Highway assets serve as an important component of transportation infrastructure, encompassing highway structures, land used for highways and various ancillary facilities. Addressing the challenge of existing lightweight models struggling to balance accuracy and efficiency in multi-scale detection of various kinds of highway asset facilities, an improved lightweight multi-scale detection model based on YOLOv8n was proposed. Three kinds of representative highway asset facilities, including streetlights, traffic signs and pavement markings were selected as detection objectives. By fusing unmanned aerial vehicle (UAV) field survey data with the VisDrone2019 data, the UAV-HIA dataset was established to enhance data diversity and model robustness. Model improvements included: replacing the backbone network with MobileNetV3-Small to reduce model parameter count and computational complexity; embedding the CBAM attention mechanism in the backbone network to enhance the ability to extract features of small targets; designing the C2iAF feature fusion module based on C2f and iAFF to improve the expression capability of multi-scale features. Experiments demonstrate that the improved model maintains accuracy improvement while significantly reducing computation and parameter quantity, especially achieving better detection performance for small targets. Compared to other existing mainstream models and the newly released YOLO model, the improved model exhibits comprehensive advantages in efficiency, accuracy and adaptability, making it suitable for practical intelligent inspection tasks of highway assets.

Freeze-thaw cycles (FTCs) exert a significant impact on the stability and safety of engineering during the construction, operation and maintenance in clod regions. To investigate the impact of FTCs on the hydro-thermal deformation interaction within gravel interlayer structures, silty clay and gravel were selected to construct the gravel interlayer structure. By monitoring the changes in soil temperature, unfrozen water volume, soil pressure, and displacement deformation of the specimens during the freeze-thaw cycle, the influence rule of the freeze-thaw cycle on the gravel interlayer structure was revealed, and three kinds of machine learning algorithms were employed to develop predictive models for vertical displacement in gravel interlayer structures, respectively. The test results reveal that notable temperature fluctuations occur within the gravel interlayer during the FTCs, and the gravel interlayer has a certain heat insulation effect on the lower soil. Additionally, under the FTCs, both the silty clay samples and the gravel interlayer samples undergo internal moisture migration and redistribution. However, the unfrozen water content in the gravel interlayer structure remains relatively stable. The gravel interlayer structure is less prone to significant deformation during the FTCs, whereas the displacement deformation and frost heaving/thaw settlement rates of silty clay samples are both higher than those of gravel interlayer samples. During a single FTC, the soil pressure of the silty clay sample initially decreases and then increases, while the soil pressure variation in the gravel interlayer exhibits two distinct peaks. The XGBoost model can effectively predict the vertical displacement of the gravel interlayer structure, with a correlation coefficient (R2) between the predicted and measured values greater than 0.93.

Loess exhibits characteristics such as large pores, weak cementation and loose structure, making it prone to deterioration under wet-dry cycles, which can trigger foundation and engineering structure diseases in loess areas. Through uniaxial compression, axial fracturing, scanning electron microscopy (SEM), and nuclear magnetic resonance (NMR) tests, the tensile and compressive mechanical properties and microstructures of lignin-modified loess under wet-dry cycles were systematically investigated. The results indicate that with an increase in lignin content, the silt content of the modified loess firstly increases and then decreases, while the clay content firstly decreases and then increases. Both the compressive and tensile strengths initially increase and then decrease with increasing lignin content, reaching an optimal dosage of 1.0%, at which point the brittleness is minimized. The structure of the modified soil is primarily characterized by point-to-surface contacts and embedded arrangements, while wet-dry cycle promotes a gradual shift in contact patterns towards point contact and aerial arrangement. The pore volume ratio of the samples firstly decreases and then increases with increasing lignin content, reaching its minimum at a dosage of 1.0%.

