Objective Conventional inversion methods for determining material parameters of arch dams are plagued by low computational efficiency, insufficient automation, and an inability to support dynamic tracking during long-term engineering service. Traditional Stochastic Subspace Identification (SSI) algorithms demand intensive manual parameter tuning and struggle with the effective elimination of spurious modes, which introduces considerable subjectivity. Likewise, trial-and-error or successive iteration-based inversion approaches require repeated finite element computations, incurring prohibitive computational costs. Moreover, most existing studies perform only static inversion at isolated time points, thus failing to capture the time-varying evolution of material properties over extended operational periods. To overcome these limitations, this paper proposes an intelligent dynamic inversion framework for the dynamic elastic moduli of arch dams and their foundations, integrating automatic modal identification with surrogate-assisted optimization. Principal Component Stochastic Subspace Identification (PCSSI) is employed to enable automatic, continuous extraction of modal parameters from long-term monitoring records. A Kernel Extreme Learning Machine (KELM) surrogate model is constructed to characterize the nonlinear mapping between dynamic elastic moduli and natural frequencies, while the Ivy Algorithm (IVYA) performs dynamic inversion and real-time tracking. The proposed methodology provides an efficient and reliable technical solution for the long-term health monitoring and safety assessment of arch dams.Methods The proposed dynamic inversion framework consisted of three sequential components: automatic dynamic identification of natural frequencies, surrogate model construction, and intelligent optimization-based inversion. First, based on long-term continuous measured vibration response data, modal parameters were automatically identified in a time-segmented manner using PCSSI. In this algorithm, Principal Component Analysis (PCA) was applied after constructing the Hankel matrix to compress the output data dimensionality while preserving essential system information, thereby reducing the computational burden of Singular Value Decomposition (SVD) and enhancing robustness against noise. The natural frequencies were subsequently obtained by solving the eigenvalues and eigenvectors of the reduced discrete state-space matrix. By segmenting the dynamically acquired vibration responses, continuous natural frequency evolution sequences were established, providing measured calibration benchmarks for the subsequent dynamic inversion. For surrogate model construction, Latin Hypercube Sampling (LHS) was utilized to generate a representative sample set within the plausible ranges of the dynamic elastic moduli for both the dam body and foundation. Each sample was input into the finite element model to compute the corresponding first three natural frequencies, forming the initial dataset. The KELM algorithm was then adopted to build a high-fidelity surrogate model that captured the nonlinear mapping between the dynamic elastic moduli and natural frequencies, effectively replacing the repeated invocation of the computationally intensive finite element model. The objective function for inversion optimization was defined as the minimization of the weighted relative errors between the calculated and identified natural frequencies, which was solved using IVYA. During the iterative process, the climbing competition mechanism was incorporated; samples of the dynamic elastic modulus were moved toward the optimal position following phototropism-based growth rules. An update was accepted if the new position yielded superior fitness; otherwise, random perturbations were introduced with a dynamically adjusted probability to escape local optima. Since the dynamic elastic modulus cannot be directly measured, an indirect validation strategy was adopted: the daily inverted dynamic elastic modulus time-series data were substituted back into the arch dam finite element model for forward analysis to obtain theoretical values of the first three natural frequencies, which were then compared day by day with the corresponding measured values.Results and Discussions Verification experiments were performed using a physical arch dam model. The dam body and foundation were fabricated using M10.0 and M7.5 special mortars, respectively. Daily vibration monitoring was conducted over one month after model completion, yielding a total of 30 test datasets. The Enhanced Frequency Domain Decomposition (EFDD) method was employed as a reference to validate the identification accuracy of PCSSI, with the results demonstrating close agreement between the first three natural frequencies obtained by the two methods, thus confirming the reliability of the automatic identification. For surrogate model performance, the correlation coefficients between predicted and actual values on both the training and testing sets for the KELM model reached 0.999 9. Compared with Random Forest (RF) and Response Surface Methodology (RSM), the KELM surrogate achieved coefficients of determination (R²) of 0.999 9 for the first three natural frequencies, with mean squared errors (MSE) below 0.000 6, 0.000 9, and 0.001 5, respectively—significantly outperforming the other two models. This confirmed that KELM could effectively fit the highly nonlinear mapping relationship between dynamic elastic moduli and natural frequencies. Comparative results between IVYA and Genetic Algorithm (GA) indicated that IVYA exhibited superior inversion accuracy across all modal orders, with relative errors for the first three frequencies of 0.786%, 0.406%, and 1.030%, respectively, all lower than those of GA. The daily dynamic inversion results revealed that the concrete dynamic elastic modulus of the dam body increased from an initial value of 18.91 GPa to 22.71 GPa, representing an overall increase of approximately 20.1%. The evolution pattern exhibited rapid growth in the early stage followed by gradual stabilization, which aligned well with the known hardening law of concrete materials. Meanwhile, the dynamic elastic modulus of the dam foundation remained stable at 15.0~15.5 GPa with fluctuations below 3%, consistent with the fact that the foundation was cast earlier and its hardening process had essentially concluded. When the inverted dynamic elastic moduli were substituted back into the finite element model for modal analysis, the comparison of the calculated and measured values of the first three natural frequencies showed that the maximum daily relative errors were 1.23%, 0.97%, and 1.23%, respectively, with average relative errors below 0.67%, 0.45%, and 0.69%, thereby validating the effectiveness and accuracy of the inverted parameters.Conclusions The proposed method enables intelligent dynamic inversion of dynamic elastic moduli for arch dams and their foundations. The adoption of PCSSI overcomes the reliance on manual intervention and the inability of traditional methods to continuously track modal variations, allowing for automatic segment-by-segment modal identification from long-term monitoring data and the continuous construction of natural frequency sequences. The coupled KELM-IVYA framework effectively captures the nonlinear relationship between material parameters and natural frequencies, circumventing the computational inefficiency associated with repeated finite element calculations in conventional inversion approaches and achieving dynamic parameter tracking. Physical model validation demonstrates that the proposed dynamic inversion method reliably reflects the evolution of natural frequencies, thereby providing a sound basis for the long-term health monitoring and safety assessment of arch dams.
Objective Plateau terraced irrigation districts commonly have cultivated land above their water sources, dispersed irrigation areas, and complex hydraulic connections among reservoirs, canals, lakes, and water-use units. These conditions make it difficult to determine how internal water-system interconnection and external diversion jointly affect the spatial distribution and reliability of water supply. The core area of the Shiping Irrigation District in Yunnan Province, China, was selected as the study area. It includes the northern mountainous and Yibao irrigation zones, covers 18,887 ha of irrigated land, and supplies domestic, industrial, agricultural, and ecological uses. Its supply system relies mainly on lake replenishment, reservoirs, and groundwater, but seasonal runoff concentration, dry-season inflow shortages, long conveyance distances, and uneven storage conditions cause persistent local deficits. This study developed a dynamic allocation model for the multi-source system under a water network pattern, quantified the effects of new reservoirs, internal interconnection works, and the Central Yunnan Water Diversion Project, and evaluated shortages and supply reliability while maintaining ecological releases.Methods A monthly dynamic water-resources allocation model based on the minimum-cost network-flow algorithm was developed. The supply system was generalized as a directed network: inflow nodes represented local runoff and external replenishment, reservoir nodes represented storage and reallocation facilities, demand nodes represented irrigation and urban-rural supply units, and arcs represented conveyance routes. The model contained 70 nodes, including 7 reservoir nodes, 10 water-use nodes, 20 inflow nodes, 25 intermediate nodes, and 8 outflow nodes. Node-balance equations described water conservation, arc capacities represented conveyance and diversion limits, and storage carryover arcs linked end-of-month reservoir storage to subsequent allocations. Arc costs and shortage penalties represented supply priorities and unmet demand. Monthly ecological-release targets were set to 10% of the long-term mean reservoir inflow for the corresponding month and were assigned lower unit costs than other supply arcs. Thus, ecological releases were allocated before domestic and agricultural uses; domestic use was then prioritized over agricultural use.Results and Discussions The baseline simulation revealed clear spatial differences in water security. The multi-year mean shortage was 1.6799 million m³ in Scenario I, corresponding to a district-wide shortage rate of 2.00%. Mean monthly and annual reliabilities across water-use units were 96.86% and 84.75%, respectively. Basic supply could therefore be maintained in many months, but interannual stability remained inadequate in several high-deficit units. Dalianzhuang and Tuoda irrigation blocks had the largest multi-year mean shortages, at 0.6942 and 0.5754 million m³, respectively, and their annual reliabilities were substantially lower than their monthly reliabilities.Conclusions The network-flow model provided a unified representation of source-project-demand relationships, reservoir carryover, conveyance limits, supply priorities, and ecological releases in the Shiping Irrigation District. It enabled long-term monthly comparison of alternative engineering systems and showed that improvement occurred in two linked stages. New reservoirs and internal interconnection works primarily reduced the spatial mismatch between water sources and demand, particularly in high-deficit units. The Central Yunnan Water Diversion Project then increased available supply and regulation margin, further reduced mean shortages and high-shortage risk, and alleviated localized adverse effects associated with internal path adjustment. The proposed network generalization and scenario-comparison approach can support engineering-effect evaluation and coordinated multi-source allocation in similar plateau mountainous irrigation districts. The results represent system-scale responses under the stated monthly series, engineering capacities, and operating rules. Detailed canal scheduling, gate operation, conveyance losses, extreme drought sequences, and abrupt operational conditions were simplified and should be examined using finer temporal resolution, representative drought years, or stochastic hydrological series.
SignificanceThe Panxi region is one of China's most important strategic mineral-resource bases, with abundant rare-earth resources and vanadium‒titanium magnetite. These two resource systems are concentrated in the same region but differ in deposit type, carrier minerals, beneficiation routes, and downstream utilization. In vanadium‒titanium magnetite, vanadium is mainly incorporated into titanomagnetite by isomorphic substitution, titanium is primarily hosted in titanomagnetite and ilmenite, and scandium and other dispersed elements occur in minerals such as clinopyroxene and ilmenite. By contrast, rare-earth resources in the Mianning‒Dechang metallogenic belt are mainly hosted in bastnaesite and are enriched in light rare-earth elements, especially cerium, lanthanum, and neodymium. Bastnaesite commonly coexists with barite, fluorite, and calcite, which complicates mineral liberation and separation. Therefore, high-value utilization of Panxi resources requires differentiated mineral-processing routes, efficient recovery of associated elements, and close integration between resource characteristics and the composition, structure, and performance requirements of advanced materials.ProgressRecent research on rare-earth functional materials has focused on permanent magnets, hydrogen-storage alloys, magnetocaloric materials, luminescent materials, and electromagnetic functional materials. In Nd‒Fe‒B permanent magnets, rapid solidification, nanocrystalline exchange coupling, grain refinement, and grain-boundary regulation have been used to reduce the consumption of critical rare-earth elements while maintaining magnetic performance. However, increasing the substitution level of La and Ce generally decreases magnetocrystalline anisotropy, coercivity, remanence, and thermal stability. Future design must therefore coordinate high-abundance rare-earth substitution with grain-boundary phase control and long-term reliability.Rare-earth hydrogen-storage alloys provide another route for utilizing La-and Ce-rich resources. Neodymium-free alloy design improves the use of abundant light rare-earth elements, while electrode engineering and conductive-network construction enhance low-temperature electrochemical kinetics. Nickel powders with different particle sizes can improve charge transfer and low-temperature discharge behavior. Remaining challenges include pulverization, corrosion, capacity decay, and deterioration of heat and mass transfer during repeated hydrogen absorption and desorption.For magnetocaloric materials, Gd‒Si‒Ge and La‒Fe‒Si systems have been studied to balance magnetocaloric performance, transition temperature, raw-material cost, processing time, brittleness, and device integration. Lower-purity gadolinium, elemental substitution, short-time high-temperature annealing, and controlled hydrogenation have been used to regulate phase transition and Curie temperature. Composite forming, porous metallic substrates, and low-melting-point metallic binders have also improved machinability, thermal conductivity, and corrosion resistance. These studies show that magnetocaloric research is shifting from intrinsic property optimization toward integrated material-heat-transfer-device design.In rare-earth luminescent materials, wet-chemical synthesis, particle-size control, and surface coating have been used to reduce synthesis temperature, improve particle uniformity, and decrease interfacial optical loss. Silica modification of YAG:Ce phosphors can enhance photoluminescence.The high-value utilization of vanadium-titanium resources has expanded from conventional metallurgy to hydrogen-storage alloys, electrochemical energy-storage materials, smart oxides, MXenes, carbonitrides, hard materials, and high-temperature self-lubricating composites. Vanadium-based body-centered-cubic alloys possess high theoretical hydrogen-storage capacities, but their applications are restricted by the high cost of pure vanadium, surface oxidation, impurity sensitivity, pulverization, and insufficient low-temperature kinetics. Metallurgical vanadium-bearing feedstocks and multicomponent alloying with Ti, Cr, Fe, Mn, and Al provide routes for regulating lattice distortion, hydride stability, plateau pressure, and cycling behavior.MXenes are promising for batteries, supercapacitors, electromagnetic shielding, and catalysis because of their high conductivity, layered structure, and tunable surface terminations. Conventional preparation usually depends on costly MAX‒phase precursors and corrosive fluorine-containing etchants, followed by complicated washing, intercalation, and delamination. One-step molten-salt and gas-phase synthesis routes have therefore been developed to shorten preparation cycles, reduce dependence on conventional etching, and improve phase purity. Remaining issues include precise control of surface terminations, oxidation resistance, batch consistency, environmental safety, and continuous large-scale manufacturing.Vanadium oxides exhibit tunable valence states and diverse crystal structures. In VO2, aliovalent doping modifies carrier concentration, local V‒V bonding, structural stability, and the semiconductor-to-metal transition, thereby lowering the phase-transition temperature while retaining infrared modulation. In perovskite AVO3 oxides, the A-site cation regulates vanadium valence, VO6 octahedral configuration, electronic structure, and transport behavior. CaVO3 shows much higher electrical conductivity than rare-earth AVO3 compounds, while its stable framework supports reversible lithium insertion with limited volume change. Fast-charging performance is therefore governed by the coupled effects of vanadium valence, electronic conductivity, lithium-ion migration, and structural stability.Titanium oxides can be modified through rare-earth doping, defect regulation, heterostructure construction, and surface functionalization to improve phase stability, light absorption, charge separation, and catalytic performance. Vanadium-titanium carbonitride powders and cermets have been developed through multicomponent solid-solution design and controlled carbothermal reduction‒nitridation. The introduction of W, Mo, Cr, and V can regulate powder size, composition uniformity, sinterability, grain growth, and interface structure. Core-rim regulation, secondary carbide addition, grain refinement, and multiphase hard-phase design have been used to improve the balance among hardness, transverse rupture strength, fracture toughness, and wear resistance. For high-temperature tribological applications, nano‒TiB₂ heterostructures have been introduced into Ti3AlC2 matrices. TiB2 particles increase density, refine grains, improve load-bearing capability, and suppress deformation and crack propagation. At intermediate temperatures, brittle oxide films may repeatedly form and spall, leading to abrasive, oxidative, and adhesive wear. At higher temperatures, TiB₂ oxidation produces B2O3, which combines with TiO2, Al2O3, and SiOₓ to form a relatively continuous tribofilm. The flow and wetting behavior of B2O3 help fill surface grooves and reduce direct contact, creating synergy between particle strengthening and semi-solid oxide lubrication.Conclusions and ProspectsThe utilization of Panxi rare-earth and vanadium-titanium resources is evolving from primary extraction and conventional alloy production toward compositionally controlled functional materials, key components, and engineering applications. Performance improvement increasingly depends on multiscale regulation extending from mineral occurrence and elemental separation to electronic structure, defects, grain boundaries, interfaces, phase transitions, and service-induced films. Future work should emphasize integrated resource‒material design, efficient recovery of associated elements, reduced dependence on critical raw materials, environmentally friendly MXene synthesis, stable fast-charging oxide frameworks, low-oxygen carbonitride powders, and reliable performance under high temperature, corrosive atmospheres, cyclic loading, and thermal shock. Advanced characterization, first-principles calculations, data-driven composition screening, process modeling, lifecycle evaluation, recycling, and industrial validation should be combined to accelerate the transformation of Panxi resources into sustainable material, technological, and industrial advantages.
