistic and application-oriented perspectives. First, considerable progress has been made in elucidating force-induced bond scission under processing conditions. Both industrial processing operations, such as extrusion and melt mixing, and equivalent mechanochemical platforms, including ultrasonication, ball milling, and model flow devices, have been widely employed to investigate processing-induced chain scission. These studies have systematically revealed how processing conditions and molecular characteristics, including flow type, deformation rate, molecular weight, molecular architecture, and polymer concentration, govern stress transmission and polymer chain scission. Advanced characterization techniques, including rheological analysis, molecular-weight determination, and electron spin resonance spectroscopy, have enabled increasingly quantitative evaluation of mechanochemical degradation. In particular, the incorporation of mechanophore-based probes into polymer chains has enabled the direct visualization of force-induced bond scission during processing. Meanwhile, molecular dynamics simulations and multiscale theoretical models have provided mechanistic understanding at the molecular level. Together, these complementary experimental and computational approaches have established a more comprehensive mechanistic framework linking macroscopic processing conditions with molecular-scale mechanical activation.Conclusions and Prospects polymer processing mechanochemistry has evolved from empirical observations of processing-induced degradation to a molecular-level understanding of force-induced bond scission, and further to processing-enabled polymer functionalization, controlled degradation, and chemical upcycling. Despite these advances, several key challenges remain. First, quantitative relationships between processing conditions, molecular stress distribution, and bond scission behavior have yet to be established, largely because direct observation of molecular events during processing remains difficult. Developing non-invasive in situ characterization techniques capable of monitoring chain rupture, radical generation, and subsequent reaction pathways will therefore be essential. Second, mechanochemical mechanisms under realistic processing conditions, where thermal, mechanical, oxidative, and chemical effects are strongly coupled, remain insufficiently understood, highlighting the need for integrated experimental, theoretical, and multiscale simulation approaches. Third, extending processing mechanochemistry beyond conventional synthetic polymers to structurally complex biomacromolecules, such as cellulose, chitosan, proteins, and DNA, may reveal new force–structure relationships and broaden its applications in sustainable materials and bio-related systems. Finally, translating laboratory-scale mechanochemical strategies into industrial polymer processing requires improved reaction controllability, scalability, and compatibility with existing manufacturing technologies. Addressing these challenges will enable more precise mechanochemical regulation during polymer processing and accelerate the development of high-performance polymeric materials, sustainable manufacturing, and efficient polymer recycling and upcycling.关键词：polymer processing;mechanochemistry;flow field;bond scission;degradation and recycling8|0|0更新时间：2026-07-17
-
- 摘要：ObjectiveRock-mass grade assessment at the tunnel face provides an important basis for excavation scheme optimization, support parameter adjustment, and construction risk control in deep underground engineering. Manual assessment based on close-range geological observation, field sketching, empirical judgment, and standardized evaluation is affected by operational risk, low efficiency, subjectivity, and inconsistent judgment. Autonomous unmanned ground vehicle (UGV) inspection can reduce personnel exposure and support close-range tunnel-face information acquisition, but deep underground tunnel environments involve Global Navigation Satellite System (GNSS) denial, narrow spaces, dust interference, low illumination, and dynamic obstacles such as workers and construction equipment. These conditions require the UGV to predict the future motion of dynamic obstacles, avoid potential collisions, acquire tunnel-face red-green-blue-depth (RGB‒D) images, and provide reliable image input for rock-mass grade classification. This study developed an autonomous UGV inspection workflow integrating multimodal trajectory prediction, prediction-based dynamic obstacle avoidance, tunnel-face RGB‒D image acquisition, intelligent rock-mass grade classification, and explainable result visualization.MethodsA multimodal trajectory prediction network, named Endpoint-Prototype-Guided Trend-State Hybrid Network (EP-TSHNet), was constructed to predict dynamic obstacle motion. Historical trajectories were transformed into relative displacement sequences, and the prediction setting adopted 8 observed frames, 12 future frames, and 20 candidate trajectories. EP-TSHNet included a trend-state temporal encoder, a hybrid graph interaction encoder, and an endpoint-prototype-guided hierarchical multimodal decoder. The trend-state temporal encoder extracted local displacement changes, velocity changes, acceleration changes, and terminal motion trends from historical trajectories. The hybrid graph interaction encoder modeled spatial proximity relationships and motion-feature similarity relationships among dynamic targets. The endpoint-prototype-guided decoder generated representative future endpoint prototypes, branch-level motion variations, anchor trajectories, residual refinements, and modal probabilities. The training objective combined multimodal trajectory distribution loss, endpoint auxiliary supervision, mean displacement auxiliary supervision, endpoint prototype matching, and prototype separation constraints. Based on EP‒TSHNet, a trajectory-prediction-based dynamic obstacle avoidance method, EP‒TSHNet‒PRA, was designed. Dynamic targets were detected and tracked from laser sensing data, and their historical trajectories were used for online trajectory prediction. The predicted trajectories were transformed into look-ahead risk information and introduced into local cost map updating, collision-risk corridor assessment, and safe velocity constraints. Dynamic obstacle avoidance experiments were conducted on a Robot Operating System (ROS)/Gazebo tunnel simulation platform using a Husky A200 UGV. Low-, medium-, and high-density dynamic obstacle scenarios were constructed, and a closed-loop inspection task was designed, including start-point departure, tunnel dynamic obstacle avoidance, tunnel-face target-region arrival, RGB-D image acquisition, return-trip obstacle avoidance, and final arrival at the start point. For tunnel-face rock-mass grade classification, a confusion-aware RepViT‒based rock classifier, CA‒RepViT‒Rock, was developed. The network used RepViT‒Rock as the feature extraction backbone and incorporated multi-scale depthwise convolution, the Convolutional Block Attention Module (CBAM), a multi-pooling cosine classification head, and confusion-category refinement branches. A tunnel-face image dataset from an under-construction tunnel project in northwest China was used. The dataset contained 1 800 high-resolution tunnel-face images of four rock-mass grades, including Ⅲ‒1, Ⅲ‒2, Ⅳ, and Ⅴ, and was divided into training, validation, and test sets at a ratio of 70%, 20%, and 10%. Accuracy, Precision, Recall, F1‒score, Macro‒F1, Weighted‒F1, and confusion matrices were used for evaluation. Gradient-weighted Class Activation Mapping (Grad‒CAM), Grad‒CAM++, and intermediate-layer feature responses were used to visualize model attention regions.Results and DiscussionsEP‒TSHNet achieved minADE20/minFDE20 values of 0.17/0.30 on ETH-UCY, 7.66/13.20 on the Stanford Drone Dataset (SDD), and 0.11/0.28 on INTERACTION, indicating its applicability to pedestrian interaction, aerial-view complex motion, and vehicle interaction scenarios. On SDD, compared with TDAGCN, EP‒TSHNet reduced minADE20 and minFDE20 by approximately 12.06% and 13.39%, respectively. On INTERACTION, compared with EPHGT, EP‒TSHNet reduced minADE20 and minFDE20 by approximately 26.67% and 33.33%, respectively. The ablation experiment on SDD showed that removing any key module or constraint degraded prediction performance. Removing the endpoint-prototype-guided hierarchical multimodal decoder increased minADE20/minFDE20 from 7.66/13.20 to 8.76/15.34, showing the importance of endpoint prototypes and branch expansion for multimodal intention coverage. Removing the hybrid graph interaction encoder increased the errors to 8.21/14.44, indicating the contribution of target interaction modeling. Removing the trend-state temporal encoder, endpoint auxiliary supervision, mean displacement auxiliary supervision, or prototype separation constraint also increased prediction errors. In the dynamic obstacle avoidance experiments, the conventional method without look-ahead prediction achieved success rates of 80.0%, 65.0%, and 35.0% in low-, medium-, and high-density scenarios, respectively, with an average success rate of 60.0%. EP-TSHNet-PRA achieved success rates of 100.0%, 95.0%, and 75.0% under the same conditions, and the average success rate increased to 90.0%. The results show that predicted trajectory cost regions and safe velocity constraints enabled the local planner to avoid future occupied areas before dynamic obstacles entered the immediate collision region. In the closed-loop inspection task, the UGV completed autonomous navigation, dynamic obstacle avoidance, RGB-D image acquisition, and return navigation under dynamic obstacle density changes and dust interference. For rock-mass grade classification, CA-RepViT-Rock achieved an Accuracy of 92.77%, a Macro-F1 of 92.11%, and a Weighted-F1 of 92.75%. The F1-scores of grades Ⅲ‒1, Ⅲ‒2, Ⅳ, and Ⅴ were 0.967 7, 0.875 0, 0.944 4, and 0.897 4, respectively. The normalized confusion matrix showed correct recognition rates of 97%, 86%, 95%, and 91% for grades Ⅲ‒1, Ⅲ‒2, Ⅳ, and Ⅴ, respectively. Misclassifications mainly occurred between adjacent grades, reflecting the continuous transition and fuzzy boundaries of rock-mass features in tunnel-face images. Grad‒CAM and Grad‒CAM++ visualization showed that the model mainly focused on rock texture, fractures, broken zones, structural planes, and texture transition regions, which were consistent with geological features used in rock-mass grade assessment.ConclusionsThe results demonstrate that EP-TSHNet can provide multimodal trajectory prediction for dynamic obstacles by integrating short-term motion trend extraction, target interaction modeling, and endpoint-prototype-guided multimodal decoding. EP-TSHNet-PRA transforms predicted trajectories into look-ahead risk information for local cost map updating, collision-risk assessment, and safe velocity control, thereby improving UGV dynamic obstacle avoidance in tunnel simulation scenarios. CA-RepViT-Rock can classify tunnel-face rock-mass grades from RGB images and provide explainable visual evidence through Grad‒CAM++. The developed workflow connects autonomous UGV navigation, dynamic obstacle avoidance, tunnel-face RGB-D information acquisition, rock-mass grade classification, and explainable visualization. The results provide a methodological basis for autonomous UGV inspection, safe tunnel navigation, close-range tunnel-face information acquisition, and intelligent rock-mass grade assessment in deep underground engineering.关键词：deep underground engineering;UGV inspection;trajectory prediction;Dynamic obstacle avoidance;intelligent rock-mass grade classification36|27|0更新时间：2026-07-02
