Abstract:In recent years, with the deep integration of unmanned aerial vehicle (UAV) technology and embodied intelligence, low-altitude UAV has demonstrated great potential in fields such as smart logistics, urban inspection, and disaster response, owing to their flexibility, multimodal perception, and autonomous decision-making capabilities. However, the complex and dynamic low-altitude environments, characterized by dense obstacles and meteorological disturbances, pose significant challenges to UAV in terms of perception accuracy, real-time decision-making, and system robustness. Although extensive research has been conducted on UAV embodied intelligence, most efforts focus on isolated technical modules, lacking a systematic overview and cross-domain integration. To address this gap, this paper provides a comprehensive review of key technologies and recent advances in low-altitude UAV embodied intelligence from the perspective of the perception-decision-control loop. It summarizes the development of multimodal perception, task reasoning, physical interaction, and simulation-based training, and further discusses promising future research directions.

Abstract:With the acceleration of urbanization and the continuous expansion of rail transit networks, rail transit systems face the demand for more efficient, refined, and intelligent solutions in operation scheduling, facility maintenance, and train control. Artificial intelligence (AI), with its advantages in data perception, pattern recognition, and intelligent decision-making, is profoundly reshaping the technical system and management mode of rail transit. From the core perspective of AI-empowering rail transit, this paper systematically reviews the current research progress and engineering practices around the three key links of operation optimization, equipment maintenance, and intelligent driving, and summarizes the main challenges of existing methods in terms of high-dimensional data, insufficient model interpretability, and limited system integration. Furthermore, the coping paths of AI technologies such as deep learning, graph neural networks, collaborative multi-agent reinforcement learning, self-supervised learning, and digital twins are discussed. Finally, the trends of future rail transit in the directions of computing power support, standard system construction, and human-machine collaborative agent development are prospected, aiming to provide theoretical reference and practical enlightenment for the deep integration of AI and rail transit.

Abstract:In order to investigate the early warning method for bridge assessment under random train loading with joint distributed macro-strain monitoring and machine learning, and to realize the distributed rapid assessment of railroad bridges, this study establish a three-dimensional refined finite element model of vehicle-rail-bridge coupled vibration. It is also apply load statistical analysis methods to construct a stochastic traffic flow model that is suitable for actual train operation, and based on the principle of distributed monitoring, propose a distributed macro-strain influence line area as the indicator design warning interval evaluation warning method for bridge warning; Furthermore, through simulation analysis of various stiffness degradation conditions, a distributed macro-strain monitoring data sample library under random train loads was constructed to compare and study the accuracy of 4 machine learning methods in quantifying and locating bridge damage. The results show that all 4 types of machine learning are able to localize and quantify the localized damage of bridge structures with an average recognition accuracy of 90.0%, with the KNN model and the SVM model performing the best in the test of quantifying bridge damage, both with 95.0% recognition accuracy, and the SVM model performing the best in the test of locating the damage of the bridge structure, with a recognition accuracy of 98.3%. The joint distributed macro-strain monitoring and machine learning approach for bridge assessment has feasibility, SVM model performs best in the test of bridge structure damage localization, KNN model and SVM model perform best in the test of bridge damage quantification, and in the comprehensive analysis, SVM performs best in bridge damage localization and damage quantification analysis.

Abstract:In order to solve the problems that traditional research methods often have limitations in dealing with high-dimensional and complex data features in complex environments, and it is difficult to achieve high-precision and robust prediction, a traffic accidents severity intelligent prediction method based on convolutional neural network and attention mechanism is proposed. A multi-scale feature extraction model with attention fusion, which is called channel and multi-head attention network (CMHANet), is constructed to make full use of the advantages of convolution and attention mechanism. The convolution layer is used to effectively extract spatial features in the data, while the channel attention mechanism can weight and enhance important features, suppress unimportant features, and emphasize the focus on key data points. In addition, in order to capture the complex dependencies between different features, a multi-head attention mechanism is also introduced. Finally, experiments are conducted on the US-Accidents dataset. The experimental results show that the prediction framework with this model as the backbone achieves improvements in F1-score, precision, recall and accuracy. While improving the effect of feature extraction and association modeling for high-dimensional and complex data, the proposed model also provides a new idea for intelligent prediction of traffic accidents.