Objective To measure the resilience of urban water supply infrastructure under risk shocks, this study established a logical framework for the resilience analysis of urban water supply system infrastructure along the lines of water supply infrastructure service failure path portrayal - failure scenario analysis - resilience measurement model development - simulation. Firstly, understanding transmission mechanisms of infrastructural service failures within urban water supply systems is crucial to assess resilience and service performance losses in urban water supply infrastructure. Therefore, this paper focuses on the core physical infrastructure to sort out the infrastructure composition of different water supply subsystems.Methods Referencing the reorganized real water supply network, we derive the general water supply simulation network and characterize the failure conduction paths. The water supply system comprises node elements interconnected in parallel and series to form a network, exhibiting functional linkage and transmission within each element. Water supply infrastructure, susceptible to diverse internal and external risk influences, undergoes service transmission failings and exhibits resilience variances. Water supply network service performance exhibits diverse failure conduction routes under a single node risk disturbance or multiple concurrent risky node disruptions. As far as the service performance of the whole water supply network is concerned, after the node components are subjected to risk, the water supply network resilience shows two modes of instantaneous failure or progressive failure along the "pressure → state → impact → response" by the damage and conduction of the node components. As a result, the evolution curves of the service performance of water supply infrastructures under different failure modes are obtained.Results and Discussions Based on the analysis of the service performance (resilience) function curve of water supply infrastructure, in the instantaneous failure mode, the service performance of water supply infrastructure exhibits an "S"-shaped variation during the recovery phase. Under the progressive failure mode, the service performance shows an inverted "S"-shaped variation during the progressive failure phase; during the recovery phase, its service performance variation is similar to that of the instantaneous mode, presenting an "S"-shaped curve. Whether in the instantaneous mode or the progressive mode, the service performance of water supply infrastructure exhibits an "S"-shaped variation in different stages and eventually tends to a new stable state. Consequently, this paper introduces the Sigmoid function. The urban water supply infrastructure's resilience evaluation models were developed, which comprise node and link models. Subsequently, the resilience models for various failure scenarios were devised and simulated. This paper uses a hypothetical case to set parameters and conduct simulation experiments. Simulation results demonstrate that: water supply resilience is related to the number of nodes in the water supply infrastructure network. The more nodes there are, the larger the function value, and the greater the resilience. This indicates that the more diverse the network is, the more resistant it is, and the stronger the resilience performance is. Overall, under the same number of nodes, the resilience performance of the instantaneous failure mode is better than that of the progressive failure mode. By comparing different scenarios under the two failure modes separately, the failure mode scenario with the maximum sustained loss stage (or called the buffer stage) has a better resilience performance than the failure scenario without the buffer stage.Conclusions In summary, this study established a logical framework for the resilience analysis of urban water supply system infrastructure along the lines of water supply infrastructure service failure path portrayal - failure scenario analysis - resilience measurement model development - simulation. A dynamic water supply infrastructure resilience models were developed, which reflect the service attributes and network conduction characteristics of water supply. And the segmented function integral model can measure the resilience performance in different time periods. The simulation results lead to the following conclusions: firstly the more diversified the water supply infrastructure network is, the more resistant and resilient the water supply infrastructure becomes. Secondly, the resilience of the water supply link network with a longer duration of post-shock loss change state is less than that with a shorter interval. Thirdly, as the service factor diminishes, the overall resilience of the water supply link network declines. Similarly, and the service performance factor and the resilience performance in the loss increase and performance recovery periods present different direction change under the progressive failure mode. The findings of the study provide a focus for optimizing the network structure of water supply infrastructure to enhance its resilience.
Objective To accurately characterize the large deformation and large motion behavior of flexible structures interacting with fluid flows, a hybrid numerical method integrating computational fluid dynamics (CFD), the nonlinear finite element method (FEM), and the immersed boundary method (IBM) is proposed. Conventional body fitted dynamic mesh methods, though capable of enforcing boundary conditions precisely, suffer from severe mesh distortion under large displacements, requiring frequent remeshing that increases computational cost and introduces numerical errors. Non body fitted overset grid methods avoid mesh regeneration but involve complex hole cutting procedures and yield uncertain interpolation accuracy at overlapping interfaces. The proposed method overcomes these issues by accurately computing the coupling forces at the fluid–solid interface and ensuring stable strong coupling through an iterative partitioned scheme, eliminating the need for mesh regeneration or overset operations.Methods The proposed method is based on a partitioned coupling framework that integrates computational fluid dynamics (CFD), nonlinear finite element method (FEM), and immersed boundary method (IBM). The fluid field is described using the Eulerian approach and governed by the Navier–Stokes equations. The large deformation nonlinear mechanical behavior of the flexible structure is simulated using the finite element method based on the updated Lagrangian formulation. The fluid–solid interface, which undergoes severe deformation and large motion, is represented by a set of immersed boundary points distributed on the surface of the deformable structure. The fluid–structure interaction forces are evaluated through the immersed boundary method. At the fluid–solid interface, the velocity boundary condition and the divergence free condition are strictly enforced. The coupled system is solved using a partitioned iterative strategy. Within each time step, multiple coupling iterations are performed, with the fluid and structural subdomains solved alternately and interface data exchanged until convergence is achieved, thereby ensuring accurate and stable treatment of the strongly coupled interaction between the fluid and the flexible structure.Results and Discussions The reliability and accuracy of the proposed method were validated through two benchmark cases: flow past a flexible beam and flow past a cylinder with a cantilevered flexible beam. In the flexible beam case, three cases with different elastic moduli were considered, and the streamwise displacements of the reference point were 0.487 cm, 0.443 cm, and 0.558 cm, respectively; the result obtained under Case 1 (0.487 cm) showed good agreement with the literature value of 0.478 cm, and the method also effectively prevented streamline penetration into the solid body, demonstrating that the velocity boundary condition was accurately enforced at the fluid–solid interface. In the cylinder with cantilever case, the maximum cross flow displacements of the reference point under the three cases were 0.084 2 m, 0.084 4 m, and 0.084 0 m, respectively, which were close to the literature value of 0.083 0 m; a comparison of the three cases revealed that, when the solid mesh resolution was kept consistent, a finer fluid mesh led to smaller displacement amplitudes of the reference point, whereas when the fluid mesh resolution was kept consistent, a coarser solid mesh resulted in smaller displacement amplitudes. A free falling flexible cross case was further conducted to examine the method's capability in handling large displacement, large deformation, and strong fluid–structure interaction; during its descent, the cross underwent significant rotation, folding, and stretching, eventually tending to become flattened, with the gravitational force and the fluid–structure interaction forces approaching equilibrium and the acceleration gradually decreasing. The results showed that the deformation became more severe as the elastic modulus decreased, and the elastic modulus predominantly affected the displacement amplitude in the x direction while having a minor influence on the y direction displacement. On the basis of these validation studies, the method was applied to a practical engineering problem: the vortex induced vibration of a 15 m free span submarine cable. The results showed that the streamwise vibration amplitude ranged from 0.35D to 0.8D (where D denotes the cable diameter), the cross flow amplitude was approximately ±1.1D, and the frequency ratio between the streamwise and cross flow responses was approximately 2.04, which was close to the literature value of 1.88. These results demonstrated the method's applicability and reliability for practical engineering flexible structures subjected to complex flow induced vibrations.Conclusions To address the large deformation and large motion fluid–structure interaction (FSI) problems of flexible structures interacting with fluid flows, this paper proposes a hybrid simulation method integrating computational fluid dynamics (CFD), the finite element method (FEM), and the immersed boundary method (IBM). The proposed method effectively resolves the difficulty in accurately capturing the rapidly evolving fluid–solid interface under large deformation and large motion conditions. In this method, the fluid field is governed by the Navier–Stokes equations within the Eulerian framework, while the large deformation and large motion behavior of the flexible structure is accurately described using the nonlinear finite element method based on the updated Lagrangian formulation. The FSI forces are computed through the direct forcing immersed boundary method without requiring empirical parameter calibration. A partitioned iterative strong coupling strategy is adopted, in which multiple sub iterations are performed within each time step to ensure the strict satisfaction of interface conditions, thereby guaranteeing the stability and accuracy of the simulation results. The validation through two classical benchmark cases—flow past a flexible beam and flow past a cylinder with a cantilevered flexible beam—demonstrates that the proposed method can accurately reproduce the large deformation and large motion response of flexible structures under fluid action, fully confirming its reliability and computational accuracy in solving such FSI problems. The simulation of a free falling flexible cross further verifies the method's capability in handling large displacement, large deformation, and strong FSI effects. Finally, the engineering application to the vortex induced vibration of a submarine cable demonstrates the method's effectiveness in practical engineering FSI problems involving flexible structures. This method provides effective technical support and a reference framework for future simulation and investigation of various FSI problems of flexible structures.