- 摘要：Objective With the strategic focus of transportation infrastructure construction in China shifting toward western mountainous and karst regions characterized by extremely complex topographic and geological conditions, tunnel construction is increasingly exposed to severe water and mud inrush hazards. When mountain tunnels pass through water-rich karst zones, excavation-induced disturbance and the exposure of concealed karst cave may reactivate groundwater hazard sources, thereby triggering sudden geological disasters such as water and mud inrush. The water-resistant rock mass between the tunnel face and the karst cave serves as a critical load-bearing structure against karst water pressure, with its stability controlling tunnel face stability and the prevention of water and mud inrush disasters. Once the water-resisting rock mass in front of the tunnel face loses its bearing capacity, high-pressure karst water and filling materials may rapidly invade the tunnel, resulting in sudden water and mud inrush disasters. Therefore, this study aims to reveal the instability mechanism of water-resistant rock mass containing random fractures and to determine the minimum safety thickness required to prevent instability failure under karst water pressure, thereby providing design guidance for the construction of karst tunnels.Methods A mechanical model of a rectangular water-resisting rock slab was established. The governing deflection equation was derived via the minimum energy method. Subsequently, the maximum tensile stress criterion was adopted as the failure condition, yielding an analytical solution for the minimum safety thickness of the water-resistant rock mass. Furthermore, sensitivity analyses were performed on key parameters, including karst water pressure, tunnel height, tunnel width, and the tensile strength of the rock mass. The influence law of all parameter was consistent with the mechanical mechanism, validating the rationality and applicability of the proposed theoretical model. In addition, considering that natural rock mass commonly contained randomly distributed fractures, a DFN-DEM numerical model of the water-resisting rock mass containing random fractures was established. The generation and structural control of the fracture network were implemented using the built-in DFN generation command "Fracture generate" in PFC. The fracture dip angles were described by a normal distribution, the fracture length L was assumed to follow a lognormal distribution, and the fracture density was quantified using P21. The displacement responses of monitoring points and crack propagation processes within the water-resisting rock mass were monitored and quantitatively analyzed, thereby revealing the damage evolution law and instability failure mechanism of the water-resisting rock mass containing random fractures under karst water pressure. Finally, a fracture influence coefficient η was introduced to quantitatively evaluate the weakening effect of random fractures on rock mass stability. Based on the numerical results, the relationship between η and fracture density P21 was established through regression analysis, revealing the effects of fracture development on the bearing capacity and minimum safety thickness of the water-resisting rock mass.Results and Discussions Theoretical analysis showed that the deflection reached its maximum at the slab center and gradually decreased toward the edges, approaching zero near the boundaries, which was consistent with the deformation characteristics under four-edge fixed boundary conditions. Parameter sensitivity analysis showed that the minimum safety thickness of the water-resisting rock mass was positively correlated with karst water pressure, tunnel height, and tunnel width, while it was negatively correlated with the tensile strength of the rock mass. Numerical simulations showed that the deformation and failure of the fractured water-resisting rock mass under karst water pressure were characterized by a staged evolution process. Under karst water pressures of 0.50-1.10 MPa, the monitoring-point displacement followed a characteristic evolution of rapid growth, gradual increase, and eventual stabilization. However, when the water pressure increased to 1.18 MPa, the displacement response changed from gradual growth to rapid acceleration, marking the transition of the water-resisting rock mass from stable deformation to instability. The crack evolution results indicated that cracks initially initiated at the tips of random fractures, subsequently propagated along pre-existing fractures and gradually coalesced, with the deformation and failure of the rock mass being primarily controlled by the rock bridges between the random fractures. Concretely speaking, maximum tangential stress induced 45° crack initiation in the upper part of the water-resisting rock mass, followed by coalescence along pre-existing fractures. Meanwhile, the lower part underwent bending failure due to overall toppling deformation, with compressive fracturing developing near the lower side of the tunnel face. A further regression analysis based on the numerical simulation results showed that the fracture influence coefficient η was positively and exponentially correlated with fracture density P21, indicating that increasing fracture density significantly enhanced the weakening effect of random fractures on the stability of the water-resisting rock mass.Conclusions This study combined theoretical analysis and numerical simulation to evaluate the stability of water-resisting rock mass between the tunnel face and the karst cave and determined its minimum safety thickness. First, an analytical solution for the minimum safety thickness of an intact rectangular water-resisting rock slab was derived. On this basis, parametric analysis showed that the minimum safety thickness of the water-resisting rock mass increased with karst water pressure, tunnel height, and tunnel width, but decreased with increasing tensile strength of the rock mass. Subsequently, a DFN-DEM model incorporating random fractures was established to investigate the effects of fracture development on deformation response, crack evolution, and progressive failure mechanisms. For fractured water-resisting rock mass, the deformation response and crack evolution were primarily governed by the rock bridges between adjacent fractures. Specifically, cracks preferentially initiated at fracture tips and local stress concentration zones within rock bridges, and subsequently propagated along pre-existing fracture paths until coalescence occurred. The failure process was characterized by a progressive instability mechanism involving crack-tip initiation, fracture propagation, rock-bridge coalescence, and bending-induced tensile cracking at fracture ends. Finally, regression analysis established a quantitative relationship between the fracture influence coefficient η and fracture density P21, which could be used to characterize the effect of fracture development on the required safety thickness of the water-resisting rock mass. The findings provided theoretical support and engineering reference for tunnel face stability assessment, water-resisting rock thickness design, and water-mud inrush prevention in water-rich karst tunnels.关键词：karst tunnel;water-resisting rock mass;minimum safety thickness;random fractures;numerical simulation;regression fit23|28|0更新时间：2026-06-30
- 摘要：Objective Conventional triaxial tests can only measure the bulk average deformation of soil specimens and fail to quantify the non-uniform deformation induced by end restraint. To address this limitation, a photogrammetric technique based on Ringed Automatically Detected (RAD) coding targets was adopted to establish a refined method for measuring the local three-dimensional deformation of sand-clay mixtures. The influence of coarse grain content (CGC) on the non-uniform deformation characteristics of triaxial specimens was revealed, providing a quantitative basis for mitigating end-restraint effects in geotechnical laboratory tests.Methods Consolidated-drained triaxial tests were conducted on sand-clay mixtures with seven coarse grain content (CGC) levels: 10%, 20%, 30%, 40%, 50%, 60% and 70%. Confining pressures of 100, 200 and 300 kPa were applied. Specimens were fabricated with consistent dimensions of 80 mm in height and 39.1 mm in diameter. Shear tests were conducted at a constant displacement rate of 0.004 mm/min. A photogrammetric system was established for non-contact full-field deformation monitoring. A total of 162 Ringed Automatically Detected (RAD) coded targets were evenly attached to the rubber membrane of each specimen in a 9×18 grid. Prior to testing, approximately 120 RAD coded targets were attached to the loading frame, the upper and lower surfaces of the triaxial cell, and three vertical columns to establish a unified world coordinate system. Optical distortion correction was performed to eliminate lens-induced systematic errors. A camera was mounted on a circular rail with 30 uniformly distributed shooting positions, and the horizontal rotation angle between adjacent positions was set to 12° to ensure an image overlap of no less than 85%. Loading was paused at each target axial displacement (1, 2, 3, …, 12 mm), and images were captured under constant strain conditions. The three-dimensional coordinates of surface points were continuously captured, identified, tracked and calculated during shearing. An axial strain non-uniformity coefficient(Cvσ) and a radial strain non-uniformity coefficient (Cvr) were defined based on the intra-layer strain standard deviation and inter-layer strain difference. These coefficients characterized the deformation dispersion and spatial heterogeneity along the specimen height. Three zones (central 3 layers, central 5 layers and central 7 layers) were compared. For specimens of 80 mm×39.1 mm, the central 20 mm height (central 3 layers) was determined as the representative zone for local stress-strain analysis, as this zone was least affected by end friction. Data from this zone were used to mitigate the adverse effects of end restraint on deformation measurement.Results and Discussions All deformation data were acquired from the central three layers of specimens using photogrammetry. A distinct mechanical threshold of coarse grain content (CGC) was identified in the range of 30%-40% for sand-clay mixtures. This threshold governed the deformation mode, stress-strain response