Abstract:This study aimed to address the reduced durability and corrosion issues of traditional multi-span simply supported beam bridges caused by concrete cracking and design flaws. A novel steel-concrete composite flat slab continuous bridge deck structure (referred to as the flat slab continuous) is proposed, which effectively prevents surface cracking by keeping the surface concrete to compression under positive bending moments. With a three-span simply supported beam bridge as the engineering context, this research conducted finite element numerical simulation on the deformation and force characteristics of the flat slab continuous and the traditional tie-rod continuous simply supported beam bridges under various vehicle loads by establishing a detailed finite element model. The results indicate that under mid-span load and off-center load, the maximum tensile stresses for the traditional tie-rod continuous were 1.16 MPa and 4.50 MPa, respectively, in contrast, the upper surface of the novel flat slab continuous was in compression. Furthermore, comparative analysis shows that under overload and anti-tipping conditions, the deflection difference at the mid-span cross-section of the main beam for the flat slab continuous simply supported beam bridge was reduced by 24.50% and 13.43%, respectively. The findings demonstrate that the flat slab continuous can effectively prevent the occurrence of deck cracking and has a significant advantage in improving the bridge's anti-tipping performance, offering a new perspective for the design and retrofitting of multi-span simply supported beam bridges.

Abstract:The objective of this study is to examine the service performance of a cantilevered enclosure during the excavation of a high slope pit under rainfall conditions. The high slope sinking plaza pit of Shenzhen Wuhe Intercity Hub Station is taken as the research object. Based on the field measurements, an analysis is conducted of the evolution of the wall top displacement of the cantilevered enclosure diaphragm wall of the high slope pit under the condition of continuous heavy rainfall. This study focuses on the ZQ-8 hazardous monitoring section. The depth of the soil body in the saturated area outside the pit was calculated, and the saturated soil and soil at the bottom of the pit outside the pit were softened using the strength reduction method. A numerical simulation analysis was conducted to study the softening of soil outside the pit. A simulation analysis was employed to investigate the impact of soil softening both within and outside the pit, as well as the reinforcement of anchor cables, on the internal force and deformation of the ground connecting wall. The results demonstrate that: In the event of soil softening within and around the pit, the maximum bending moment of the ground-connected wall reaches 1 600 kN·m, which is in close proximity to the ultimate design bending moment. Additionally, the maximum horizontal displacement of the enclosure reaches 76.0 mm; In the case of significant deformation of a cantilever-supported foundation pit with high slopes under continuous rainfall conditions, the use of prestressed anchor cables for emergency treatment is recommended, while backfill counterpressure has limited efficacy. The research findings can provide theoretical support for safe construction in the field and can also serve as a reference for other similar projects.

Abstract:To study the full frequency domain vibration of track structures, a frequency domain vibration model of ballastless track structures is established based on spectral geometry method. Firstly, the track structure is discretized into several spectral geometric element,which are coupled to each other by setting virtual springs. Secondly,the characteristic matrix,element coupling stiffness matrix, and load vector of the spectral geometric element of the ballastless track structure is derived by the Rayleigh-Ritz method. Adopting the principle of element matching, the spectral geometric element characteristic matrix and element coupling stiffness matrix of the track structure are combined into the overall characteristic matrix of the track structure. By solving the spectral geometric dynamic equation of the track structure, the full frequency vibration response of the track structure is obtained. Finally, a program is developed using Matlab to verify the feasibility of the spectral geometry method. By comparing with the finite element method, the feasibility and high efficiency of the spectral geometry method is verified. The influence of track structure parameters on the vibration characteristics of track structures in the full frequency domain has been studied. The calculation results show that the computational speed of the spectral geometry method is 8 times faster than that of the finite element method within the frequency range of 1~2 000 Hz; fastener stiffness mainly affects the third-order natural frequency of rail; the CA mortar stiffness mainly affects the second-order mainly natural frequency of the track structure; the subgrade stiffness mainly affects the first-order natural frequency of the track structure; the fasteners damping mainly attenuates the peak values at the second-order natural frequencies of track structures and the third-order natural frequencies of rails; the damping of CA mortar mainly attenuates the peak value at the second-order natural frequency of the track structure.The subgrade damping mainly attenuates the peak value at the first-order natural frequency of the track structure; the Pinned-Pinned frequency of rails is not affected by the stiffness of fasteners, CA mortar stiffness,and subgrade stiffness,and is mainly related to the spacing between the rail fasteners.The research results can provide efficient computational methods and technical support for vibration and noise reduction within the wide-frequency range of track structures.

Abstract:To analyze the effect of fire temperature gradient on shear lag effect of the corrugated steel web composite box girder, based on the energy variational method and combined with the temperature stress theory of steel-concrete composite beams, this paper deduces the theoretical analysis formula of shear hysteresis-effect of composite box beams with corrugated steel webs under fire temperature gradient, and establishes the thermal coupling numerical analysis model of composite box beams with ANSYS finite element software to verify the applicability of the theoretical formula. The variation of shear lag effect of composite box girder with corrugated steel web under different fire time and load ratio is further analyzed. The error of the analytical solution, finite element numerical solution and measured values of the shear lag effect of the composite box girder with corrugated steel webs derived in this paper is less than 7%, which indicates the good applicability. Under fire, the shear lag effect of composite box girders is significant, and it becomes more pronounced as the temperature increases. The maximum shear lag coefficient λ_max of the composite box girder increases first and then decreases with the increase in fire time. The shear lag effect of the composite beam decreases with the increase of load ratio. λ_max is greater than that under normal temperature when the load ratio is 0.2 and 0.3, and less than normal temperature when the load ratio is 0.4. The research results can provide a reference for the analysis and calculation of the shear lag effect of corrugated steel web box girder under fire temperature gradient.