Objective Mountain rivers are frequently subjected to short-term and episodic sediment input due to rainfall, earthquakes, bank collapses, landslides, or debris flows, causing rivers rapidly shift from a relatively stable state to a non-equilibrium sediment transport state, further leading to sharp increase in bedload transport rate, local aggradation in riverbed, and continuous sediment transport after those events. Previous studies have demonstrated that episodic sediment supply can substantially change the bedload transport and river morphology. However, many studies have applied continuous sediment pulses during experiments, meaning that each subsequent pulse acts on a riverbed which was already modified by previous events. Consequently, the effects of sediment supply magnitude and frequency are difficult to isolate and compare directly. Moreover, limited attention has been paid to how different magnitude and frequency combinations influence the bedload transport rate, grain size selectivity, response lag, sediment transport and storage, and post-supply recovery process. Therefore, the objective of this study was to investigate the non-equilibrium response of bedload transport of mountain river to episodic sediment supply, and to clarify how sediment supply magnitude and frequency control the bedload transport rate, transported sediment grain size, and the relationship between sediment transport and storage.Methods Five experimental runs were conducted in a 10.0 m long, 0.6 m wide, and 0.8 m deep rectangular glass-walled water recirculating flume at the State Key Laboratory of Hydraulics and Mountain River Engineering, Sichuan University, China. Both the bed materials and supplied sediments ranged from 0.125 to 16.000 mm, with the median grain size D50 = 4.3 mm and geometric standard deviation σg = 2.54. The sediment mixture can represent the broad grain size distributions of many gravel-bed mountain rivers. Each run lasted 40 h and consisted of a 10 h clear water stage and followed by 30 h sediment supply and adjustment stage. Run1 employed continuous sediment supply regime with sediment supply rate per unit width of 8.2 g/m/s, and the total sediment input mass is 530 kg. Run2-Run5 used the same total sediment mass but divided it into one, two, three, and four episodic inputs, respectively. The corresponding sediment supply rate per unit width were 245.4 g/m/s, 122.7 g/m/s, 81.8 g/m/s, and 61.3 g/m/s, respectively, equivalent to 4.8, 2.4, 1.6, and 1.2 times the bedload transport capacity of the flow which were measured in preliminary experiments. These experimental runs formed a controlled sequence from large-magnitude and low-frequency sediment supply regimes to small-magnitude and high-frequency supply sediment supply regimes. Bedload transport was continuously measured at the flume outlet using an automatic weighing system, and the data were recorded at 30 s intervals. Transported sediment was collected and sieved every 60 min to determine the characteristic grain sizes D16、D50 and D84. According to the sediment mass conservation, sediment storage was calculated using the difference between sediment input and output. The cumulative bedload departure (CBLD) was further calculated to identify the high or low transport stages, estimate the lag between sediment input and transport response, and characterize the post-supply adjustment process.Results and Discussions 1) Episodic sediment supply can rapidly shift the riverbed from a stable or low transport state to a high transport state. After sediment supply stopped, the bedload transport rate gradually decreases, exhibiting a significant non-equilibrium response. The larger sediment supply magnitude and lower supply frequency result in a higher bedload transport rate and greater fluctuations, with the bedload transport rate varying over approximately four to five orders of magnitude. 2) During episodic sediment supply period, not only the bedload transport rate increase rapidly, but the transported sediment grain sizes D16、D50 and D84 increase with varying degrees and then gradually decrease after the sediment supply ended. For different grain size fractions, finer sediment (D84) are predominantly stored in the channel, whereas the intermediate fractions (D16~D50 and D50~D84) are mainly transported downstream. 3) Under episodic sediment supply regimes, sediment storage increases rapidly after sediment feeding begins, followed by an increase in the bedload transport rate. During the intervals between sediment supply events, the stored sediment continues to be transported downstream. In addition, sediment transport and storage showed counterclockwise hysteresis loops, and the loop scale increased with increasing sediment supply magnitude. 4) The cumulative bedload departure (CBLD) curve can effectively identify the high and low transport stages induced by episodic sediment supply. The larger sediment supply magnitude and lower frequency resulted in a larger CBLD fluctuation range, a shorter lag time of the bedload transport response, and a longer the recovery time of sediment transport of channel after sediment input.Conclusions Episodic sediment supply can rapidly transform a mountain river from a low transport state to a high transport state, followed by a gradual and storage-controlled decline after sediment input stopped. For the same total sediment input, large-magnitude and low-frequency sediment supply regimes produced higher bedload transport rate, wider fluctuations, larger transport-storage hysteresis loops, shorter response lags, and more persistent post-supply adjustment than those under small-magnitude and high-frequency sediment supply regimes. Episodic sediment inputs also temporarily enhance the transport of intermediate fractions. However, the finer and coarse particles tend to be stored in the channel. These findings can provide some references for predicting bedload transport, identifying high transport period and the aggradation risks of riverbed, and sediment-related disasters control in mountain rivers under sudden changes in sediment supply.
Objective Accurate identification of underwater cracks in dams is important for structural safety assessment and intelligent inspection. However, the generalization capability of intelligent models for dam underwater crack inspection is often limited by the scarcity of representative underwater crack samples. Moreover, domain differences among above-water crack images, generated underwater-style images, and real underwater dam crack images make it difficult for transfer learning and detection models to extract consistent crack features. To address these problems, this paper proposes a multi-source domain transfer learning method for underwater crack identification in dams, which combines CycleGAN-based image generation with an improved YOLOv8n detection model to enrich underwater crack samples and improve cross-domain identification performance.Methods Firstly, above-water crack images from campus buildings and bridge structures were assembled as multiple source domains, while field-acquired underwater dam crack images were used as the target domain. CycleGAN was then employed to translate the multi-source above-water crack images into underwater-style images through unpaired image translation. The translation process learned the color, illumination, and background characteristics of underwater environments while preserving crack morphology and structural information. Secondly, a cross-domain crack identification model was constructed based on YOLOv8n. An Asymmetric Cross-domain Attention Module (ACAM) was introduced at the backbone output to enhance directional crack features and suppress image-generation artifacts and background interference, while a Cross-domain Guided Fusion Module (CGFM) was incorporated into the neck network to strengthen the interaction between shallow spatial details and deep semantic features. Finally, the generated underwater-style samples were combined with real underwater dam crack images to train the improved YOLOv8n model, forming an integrated framework for multi-source sample generation, cross-domain feature enhancement, and underwater dam crack identification.Results and Discussions Experiments used 500 above-water crack images from two source domains and 500 real underwater dam crack images. For CycleGAN, the above-water and underwater image sets were each split into training and test subsets at an 8:2 ratio. For crack detection, the 500 real underwater images were further divided into 400 training images and 100 validation images, and an additional 91 independent underwater dam crack images were used to evaluate model generalization. The experiments were performed using an Intel Core i7-13700KF CPU and an NVIDIA GeForce RTX 4070 GPU with 12 GB of memory. Generated-image quality was evaluated using Fréchet Inception Distance (FID), Learned Perceptual Image Patch Similarity (LPIPS), and Structural Similarity Index Measure (SSIM), while detection performance was evaluated using mAP50, mAP50-95, GFLOPs, and FPS. The quality of the CycleGAN-generated images improved as the training duration increased from 50 to 150 epochs but deteriorated at 200 epochs. At 150 epochs, CycleGAN achieved the best generation performance, with an FID of 129.68, an LPIPS of 0.7636, and an SSIM of 0.5997, outperforming CUT and DCLGAN in image distribution consistency and crack structure preservation. Therefore, CycleGAN trained for 150 epochs was selected for sample generation. Among the 500 generated underwater-style images, 428 high-quality samples were retained, expanding the detection training set from 400 to 828 images. After incorporating the generated samples, the mAP50 values of YOLOv5n, YOLOv8n, YOLOv11n, YOLO26n, and RT-DETR-l increased by 2.63, 2.73, 3.01, 2.42, and 2.59 percentage points, respectively, demonstrating the applicability of the generated samples to different detection architectures. The proposed cross-domain identification model achieved mAP50 and mAP50-95 values of 88.19% and 56.74%, respectively, outperforming YOLOv5n, YOLOv8n, YOLOv11n, YOLO26n, Faster R-CNN, and RT-DETR-l. Compared with the baseline YOLOv8n model, mAP50 and mAP50-95 increased by 2.88 and 2.70 percentage points, respectively. The proposed model required 8.60 GFLOPs and achieved 213.63 FPS, indicating a balance between detection accuracy and computational efficiency. Ablation experiments quantified the contributions of ACAM and CGFM. Compared with the baseline YOLOv8n, ACAM increased mAP50 1.82 percentage points to 87.13%, while CGFM increased mAP50-95 by 2.09 percentage points to 56.13%. Combining ACAM and CGFM increased mAP50 and mAP50-95 by 2.88 and 2.70 percentage points to 88.19% and 56.74%, respectively. Grad-CAM++ visualizations showed that ACAM enhanced responses to crack bodies and extensions, CGFM improved the aggregation of crack features, and their combination focused more accurately on crack regions while suppressing background responses. Compared with CBAM, Coordinate Attention (CA), and SimAM, ACAM achieved the highest mAP50 and mAP50-95 values of 87.13% and 55.72%, respectively, while maintaining an inference speed of 275.99 FPS. Qualitative comparisons showed that the proposed model localized cracks more accurately and produced fewer missed and false detections under strong reflections, complex textures, low contrast, multiple cracks, and motion blur.Conclusions The proposed method integrates multi-source image translation with cross-domain feature learning to alleviate the scarcity of underwater dam crack samples and improve cross-domain identification performance. CycleGAN generated underwater-style samples while preserving crack morphology, and the generated samples consistently improved different detection models. ACAM enhanced directional crack features and suppressed image-generation interference, whereas CGFM strengthened the fusion of semantic and spatial information across domains. Their integration enabled the improved YOLOv8n model to achieve accurate and real-time underwater dam crack identification. The method can support automatic crack detection using optical images acquired by underwater inspection equipment, providing technical assistance for dam defect investigation and condition assessment.
SignificanceLiquid sloshing is the free-surface oscillation of a liquid within a partially filled container subjected to external excitation, a phenomenon whose dynamics range from linear standing waves governed by potential flow theory, through weakly nonlinear multi-modal interactions, to violent strongly nonlinear motions accompanied by wave breaking, spray, and gas-liquid entrainment, with impact pressures that can exceed the hydrostatic pressure by several orders of magnitude near resonance. This spectrum is regulated by dimensionless parameters including the Froude number, the Bond number, the Reynolds number, and the depth-to-length ratio, which jointly determine the transitions between distinct regimes. Because partially filled containers are ubiquitous in aerospace, marine, mechanical, and chemical engineering, liquid sloshing carries both destructive and exploitable implications. On the destructive side, membrane-type liquefied natural gas (LNG) carriers are vulnerable to sloshing impact damage near a 50% filling level under cryogenic conditions; propellant sloshing couples with flexible spacecraft structures, shifting natural frequencies and causing underestimation of the control torque demanded by attitude control systems; and liquid storage tanks must withstand both the impulsive hydrodynamic component and the convective sloshing component, as demonstrated by the tank failures during the 1964 Alaska and 1999 Kocaeli earthquakes. On the exploitable side, tuned liquid dampers (TLDs) convert sloshing into a mechanism for structural vibration control, and the phase coupling between waves, floating bodies, and internal sloshing has been harnessed for the motion control of floating production storage and offloading units, semi-submersible platforms, and floating wind turbine foundations. These dual implications establish the engineering significance of sloshing research.ProgressDrawing upon more than 220 publications, this review summarizes the progress of liquid sloshing research along three complementary lines. Mechanistically, the linear potential-flow framework provides the dispersion relation and the natural modal structure of a partially filled tank, which form a complete orthogonal basis for describing free-surface motion and support resonance prediction in preliminary design; this framework has been extended from simple rectangular tanks to elliptical, spherical, annular, and aqueduct-shaped geometries. Nonlinear multi-modal methods further capture amplitude-dependent frequency shift, internal resonance, modal energy cascades, subcritical bifurcations, and parametric resonance phenomena such as Faraday waves, revealing that resonance can arise even within parameter ranges judged safe by linear theory and that impact regimes can emerge abruptly through subcritical bifurcation without obvious precursors. Multi-physics coupling effects have been progressively clarified, among which cryogenic thermodynamics governs the acceleration of boil-off gas (BOG) generation induced by sloshing-enhanced mixing and heat transfer, fluid-structure interaction with elastic baffles introduces new natural frequencies below the lowest sloshing mode and hence potential resonance risks, and density stratification in layered liquids substantially alters the modal structure through buoyancy-driven coupling; in two-layer and multi-layer systems the density ratio controls the intensity of interfacial wave excitation, and internal interfaces can attain amplitudes comparable to the free surface, with beating phenomena absent in homogeneous configurations. Methodologically, analytical and semi-analytical methods anchored on the potential-flow assumption, computational fluid dynamics (CFD) techniques spanning volume-of-fluid (VOF), smoothed particle hydrodynamics (SPH), and moving particle semi-implicit (MPS) methods, and model experiments equipped with particle image velocimetry (PIV) and multi-source synchronous sensing have evolved into a complementary toolbox, although the accurate representation of damping remains a common challenge across all approaches. In engineering practice, sloshing suppression has been pursued through rigid, perforated, porous, elastic, and floating baffles as well as active control, and has been applied to spacecraft propellant management, LNG maritime transportation safety, seismic and nuclear safety including nuclear spent-fuel pools, and floating production and storage systems. In parallel, sloshing utilization has been developed through TLDs, tuned liquid column dampers (TLCDs), and tuned multi-liquid-column dampers (TLMCDs) for structural vibration control, ballast-water-filled floating breakwaters that exploit internal sloshing nonlinear damping and phase resonance to reduce wave transmission, oscillating water column (OWC) wave energy converters whose internal water column behaves as a vertical sloshing motion, and aquaculture vessels in which sloshing dynamics are regulated to decouple internal liquid motion from hull motion. At the interdisciplinary frontier, data-driven modeling with physics-informed neural networks and operator learning frameworks such as DeepONet and Fourier neural operators, the extension toward liquid hydrogen and ammonia supported by digital twin paradigms, bio-inspired surface suppression, non-Newtonian fluid sloshing, and sloshing scale effects have emerged as directions that may reshape the research landscape.Conclusions and ProspectsThe accumulated understanding of linear resonance, nonlinear multi-modal coupling, and multi-physics coupling constitutes a common basis for both sloshing suppression and sloshing utilization, with analytical methods, computational fluid dynamics, and model experiments forming an interdependent methodological toolbox. Nevertheless, several issues remain unresolved, including the accurate prediction of sloshing loads under strongly nonlinear conditions with wave breaking and gas-liquid entrainment, the development of mitigation techniques effective across broadband excitation, arbitrary filling levels, and arbitrary tank geometries, and the construction of complete cross-scale validation datasets linking laboratory tests with full-scale design. It is also recognized that sloshing suppression and utilization are not strictly opposed, since internal sloshing can reduce ship motion amplitude and added resistance under certain conditions, leaving an exploitable intermediate regime. Future research is expected to concentrate on the fusion of physics-based modeling with data-driven methods, multi-fidelity modeling of cryogenic multi-physics coupling, operator learning frameworks for real-time load prediction, bio-inspired baffle-free suppression, and cross-scale validation. The introduction of interdisciplinary perspectives such as bio-fluid-structure coupling, materials science, and experimental similarity theory is expected to open new strategies for sloshing suppression and to accelerate the translation of fundamental sloshing research into safe, economical, and sustainable engineering solutions.