and strength behavior of the mixtures. For specimens with CGC below this threshold, the clay matrix acted as the predominant load-bearing component, and typical strain-hardening behavior was observed during triaxial shearing. Axial strain fluctuated slightly, while radial strain increased gradually. Uniform deformation behavior was recorded within the selected representative region. When CGC exceeded the threshold, sand particles formed a continuous interlocking skeleton to bear external loads. The radial strain increased significantly and reached 19% at a CGC of 70%. The stress-strain curves transformed from strain-hardening to strain-softening, and the dependence of peak deviator stress on confining pressure was enhanced. For instance, the peak deviator stress of specimens with a CGC of 70% under 300 kPa was approximately 1.4 times that under 100 kPa. A comparative analysis was conducted between photogrammetric measurements and conventional triaxial test results. The radial strain measured by photogrammetry was 1.73%-3.34% higher, while the axial strain was 0.24%-0.48% lower, and the failure deviator stress was 2.91%-10.71% lower. These discrepancies originated from the inherent limitations of conventional testing apparatus. Traditional equipment only monitored bulk deformation through external displacement sensors and did not capture the local bulging at the mid-height of the specimen. As a typical heterogeneous deformation induced by end restraint, local bulging could not be detected by conventional triaxial devices. The neglect of this deformation characteristic led to systematic underestimation of radial deformation and overestimation of deviator stress. The results confirmed that selecting the central three layers (20 mm in height) mitigated the interference of end restraint and improved the measurement accuracy of mechanical parameters. The photogrammetric method with RAD coded targets effectively characterized the local deformation behavior of sand-clay mixtures.Conclusions The photogrammetric technique using Ringed Automatically Detected (RAD) coded targets was employed to quantify the local three-dimensional deformation characteristics of sand-clay mixtures during triaxial shearing. The defined axial and radial strain non-uniformity coefficients (Cvσ and Cvr) were used to identify the optimal analysis zone with a central height of 20 mm, where uniform deformation behavior was observed and the influence of end restraint was significantly reduced. A distinct mechanical threshold of coarse grain content (CGC) was identified in the range of 30%-40%, which controlled the deformation mode and strength characteristics of sand-clay mixtures. The selection of the central three layers as the representative analytical region improved the monitoring scheme of triaxial tests and reduced measurement errors induced by end restraint. The research findings provide reliable experimental references for refined mechanical testing and deformation analysis of sand-clay mixtures in geotechnical engineering.关键词：triaxial test;strain inhomogeneity coefficient(Cvσ、Cvr);photogrammetry;sand-clay mixtures;end constraint14|33|0更新时间：2026-06-30
- 摘要：Objective The energy deposition rate in plasma channels formed by liquid-phase discharges critically influences rock fragmentation dynamics and energy utilization efficiency. However, the combined effects of charge voltage and electrode spacing on this rate, and its quantitative description, had not been systematically clarified. This study aimed to reveal the synergistic influence of these two parameters, characterize the temporal stages of energy deposition, and develop a predictive model for precise energy control in liquid-discharge rock breaking.Methods A transient voltage–current measurement system for plasma channels formed by liquid-phase discharge was established, consisting mainly of a charge unit, a discharge unit, and a measurement unit. Experiments were conducted with charge voltages of 6–9 kV and electrode spacings ranging from 5 mm to the critical breakdown gap. Based on the voltage and current time-history data measured in each test, the deposition energy curve for a single discharge was obtained by integrating the instantaneous power. The average energy deposition rates of the fast deposition stage and the slow deposition stage were determined from the slopes of the corresponding linear segments on the energy–time curve. The influences of charge voltage and electrode spacing on these deposition rates were then analyzed.Results and Discussions The energy deposition time-history curves of the plasma channel exhibit a pronounced two-stage characteristic, which can be divided into a fast deposition stage and a slow deposition stage according to the distinct difference in their slopes. The fast stage features an extremely high deposition rate and a very short duration, whereas the slow stage is characterized by a drastically reduced rate but a significantly prolonged duration. For charge voltages of 6–9 kV, the average energy deposition rates of the fast stage ranged from 9.03–19.91 MJ/s at 6 kV, 8.68–27.42 MJ/s at 7 kV, 20.94–51.37 MJ/s at 8 kV, and 18.91–61.35 MJ/s at 9 kV; the corresponding rates of the slow stage were 0.81–2.12 MJ/s, 0.95–2.61 MJ/s, 1.08–4.57 MJ/s, and 2.55–7.55 MJ/s. Analysis of the current waveforms indicated that the duration of the fast stage was determined by the width of the main current pulse: the main pulse width remained nearly constant with rising charge voltage but varied with electrode spacing due to changes in circuit damping and discharge mode. The duration of the slow stage depended on the decay time of the current tail; the charge voltage had little influence on this characteristic time, whereas the electrode spacing caused its change to lose monotonic regularity by affecting plasma cooling and deionization processes. From the perspective of rock fragmentation mechanisms, the ultra-high deposition rate of the fast stage created a steep shock-wave front that pulverized the rock around the borehole wall to form a crushed zone, while the subsequent slow stage sustained bubble pulsation and quasi-static loading, driving cracks to propagate outward and creating fractured and elastic vibration zones. The rates of the two stages and their relative proportions determined the peak load, rise time, and energy partitioning of the shock wave, thereby directly controlling the extents of the crushed and fractured zones and the overall energy utilization efficiency. The charge voltage and electrode spacing jointly determined the energy deposition rate by regulating the electric field strength: increasing the charge voltage not only produced a quadratic growth in capacitive energy storage, but also enhanced the average electric field, thereby accelerating pre-breakdown, raising the driving voltage of the main discharge, and significantly increasing the deposition rate; conversely, enlarging the electrode spacing weakened the average electric field, lengthening the pre-breakdown path and time, raising the breakdown threshold, and suppressing the development speed and instantaneous power of the plasma channel during the main discharge, ultimately causing the deposition rate to decline. The non-monotonic variation of the deposition energy with electrode spacing can be explained by two competing effects: the positive effect is that a larger gap increases the plasma channel resistance, which promotes Joule heating deposition; the negative effect is that a larger gap markedly increases the difficulty of channel formation, leading to greater energy consumption during the pre-breakdown stage for establishing a conductive channel. The initial increase and subsequent decrease of the fast-stage energy as the gap widens is a direct result of the shifting balance between these positive and negative effects across different gap intervals.Conclusions These findings clarify the synergistic regulation of the energy deposition rate by charge voltage and electrode spacing in liquid-phase discharge plasma channels and establish the two-stage nature of the energy deposition process. Under identical conditions, the fast-stage deposition rate was approximately eleven times that of the slow stage. The average energy deposition rates of both stages increased linearly with increasing charge voltage and decreased linearly with increasing electrode spacing. The fast-stage rate exhibited independent linear dependences on the two parameters, resulting in a planar distribution, based on which a bivariate linear predictive model was constructed. The deposited energy in both stages increased monotonically with charge voltage; with increasing electrode spacing, the fast-stage energy first increased and then decreased, whereas the slow-stage energy decreased monotonically. The fast stage contributed 66.1%–100% of the total deposited energy, dominating the energy injection, with its fraction approaching 100% at the critical electrode spacing and the energy ratio between the two stages increasing accordingly. Consequently, the proposed model provides an effective tool for predicting and actively tuning the fast-stage deposition rate, and the energy deposition rate, as a critical link between electrical input parameters and rock dynamic response, can be optimized to achieve on-demand fragmentation zones and improved energy efficiency.关键词：Liquid phase discharge;Energy deposition rate;Deposition energy;Charge voltage;Electrode spacing21|24|0更新时间：2026-06-30