To achieve the resource utilization of solid waste from waste rubber powder and plastic, and to enhance the road performance of rubber-plastic alloy modified asphalt pavement, a pretreatment technology for rubber-plastic alloy modifiers was proposed, which involved desulfurization activation of rubber powder using composite waste edible oil and microwave irradiation, as well as grafting of maleic anhydride onto plastic. The rules of different pretreatment methods influencing the rheological behaviors of rubber-plastic alloy modified asphalt were analyzed by adopting conventional asphalt performance test, dynamic shear rheometer (DSR), multiple stress creep recovery (MSCR) test, and bending beam rheometer (BBR) test. Meanwhile, the microstructural characteristics and modification mechanisms were investigated by using Fourier transform infrared spectrometer (FTIR) and fluorescence microscope (FM). The results indicate that the optimal dosage of the rubber-plastic alloy modifier is 32%. After activation and grafting pretreatment, the softening point difference of the rubber-plastic alloy modified asphalt decreases, the rutting factor increases, the bending creep stiffness decreases while the creep rate increases. Pretreatment effectively improves the compatibility of the rubber-plastic alloy modified asphalt and enhances its capability of high-temperature resistance to permanent deformation and performance of low-temperature crack resistance. The pretreated rubber-plastic alloy modifier is more uniformly dispersed in asphalt with reduced agglomeration, and the modification process is physical. The rubber-plastic alloy modified asphalt prepared with rubber powder activated by microwave irradiation and plastic grafted with maleic anhydride exhibits superior comprehensive performance.

Corrosion of long-distance water pipelines not only affects normal operation of the project but also poses a direct threat to structural safety. Taking a certain water pipeline material as the research subject, specimens with different theoretical corrosion rates were prepared by the electrochemically accelerated corrosion method. The electrochemical behaviors of specimens in freshwater and 3% NaCl solution were comparatively analyzed by open circuit potential (OCP), potentiodynamic polarization (PDP) curve, electrochemical impedance spectroscopy (EIS), and scanning electron microscopy (SEM). The research results indicate that the relationship between corrosion current density and corrosion degree in freshwater environment is nonlinear. In the NaCl environment, the strong erosive effect of Cl- ions weakens the protective capacity of the corrosion products, leading to an approximately linear variation between both the total impedance and the corrosion current density changing with the corrosion degree. At the same corrosion rate, the corrosion rate of the steel sheet specimen in the saline solution is significantly higher than that in freshwater. SEM observation further confirms that the microstructural changes of corrosion products correspond well with electrochemical characteristics such as corrosion current density and total impedance value.

Railway tunnel lining construction is a crucial part of tunnel construction, and its carbon emissions cannot be ignored. In order to solve the problems of insufficient accuracy and poor generalization ability of carbon emission prediction results caused by unclear key carbon emission sources and influencing factors of railway tunnel lining construction, a screening method and its prediction model of carbon emission influencing factors of railway tunnel lining construction based on feature engineering and XGBoost were proposed. Firstly, the calculation boundary of the railway tunnel lining construction stage was defined, and the carbon emission calculation model of modular lining construction based on process unit was constructed. Secondly, the out-of-bag (OOB) estimation and mutual information algorithms in the random forest were used to remove redundancy from the initial feature set, and the OOB error rate was used as the evaluation index to screen out the optimal influencing factor set. Finally, the extreme gradient boosting (XGBoost) algorithm was used to predict carbon emissions, and the partial dependence plot (PDP) was introduced to reveal the marginal influence effect between feature variables and carbon emissions. Taking a railway tunnel in southwest China as a case study for verification, the results show that in the case tunnel, the carbon emissions from shotcrete, steel frames, connecting steel bars and bolt support account for the highest proportion, totaling over 70%, and in the consumption of energy materials, concrete and steel contribute the most to carbon emissions, accounting for over 80% in total. The feature engineering-XGBoost model was verified. The numerical values of various evaluation indicators showed that the proposed model had good results, the optimal subset C={surrounding rock grade, construction method, buried depth, steel frame type, reserved deformation} was finally determined, which visually analyzed the influence mechanism of different features.