Objective Airborne electronic devices operate for long durations in harsh marine environments with high humidity and dense salt spray. Exposed components, such as metal shells, signal connectors, and radio-frequency interfaces, are eroded by corrosive media and are therefore considered corrosion-sensitive parts. Their shielding effectiveness degrades as service time increases. In complex battlefield electromagnetic environments, this degradation exacerbates electromagnetic pulse-induced damage, performance degradation, disturbance, and interference to electronic systems, leading to potential safety hazards and economic losses. H62 brass is used as the matrix material in these corrosion-sensitive areas. When brass corrodes in salt spray environments, additional electromagnetic leakage paths are formed. However, systematic studies on the variation of corrosion-induced shielding effectiveness, the quantitative mapping between corrosion degree and equivalent conductivity, and an analytical model for rapid evaluation are still lacking. Moreover, the internal correlation between corrosion behavior and shielding effectiveness degradation remains unclear. In this study, we established the mapping relationship between corrosion degree and equivalent conductivity and constructed an analytical model for shielding effectiveness. The evolution law of corrosion-induced shielding performance degradation was investigated, providing a theoretical basis and engineering references for the anti-corrosion design and electromagnetic compatibility evaluation of airborne electronic devices.Methods A hierarchical experimental strategy, ranging from material-scale mechanism analysis to component-level performance verification, was adopted to investigate the marine atmospheric corrosion behavior of typical airborne electronic equipment and its effect on shielding effectiveness. H62 brass, widely used for corrosion-prone positions, was selected as the test material. Acid salt spray corrosion tests were conducted for 24 h, 60 h, and 100 h using a Z-UHA-225A-X10-W salt spray testing machine in accordance with the military equipment environmental test standard GJB150A-2009, and specimens with different corrosion degrees were prepared. A JSM6700F scanning electron microscope-energy dispersive X-ray spectroscopy system was used to observe corrosion product morphologies and detect elemental compositions. A HORIBA HR Evolution Raman spectrometer was employed to analyze the product components. A three-electrode test system was built based on a PARSTAT 4000A electrochemical workstation. Potentiodynamic polarization and electrochemical impedance spectroscopy tests were performed in 3.5 wt% sodium chloride solution to analyze the electrochemical corrosion behavior and kinetic evolution characteristics of the material. Based on electromagnetic topology theory and the Baum-Liu-Tesche equation, an analytical model for the shielding effectiveness of slotted shielding cavities was established with the introduction of transfer impedance. The measured interfacial contact impedance was mapped to equivalent conductivity, enabling quantitative characterization and equivalent modeling of corrosion severity. Shielding effectiveness values under different corrosion durations were calculated using both the constructed analytical model and CST full-wave simulation. Electromagnetic shielding performance verification tests were carried out in a semi-anechoic chamber. The correlation between microscopic corrosion behavior and macroscopic shielding effectiveness degradation was determined through mutual verification of experimental and simulation results, and the variation characteristics of corrosion-induced shielding performance were comprehensively analyzed.Results and Discussions The EDS and Raman spectroscopy results showed that the corrosion products generated on H62 brass were cuprous oxide (Cu₂O), zinc oxide (ZnO), and copper hydroxychloride (Cu₂(OH)₃Cl). Potentiodynamic polarization test results indicated that the corrosion current density first increased and then decreased with extended corrosion time, while the corrosion potential shifted significantly in the negative direction and then remained stable. The capacitive reactance arc radius obtained from EIS tests decreased sharply after 24 h of corrosion, indicating intensified interfacial electrochemical reactions. The arc radius then increased gradually from 24 h to 100 h of corrosion. The surface-deposited films hindered charge transfer and continuously reduced the corrosion rate. Parameter analysis of the analytical model showed that shielding effectiveness decreased as equivalent conductivity reduced. The attenuation rate of shielding effectiveness gradually increased, and the degradation reached a saturated lower limit when the equivalent conductivity approached 0 S/m. At 1.0 GHz, the shielding effectiveness decreased by 38.07 dB when the equivalent conductivity was reduced from 200 S/m to 0 S/m. Shielding effectiveness declined with increasing transfer impedance, presenting an attenuation trend that was fast initially and then slow, consistent with the influence of equivalent conductivity. At 1.0 GHz, the shielding effectiveness decreased by 8.67 dB when the transfer impedance increased from -30 dBΩ·m to -20 dBΩ·m, and by 3.76 dB when it increased from -20 dBΩ·m to -15 dBΩ·m. Contact impedance increased significantly with corrosion progression, with an increment of 13.33 mΩ from 24 h to 100 h of corrosion, while the contact resistance increased by only 0.1 mΩ, proving that contact impedance has higher corrosion sensitivity. Shielded structures with and without apertures exhibited different attenuation trends. A fast-growth-then-saturated attenuation was observed in aperture-free cavities, while aperture-containing samples showed a slow-fast-saturated variation trend. Corrosion-induced electromagnetic leakage caused large-amplitude shielding effectiveness loss in aperture-free shielding cavities, with a decline of 31 dB when the equivalent conductivity decreased from 100 S/m to 0 S/m. Apertures served as the dominant electromagnetic leakage channels and weakened the additional leakage caused by corrosion, resulting in only a 6 dB drop under the same test conditions. The calculation results of the analytical model were in good agreement with CST simulation and experimental data, with maximum absolute errors of 0.80 dB and 0.77 dB, respectively. The analytical model required a single calculation time within 10 s, offering significantly higher computational efficiency than CST full-wave simulation. The shielding effectiveness of 100 h corroded specimens decreased by 1.6 dB compared with that of 24 h corroded specimens. The experimental measurements, CST simulations, and analytical model calculations maintained consistent variation trends.Conclusions The kinetic evolution law and shielding effectiveness degradation mechanism of typical airborne electronic devices under acid salt spray conditions were clarified. The formation and accumulation of corrosion products increased the interfacial contact impedance, reduced the slit equivalent conductivity, destroyed the electrical continuity of structural surfaces, and formed additional electromagnetic leakage channels, thereby reducing the shielding effectiveness of the equipment. The established analytical model based on equivalent conductivity mapping was verified by CST simulation and semi-anechoic chamber experiments. The model achieved high calculation accuracy and computational efficiency in characterizing corrosion-induced shielding performance degradation. Aperture structures are the main electromagnetic leakage channels and can mask the additional leakage caused by corrosion, which should therefore be considered separately in shielding effectiveness evaluation. The research results provide a theoretical basis and engineering references for material selection, joint structure design, and shielding degradation evaluation of airborne electronic device shells in corrosive environments, and they support the development of corrosion inhibition methods and electromagnetic shielding performance optimization technologies.
SignificanceSpace has become a decisive arena for major-power strategic competition and technological supremacy. The United States, leveraging its low-cost reusable launch vehicles, is rapidly deploying mega-constellations such as Starlink to seize scarce orbital slots and frequency resources. It also imposes component embargoes, establishes exclusionary international cooperation frameworks, and dominates the formulation of outer space governance rules. These actions collectively constrain China's space development. In response, China adheres to an independent innovation path. Its space program is anchored by the Beidou navigation system, crewed spaceflight, lunar and deep-space exploration, and civil space infrastructure. These efforts continuously reinforce China's technological base to safeguard national security. To meet the strategic demands of high-quality aerospace development, China is prioritizing innovation capability enhancement, industrial ecosystem consolidation, and the cultivation of global competitive advantages. This paper systematically reviews worldwide space technology trends across the three pillars of access to space, utilization of space, and exploration of space. It summarizes China's landmark achievements, critically examines industrial bottlenecks, and proposes strategic pathways for frontier areas. These areas include heterogeneous mega-constellations, high-performance computing satellites, on-orbit servicing, embodied-intelligence spacecraft, nuclear propulsion and power, and cis-lunar resource development.ProgressGlobal space technology has entered a new stage of intensifying competition, accelerating technological iteration, and deepening commercialization. Recent years have witnessed an order-of-magnitude reduction in launch costs, deep involvement of commercial actors, and widespread application of artificial intelligence, all of which are profoundly reshaping the space industrial system. In the domain of access to space, space transportation is transitioning from low-frequency expendable launches to high-density, reusable airline-style operations. The United States leads in reusable technology, while China ranks alongside it in launch frequency but lags in overall lift capacity. In the domain of space utilization, space applications are evolving from standalone single-satellite services to systematic construction and integrated deployment. In the domain of space exploration, crewed spaceflight and deep-space exploration are expanding from low Earth orbit habitation to cis-lunar space and further toward distant resource development. These global trends provide both reference and impetus for China's strategic planning.Conclusions and ProspectsTo secure strategic initiative in the evolving space order, China must adopt a systematic, domain-specific layout covering the entire spectrum from very-low-Earth-orbit to deep space. This framework encompasses six major orbital domains, each with distinct strategic roles and core capability requirements.In very-low-Earth orbit, the focus is on ultra-low-latency communication, ultra-high-resolution remote sensing, and persistent surveillance, exploiting its clean environment while addressing atmospheric drag and rapid orbital decay. In low Earth orbit, the priority is large-scale constellation deployment with integrated communication, navigation, remote sensing, and computing services, managing spectrum scarcity and collision risks. In medium Earth orbit, enhanced positioning, navigation, timing, and cross-orbit relay functions are essential, calling for a backbone system bridging LEO and GEO operations. In geostationary orbit, securing scarce slots and providing wide-area information transmission and deep-space telemetry are critical, alongside debris and latency challenges. In cis-lunar space, the vision includes routine round-trip transportation, permanent habitation, and high-quality communication and navigation services, positioning this region as a new frontier of competition. In deep space, sustained planetary exploration capabilities are required, including autonomous navigation, long-distance communication, and nuclear power and propulsion systems.Building upon this spatial framework, the paper further identifies five key areas for high-quality aerospace development. The first is space-based network integration, which requires top-level coordination of communication, navigation, remote sensing, and computing constellations, establishing standardized cross-orbit links to form a unified infrastructure system. The second is space-based intelligent computing, addressing the exponential growth of satellite data and the bottlenecks of ground-based processing through deploying computational resources across orbital nodes for real-time data processing and intelligent decision-making. The third is on-orbit servicing and maintenance, transitioning from fixed, non-upgradeable hardware to maintainable, extendable spacecraft through technologies such as refueling, module replacement, and in-orbit assembly. The fourth is the intelligent evolution of spacecraft, shifting from ground-commanded operations to autonomous perception, decision-making, and execution, with embodied-intelligence spacecraft playing a central role in debris removal and space security. The fifth is space exploration and development, emphasizing routine cis-lunar transportation, long-term habitation support, and in-situ resource utilization.Looking further ahead, the paper outlines six strategic directions with phased roadmaps. The first is complex heterogeneous mega-constellations, progressing from inter-layer interoperability and autonomous networking to a seamless on-demand heterogeneous network. The second is high-performance computing satellites, evolving from single-satellite computing capability through kilowatt-class power platforms to a global space-based data-center network. The third is on-orbit servicing and maintenance, advancing from life-extension capabilities through modular construction to intelligent cluster-based maintenance services. The fourth is embodied-intelligence spacecraft, moving from prototype validation through non-cooperative target handling to systematic capabilities for large-scale debris clearance. The fifth is advanced nuclear propulsion and power, progressing from small-scale verification through kilowatt-class extreme-environment supply to an integrated propulsion-power platform. The sixth is cis-lunar resource exploration and development, advancing from relay networks and safe landing through reusable transfer stages to a full-coverage development system with prospecting, logistics, and permanent habitation.By systematically advancing these frontier areas and integrating national and commercial efforts, China will provide solid support for seizing the initiative in future space competition and achieving high-quality development of its space power.