- 摘要：Objective The Yebatan Hydropower Station is the largest hydropower project in the upper reaches of the Jinsha River, and serves as a landmark engineering achievement contributing to China's "West-to-East Power Transmission" and "dual carbon goals". The construction of this project faced extreme challenges under high altitude, a frigid climate, and complex geological condition. This paper systematically summarizes the key technological innovations and engineering practices implemented to ensure the safe construction of the ultra-high arch dam.Methods Firstly, in face of the challenge of foundation surface selection for ultra-high arch dam under the influence of deep unloading zone, the comprehensive quality assessment of rock mass was conducted. Based on the assessment results, the foundation surface scheme was compared and selected based on engineering geology, technical difficulties, and engineering investment. Secondly, focus on the geological defects at dam foundation, numerical simulations were conducted to study the impact of geological defects on the safety and stability of dam. Referring to the experience of similar ultra-high arch dam projects, geological defect disposal measures were determined, including concrete replacement and consolidation grouting. Numerical simulations and geomechanical model tests were conducted to research the function of disposal measures. The mechanical behavior of dam body was analyzed, and the dam safety was evaluated. Then, faced with the safety issue of ultra-high arch dam under high seismic intensity and complex geology, a comprehensive dynamic response analysis of the dam was carried out. Arch beam load distribution method, nonlinear finite element method, and rigid limit equilibrium method were employed. Based on the analysis results, seismic measures for the dam were designed, and targeted measures were proposed for potential risk areas to ensure safety. Finally, a temperature control standard for dam concrete was established to address the issue of temperature control and crack prevention in extremely cold environment. Numerical simulations of concrete pouring as well as field concrete model tests were conducted. Based on the results, temperature control and crack prevention measures for dam concrete were determined, and a refined control system with polyurethane insulation and staged water cooling technology as the core was established.Results and Discussions The complex geology and strong unloading of this project had an impact on the rock mass quality. Within Class III and Class IV rock mass at the dam area, deep unloading zone was regarded as independent geological unit for rock mass classification. The moderate unloading zone was classified as Class III2s, and the strong unloading zone was classified as Class IVs. This was the unique feature of the rock mass classification in this project compared to other projects. The comprehensive selection results of dam foundation showed that the EX2 scheme could reduce the span of arch ring, decrease the water pressure, and reduce the concrete amount and foundation excavation. So EX2 is the recommended arch dam foundation surface scheme. The disposal measures taken for faults included concrete replacement, consolidation grouting, and groove replacement. Measures such as deep consolidation grouting and concrete replacement were adopted for deep unloading zone. Measures such as concrete replacement and resistance anchor reinforcement were adopted for crushed zones. The anti-seepage and drainage measures were a combination of multi-layer curtain drainage holes and resistance body drainage holes. After taking the above measures, the displacement of the dam was reduced, and the mechanical state was improved, with the tensile stress reduced. In response to the potential tensile stress zone of dam under strong earthquake, reinforcing bar was taken in the direction of dam beam. Seismic safety measures were conducted in terms of shape design, structural design, and foundation treatment. Such as a thicker dam shape, setting corner stickers, locally increasing concrete strength grades, strengthening foundation grouting, and enhancing drainage of resistance bodies on both sides. The calculation results of arch beam load distribution method showed that the dynamic response of the dam of this project is similar to that of Xiluodu Project, Dagangshan Project, and Baihetan Project, and it had comparable seismic resistance. The finite element calculation results showed that the Yebatan arch dam was safe under design earthquake and verification earthquake. The analysis of ultimate seismic resistance showed that the arch dam has good seismic overload capacity. The dynamic stability analysis of the dam shoulder showed that the safety factor of the dam shoulder's dynamic anti sliding stability met the requirements of specifications. The numerical simulation results showed that after the concrete pouring of dam, the highest surface temperature was 25.5 ℃, and the highest internal temperature was 27.5 ℃. The maximum tensile stress of dam was located in the strong constraint zone of section 12, which was 1.53 MPa with a safety factor of 2.24. The experimental results showed that the insulation effect of spraying polyurethane on the surface of concrete was significantly better than that of polystyrene insulation board and insulation blanket. When spraying 10 cm thick polyurethane, the surface heat release coefficient of concrete is 28.5 Kj/(m2·℃·d), while that under 10 cm thick polystyrene insulation board and 10 cm thick insulation cover were 67.5 Kj/(m2·℃·d) and 70 Kj/(m2·℃·d), respectively. Based on simulation and experimental results, a refined temperature control and crack prevention technique system for the entire process of concrete under high altitude and large temperature difference was established. A world record 38.1m complete concrete core sample was successfully extracted from the interior of dam, demonstrating the effectiveness in concrete temperature control and crack prevention.Conclusions The construction practice of Yebatan Hydropower Station has shown that the safe and efficient construction of hydropower projects in extreme environment can be achieved through deep integration of precise geological evaluation, innovative construction technology, and intelligent monitoring technology. The successful completion of the Yebatan project provides a significant technical paradigm and reference for future hydropower development in high-altitude, cold, and complex geological rcondition.关键词：Yebatan Hydropower Station;Ultra-high arch dam;Deep unloading zone;seismic safety;Concrete temperature control52|24|0更新时间：2026-06-30
- 摘要：Objective Urban flooding has become an increasingly prominent challenge in rapidly urbanizing coastal cities, where drainage systems are frequently affected by multiple hydrodynamic boundary conditions, including extreme rainfall, river water levels, and external backwater effects. Traditional drainage system evaluation methods mainly focus on individual engineering facilities or single hydrological processes and often fail to quantitatively assess the overall performance of urban drainage systems under complex boundary conditions. To address this issue, a multi-source collaborative optimization method for urban drainage systems under compound boundary conditions was proposed. The objective was to establish an integrated framework that combines urban flood simulation, system performance evaluation, and drainage system optimization, thereby providing technical support for urban flood mitigation and drainage management.Methods A coupled hydrological-hydrodynamic model integrating river networks, drainage pipe networks, and surface runoff processes was established to simulate urban rainfall-runoff generation and flood inundation dynamics. The surface flow process was described using two-dimensional shallow water equations, while the drainage pipe network and river network were simulated using one-dimensional hydrodynamic equations. Water exchange among different subsystems was achieved through coupling nodes, enabling unified computation of flow and water levels under compound boundary conditions. Based on the coupled model, a comprehensive performance evaluation system was developed. Multiple indicators, including inundation area, water depth, overflow node number, overflow volume, drainage efficiency, and recession time, were selected to quantitatively characterize the operational performance of urban drainage systems. Furthermore, an optimization framework based on the collaborative configuration of green, blue, and grey infrastructure was established. Green infrastructure was used to enhance infiltration and reduce runoff generation, blue infrastructure was employed to increase storage and detention capacity, and grey infrastructure was adopted to improve drainage conveyance capability. Different infrastructure combinations were evaluated through scenario simulations to identify effective optimization strategies. The proposed method was applied to the central urban area of Lianyungang City, Jiangsu Province, China. Lianyungang is a typical coastal city affected by rainfall, river water levels, and tidal backwater conditions. Multiple rainfall return periods and external water-level scenarios were designed to investigate drainage system responses and optimization effects under different hydrodynamic conditions.Results and Discussions Simulation results indicated that compound boundary conditions significantly influenced the operational performance of urban drainage systems. As rainfall intensity increased, both inundation severity and overflow risk became more pronounced. Under high-return-period rainfall scenarios, surface runoff generation increased rapidly, leading to increased surcharge conditions within the drainage network and a substantial rise in overflow volume. The implementation of the proposed optimization scheme effectively improved system performance under all investigated rainfall scenarios. Compared with the existing drainage configuration, the number of overflow nodes was reduced by 6.57%–16.44%, while the total overflow volume decreased by 20.80%–21.78%. Meanwhile, the total system outflow increased slightly, indicating that the drainage conveyance capacity was enhanced after optimization. The optimization measures also significantly reduced flood inundation impacts. The inundated area decreased considerably, and maximum water depths in critical flooding locations were effectively controlled. From the perspective of hydrological and hydraulic mechanisms, different infrastructure measures contributed to flood mitigation through distinct pathways. Green infrastructure reduced runoff generation by increasing infiltration and temporary storage capacity. Blue infrastructure attenuated peak flows through detention and retention functions, thereby reducing the burden on downstream drainage facilities. Grey infrastructure improved hydraulic conveyance capacity and accelerated the drainage process. The coordinated operation of these three infrastructure categories formed a multi-scale regulation mechanism that acted throughout the runoff generation, flood propagation, and drainage processes. The results further demonstrated that the proposed framework can effectively identify system weaknesses and evaluate the performance of different drainage improvement strategies. Compared with conventional approaches that mainly focus on local engineering measures, the proposed method provided a system-level perspective for urban flood mitigation and drainage planning.Conclusions A multi-source collaborative optimization method for urban drainage systems under compound boundary conditions was developed by integrating hydrological-hydrodynamic simulation, performance evaluation, and infrastructure optimization. The proposed framework successfully established a closed-loop process from flood simulation to optimization decision-making. Application to the central urban area of Lianyungang demonstrated that the collaborative configuration of green, blue, and grey infrastructure can effectively reduce overflow risks, decrease inundation impacts, and improve overall drainage performance. The proposed method is particularly suitable for coastal cities and plain river-network regions affected by multiple hydrodynamic boundary conditions. The study provides a practical technical framework for urban flood risk management, drainage system planning, and resilient city construction under changing climatic and hydrological conditions.关键词：urban flooding;drainage system;compound boundary conditions;hydrological-hydrodynamic model;optimization configuration31|34|0更新时间：2026-06-24