Bridge inspection reports are typically stored in electronic documents, and the utilization rate of information such as defect descriptions, damage causes, and technical indicators contained therein is not high. Existing methods often rely on general-purpose pre-trained language models like BERT. Due to the lack of professionalism in the bridge field in the training corpus, it is easy to cause incomplete or incorrect recognition of professional terminology. To address this issue, an information extraction method based on the NuNER model was proposed. In the proposed method, large language models were used for automatic data annotation and multi-level semantic features were integrated with concept encoders, thereby enhancing the ability to model professional entities and long-range dependencies. A bridge inspection corpus was constructed, containing 9 types of information, 1 624 samples, and a total of 11 450 key information. Domain fine-tuning for NuNER was conducted on the basis of this corpus. Research results show that the proposed method significantly outperforms baseline models in domain entity recognition, with the F1 score increased to 0.920 6. The proposed model exhibits excellent accuracy and recall rates in extracting key information such as the quantity and distribution of diseases, verifying its effectiveness in professional information extraction for bridge inspection. The proposed method can effectively improve the efficiency of extracting bridge management and maintenance information and lays a solid foundation for subsequent construction of knowledge graph, decision support systems and intelligent question-answering platforms, showing broad application prospects.

Aiming at the inherent limitation that conventional traffic signal control systems are difficult in adapting to dynamic traffic flow, as well as the problems such as insufficient model generalization capability of existing reinforcement learning methods and high deployment complexity of large language models (LLM), a lightweight large language model-based traffic signal control system (L3M-TCS) was proposed. Firstly, an instruction-based fine-tuning dataset specifically designed for traffic signal control was constructed. Secondly, through fine-tuning and parameter quantization techniques, the LLM was compressed into a lightweight architecture suitable for roadside devices. Finally, a verification system architecture with real-time feedback mechanisms was designed to validate its effectiveness and reliability in real-world traffic environments. Research results demonstrate that: compared to the traditional fixed-timing scheme, L3M-TCS reduces traffic delays by 60.6% and queue lengths by 50.2%. Compared to reinforcement learning methods, L3M-TCS reduces the delay at untrained intersections by 64.7%, while providing natural language-based explanations for decision-making. The proposed model achieves an inference speed of 11.41 tokens/s on roadside device side, with memory footprint compressed to 81.7% of the original model, while maintaining control decision latency within 2 500 ms.

Accurate traffic state prediction is crucial for alleviating traffic congestion and advancing intelligent traffic management. Existing methods primarily rely on sectional traffic flow parameters obtained from ground sensors, neglecting differences in states among multiple lanes and the dynamic changes in the overall traffic state within the area, which results in limited prediction accuracy in complex traffic environments such as weaving areas. To address this, TraP-VisNet, a short-term traffic state prediction model based on traffic flow parameters and image latent features, was proposed from a UAV perspective. Firstly, by analyzing vehicle trajectories in the video, the spatial average speed and spatial occupancy rate were extracted to characterize the traffic state within the area. Simultaneously, an improved variational autoencoder (VAE) with temporal modeling and semantic guidance mechanisms was used to encode image frame sequences and extract latent features that could reflect traffic flow evolution. Then, to achieve multimodal feature fusion, attention mechanisms were introduced in both temporal and feature dimensions, and the data scale was unified through temporal alignment and Z-score normalization. Finally, the fused feature vectors were input into a multi-layer perceptron (MLP) for regression prediction of traffic state, and the state level classification was achieved based on random forest (RF). Experiment results demonstrate that the proposed model exhibits superior prediction capability and stability across various scenarios. Compared to models based solely on traffic flow parameters, models relying solely on image latent features and mainstream baseline methods such as Transformer, the proposed model achieves an average improvement of 19.96%, 15.15%, and 17.70% in accuracy, F1 score, and Matthews correlation coefficient (MCC), respectively, fully demonstrating its robustness and practical application potential in complex traffic environments.