Objective Composite modification of soil with steel slag and municipal solid waste incineration bottom ash serves as a feasible pathway for solid waste recycling and soft soil improvement. However, the triaxial mechanical behavior of the composite improved soil under varying confining pressures, the response mechanism of shear strength parameters to mix proportions, and the macro-micro coupling mechanism of synergistic solidification remain insufficiently interpreted. The internal cause of the non-monotonic strength variation with solid waste dosage has not been clearly clarified, which restricts the engineering design and application of such modified materials. This work systematically investigates the triaxial mechanical properties of steel slag-MSWIBA composite improved soil with different mix ratios, reveals the microscopic solidification mechanism, and establishes the correlation between mix proportion and shear strength performance.Methods Low-liquid-limit clay, sampled from a highway construction site in Ma'anshan at a depth of 2–3 m, was used as the base soil. The clay had a plastic limit of 20.8%, a liquid limit of 39.6%, a plasticity index of 18.8, an optimum moisture content of 18.2% and a maximum dry density of 1.85 g/cm³. Municipal solid waste incineration bottom ash (MSWIBA) was collected from a waste incineration plant in Anhui Province, with particle size controlled below 2 mm. Its main chemical components included SiO₂, CaO and Al₂O₃, and its heavy metal leaching concentrations met the national standard limits. Steel slag powder, obtained from a steel enterprise in Ma'anshan with an aging period of over 2 years, was ground and sieved through a 0.075 mm sieve. It contained active minerals such as C₂S and C₃S, with a free CaO content of 0.7% and a basicity of 4.1.Results and Discussions The morphology of stress-strain curves was significantly regulated by confining pressure. Under 100 kPa confining pressure, all specimens reached peak deviatoric stress at axial strains of 8%–10% and then exhibited strain softening. Under 200 kPa confining pressure, the post-peak attenuation weakened, and the peak strain range expanded to 8%–12%. Under 300 kPa confining pressure, all specimens showed strain hardening behavior within 20% axial strain. Elevated confining pressure enhanced normal contact and interlocking friction between particles, restricted lateral deformation and crack propagation, and shifted the failure mode from structural peak-controlled to continuous compaction and friction-controlled.Conclusions Confining pressure acts as a key factor controlling the stress-strain response mode of steel slag-MSWIBA improved soil. With rising confining pressure, the specimens transform gradually from strain softening to strain hardening. The shear strength of the improved soil shows a non-monotonic trend of first increasing and then decreasing with the dosage of both solid wastes. The mix proportion of 50% MSWIBA replacement and 15% steel slag dosage delivers the optimal performance, enhancing both cohesion and internal friction angle simultaneously, and its strength advantage is further amplified with increasing confining pressure.
Significance Terahertz (THz) wave technology demonstrates remarkable application potential in high-speed wireless communication, high-resolution imaging, non-destructive testing, and security screening. However, the scarcity of naturally occurring materials with strong THz response, combined with inherent drawbacks of traditional modulation techniques such as narrow operating bandwidth, slow response speed, and high power consumption, has become a critical bottleneck restricting the practical development of THz technology. Integrating metamaterials with functional materials to fabricate active tunable devices serves as a core technical approach to achieve dynamic manipulation of THz waves. Phase-change materials (PCMs) are ideal candidates for reconfigurable THz devices, as they undergo significant and reversible changes in electrical conductivity and dielectric constant under external stimuli including temperature, electric field, and light field. Among various PCMs, vanadium dioxide (VO2) and germanium-antimony-telluride (Ge2Sb2Te5, GST) have attracted extensive attention due to their unique and complementary response characteristics in the THz band. Systematic comparison of their phase-change mechanisms, response properties, and device applications provides essential theoretical and technical support for the design of high-performance THz reconfigurable devices.Progress The fundamental distinction between VO₂ and GST lies in their physical phase-change mechanisms. VO2 undergoes an insulator-to-metal transition (IMT) driven by the synergy of strong electron correlation and lattice distortion, featuring volatile reversibility, a high electrical conductivity switching ratio of 104 to 105 times, and ultrafast intrinsic response at picosecond level in theory. In practical applications, its response speed varies with excitation modes: millisecond level for thermal excitation, nanosecond level for electrical excitation, and picosecond level for femtosecond laser excitation. In contrast, GST achieves a non-volatile amorphous-to-crystalline transition through lattice reconstruction induced by atomic rearrangement, enabling simultaneous wide-range tuning of both the real and imaginary parts of the refractive index and supporting multi-state storage including amorphous, face-centered cubic, and hexagonal close-packed phases. Its crystallization process takes microseconds, while the amorphization process via melt-quenching can be completed within tens of nanoseconds. Under thermal excitation, VO2 exhibits a reversible phase transition at approximately 68 °C with a modulation depth exceeding 90%, but its response speed is constrained by the thermal conduction rate. Element doping and substrate stress regulation are widely adopted to adjust the phase transition temperature of VO2 and reduce the required excitation energy. Impedance matching design based on conductivity variation during phase transition further improves the modulation depth to over 99%. GST presents a two-stage crystallization behavior, transitioning from the amorphous state to the face-centered cubic phase at around 150 °C and further to the hexagonal close-packed phase at around 260 °C. It shows non-volatile multi-state tunability and high stability, while the thermal stability of its amorphous phase requires improvement. Under electrical excitation, VO2 achieves nanosecond-level response and a maximum modulation depth of 99.9%. Nevertheless, the relative contributions of electric field-induced carrier injection and Joule heating to the phase transition remain controversial, and thermal accumulation prolongs the recovery process from metallic state to insulating state. Optimized electrode structures and carbon nanotube-based nanoheaters are applied to reduce energy consumption and improve response speed. The phase transition of GST is mainly triggered by Joule heating from electric pulses, with favorable CMOS compatibility. Its crystallization process requires relatively long low-amplitude pulse durations to ensure ordered atomic rearrangement, and repeated electrical cycles easily cause element segregation and electromigration. Under optical excitation, VO2 realizes picosecond-level non-thermal phase transition under femtosecond laser pulses via photogenerated carrier injection, while the strong coupling between photothermal and non-thermal effects hinders precise and uniform phase control. High-frequency chopping above 2 kHz can effectively suppress the photothermal effect to isolate the carrier injection mechanism. GST supports all-optical non-volatile modulation and multi-level refractive index tuning, but its reversible modulation depth is generally limited to around 30%. In reconfigurable device applications, VO2-based amplitude modulators achieve a modulation depth of over 99% and nanosecond-level response via impedance matching, electromagnetically induced transparency, and Fano resonance mechanisms, making them suitable for high-speed THz switches. GST-based amplitude modulators support multi-level non-volatile modulation with stable reversible switching for dozens of cycles, but suffer from non-uniform phase transition in large-area arrays. For wavefront modulation devices, VO2-based metasurfaces realize 4-step phase coverage of 2π range and 95.01% amplitude modulation depth at 0.8 THz, enabling fast phase switching and multi-channel wavefront reconstruction. However, high loss in the metallic state leads to a trade-off between phase modulation efficiency and transmitted energy. GST-based metasurfaces achieve 8-level phase coverage of 0 to 2π in the 0.4–0.8 THz band, supporting non-volatile broadband wavefront control and reconfigurable planar lens functions, yet face challenges in wideband full-phase coverage and large-area uniform crystallization. In programmable metasurfaces, VO2-based devices with 8×8 pixel arrays driven by field-programmable gate arrays (FPGAs) realize real-time function switching and kilohertz-level modulation speed, with a state retention time of over 5 hours, but are limited by thermal crosstalk between pixels and coding distortion. Liquid metal thermal management layers are adopted to suppress inter-pixel thermal crosstalk.Conclusions and Prospects VO2 and GST exhibit high complementarity in THz dynamic manipulation. VO2 is applicable to high-speed dynamic switching scenarios due to its ultrafast response and ultrahigh modulation depth, while GST excels in low-power non-volatile information processing and long-term state maintenance owing to its non-volatile nature and multi-state tunability. Their respective performance advantages and application boundaries are determined by the essential differences in phase-change mechanisms. Future research can be expanded from four directions. First, novel composite materials, heterostructures, and doping modification schemes can be developed to improve response speed, modulation depth, and cycling stability while reducing power consumption and driving threshold. Second, anti-crosstalk device structures and photo-electro-thermal multi-field synergistic driving mechanisms can be designed to realize integrated devices with multiple functions such as modulation, wavefront control, and logic operation. Third, advanced algorithms such as deep learning can be integrated to optimize the coding strategy of programmable metasurfaces, and on-chip integration with THz sources, detectors, and signal processing circuits can be promoted to build compact intelligent THz systems. Fourth, ultrafast phase transition dynamics and new physical mechanisms can be further explored to expand applications in high-speed THz communication, real-time imaging, biosensing, quantum information processing, and neuromorphic computing.
Objective Complete coverage path planning for indoor mobile robots suffers from localization degradation in long corridors and weak-feature areas, high path redundancy in irregular regions, and unstable return to the global coverage path after dynamic obstacle avoidance. These problems directly cause map misalignment, trajectory tracking deviation, coverage missing, repeated coverage and collision risks, and restrict the reliability of routine indoor coverage tasks. A complete coverage path planning framework based on multi-sensor fusion is constructed to enhance localization stability, generate low-redundancy structured coverage paths, and maintain task continuity under dynamic disturbances.Methods A four-wheel mecanum mobile robot was adopted as the experimental platform, and a hierarchical architecture was built, including perception, localization and mapping, global coverage planning, local obstacle avoidance and chassis control. Two Livox Mid-360 Light Detection and Ranging (LiDAR) sensors were mounted at the front and rear of the robot respectively. After time synchronization, extrinsic calibration and coordinate transformation, two groups of point clouds were fused into the chassis coordinate frame as the main input for Simultaneous Localization and Mapping (SLAM). Wheel odometry and Inertial Measurement Unit (IMU) were introduced to provide motion prediction and attitude constraints, and ultrasonic sensors were applied for near-field safety compensation.The Cartographer algorithm was adopted to build the occupancy grid map and estimate robot poses. A factor-graph-based multi-source constraint model was established by fusing dual-LiDAR scan matching, wheel odometry, IMU preintegration and loop closure factors. The front-rear dual-LiDAR configuration provides multi-directional geometric observations, and alleviates localization degradation in long corridors and partially occluded environments. On the grid map, occupied, free and unknown cells were classified, and obstacle inflation was performed according to robot size and safety margin. The obtained reachable free space was taken as the input of the improved Iterative Partitioning Algorithm (IPA) for complete coverage path planning. The planning process included uncovered region extraction, connected-component analysis, scanline topological decomposition, fragmented region merging, principal direction extraction, boustrophedon trajectory generation and reachability-based path splicing. For each connected region, scanning was performed along candidate directions, and topological decomposition boundaries were detected according to the number change of traversable line segments. Small fragmented regions were merged based on area and directional consistency. The principal direction of each subregion was extracted from the covariance matrix of grid centers, and parallel coverage lines were generated along this direction. Reachability constraints were introduced during region splicing to prevent transition paths from crossing obstacles or unknown areas. For dynamic environments, a hybrid local planning layer integrating the Timed Elastic Band (TEB) method and the Dynamic Window Approach (DWA) was designed. Local trajectories were optimized by TEB with time, kinematic, obstacle avoidance and smoothness constraints. Longitudinal velocity, lateral velocity and angular velocity were sampled by DWA according to the omnidirectional motion capability of the mecanum chassis. Velocity commands were selected by a fused evaluation function, which contained goal-approaching, obstacle avoidance, speed-reward and trajectory-tracking terms. The trajectory-tracking term measured the deviation between the predicted local trajectory and the global coverage reference path, which allowed temporary deviation for obstacle avoidance, suppressed long-term drift, and guided the robot back to the global path after obstacles disappeared.Results and Discussions Simulation and real-robot experiments were conducted to verify the performance of the proposed framework. In the simulation, an indoor office map with irregular boundaries, L-shaped partitions and narrow passages was adopted. The improved IPA planner generated structured parallel trajectories in open areas, and maintained continuous coverage near partitions and narrow passages, with a static coverage rate of over 95%.Conclusions The proposed framework improves the execution performance of indoor complete coverage tasks through perception enhancement, structured coverage planning and hybrid path planning. The front-rear dual-LiDAR configuration expands the observation range, and improves the localization robustness of robots in long corridors and partially occluded environments. The improved IPA coverage planning method generates continuous coverage paths by combining scanline topological decomposition, principal direction extraction and reachability-based path splicing. The hybrid execution layer integrating TEB and DWA enables robots to avoid moving obstacles and return to the global coverage path after disturbances. Experimental results show that the proposed method achieves high coverage rate, low redundancy and collision-free execution in dynamic indoor environments, and can provide technical reference for indoor inspection, cleaning, disinfection and security patrol robots.