- 摘要：Objective To address the abrupt erosion of lake shores under extreme wind conditions, this study investigated the impact of flexible vegetation on the evolution of lake shore profile. Flexible vegetation represented by reeds can dissipate wave energy and weaken nearshore grain transport, but the erosion–accretion response of lake shore profiles behind flexible vegetation under wind wave action has not been sufficiently quantified. The objective of this study was to investigate the influence of flexible vegetation on the lake shore profile evolution under wave action. Physical model experiments were used to reveal the effects of vegetation zone width and vegetation density on the profile evolution, and the XBeach-NH numerical model was further used to analyze both effects of incident wave parameters and grain characteristics parameter on the lake shore erosion behind the vegetation.Methods Physical model experiments and numerical simulations were used in combination. The physical experiments were carried out in a wave flume 40.0 m long, 0.5 m wide, and 0.8 m high. The experimental model consisted of a fixed bed and a movable lake shore with an initial slope of 1.0:2.6. The experiment was designed according to Froude similarity with a geometric scale of 1:10. The test regular wave condition consists of a wave height of 0.1 m, a wave period of 1.0 s, and a water depth of 0.45 m. The lake shore slope was constructed using non-cohesive fine sand with a median grain size of 0.2 mm. Reeds, which are common vegetation on lake shoals were selected as the flexible vegetation model. The reed panicle part was used, with a height of 0.65 m, an equivalent diameter of approximately 0.06 m, a measured single-plant volume of approximately 48 cm³, and a stem bending stiffness of approximately 0.08 N·m². Four cases were tested, including a non-vegetated case and three vegetated cases, namely M0, M1, and M2. In the M0 case, the vegetation-zone width was 0.5 m and the vegetation density was 100 plants/m². In the M1 case, the vegetation density was increased to 164 plants/m² based on M0. In the M2 case, the vegetation-zone width was increased to 1.0 m based on M0. Seven resistance-type wave gauges were used to record free-surface elevations at a sampling frequency of 50 Hz, and a laser topography measurement system was used to obtain the centerline elevation of the lake shore slope. The XBeach-NH non-hydrostatic phase-resolving model was established based on the physical model experiments to simulate coupled wave–vegetation–sediment processes. The numerical flume layout was kept consistent with that of the physical experiment. Grid convergence testing was conducted, and a uniform grid spacing of 0.02 m was finally adopted. Parameter sensitivity analysis was performed, and a bunch of calibrated values was obtained as the parameter combination for subsequent simulations. The fixed-bed vegetation zone and movable-bed lake-shore zone were consistent with those of the M0 vegetation model, with a vegetation density of 100 plants/m², a stem diameter of 0.06 m, and a vegetation height of 0.65 m. The equivalent vegetation drag coefficient CD = 3.0 was obtained by calibration against the equilibrium lake-shore profile under the physical experimental condition.Results and Discussions The physical experiments reproduced the erosion–deposition process of lake-shore profiles under the regular waves. With continuous wave action, erosion occurred in the upper part of the slope, sediment was transported offshore, and deposition developed both near the slope toe and in the offshore region. A bar-type equilibrium profile characterized by upper part erosion and lower part deposition was finally formed on the slope. The presence of flexible vegetation did not change the evolution pattern of the equilibrium profile, but it clearly reduced scouring depth, accretion depth and horizontal shoreline migration. Results show that as the width and density of vegetation zone increased by factors of 2 and 1.6, the maximum scour depth was reduced by 27.17% and 9.25%, respectively. Meanwhile, the horizontal migration distance of shoreline was decreased by 13.01% and 6.50%, respectively. The XBeach-NH numerical model which was suitable for simulating the coupled wave–vegetation–sediment processes was validated using the experimental data. The calibrated XBeach-NH model reasonably reproduced the main erosion–accretion characteristics of lake-shore profiles behind the flexible vegetation. However, local differences remained in the swash zone, berm region, and underwater bar area. These differences were possibly related to wetting–drying boundary movement, intermittent sediment transport, bed ripple development, local slope failure, and the simplified treatment of flexible vegetation motion using an equivalent drag term. After validation, the model was used to further analyze the effects of incident wave parameters and grain characteristic parameter on the lake shore erosion. Specifically, the scour depth and the horizontal shoreline migration distance were increased by 28.05% and 28.87%, respectively, as the incident wave height increased from 0.6 m to 1.4 m; while they were increased by 631.25% and 752.94%, respectively, as the incident wave period increased from 1.9 s to 4.4 s. Moreover, they were increased by 21.21% and 13.22%, respectively, as the median grain diameter increased from 0.1 mm to 0.5 mm. Variations in water depth had limited impacts on the scour depth, but a more evident shoreline retreat was observed under high water-level conditions within the tested range.Conclusions Flexible vegetation enhanced lake-shore protection by reducing scouring depth and horizontal shoreline migration distance. In this regard, increasing vegetation-zone width produced a stronger protective effect than increasing vegetation density under the tested conditions. Incident wave height and wave period were the key factors controlling lake-shore erosion. Water-depth variation had a limited influence on scour depth, but shoreline retreat became more evident under high-water-level conditions. Increasing the sediment median grain size intensified local scour and accretion to some extent, but did not change the evolution pattern of the equilibrium profile. The findings in this study can provide a theoretical guidance and methodological reference for evaluating the effectiveness of ecological lake-shore protection projects.关键词：Lake shore;flexible vegetation;wave-flume experiment;XBeach-NH;morphological evolution25|37|0更新时间：2026-06-24
- 摘要：Objective Seepage analysis is a fundamental research topic in geotechnical and hydraulic engineering, where the accuracy and efficiency of numerical solutions directly determine the long-term safety of engineering structures. Conventional numerical methods, such as the finite element method (FEM) and finite difference method (FDM), suffer from inherent mesh dependency. While physics informed neural networks (PINN) require extensive iterations and incur high computational costs, physics informed extreme learning machines (PIELM) rely on strong-form partial differential equation (PDE) residuals to construct objective functions. This leads to difficulties in tuning weights for multiple physical constraints, the inability to inherently satisfy Neumann boundary conditions, and increased modeling complexity. To address these limitations, this paper proposes an energy-based physics-informed extreme learning machine (EPIELM) for the efficient and accurate solution of steady-state seepage problems.Methods Built upon the extreme learning machine (ELM) architecture, the proposed method employs ELM as trial functions for the seepage head field and transforms the governing equations of steady-state seepage into an energy functional minimization problem based on the energy variational principle. By constructing differential operators and applying meshless Monte Carlo integration to discretize both the domain physical constraints and Dirichlet boundary constraints, these conditions are unified into a single system of linear equations. The output layer weights are then analytically solved in a single step via the least squares method, thereby achieving an iteration -free and mesh-independent seepage field solution process. Furthermore, this approach inherently satisfies Neumann natural boundary conditions without the need for manual tuning of loss term weights.Results and Discussions The proposed method is comprehensively validated through three typical case studies: steady-state seepage in a 2D soil column, seepage in a water conveyance canal, and seepage in an underground tunnel, with systematic comparisons against PINN, PIELM, energy-based physics-informed neural network (EPINN), and high-fidelity finite element method (FEM) benchmark solutions. Results from the 2D soil column case indicate that gradient-iteration-based neural network methods (PINN and EPINN) require training times of 307.77 s and 174.23 s, respectively, whereas analytical-solution-based ELM methods (PIELM and EPIELM) require only 1.42 s and 0.62 s. EPIELM achieves a computational efficiency approximately 500 times higher than that of PINN and a 56% acceleration over conventional PIELM, demonstrating that energy-form constraint construction effectively optimizes the solution path and accelerates convergence. In terms of predictive accuracy, energy-based methods (EPINN and EPIELM) significantly outperform the baseline methods, with the accuracy ranking as EPINN ≈ EPIELM > PIELM > PINN. Specifically, EPINN exhibits the lowest relative L₂ error (0.003248) and root mean square error (RMSE, 0.000871), demonstrating superior global fitting capability; meanwhile, EPIELM achieves the lowest maximum absolute error (0.014421) among the four comparative methods, showing optimal local error control in boundary regions. The mean absolute errors (MAE) of both EPINN and EPIELM are maintained at the scale of 0.0003–0.0004, which is lower than that of PINN (0.001 46). Visualizations of the head and velocity fields reveal that while all four methods accurately reproduce the head distribution, the energy-based variants yield smoother streamlines with fewer numerical perturbations, indicating better solution smoothness and physical consistency. Error distribution contour maps further demonstrate that errors across all methods are concentrated in the bottom-left boundary region; however, EPIELM exhibits the smallest error field extent and lowest intensity, effectively suppressing error diffusion. Results from the water conveyance canal and underground tunnel engineering cases show that the head field contours obtained by EPIELM are smooth and continuous, the flow velocity streamlines are highly consistent with the head gradient direction, and high-velocity zones correspond precisely to regions of geometric discontinuity, demonstrating good agreement with the FEM benchmark solutions. Specifically, the RMSE and maximum absolute error for the canal case are 0.008 924 and 0.088 865, respectively, while those for the tunnel case are 0.009 442 and 0.083 147, respectively; the solution accuracy in both cases fully meets the practical requirements for seepage analysis in hydraulic engineering.Conclusions The proposed EPIELM method not only achieves exceptional computational efficiency and excellent solution accuracy but also eliminates the constraints of complex mesh partitioning, demonstrating strong adaptability to complex computational domains with geometric discontinuities; it effectively resolves the challenge of optimizing multiple loss term weights in conventional physics-informed models, featuring a concise modeling process with low implementation complexity. This approach provides a novel, efficient, and accurate solution for seepage analysis in geotechnical and hydraulic engineering, with promising applicability in scenarios demanding both computational efficiency and precision, such as rapid simulation of engineering seepage and the optimization of anti-seepage structures.关键词：seepage analysis;physics-informed extreme learning machines;energy method;Meshfree;physics-informed neural network28|64|0更新时间：2026-06-23