Abstract:In order to study the temperature warping effect of CRTS Ⅲ slab ballastless track under continuous high temperature environment, the field test method was used to monitor the temperature and strain of track slab for a long time. Based on the monitoring data, the temperature distribution law of the track slab under continuous high temperature was studied, and the temperature warping effect of the track slab was analyzed. The results show that the temperature of the track slab shows a wave-like upward trend in the continuous high temperature environment. Among them, the bottom surface temperature lags behind the top surface temperature, the phase difference is about 1.5~4 h, and the positive temperature gradient time in the plate is greater than the positive temperature gradient time at the edge of the plate. Under the positive temperature gradient load, the curvature is positive, and the track slab is arched; under the negative temperature gradient load, the curvature is negative, and the track slab warps downward. The linear function relationship between the warping curvature of the track slab and the vertical temperature gradient is fitted. The fitting coefficient R² is above 0.85, and the temperature strain change value and the warping curvature value increase with the increase of the vertical temperature gradient. The research results can provide experimental basis and data support for the temperature warping deformation of CRTS Ⅲ slab ballastless track under continuous high temperature environment, and provide important guarantee for ensuring the safe, stable and efficient operation of high-speed railway.

Abstract:To explore the effects of synchronous grouting on tunnel structure uplift and ground settlement, this study proposes an improved grouting material to optimize construction performance. Microbubbles were introduced into the grout by adding an air entraining agent, endowing the grout with lightweight, elastic and compressible properties. A combination of laboratory experiments and field monitoring was employed to evaluate the influence of air entraining agent content on grout properties and construction performance. A lightweight, elastic, and compressible synchronous grouting material for shield tunnels based on the air entraining agent was developed. The study revealed that under identical grouting pressure and volume conditions, the lightweight grouting material effectively suppressed tunnel uplift during construction, reducing the average uplift from 8.7 mm to 0.1 mm. Additionally, ground settlement over 14 d decreased from 10.5 mm to 4.5 mm, a reduction of approximately 57%. The air-inducing effect of the air entraining agent enables grout lightweighting and enhances its performance, significantly mitigating tunnel uplift and ground settlement during shield tunnel construction. This provides an efficient grouting material solution for tunnel engineering projects.

Abstract:Existing defect detection algorithms for power equipment are difficult to ensure both detection accuracy and speed, and the large scale of model parameter redundancy is a challenge for deployment in edge-side embedded devices. In this paper, we propose a power equipment defect detection technique based on federated gradient score correction(FedGSC) algorithm and real-time-detection Transformer(RT-DETR). First, the lightweight backbone network GhostNet is used to replace the original backbone network of the model in RT-DETR, and the model volume is further compressed using channel pruning, which significantly reduces the redundant parameters and improves the inference speed; on this basis, the federated learning architecture is introduced into the cloud server for the distributed training of the lightweight RT-DETR model at the edge end in order to solve the problem of non-independent homogeneous distribution(NID) in the federated learning training process. Non-independent and identically distributed(Non-IID) data exists, FedGSC is introduced to correct the gradient of each round of model update. Comparing the lightweight RT-DETR with traditional RT-DETR and YOLOv8, the algorithm model size is only 47 MB, and the mean average precision(mAP) is 90.46%, which can quickly and accurately identify the defects of power equipment; and the FedGSC algorithm reduces communication costs by approximately 40% and 20% compared to the FedAvg and FedFV algorithms, respectively.

Abstract:Large language models (LLMs) are often trained with instruction fine-tuning to adapt to downstream tasks to enhance their generalization ability, but this method has certain performance limitations for LLMs' classification tasks, and sometimes cannot meet the task requirements. To address this issue, a global feature extraction classification large model framework is proposed. This framework uses the global feature extraction enhancement method proposed in this paper to release global features in the attention layer, then enhance the features, and apply the depth low-rank fine-tuning optimization loss proposed in this paper during fine-tuning. Finally, a global feature extraction classification large model is constructed. Compared with the baseline model RoBERTa, the accuracy on the general sentiment analysis dataset SST-2 and AGNews was improved by 1.44 and 0.95 percentage points, respectively. Compared with the baseline model PIQN, the F₁ score on the general named entity recognition (NER) dataset OntoNotes and CoNLL2003 was improved by 0.79% and 1.99%, respectively. The experimental results show that, under the condition of not requiring complex prompt engineering or external knowledge, the performance of the large model using this framework is significantly better than that of its several times larger LLMs.