Objective Accurate prediction of seepage pressure is important for identifying abnormal changes in the seepage state of earth-rock dams. Purely data-driven deep learning models can learn nonlinear temporal relationships from long-term monitoring data, but their predictions may still exhibit abnormal local fluctuations, inconsistent response directions, or limited engineering interpretability because model optimization is dominated by data-fitting errors. To improve physical consistency and interpretability while retaining the nonlinear temporal modeling capability of data-driven methods, a mechanism knowledge-guided bidirectional long short-term memory (Bi-LSTM) framework was developed. The framework was intended to incorporate engineering knowledge available from routine monitoring systems without requiring complete seepage governing equations or detailed material parameters, and to examine whether different mechanism combinations should be selected adaptively for monitoring points with different locations and response characteristics.Methods Four types of mechanism knowledge were introduced into a Bi-LSTM baseline: the calculation rule of vibrating-wire seepage-pressure sensors, a seepage-pressure variation-coefficient constraint, a water-level–seepage-pressure trend-consistency constraint, and a temperature-related response constraint. Because these mechanisms differed in expression form and applicable stage, a differentiated embedding strategy combining a physical-loss constraint with rule-triggered post-processing correction was adopted. The vibrating-wire sensor rule provided a physical reference value and was embedded in the training stage through a joint loss consisting of the data-fitting loss and physical-reference loss. Its weight was determined by sensitivity analysis to balance prediction accuracy and physical consistency. The other three mechanisms were formulated as empirical or trend-type rules and were applied after the Bi-LSTM produced its initial prediction. When a prediction violated the reasonable variation-coefficient range, the expected water-level–seepage-pressure response trend, or the dominant temperature-related response, the corresponding rule was triggered and the output was corrected using a small-step relaxation coefficient. The correction coefficients were calibrated separately for different monitoring points on the validation set and selected jointly according to root mean square error (RMSE) and trend consistency rate (TCR), thereby limiting excessive interference with the temporal patterns learned from monitoring data. Daily monitoring data from a concrete-face rockfill dam in China were used for verification. The monitoring period extended from September 5, 2010 to June 2, 2020. The dataset contained daily seepage-pressure measurements from 10 gauges installed inside the dam and three downstream standpipe piezometers, together with upstream water level and temperature measurements. Three representative points, p07, Du1 (deep), and Du6 (shallow), were selected. The data from September 5, 2010 to April 18, 2017 were used for training, those from April 19, 2017 to April 18, 2019 were used for validation, and the remaining data were used for testing. Local gross errors were processed using the interquartile range method, missing values were completed by linear interpolation, and variables were normalized by Min-Max scaling. To represent cumulative and lagged hydraulic responses, the inputs included the previous-day upstream water level, 3-day and 6-day average water levels, temperature, sensor frequency, and historical seepage pressure. Multivariate samples were constructed using a 30-day sliding window. Model performance was evaluated using RMSE, mean absolute error (MAE), normalized mean absolute percentage error (nMAPE), coefficient of determination (R²), TCR, and abnormal first-difference ratio (AFDR). Single-mechanism, dual-mechanism, triple-mechanism, and full four-mechanism experiments were conducted using the purely data-driven Bi-LSTM as the benchmark. In addition, polynomial regression and XGBoost were introduced as conventional comparison methods to further evaluate the effectiveness of mechanism knowledge embedding.Results and Discussions The single-mechanism experiments showed that all four types of mechanism knowledge improved prediction accuracy or physical consistency to different degrees, but their dominant effects differed among monitoring points. The sensor-rule constraint improved agreement between the model output and the physical conversion relationship of the instrument; the variation-coefficient constraint reduced unreasonable seepage-pressure amplitudes; the water-level–seepage-pressure trend constraint improved tracking of directional changes; and the temperature-response constraint was more effective for local rapid responses at shallow points. These differences indicated that the contribution of a given mechanism depended on the monitoring-point location and response characteristics rather than being uniformly effective. The multi-mechanism experiments further revealed complementary effects among different constraints. For p07, the combination of the water-level–seepage-pressure trend constraint and the temperature-response constraint was optimal. RMSE decreased from 0.3566 for the data-driven Bi-LSTM to 0.3294, MAE decreased from 0.2809 to 0.2523, R² increased from 0.9535 to 0.9603, TCR increased from 50.24% to 64.93%, and AFDR decreased from 28.54% to 22.04%. For Du1 (deep), the combination of the sensor calculation rule, variation-coefficient constraint, and temperature-response constraint gave the best overall performance. RMSE decreased from 1.0330 to 0.8076, MAE decreased from 0.6997 to 0.5308, R² increased from 0.9784 to 0.9868, TCR reached 87.32%, and AFDR decreased from 25.37% to 18.46%. For Du6 (shallow), the same three-mechanism combination was optimal, reducing RMSE from 1.5315 to 1.2644 and MAE from 1.1459 to 0.8736, increasing R² from 0.9644 to 0.9757 and TCR from 71.95% to 85.49%, and reducing AFDR to 6.70%. Further comparison with polynomial regression and XGBoost showed that the optimal mechanism combinations consistently achieved lower RMSE and higher TCR at all three representative monitoring points, further confirming the effectiveness of mechanism knowledge embedding in improving both prediction accuracy and trend consistency. However, increasing the number of mechanisms did not produce monotonic improvement. Relative to the optimal combinations, the full four-mechanism combination increased RMSE by 6.56%, 8.57%, and 6.59% at p07, Du1 (deep), and Du6 (shallow), respectively, with corresponding decreases in TCR. These results indicate that adding more mechanism constraints does not necessarily provide effective incremental information and may instead introduce constraint redundancy or over-constraint, thereby weakening the model's adaptability to local temporal variations.Conclusions A mechanism knowledge-guided Bi-LSTM framework was established by embedding heterogeneous engineering knowledge at different stages according to its form and function. The training-stage physical reference and prediction-stage trigger-based corrections improved the balance among numerical accuracy, trend consistency, and suppression of abnormal fluctuations without requiring a complete seepage governing equation. Comparisons with polynomial regression and XGBoost further verified the overall advantages of the proposed framework in prediction accuracy and trend consistency. The optimal mechanism combination was monitoring-point dependent: the water-level trend and temperature constraints were more suitable for p07, whereas the combination of sensor calculation, variation coefficient, and temperature response was more suitable for Du1 (deep) and Du6 (shallow). Full superposition of all available mechanisms was not necessarily beneficial; mechanism integration should therefore follow an adaptive and sparsely effective principle, with constraints selected according to monitoring-point characteristics and validation performance. Because the required knowledge and variables can be obtained from routine dam monitoring systems, the framework provides an engineering-oriented approach for integrating mechanism knowledge with data-driven time-series models and can support seepage-pressure prediction, trend assessment, and safety monitoring in similar earth-rock dam projects.
Objective Stable contact recognition is important for tactile sensing and human–machine interaction systems based on triboelectric nanogenerators (TENGs). However, TENG-based contact recognition is limited by the strong dependence of output signals on the applied contact force, because force-induced signal fluctuation may mask the signal differences between contact objects with similar triboelectric properties. This work proposes a low-force-dependent triboelectric nanogenerator (LFD-TENG) based on a flat and dense multi-walled carbon nanotubes/polydimethylsiloxane (MWCNTs/PDMS) composite film. The device was designed to limit deformation under pressure and to form sufficient interfacial contact at relatively low pressure, thereby reducing the influence of force variation on voltage output. The pressure-dependent responses of voltage and current signals were compared, and the low-force-dependent voltage output was selected as the recognition signal for stable contact sensing. The output stability, contact recognition capability, information coding performance, and array-based tactile sensing ability of the LFD-TENG were investigated.MethodsMWCNTs/PDMS composite films with different MWCNT mass fractions were prepared by a coating method. The film morphology was characterized by scanning electron microscopy (SEM). Electrical tests were carried out in a contact–separation mode using polyethylene terephthalate (PET) as the positive triboelectric layer. Subsequently, the output voltages of LFD-TENGs with different MWCNT mass fractions were compared, and the optimized device was tested under different pressures to compare the pressure dependences of the voltage and current outputs, thereby identifying the voltage output with lower force dependence as the recognition signal for contact sensing. The stability of the voltage output was further evaluated under different contact frequencies. For textile recognition, the peak-to-peak value of each contact signal was extracted as the input feature and classified using a random forest (RF) model. For short-term individual identification, finger tapping test (FTT) signals from five volunteers were collected, and time- and amplitude-related features were extracted for RF classification. In addition, Morse code coding, array crosstalk evaluation, contact trajectory tracking, and directional key input tests were carried out.Results and DiscussionsSEM images showed that the MWCNTs/PDMS composite film had a flat surface and a dense internal structure without obvious pores or defects. MWCNTs were distributed in the PDMS matrix without large-scale aggregation, indicating effective filler dispersion. The film thickness was approximately 300 μm. In electrical output tests, the output voltage increased after MWCNT incorporation compared with the pure PDMS-based device. With increasing MWCNT content, the output voltage first increased and then decreased. The maximum output was obtained at an MWCNT mass fraction of 0.75%, and the open-circuit voltage increased by approximately 141% compared with that of the pure PDMS film. Under different pressures, both the voltage and current outputs changed to some extent, but their pressure dependences were clearly different. The normalized pressure sensitivity of the current signal was about 4.3 times that of the voltage signal, indicating that the voltage output had a lower dependence on contact pressure. Therefore, the voltage output was selected as the low-force-dependent recognition signal for subsequent contact sensing. The voltage output was then used to evaluate the low-force-dependent behavior of the LFD-TENG under different pressures and frequencies. When the pressure increased from 15 kPa to 187 kPa, the average output voltage was about 42.36 V, and the maximum fluctuation was approximately 3.51 V, corresponding to a relative variation of about 8.28%. Under different frequencies, the average output voltage was about 42.20 V, and the maximum fluctuation was approximately 1.63 V, corresponding to a relative variation of about 3.86%. These results indicate that the LFD-TENG maintained stable voltage output under different pressure and frequency conditions. Based on the theoretical model of contact-separation TENGs, the low-force-dependent voltage output was mainly related to the limited variation in the air gap distance and the effective surface charge density. The dense MWCNTs/PDMS layer limited the pressure-induced thickness variation of the dielectric layer, while the flat surface allowed the effective contact area to approach saturation at relatively low pressure. As a result, the variations in dielectric-layer thickness and newly generated effective contact area were both restricted, jointly reducing the influence of applied force on voltage output. For contact recognition applications, corduroy, polyester, polyester-wool, cotton-linen, and cotton were used as contact objects. Different textiles produced distinguishable voltage responses when contacting the LFD-TENG. Based on the peak-to-peak feature and RF classifier, the recognition results were concentrated on the diagonal of the confusion matrix, and the overall recognition accuracy reached 100%. The FTT results further demonstrated the ability of the LFD-TENG to detect dynamic tapping signals. Five volunteers generated periodic voltage pulses during continuous tapping within 5 s. When only the peak-to-peak feature was used, the overall individual identification accuracy was 77.54%. After three features were used, including the adjacent opposite-peak interval, peak-to-peak value, and adjacent same-direction peak interval, the accuracy increased to 97.83%. Furthermore, short and long tapping pulses were distinguished and used for Morse code coding, and the "SOS" signal was converted into a stable voltage response sequence. An LFD-TENG array was constructed for spatial contact detection. When a local unit was pressed, non-contact units generated only weak voltage signals, with peak-to-peak values mainly distributed in the range of 0.03–0.08 V, indicating low crosstalk. Contact trajectory tests showed that the response signals appeared sequentially according to the contact path, and a directional key input experiment verified its feasibility for human–machine interaction.Conclusions A low-force-dependent TENG based on a flat and dense MWCNTs/PDMS composite film was developed for stable contact sensing and recognition. By comparing voltage and current responses under different pressures, the voltage signal was confirmed to have lower force dependence and was selected as the recognition signal for contact sensing. The dense structure and flat interface reduced the influence of pressure variation on voltage output by limiting thickness deformation and effective contact area variation. The optimized device with an MWCNT mass fraction of 0.75% exhibited enhanced output performance and stable voltage response under different pressures and frequencies. Textile recognition, finger tapping identification, Morse code coding, array trajectory tracking, and directional key input experiments demonstrated that the LFD-TENG could provide stable and distinguishable contact signals without intentional force control. These results indicate that the proposed LFD-TENG provides an effective approach for contact sensing when the magnitude of the contact force is not deliberately controlled and shows potential for self-powered sensing and human-machine interaction systems.