- 摘要：水工隧洞混凝土衬砌在长期服役过程中受围岩压力、内水压力及高速水流冲刷等多因素作用，易形成多重缺陷工况，传统基于经验或单一缺陷分析的修复方法难以满足复杂工程条件需求。本研究基于包含102项有效案例的多源工程案例库，通过Apriori关联规则挖掘方法对水工隧洞多重缺陷与修复材料之间的关系进行统计分析，阈值设定最小支持度为0.06、最小置信度为0.40、最小提升度为1.50，对多重缺陷条件下材料选用的集中度与提升度特征进行评价。结果表明：多重缺陷占比为63.7%，其中“冲磨+空蚀”和“裂缝+渗漏”为高频组合；多重缺陷条件下材料选用更集中，主导材料集中度由47%提升至74%。基于案例数据构建的缺陷–材料关联模型，表明在多重缺陷约束下，材料选用呈现出一定的统计收敛特征，可为水工隧洞复杂病害条件下的修复材料选型提供数据参考。关键词：水工隧洞;缺陷修复;关联规则算法;多重缺陷26|40|0更新时间：2026-06-23
- 摘要：Cooperative smoke-screen shielding by unmanned aerial vehicle (UAV) swarms is a soft-kill protection problem in which multiple UAVs deliver smoke munitions to form obscurant regions between protected targets and incoming threats. This task couples target assignment, release sequence, detonation timing, cloud motion, line-of-sight geometry, and continuous flight parameters. Since shielding intervals generated by different clouds may overlap, directly summing local shielding durations can overestimate the system-level effect. To address these issues, this paper establishes a unified optimization model and proposes a hierarchical algorithm for UAV-swarm cooperative smoke-screen shielding.A kinematic model is first constructed for UAVs, smoke munitions, smoke clouds, and incoming missiles. Each UAV flies at a constant altitude, speed, and heading after task assignment. A smoke munition inherits the carrier velocity after release and moves under gravity before detonation. After detonation, the smoke cloud is modeled as a spherical obscurant region whose center descends vertically during its effective duration. This model gives the release point, detonation point, smoke-cloud center, and missile position as functions of time.An effective shielding criterion is then established from the missile viewpoint, smoke-cloud center, and protected true target. The true target is modeled as a fixed cylinder near the false target. A far-field parallel projection approximation is adopted, and the visible target contour is represented by an equivalent projected rectangle. Four key line-of-sight segments are formed by connecting the missile position to the four vertices of this rectangle. A smoke cloud is effective only when it is within its valid duration and its center is no farther than the effective smoke radius from all four segments. In the numerical scenario, the minimum viewing distance is about 200 m, the target scale ratio is about 4.3%, and the estimated boundary error is about 0.37 m, about 3.7% of the effective smoke radius.An interval-union-based effectiveness index is introduced. For each incoming missile, the total effective shielding duration is defined as the union length of all effective shielding intervals assigned to that missile. The global cooperative effectiveness is obtained by summing the weighted shielding durations of all missiles. This index counts only the actual effective shielding time and avoids repeated counting of overlapping intervals. Based on this index, the cooperative shielding task is formulated as a mixed discrete-continuous optimization problem. The discrete variables determine the service target of each smoke munition, and the continuous variables include UAV heading angle, flight speed, release time, and detonation delay.To solve this high-dimensional and non-smooth problem, a three-level hierarchical framework is designed. The upper level estimates UAV service potential according to geometric reachability and remaining time margin, and generates target priorities. The middle level performs shell-by-shell temporal coordination. For each smoke-munition node, a coarse action time window is constructed, and the marginal gain is defined as the increase in the union length of the accumulated coarse shielding interval after adding this node. The lower level optimizes continuous parameters to generate executable deployment parameters satisfying the strict shielding criterion.An improved differential evolution (DE) algorithm is used for lower-level optimization. A smooth proxy fitness function is constructed from the distance margin between the smoke-cloud center and the four key line-of-sight segments to guide the search, while final shielding durations are evaluated by the strict criterion. Compared with standard DE, the improved DE introduces directed initialization, adaptive parameter adjustment, and elite preservation to improve search efficiency and stability.Numerical experiments are conducted with 3 incoming missiles, 5 UAVs, and 15 smoke munitions. For target M1, shell-by-shell marginal coordination increases the coarse shielding interval union length from 21.96 s to 33.71 s, an improvement of 53.51%. After continuous optimization, standard DE and improved DE obtain total cooperative effectiveness values of 37.35 s and 40.39 s, respectively. Additional scenarios with crosswind disturbance, missile maneuver, reduced smoke radius, and shortened smoke duration show that the proposed framework remains applicable under representative engineering disturbances. Timing statistics show that platform guidance and shell coordination require less than 0.01 s, and one complete scheme using improved DE is generated in about 227.72 s. Overall, the proposed method improves shielding temporal succession and generates executable cooperative deployment schemes under strict shielding criteria.关键词：UAV swarm;cooperative smoke-screen shielding;hierarchical optimization;temporal coordination;differential evolution;mixed discrete-continuous optimization54|31|0更新时间：2026-06-17
- 摘要：Objective Attribute reduction aims to identify relevant attributes, eliminate redundant features and preserve classification capability, and thereby improve the generalization ability of machine learning models. Traditional neighborhood rough set (NRS) models rely on a fixed neighborhood radius or a unidirectional k-nearest neighbor relation. When data are unevenly distributed, two critical limitations arise. First, a fixed-radius neighborhood cannot simultaneously adapt to sparse and dense regions, particularly at class boundaries where local densities differ significantly. Second, the unidirectional k-nearest neighbor relation lacks reciprocal verification and consequently fails to distinguish true intra-class neighbors from inter-class false neighbors. These limitations lead to inaccurate characterization of boundary samples, biased evaluation of attribute significance, and suboptimal reduction performance. To overcome these challenges, this work, this work proposes a mutual nearest neighbor-based attribute reduction method that enhances discriminative ability and robustness for data with heterogeneous local density distributions.Methods First, the inherent limitations of three classic rough set models, namely the standard
- 摘要：Objective Gearboxes operating under time-varying rotational speed are affected by changing mesh excitation, limited sensor installation positions, and incomplete acquisition of state information. These factors weaken the reliability of vibration-based fault diagnosis, especially when the measured data cannot fully cover different fault types and degradation levels. A digital twin model can supplement measured data by generating dynamic response data consistent with the physical gearbox. However, the fidelity of twin data depends on the dynamic consistency between the virtual model and the physical entity. When the model parameters are fixed, the simulated response may deviate from the measured response under variable-speed conditions. Therefore, a gearbox digital twin modeling method with dynamic parameter correction was developed to reduce the mismatch between simulated and experimental signals and to improve the reliability of fault diagnosis for pitting, tooth breakage, and cracking.Methods A two-stage gearbox was taken as the research object. A lumped-parameter dynamic model was established to describe the torsional–transverse coupled vibration characteristics of the transmission system. The intermediate shaft was simplified as a rigid mass moving in three translational directions, and the bearings were represented by linear stiffness and damping elements. Gear meshing was equivalent to a time-varying spring–damper system. The normal meshing excitation was described by time-varying mesh stiffness and static transmission error. On this basis, three typical gear faults were introduced into the virtual model. For pitting, a periodic pulse excitation synchronized with the faulty gear rotation was constructed to describe the impact generated when the pitting region entered the mesh zone. The pitting severity was represented by the pulse amplitude and width related to the pitting diameter. For tooth breakage, a stiffness loss function was introduced to describe the periodic decrease and recovery of load-carrying capacity when the broken tooth passed through the mesh zone. For severe tooth breakage, an additional transient impact term was used to characterize the secondary impact caused by stiffness recovery and load redistribution. For cracking, a stiffness attenuation modulation function related to the gear rotation angle and crack depth was constructed to describe the periodic weakening of tooth-root bending stiffness. The dynamic equations were solved using numerical integration to generate twin vibration responses.To improve the consistency between the virtual model and measured data, an improved particle swarm optimization (PSO) algorithm was used for parameter identification. Time-varying mesh stiffness, damping coefficients, and fault characteristic parameters were selected as optimization variables. A composite loss function was designed by combining the time-domain correlation coefficient and the relative error of order-spectrum amplitudes, so that both waveform similarity and fault characteristic consistency were considered. Four strategies were integrated into the PSO framework: a linearly decreasing dynamic inertia weight was used to balance global exploration and local exploitation; a constriction factor was introduced to improve convergence stability; a multi-swarm