Objective Concrete-filled double skin steel tubular (CFDST) structures offer superior mechanical performance, high ductility, lighter self-weight, and enhanced durability compared to conventional CFST members. However, current studies on CFDST mainly focused on the behavior of members and welded joints. Little attention had been paid on the bolted assembled joint owing to the fact that the closed-section feature of CFDST column hindered the bolted connection to steel beam. To transition CFDST members from isolated column elements into broad engineering practice, robust structural jointing systems utilizing mechanical connectors such as end plates, blind bolts, shear studs, and reinforced concrete floor slabs are paramount. This paper aims to investigate the cyclic behavior of the assembled joint between CFDST column and beam using blind bolts. For this type of joint, the complex force transferring mechanisms of assembled CFDST joints under cyclic loading remain a critical application bottleneck. In this study, the mechanical performance, hysteretic behavior, and complex multi-component failure mechanisms of assembled CFDST beam-to-column joints were systematically discussed subjected to cyclic loading. By addressing the critical design challenge, specifically the accurate prediction of flexural capacity under both sagging and hogging bending moments, this work establishes a comprehensive analytical framework and practical design methodology to guide real-world engineering applications and code implementations.Methods To achieve these objectives, an integrated methodology combining quasi-static cyclic testing, high-fidelity three-dimensional non-linear finite element (FE) modeling, and mechanical analytical modeling was executed. A series of full-scale cyclic loading tests were carried out to evaluate seismic performance across key design parameters, including extended versus flush end-plate configurations, varying column hollow ratios, and different beam types including steel - concrete composite beams and steel H-beams. In parallel, advanced 3D numerical models were developed using finite element software. Concrete damaged plasticity models were assigned to the core concrete, while progressive damage criteria were applied to the outer and inner steel tubes. Surface-to-surface contact formulations with finite sliding were implemented across all interfaces in the joint core, incorporating detailed sensitivity analyses on element types and contact friction coefficients. Crucially, non-linear discrete spring elements were employed to explicitly simulate the force-slip relationship and progressive shear degradation/failure of the shear studs. Based on the stress distributions and interaction mechanisms revealed by the experimental and numerical findings, a component-based analytical model was formulated, in which the force states and work mechanism between multi components were considered.Results and Discussions The results demonstrate that the assembled CFDST joints exhibit exceptional seismic performance, characterized by full, stable hysteretic loops with high energy dissipation capacity and minimal pinching behavior. The primary failure modes observed in both tests and simulations included yielding of the end plate, slip of high-strength bolts, and shear failure of shear studs. The inner steel tube played a critical structural role by acting as an internal radial restrainer that effectively suppressed inward concrete crushing and local instability. The detailed 3D FE model demonstrated high numerical precision, capturing load-displacement hysteretic responses, ultimate load capacities, stiffness degradation, and local failure modes within an error margin of 8% compared to experimental measurements. Stress state evaluations at key loading stages further highlighted the strong composite action within the components in joint core. The inner steel tube, sandwiched concrete together with the outer tube work interactively to resist the internal force, and the tensile force of blind bolt acted on the inner tube dispersed significantly when transferring to the outer tube. The proposed design formulas for sagging and hogging moment resistance were validated against experimental and parametric numerical datasets, achieving excellent agreement with variances remaining consistently below 10%.Conclusions In conclusion, assembled CFDST beam-to-column joints possess robust rotational ductility, high initial stiffness, and substantial energy dissipation capacity, rendering them exceptionally well-suited for seismic-resistant multi-story and high-rise structural systems. The inner steel tube in CFDST columns serves a vital dual function by providing additional flexural resistance and maintaining an internal boundary that mitigates concrete damage under severe cyclic demands. The established 3D non-linear FE modeling technique incorporating discrete non-linear spring elements for stud failure and refined friction contact interfaces provides a proven, highly accurate tool for virtual testing and structural optimization. Furthermore, the proposed component-based design method yields accurate, explicit equations for determining moment resistance, offering practicing structural engineers a direct, trustworthy analytical framework for semi-rigid joint design in civil engineering projects. Beyond the core design framework and hysteretic evaluations, this research yields several additional insights of significant academic and practical value. Parametric evaluations revealed that maintaining an optimal column hollow ratio between 0.5 and 0.7 achieves the ideal balance between self-weight reduction, structural load-bearing capacity, and local buckling prevention in the connection zone. To assist in full-frame structural modeling, empirical calculation formulas were developed to estimate the initial rotational stiffness of assembled CFDST connections, enabling accurate semi-rigid frame analysis within commercial structural design software. Finally, constructability and detailing guidelines are established regarding minimum end-plate thickness and optimized bolt layout geometries to prevent premature brittle shear or tension failure, thereby ensuring that the assembled joint reliably achieves the desirable "strong-joint, weak-member" seismic design hierarchy required by modern building codes.
SignificanceInterconnected control systems have been widely encountered in modern engineering applications, including energy networks, industrial processes, and multi-robot systems. These systems are generally composed of multiple relatively independent subsystems, which are coupled through physical connections, material flows, energy exchanges, or other dynamic interactions, thereby forming an integrated large-scale dynamical system. Compared with conventional high-dimensional systems, the main difficulty of interconnected systems does not merely arise from the increase in system dimensions, but from the strong coupling relationships among subsystems and the resulting challenges in stability analysis and controller design. A fundamental problem in such systems is that satisfactory stability of individual subsystems cannot directly guarantee the stability and performance of the overall system. Dynamic interactions among subsystems may modify the closed-loop characteristics, reduce stability margins, deteriorate transient responses, and even cause instability under certain operating conditions. With the continuous expansion of engineering system scales and the increasing complexity of control objectives, traditional centralized control strategies have encountered considerable limitations in communication requirements, online computational burden, and scalability. The requirement of global system information and centralized decision-making mechanisms makes such approaches difficult to implement in large-scale systems with limited communication resources and real-time constraints. Decentralized control strategies, which have been developed by designing local feedback controllers based only on subsystem-level states or outputs without requiring global state information or information exchange among controllers, provide an effective solution with simple structures, low implementation costs, and favorable scalability. However, due to the existence of physical couplings and dynamic interactions, decentralized controllers designed independently for individual subsystems may still suffer from degraded transient performance, including increased settling time, excessive overshoot, and intensified oscillations. Therefore, improving closed-loop transient performance while maintaining stability under interconnection effects has become an important research issue in decentralized control of large-scale interconnected systems.ProgressIn this paper, representative theoretical achievements and technical developments in decentralized control for large-scale interconnected systems have been systematically reviewed from three perspectives: interconnected system modeling and characterization, coupling effect mitigation, and closed-loop performance analysis. First, centralized, distributed, and decentralized control architectures have been compared from the perspective of information structures. The differences in information requirements, communication mechanisms, computational modes, scalability, and typical application scenarios have been clarified. It has been indicated that centralized control relies on global system information and centralized computation, distributed control requires information exchange among neighboring subsystems through communication networks, while decentralized control mainly relies on local subsystem information without real-time communication among controllers. Then, typical modeling approaches for interconnected systems have been summarized. State interconnection, input interconnection, output interconnection, dynamic interconnection, and time-delay interconnection models have been reviewed according to different physical coupling mechanisms. The characteristics, modeling assumptions, and limitations of these interconnection structures have been analyzed, and their influences on stability analysis and controller design have been discussed. State interconnection models have been mainly used to describe coupling effects caused by subsystem states, while input and output interconnection models have been adopted to characterize interactions through control actions and measured variables. Dynamic and time-delay interconnection models have further been considered to describe more complex physical processes involving internal coupling dynamics, transmission delays, and memory effects. Furthermore, coupling-handling methods under different levels of interconnection information availability have been reviewed. For systems with completely known interconnections, exact compensation, decoupling control, and dynamic compensation approaches have been developed by directly utilizing structural information of coupling terms. For systems with bounded but uncertain interconnections, robust control, sliding-mode control, input-to-state stability analysis, and small-gain techniques have been widely investigated to guarantee stability without requiring accurate coupling models. For systems with unknown parameters, adaptive control, backstepping methods, and dynamic surface control approaches have been adopted to estimate uncertain parameters online and reduce the influence of modeling errors. When nonlinear interconnections are completely unknown, neural-network approximation, fuzzy approximation, and observer-based reconstruction methods have been explored to approximate unknown dynamics. The applicable conditions, theoretical advantages, and inherent limitations of these approaches have been compared, particularly regarding conservatism, computational complexity, and practical implementation difficulties. In terms of closed-loop performance improvement, the evolution of decentralized control objectives from asymptotic stability to finite-time control, fixed-time control, prescribed performance control, and prescribed-time control has been reviewed. Different control paradigms have been analyzed according to convergence characteristics, dependence on initial conditions, transient response improvement, and engineering implementation constraints. In addition, considering the safety requirements of practical systems, state constraints, output constraints, and safety control methods have been discussed. The integration of constraint handling and safety mechanisms into decentralized control has extended the research objective from guaranteeing final stability to ensuring safe operation throughout the entire control process.Conclusions and ProspectsAlthough significant advances have been achieved in decentralized control of interconnected systems, several challenges remain open. For strongly coupled and heterogeneous systems, unified modeling frameworks and effective analysis methods are still required to describe complex interconnection structures while reducing conservatism in stability conditions. The treatment of unknown interconnections remains challenging because existing approaches often rely on predefined bounds or structural assumptions, which may limit their applicability in uncertain operating environments. Meanwhile, improving transient performance through rapid convergence mechanisms may introduce large control gains, noise amplification, and actuator saturation problems, leading to difficulties in practical implementation. The integration of intelligent control methods also requires further investigation to ensure stability, interpretability, and reliable performance guarantees rather than relying solely on data-driven optimization. Future research is expected to focus on several important directions, including modeling and analysis of strongly coupled heterogeneous systems, adaptive compensation of unknown interconnections, model-data fusion-based decentralized control, intelligent enhancement with stability guarantees, and resilient control under safety constraints and external disturbances. These developments are expected to promote decentralized control theory toward more scalable, reliable, and practical solutions for future large-scale complex engineering systems.
Objective Floating-object detectors deployed in complex water environments must recognize small, weak-texture, and dynamically deformed targets under changes in camera location, weather, surface ripples, illumination, and occlusion. Existing deep detectors often depend on large fully annotated datasets, incur excessive computational and storage costs, and lose accuracy in unseen water scenes. Conventional domain generalization also requires multiple completely labelled source domains, whereas single-labelled domain generalization reduces annotation work but is vulnerable to source-specific bias. A double-labelled domain generalization (DLDG) method was therefore developed to balance detection accuracy, real-time efficiency, annotation cost, and generalization. The method used only two labelled source domains and several unlabelled source domains while explicitly correcting feature-extraction, classification, and localization biases.Methods A lightweight multiscale detector was constructed by coupling MobileNetV3, a Dynamic Feature Pyramid Network (DyFPN), and a Single Shot MultiBox Detector (SSD). The VGG16 backbone of SSD was replaced by MobileNetV3, and feature maps from different stages were delivered to DyFPN. Dynamic modules containing a gate, an Inception unit, and a skip connection were inserted into the lateral connections. A gating signal and Gumbel-Softmax one-hot decision determined whether each lateral convolution was executed. Useful cross-scale information was therefore selected according to the input, while redundant reflections, ripples, shoreline textures, and computation were reduced. Depthwise separable convolutions were also applied to the classification and box-regression layers. DLDG training was organized as a progressive "feature generation-category discrimination-spatial localization" process. Two labelled domains were first used to initialize the feature generator, classifier, and locator through cross-entropy and Smooth L1 losses. The remaining unlabelled domains were then used to filter three types of bias. For feature-extraction bias, cluster centers were calculated from SoftMax outputs and bounding-box regression results. Pseudo-labels were assigned according to cosine distance, and information maximization improved cluster separability and prevented concentration in a few classes. Both classification and localization information constrained pseudo-label generation, after which the feature extractor was updated using classification and localization losses from unlabelled domains. For classification bias, a conditional feature-projection network mapped unlabelled features into the discriminative space jointly defined by the two labelled domains. Consistency between true labels and pseudo-labels constrained projection, and domain-similarity attention reweighted common features across domains. For localization bias, foreground-anchor regression was trained with pseudo-labels. An adversarial regressor maximized its prediction discrepancy from the main regressor on source samples, and generalized intersection over union measured box differences. The three stages shared features and pseudo-labels and were optimized sequentially under a unified objective. Experiments were conducted using fixed-camera images from the modern water-conservancy demonstration area in Deqing County, Zhejiang Province, China. Sixty representative video sequences were sampled. After redundant frames were removed, 526 water-hyacinth images, 340 floating-weed images, and 300 plastic-bottle images were retained; training-set augmentation produced 11,623 samples. The 1,920 × 1,080-pixel images were divided into five scene domains: normal water surfaces, weather variation, wave interference, illumination variation, and target-level occlusion. The dataset was split 9:1 for training and validation, and 30 labelled-source/target combinations were generated. Experiments used PyTorch 1.10, Ubuntu 18.04, an Intel i7 processor, and an NVIDIA RTX 3080 graphics card. Stochastic gradient descent was configured with a momentum of 0.9, an initial learning rate of 0.01, a batch size of 16, and weight decay of 0.001. Performance was evaluated using mean average precision (mAP), F1 score, frames per second (FPS), floating-point operations, parameter count, and model size.Results and Discussions For double-labelled domain generalization, DLDG achieved 70.33% mAP, 71.02% F1, and 22.38 FPS on the GPU. It required 3.01 billion floating-point operations, 4.98 million parameters, and 21.24 MB of storage; CPU inference reached 7.98 FPS. The strongest compared method, domain-invariant feature enhancement domain adaptation, obtained 66.48% mAP, 66.85% F1, and 20.10 FPS. DLDG therefore improved mAP by 3.85 percentage points, F1 by 4.17 percentage points, and speed by 2.28 FPS, while using fewer parameters and less storage. Under conventional domain generalization with fully labelled source data, the method achieved 85.29% mAP, 86.28% F1, and 17.81 FPS, indicating that the progressive filtering mechanism was applicable under both limited-label and fully labelled settings. The network ablation showed that MobileNetV3-DyFPN reached 86.28% mAP, 87.33% F1, and 19.15 FPS, compared with 70.02% mAP, 71.28% F1, and 12.99 FPS for VGG16 with DyFPN. Thus, mAP increased by 16.26 percentage points and speed by 6.16 FPS. Bias-filtering ablations showed that the unfiltered model produced 62.29% mAP and 63.02% F1 at 24.37 FPS. Introducing all three filters increased mAP by 8.04 percentage points and F1 by 8.00 percentage points, while speed decreased by 1.99 FPS. Feature-extraction filtering produced the largest single-module improvement, increasing mAP by 2.43 percentage points. Localization filtering contributed more than classification filtering because adversarial regression and generalized intersection over union directly corrected box displacement. In wave-interference and low-illumination scenes, DLDG produced boxes more consistent with target boundaries, whereas FixMatch and open compound domain adaptation showed missed detections or localization shifts. Stable detections were also obtained for small plastic bottles and floating weeds whose edges were mixed with reflections and ripples.Conclusions Combining lightweight dynamic multiscale feature extraction with progressive filtering of feature, classification, and localization biases reduced dependence on extensive bounding-box annotation while maintaining real-time cross-domain detection. Two labelled domains provided model initialization, and additional unlabelled domains were exploited through classification-and-localization-constrained pseudo-labels, conditional feature projection, domain-similarity reweighting, and adversarial box regression. The method can support fixed-camera identification of floating-object accumulation, cleaning prioritization, and continuous inspection in heterogeneous water environments. The current domain definition was based mainly on visible water-surface states and object distributions because synchronized flow velocity, water level, wind speed, and rainfall were unavailable. Future integration of hydrodynamic and meteorological measurements, unmanned surface vehicles, and hydrological monitoring data could clarify the relationships among water conditions, floating-object transport, and detection performance.
Objective Precise tracking of continuous trajectories on irregular surfaces by robotic manipulators is essential for industrial applications such as free-form surface grinding, polishing, and welding of complex workpieces. Existing trajectory planning methods are predominantly designed for regular paths or point-to-point motions, and cannot be directly applied to complex surface processing tasks. This paper proposes an adaptive joint trajectory planning method for six-axis robotic manipulators that achieves continuous trajectory tracking on irregular surfaces while satisfying joint-level performance requirements including smoothness and safety.Methods The proposed method consists of three core components. First, the target trajectory on an irregular surface is uniformly sampled at a specified interval. A low-dimensional continuous basis function is constructed by mapping the end-effector coordinate frame space, parameterized by three spherical angles (azimuth, elevation, and rotation about the radial vector), to the joint variable space through binding inverse solutions obtained via the Jacobian iteration method. The Jacobian iteration method computes the differential change in joint variables based on the pose error between the current end-effector configuration and the target, using the manipulator Jacobian matrix. By varying the initial joint configuration across multiple iterations, multiple distinct inverse solutions for each end-effector pose are obtained. The basis function is designed to be continuous and distinguishable across different inverse solution branches, satisfying the requirements of subsequent optimization algorithms. A dimensional reduction function further maps the six-dimensional joint variables to a one-dimensional representative value, enabling the optimization algorithm to efficiently differentiate among multiple inverse solutions for the same end-effector pose. In cases where distinct solution branches intersect, a piecewise binding strategy is employed to maintain continuity. Second, evaluation functions for joint trajectory smoothness and safety are designed. The smoothness evaluation function quantifies the rates of change in joint position and velocity between adjacent sampling points using weighted summations across all joints, where weighting factors balance the dimensional differences between prismatic and revolute joints. The safety evaluation function checks for model interference between the manipulator, the workpiece, and the environment by verifying whether simplified geometric envelope models of the manipulator collide with the workpiece or surroundings at each selected joint configuration. A survival flag is introduced to exclude infeasible solutions from the optimization process. These evaluation functions are combined with the basis function to form a composite value function. When the optimal solutions of different performance indicators are closely located, the multi-objective optimization can be simplified to a single-objective formulation with weighted summation to improve computational efficiency. A Multi-Objective Particle Swarm Optimization (MOPSO) algorithm, incorporating Pareto dominance, an external archive, adaptive grid division, and roulette wheel selection, is then employed to traverse all sampling points and select the approximately optimal joint variable at each point. Third, a quadratic interpolation strategy is adopted to control TCP velocity. After the first cubic spline interpolation generates an initial joint trajectory, a high-density resampling of the joint variables is performed, and a second cubic spline interpolation with refined time-node allocation is executed to significantly suppress TCP velocity fluctuations while maintaining trajectory tracking accuracy. The boundary conditions for both interpolation passes are set as the average slopes between the first two and last two points of the respective sequences.Results and Discussions Simulation experiments were conducted on two manipulators of different configurations—a six-axis articulated manipulator (Manipulator 1) and a Cartesian manipulator (Manipulator 2)—along two complex workpiece trajectories. The MOPSO algorithm was configured with a swarm size of 40 and 30 iterations per sampling point. The basis function was defined by binding the end-effector frame to the inverse solution with the minimum representative value. Results demonstrate that for Manipulator 1, the maximum and average joint velocities did not exceed 0.30 rad/s and 0.03 rad/s, respectively, and the maximum and average joint accelerations remained below 0.61 rad/s² and 0.04 rad/s². For Manipulator 2, the maximum linear joint velocity was within 3.08 cm/s with an average of 0.57 cm/s, the maximum linear joint acceleration was 22.01 cm/s² with an average of 0.47 cm/s², and the maximum rotational joint velocity was within 1.32 rad/s with an average of 0.13 rad/s. The maximum TCP trajectory error across all four test scenarios (two manipulators × two trajectories) was 2.00 mm, with an average error of 0.14 mm. The maximum TCP velocity error was 0.021 mm/s with an average of 0.000 3 mm/s. All joint trajectories remained strictly within their respective joint limits, and no model interference was detected throughout the entire motion process in any scenario. Physical experiments were performed using a DR270 manipulator (identical in configuration to Manipulator 1) and a Leica AT403 laser tracker to further validate the method. A spherically mounted retroreflector (SMR) was attached to the end-effector, and its position was tracked in real time as the manipulator followed the planned trajectory. The measured maximum SMR trajectory error was 3.26 mm with an average of 2.34 mm, and the maximum velocity error was 3.20 mm/s with an average of 0.91 mm/s. The increased errors compared to simulation are attributed to laser tracker measurement precision, sampling density, and manipulator control errors. Nevertheless, the achieved accuracy satisfies the requirements of most industrial applications, confirming the feasibility and practicality of the proposed method in real-world engineering environments.Conclusions An adaptive joint trajectory planning method for six-axis robotic manipulators was proposed and validated for continuous trajectory tracking on irregular surfaces. The method establishes a complete technical framework encompassing end-effector coordinate frame space mapping, composite value function construction, and trajectory generation via quadratic interpolation. The key contributions include: (1) a low-dimensional continuous basis function that enables unified representation and differentiation of multiple inverse solutions for the same end-effector pose, providing a solid foundation for optimization-based joint variable selection; (2) a composite value function mechanism that adaptively satisfies joint trajectory performance requirements such as smoothness and safety within a unified optimization framework, accommodating manipulators of different configurations and varying process-specific demands; and (3) a quadratic interpolation strategy that effectively resolves the velocity fluctuation problem inherent in single-pass interpolation, achieving a balance between trajectory tracking accuracy and velocity stability through dense resampling and refined time-node allocation. Both simulation and physical experiments confirm the method's versatility across different manipulator configurations and trajectory complexities, demonstrating good engineering practicality for industrial applications involving irregular surface processing.
Objective Accurate characterization and reconstruction of manufacturing deviation fields are essential for geometric quality evaluation of complex components considering actual manufacturing conditions. However, mapping non-developable surfaces whose Gaussian curvature is not identically zero into regular two-dimensional parameter domains inevitably introduces metric distortion, which affects the representation of spatial distribution and reconstruction accuracy of manufacturing deviation fields. Existing deviation representation methods based on regular parameter domains still have difficulties in fully considering the influence of parameterization distortion on the preservation of deviation characteristics. Therefore, a manufacturing deviation reconstruction method for non-developable surfaces based on quasiconformal mapping was proposed to improve the stability of manufacturing deviation representation in two-dimensional parameter domains.Methods First, the manufacturing deviation field of the measured component was constructed based on the theoretical geometric model. The measured point cloud was registered with the theoretical CAD model, and the projection distance of measured points along the unit normal direction of the corresponding theoretical surface was defined as the normal manufacturing deviation. Accordingly, the three-dimensional manufacturing deviation field of the actual component was represented by discrete normal deviation values, providing the data basis for subsequent parameter-domain representation and reconstruction. Second, a quasiconformal rectangular parameterization method based on Teichmüller-type constraints was established for non-developable surfaces. The local parameterization distortion was characterized by the Beltrami coefficient field, where the modulus of the Beltrami coefficient was used to describe the local anisotropic stretching degree. According to the conformal modulus relationship between the surface quadrilateral and the target rectangular domain, the theoretical extremal dilation ratio and the corresponding target Beltrami modulus were determined. For the hemispherical shell used in the validation, the conformal modulus relationship between the source surface quadrilateral and the target square domain was calculated, and the target Beltrami modulus was obtained as 0.07477. In addition, the target direction of the Beltrami coefficient was constructed using the discrete quadratic differential derived from the harmonic coordinates of the surface quadrilateral. Based on the target modulus and target direction, the Beltrami coefficient field was decomposed into modulus and direction components, and the corresponding constraints were introduced to regulate the parameterization distortion toward the target state. The modulus-deviation energy and direction-deviation energy were defined to evaluate the convergence of the iterative optimization process. The updated Beltrami field was then inversely solved through the discrete Beltrami equation to obtain the corresponding rectangular parameter mapping. Third, the manufacturing deviation field represented in the optimized parameter domain was converted into a regular grid matrix and decomposed into orthogonal frequency-domain modes using the two-dimensional discrete cosine transform (2D-DCT). According to the energy contribution of each DCT mode, key modes were selected to achieve sparse representation of the manufacturing deviation field. The reconstructed parameter-domain deviation field was finally mapped back to the three-dimensional surface along the normal direction of the theoretical model, and the reconstruction accuracy was evaluated using the root mean square error (RMSE) and relative reconstruction error (RRE). Finally, an aluminum alloy hemispherical shell with a theoretical radius of 250 mm was selected as the validation object. A high-precision three-dimensional scanning system was used to obtain the measured surface point cloud, and the proposed method was applied to reconstruct its manufacturing deviation field.Results and Discussions The experimental results demonstrated that the proposed quasiconformal parameterization method reduced the distortion variation caused by mapping the non-developable surface into the regular parameter domain. During the iterative optimization process, the average Beltrami modulus decreased from 0.43640 to 0.07566, approaching the target modulus of 0.07477. The standard deviation of the Beltrami modulus decreased from 0.27491 to 0.01557, and the 99th percentile value decreased from 0.97108 to 0.12023, indicating that the local distortion distribution became more concentrated and the high-distortion regions were effectively suppressed. Meanwhile, the modulus-deviation energy decreased from 2.0635×10-1 to 2.4317×10-4, and the direction-deviation energy decreased from 1.92635 to 0.02101, demonstrating that the optimized parameterization gradually approached the target modulus and direction constraints. The average extremal dilation ratio of the final mapping was 1.16453, which was close to the theoretical value of 1.16163. Based on the optimized parameterization, the manufacturing deviation field was transformed into a regular parameter domain and represented using DCT modes. The influence of different parameter grid resolutions on the number of retained modes and reconstruction error was analyzed. When a 401×401 regular grid was adopted, the reconstruction error became stable while the number of retained modes remained acceptable. Under a cumulative energy threshold of 99.9%, the number of DCT modes required for reconstruction was reduced from 160801 to 6414, corresponding to a mode compression ratio of 96.011%. The reconstructed three-dimensional manufacturing deviation field preserved the main morphological characteristics and spatial distribution trends of the original deviation field. The global RMSE, RRE, and mean absolute error were 0.01909 mm, 4.084%, and 0.01244 mm, respectively, and the absolute reconstruction errors of 95% of the sampling points were less than 0.03505 mm. The ablation analysis further indicated that introducing the Beltrami modulus constraint reduced the retained DCT modes from 12079 to 6499 and decreased the RMSE from 0.04734 mm to 0.01952 mm. After incorporating the direction constraint based on the discrete quadratic differential, the retained modes and RMSE were further reduced to 6414 and 0.01909 mm, respectively. These results indicate that simultaneously controlling the modulus distribution and directional structure of the Beltrami coefficient field contributes to improving the compactness and accuracy of manufacturing deviation representation.Conclusions A manufacturing deviation reconstruction method combining Teichmüller-type constrained quasiconformal parameterization and DCT-based modal representation was proposed for non-developable surfaces. By regulating parameterization distortion before frequency-domain decomposition of deviation fields, the stability of manufacturing deviation representation in regular two-dimensional parameter domains was improved. The combination of the Beltrami modulus constraint and the discrete quadratic differential-based direction constraint enabled the parameterization result to approach the target quasiconformal state, thereby reducing the influence of geometric distortion on deviation reconstruction. The proposed method provides an approach for sparse representation and accurate reconstruction of manufacturing deviation fields of complex non-developable components.