cooperative search strategy was adopted to reduce the probability of local convergence; and an adaptive boundary handling strategy was used to improve the efficiency of valid particle search. The optimized parameters were fed back to the dynamic model to update the virtual gearbox model. A digital twin visualization platform was also developed to display rotational speed, three-directional vibration responses, and characteristic curves. The fidelity of the twin data was evaluated by time-domain correlation coefficients and relative errors of order-spectrum amplitudes.Results and Discussions Experimental verification was carried out on a two-stage gearbox test rig under time-varying speed conditions. The measured vibration signals in the x, y, and z directions were compared with the twin signals generated by the updated virtual model. The results showed that the twin signals had similar time-domain trends to the measured signals under different health and fault states. Local residuals still existed in some waveform segments, mainly because the lumped-parameter model simplified complex factors such as micro-contact conditions, assembly errors, lubrication variation, sensor noise, and environmental disturbance. These residuals were mainly reflected in local peak amplitudes and phase deviations, while the main vibration trend and fault-related impact characteristics were retained.Order analysis was used to evaluate the fault characteristic components under variable-speed conditions. For the severe pitting state, the fault characteristic order and its multiple orders were clearly observed in both measured and twin signals. The relative errors of the characteristic order, first multiple order, and second multiple order amplitudes were 3.12%, 3.29%, and 4.05%, respectively. For the medium tooth-breakage state, the characteristic order and its multiple orders were also consistent between experimental and twin data, and the relative errors of the main amplitude components remained within a small range. These results indicate that the digital twin model can reproduce the main order-domain fault characteristics under time-varying rotational speed.The diagnostic performance was further evaluated for pitting, tooth breakage, and cracking. The average diagnostic accuracy reached 96.92%. Compared with the conventional dynamic model without parameter optimization, the digital twin model improved the average diagnostic accuracy by approximately 6.56%, indicating that parameter feedback correction effectively enhanced the fidelity of the simulated data. Ablation experiments showed that the complete improved PSO algorithm achieved better convergence speed and lower loss function values than standard PSO and single-strategy variants. This result verified the cooperative effect of dynamic inertia weighting, constriction factor, multi-swarm search, and adaptive boundary handling. In addition, the proposed digital twin method was compared with typical data-driven diagnosis models, including convolutional neural network (CNN), long short-term memory (LSTM), and gated recurrent unit (GRU). The diagnostic accuracy of the digital twin method on three-directional vibration signals was higher than that of these data-driven models, with an improvement of at least 3.78%. This comparison suggests that the mechanism-guided digital twin framework can provide reliable diagnosis support when measured data are limited and operating conditions vary.Conclusions A gearbox digital twin model with improved PSO-based parameter updating was developed for fault diagnosis under time-varying speed conditions. The model combined lumped-parameter dynamic modeling, fault excitation description, parameter feedback correction, and order-domain evaluation. The results showed that the updated twin model could effectively reduce the mismatch between simulated and measured vibration responses and generate high-fidelity twin data for diagnosis. The method improved diagnostic accuracy compared with an unoptimized dynamic model and typical data-driven models. The study indicates that a mechanism-dominated and data-corrected digital twin framework is suitable for describing typical gear faults and supporting fault diagnosis when measured data are limited. The current validation mainly focused on three single faults: pitting, tooth breakage, and cracking. Compound faults, unknown fault evolution, lightweight online optimization, and industrial field deployment remain important issues for further study.关键词：Digital Twin Model;Gearbox;Time-Varying Rotational Speed;Improved PSO Algorithm;fault diagnosis37|29|0更新时间：2026-06-17
- 摘要：Objective Accurate and quantitative evaluation of equipment contribution rates in space target detection systems is essential for equipment development planning and system architecture optimization. With the exponential growth of space objects—including high-value satellites, space debris, and maneuverable spacecraft—the risk of space collisions and hostile threats has increased dramatically. Traditional evaluation methods focus on individual equipment performance and fail to characterize the synergistic effects and structural adaptability of equipment within complex systems. Three core challenges remain unresolved: modeling the distortion of multifunctional equipment, the ambiguity of evaluation indicators, and the difficulty in analytically determining the connectivity reliability of complex networks. Traditional approaches relying on isolated performance metrics cannot capture the marginal contribution of equipment to overall mission success or identify critical bottlenecks, which has become a major obstacle to rational resource allocation and iterative system upgrades. A method integrating OODA Loop (Observe-Orient-Decide-Act Loop) theory, Fuzzy Comprehensive Evaluation (FCE), and Monte Carlo simulation is proposed to precisely quantify the value of equipment systems and address the adaptability deficiencies of traditional methods.Methods First, an OODA loop network model of the space target perception system was constructed, which abstracts system components into four types of nodes: perception and early warning nodes, management and control nodes, execution nodes, and target nodes. An entity-function-indicator three-layer mapping mechanism was adopted to address the modeling distortion issue associated with multifunctional integrated equipment, enabling a single piece of physical equipment to be mapped to multiple functional nodes based on its task capabilities. For example, the multifunctional ground-based radar S1, which integrates target detection and preliminary data processing capabilities, is mapped to both a perception node (S1) and a decision node (D3) in the network model. This approach accurately reflects its dual role in the OODA loop and avoids the underestimation of its system value that would occur with a single-node mapping. Second, the FCE method combined with the Analytic Hierarchy Process (AHP) was used to quantify the connectivity probability of links between nodes. Key influencing factors for different types of links were selected, and qualitative expert judgments were converted into quantitative link weights through membership functions and weight matrices, effectively addressing the ambiguity and uncertainty of evaluation indicators. Third, the evaluation of system task effectiveness was reformulated as a connectivity reliability problem for the OODA loop network. A Monte Carlo simulation process based on the Depth-First Search (DFS) algorithm was designed to traverse all possible OODA loop paths and calculate system task effectiveness through large-scale random sampling. This simulation-based approach overcomes the combinatorial explosion problem encountered by analytical methods in large-scale networks with multiple overlapping OODA loops. The DFS algorithm efficiently enumerates all valid closed-loop paths, ensuring that no potential task execution route is overlooked during connectivity assessment. Finally, the presence-absence comparison method was applied to quantify the system contribution rate of each piece of equipment by comparing system effectiveness before and after removing the equipment. Simulation experiments were conducted with 10⁶ simulations, and task weights of 0.6 for fast-maneuvering targets and 0.4 for high-value satellites were set. Robustness tests were performed under different task weight combinations (0.5/0.5 and 0.7/0.3) and different simulation durations (10³, 10⁴, 10⁵).Results and Discussions The evaluation results show that management and control nodes (space target control center D1 and regional control center D2) have a system contribution rate of 100%, and their removal directly reduces the system's effectiveness to zero, confirming their central role as the "brain" of the system. Execution nodes (interceptor I1 and electronic jamming equipment I2) have contribution rates exceeding 50%, serving as the core carriers for completing task closed loops. The multifunctional ground-based radar S1 (including the derived decision node D3) has the highest contribution rate (40.66%) among all perception equipment, significantly higher than that of single-function perception equipment. Space-based observation equipment generally has higher contribution rates than ground-based observation equipment, with the high-orbit perception satellite S4, low-orbit perception satellite S3, ground-based photoelectric system S2, and radio detection system S5 having contribution rates of 28.75%, 27.92%, 27.04%, and 23.07%, respectively. Notably, under the 0.7/0.3 weight combination—which prioritizes missions involving fast-maneuvering targets—the contribution rates of all perception equipment increase by approximately 3–5 percentage points compared to the 0.5/0.5 combination, reflecting the greater importance of real-time detection capabilities in high-threat scenarios. Robustness tests indicate that the ranking of equipment contribution rates remains completely consistent under different task weight combinations, with absolute value changes of less than 2 percentage points. When the simulation count is 10⁶, the relative error of the benchmark effectiveness is only 0.22%, which meets engineering accuracy requirements. Further analysis reveals a significant positive correlation between the equipment contribution rate and both the number of OODA loops in which the equipment participates and the loop contribution value. Multifunctional equipment can participate in multiple task links and yield more significant gains in system effectiveness. The space-ground integrated perception system forms a complementary capability, while the S5 radio detection system is identified as the weak link in the current system.Conclusions The proposed method effectively addresses the issues of modeling distortion in multifunctional equipment, ambiguity in evaluation indicators, and the calculation of reliability in complex network connectivity. It can accurately identify key core equipment and weak links in the space target perception system, providing quantitative decision support for equipment development planning, resource allocation optimization, and system architecture upgrades. The results validate the rationality and effectiveness of integrating OODA loop theory, fuzzy comprehensive evaluation, and Monte Carlo simulation for system contribution rate evaluation. Future research will extend the method to dynamic confrontation scenarios, multi-task comprehensive evaluation, full-life-cycle dynamic contribution rate evaluation, and node-link joint failure models to further improve the method's practical adaptability and evaluation accuracy.关键词：system contribution rate;combat loop;fuzzy comprehensive evaluation;monte carlo simulation;space target perception31|51|0更新时间：2026-06-17
- 摘要：Objective Oil spill regions in complex scenes usually exhibit high background similarity, blurred boundaries, and significant scale variations, making it difficult for existing segmentation methods to accurately extract target structures and boundary information. Boundary discontinuity, small-target omission, and background interference frequently occur in conventional semantic segmentation networks when oil spill regions are detected under complex environmental conditions. To address these problems, a Dual-Stage Edge-Guided Feature Enhancement Network (DSEF-Net) was proposed. The proposed method achieved collaborative optimization between structural information and semantic information through cross-scale edge difference modeling and adaptive edge-guided enhancement mechanisms, thereby improving boundary localization accuracy and segmentation quality of oil spill regions in complex scenes.MethodsDSEF-Net was constructed based on an encoder-decoder architecture and incorporated a dual-stage enhancement framework centered on edge structure modeling. In the first stage, namely the Edge Difference Modeling (EDM) stage, structural information and semantic information were extracted from shallow features and deep features, respectively. Channel compression was performed through 1×1 convolutions to reduce redundant semantic interference while preserving structural representations. Bilinear interpolation was then utilized to spatially align deep semantic features with shallow structural features.After feature alignment, a dual-branch edge modeling strategy composed of a main branch and a difference branch was adopted. In the main branch, aligned shallow and deep features were concatenated and progressively processed by multiple convolutional layers to extract local edge responses. In the difference branch, edge responses were independently generated from shallow and deep features, and the absolute difference between the two responses was calculated to explicitly characterize cross-scale semantic inconsistency regions. This strategy enhanced the representation capability of potential boundary regions and enabled the network to focus on ambiguous structural contours. The outputs of the two branches were subsequently fused to obtain discriminative edge priors.In the second stage, an Edge Fusion Sharpening Module (EFSM) and an Edge Feature Adaptive Unit (EFAU) were introduced to further refine edge representations. EFSM fused multi-scale edge information through stacked convolutional operations and constructed a local gradient enhancement branch using bias-free convolutions. Residual enhancement was then performed to improve local detail responses around edge regions. EFAU dynamically predicted spatial enhancement weight maps from input features through channel compression and nonlinear mapping operations. The adaptive weights were utilized to regulate enhancement intensity at different spatial locations, thereby suppressing background interference while strengthening important edge regions. The enhanced edge priors were finally fused with semantic features to guide the network toward more accurate structural perception of oil spill regions and achieve fine-grained segmentation.Experiments were conducted on the Ground Oil Spill Dataset (GOSD) and the SENTINEL remote sensing oil spill dataset. Several representative semantic segmentation models, including UNet, ENet, SegNet, DeepLabV3, DeepLabV3+, PSPNet, KCNet, and BGNet, were selected as comparison methods. Mean Absolute Error (MAE), Accuracy, Precision, Recall, and Intersection over Union (IoU) were adopted as evaluation metrics to comprehensively assess segmentation performance.Results and Discussions Experimental results demonstrated that DSEF-Net achieved superior segmentation performance on both datasets. On the GOSD dataset, MAE, Accuracy, Precision, Recall, and IoU reached 0.032, 0.982, 0.862, 0.834, and 0.737, respectively. Compared with the second-best model KCNet, MAE decreased by 0.3%, while Accuracy, Precision, Recall, and IoU improved by 1.7%, 1.2%, 1.9%, and 0.5%, respectively. On the SENTINEL dataset, DSEF-Net achieved MAE, Accuracy, Precision, Recall, and IoU values of 0.039, 0.913, 0.910, 0.903, and 0.826, respectively. Compared with KCNet, MAE decreased by 0.6%, while Precision, Recall, and IoU improved by 3.1%, 0.8%, and 3.0%, respectively.Qualitative comparisons further showed that DSEF-Net preserved structural integrity more effectively under complex backgrounds, low-contrast conditions, and blurred boundary scenarios. Compared with existing segmentation methods, the proposed model reduced boundary discontinuity and background false detection while maintaining more continuous edge structures. For small-scale oil spill regions and elongated boundary structures, DSEF-Net exhibited stronger boundary continuity and finer structural perception ability.Ablation experiments further verified the effectiveness of EDM, EFSM, and EFAU. EDM enhanced the representation capability of cross-scale boundary discrepancies. EFSM improved edge continuity and local detail representation, while EFAU strengthened adaptive boundary perception capability in complex regions. The collaborative integration of these modules significantly improved the overall structural modeling ability of the network.Conclusions A Dual-Stage Edge-Guided Feature Enhancement Network was proposed for oil spill segmentation in complex scenes. Through edge difference modeling and adaptive edge-guided enhancement mechanisms, collaborative optimization between structural information and semantic information was effectively achieved. The proposed method enhanced boundary representation capability and improved segmentation accuracy for blurred boundaries, small targets, and scale-varying oil spill regions. Experimental results on the GOSD and SENTINEL datasets demonstrated that DSEF-Net outperformed several mainstream semantic segmentation models in both quantitative and qualitative evaluations. The proposed framework provides an effective technical solution for intelligent monitoring and fine-grained identification of oil spill regions in complex environments.关键词：image segmentation;Oil spill;Edge guidance;feature enhancement28|48|0更新时间：2026-06-17
- 摘要：如何提取水电机组的运行特征信息，实现多类型故障诊断是水电站智能运维的关键，本文提出一种集信号处理、特征提取和故障诊断于一体的水电机组故障诊断模型。首先，针对多尺度熵粗粒化不足易丢失信号特征的缺陷，提出新的特征提取工具—时移多尺度近似熵，仿真实验证明该方法具有良好、稳定的特征提取性能。其次，针对鲸鱼优化算法易陷入局部最优且对初始解的依赖性较强的问题，融合多策略改进该算法并测试其性能，仿真实验证明改进的算法具有最快的收敛速度和求解精度。然后，针对核极限学习机超参数难以调节的问题，利用改进算法优化其主要参数并对提取的特征进行故障识别和分类。最后，通过不同数据对所提方法进行对比验证，结果表明本文方法在转子振动试验台和真机数据的诊断中精度提高了1.46%和2.86%，对比其他方法具有最佳特征提取能力和最优诊断效果。关键词：multi-strategy;fault diagnosis;time-shift multiscale approximate entropy;whale optimization algorithm;kernel based extreme learning machine119|635|0更新时间：2026-06-15
- 摘要：Objective The initial stiffness of loess in the small-strain range is important for deformation analysis and constitutive modeling. For remolded loess, the effects of specimen preparation, strain rate, stress state, and stress history on initial stiffness are still not fully clarified. Previous studies have mainly focused on strength or large-strain behavior, while the identification of stiffness in the strictly controlled small-strain range has received less attention. This study aimed to clarify the governing factors of the initial stiffness of remolded loess by comparing different specimen preparation methods, evaluating the influence of strain rate, and examining the dependence of stiffness on mean effective stress and over-consolidation ratio (OCR). Particular attention was paid to the difference between local and external deformation measurements.Methods Laboratory tests were conducted using a GDS stress-path triaxial apparatus equipped with local linear variable differential transformer (LVDT) transducers and bender elements. Jingyang loess from Shaanxi Province, China, was used to prepare three types of saturated remolded specimens: slurry-consolidated specimens, layered tamped specimens, and layered statically compacted specimens. The target dry density of the three specimen types was controlled at about 1.59 g/cm3.Results and Discussions The results showed that the deformation measurement method had a significant effect on the identified initial stiffness. In the very small strain range, the stiffness determined from local LVDT measurements was markedly higher than that obtained from the external displacement sensor, in some cases by about 1-2 times. The stress-strain curves based on local measurements were nearly linear at the initial stage, whereas those based on external displacement showed an initially concave shape. This indicated that conventional external measurements were affected by end restraint and system deformation and therefore underestimated the true small-strain stiffness of remolded loess.Conclusions The initial stiffness of remolded loess in the strictly controlled small-strain range was governed mainly by stress state and stress history rather than by strain rate within the tested interval. Local LVDT measurements were more reliable than conventional external displacement measurements for identifying the true initial stiffness. For specimens prepared by slurry consolidation, layered tamping, and layered static compaction, the initial stiffness showed only weak sensitivity to strain rate between 0.001 and 0.1 %/min. For slurry-consolidated remolded loess, unloading-reloading paths generated over-consolidated states that further reduced the rate sensitivity of initial stiffness. The evolution of initial stiffness under different states could be described effectively by mean effective stress and OCR.关键词：specimen preparation method;small-strain stiffness;strain-rate effect;over-consolidation ratio;mean effective stress;bender element test;remolded loess12|51|0更新时间：2026-06-09
- 摘要：对考虑土体参数空间变异性的工程切坡进行高效可靠度分析是岩土工程领域具有挑战性的问题，其中如何合理表征土体抗剪参数的非平稳分布特征尤为关键。为此，本文提出一种结合t分布-随机邻近嵌入（t-SNE）、反向传播神经网络（BPNN）和拉丁超立方抽样（LHS）的工程切坡可靠度分析新方法。首先，利用中点法离散黏聚力与内摩擦角随机场，通过强度折减法获取训练样本中对应的边坡稳定系数。接着，使用t-SNE-BPNN构建随机场与稳定系数之间的代理模型，并结合拉丁超立方抽样进行边坡可靠度分析。最后，选用左畲小学工程切坡验证所提方法的有效性。结果表明，t-SNE具有较好的降维能力，所提方法可准确构建随机场与稳定系数之间的关系，能以较少的训练样本条件实现工程切坡可靠度的准确评估，可有效解决工程实践中失效概率计算耗时长的难题。同时，采用非平稳随机场能更好地表征切坡土体参数空间变异性。研究成果可为工程切坡稳定性评价和加固提供参考，同时为滑坡灾害防治提供理论依据。关键词：spatial variability;reliability analysis;random field theory;surrogate model;engineered cut slope13|43|0更新时间：2026-06-08
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0