In order to meet the development needs of highly automated driving on freeways, taking off-ramp lane changes on freeways as a research scenario, an off-ramp lane-changing timing decision making model for intelligent connected vehicle platoons (ICVPs) on freeways based on the number of fitted gaps was proposed, to address the problem when vehicle platoons changed lanes before entering exit ramps. Firstly, probability distributions of the number and length of gaps were established, and based on this, the number of fitted gaps was estimated. The number of fitted gaps contained in search space was accumulated by continuously expanding it, the shortest search distance was estimated according to the level of the number of fitted gaps. Secondly, the probability distribution of lane-changing distance was established, and reasonable lane-changing distance was defined as lane-changing distance corresponding to a suitable percentile. Then, based on the shortest search distance and reasonable lane-changing distance, an optimization model for the shortest reserved distance was established to realize lane-changing timing decision-making for vehicle platoons. Finally, by conducting comparative experiments, the operability and feasibility of the proposed model were verified. Simulation analysis results show that compared to the method in which the platoon starts searching for gaps and changing lanes as soon as it enters roads, the lane-changing success rate of both above models is on par when the number of vehicles in the vehicle platoon is 2, and when the number of vehicles in the platoon is 4, the lane-changing success rate of both models increases by 82%. Compared to the method in which the platoon starts searching for gaps and changing lanes when vehicle platoons approach the allowed lane-changing termination position, when the number of vehicles in the platoon is 2, the distance from the lane-changing position of the two models to the allowable lane-changing termination position is reduced by 61% and 43% respectively; when the number of vehicles in the platoon is 4, the distance from the lane-changing position of the two models to the allowable lane-changing termination position is reduced by 19% and increased by 37% respectively. Therefore, compared to lane change without lane-changing timing decision making for vehicle platoons, lane change with lane-changing timing decision making for vehicle platoons can effectively balance lane-changing success rate and distance from lane-changing position to the allowed lane-changing termination position.

The Chengdu-Chongqing Economic Circle, as a core area of the national strategic hinterland, holds significant importance for regional economic development and national security, where accurate prediction of highway freight volume is crucial. Based on panel data from 16 prefecture-level cities in the Chengdu-Chongqing region from 2010 to 2023, a prediction model for highway freight volume was constructed and key influencing factors were identified. Initially, 16 explanatory variables were preliminarily selected from dimensions such as socio-economic activities. Through Spearman rank correlation analysis, significance testing, and city fixed-effects models, seven core variables were screened out, including the number of employed personnel, total number of tourists and regional heterogeneity coefficients. Then, a systematic comparison of the performance of eight machine learning models, such as extreme random trees and gradient boosting decision trees, was conducted, using hyperparameter optimization methods to select the most suitable prediction model for highway freight in the Chengdu-Chongqing region. Finally, feature importance analysis was performed based on SHAP values. The research results indicate that the extreme random trees and gradient bosting decision trees models perform optimally, with coefficients of determination (R2) above 0.95, errors below 10% and no underfitting or overfitting issues, demonstrating the superior applicability of the extreme random trees model for predicting highway freight volume in the Chengdu-Chongqing region. The number of employed personnel, total number of tourists and regional heterogeneity coefficients were identified as key influencing factors. Among them, the number of employed personnel shows an approximately linear positive correlation with highway freight volume, the total number of tourists drives freight growth through demand-side effects, while the regional heterogeneity coefficients exhibit complex nonlinear effects due to differences in urban resources and industrial structure.

To explore the characteristics of motor vehicle-two-wheeler accidents and reveal the causes of run-over accidents and the injury causation pathways of riders, 2,281 motor vehicle-two-wheeler accidents in China In-Depth Accident Study Database (CIDAS) were taken as the data basis. Relevant factors related to drivers, riders, two-wheeler, motor vehicles, roads, and the environment were taken as background variables, AutoGluon automated machine learning method was employed for path analysis to extract the influence relationships among background variables, whether or not to be run over and injury serverity. Firstly, AutoGluon was utilized to establish models of run-over accidents and injury severity, quantifying the marginal effects of background factors on both outcomes. Then, the path analysis method was employed to identify the key pathways that background variables influenced injury severity by being run over whether or not. The results show that the probability of death for riders who are run over increases by 16.75%, the probability of serious injury increases by 4.38%, and the distribution of their injuries is significantly different from that in non-run-over accidents. Driver age, cyclist age, loaded vehicle, collision type, and length of motor vehicle are all the main reasons for the increase in the probability of run-over accidents. There are three types of paths to injury causation: indirect effect-dominated type (work zones indirectly enhancing injury through increasing the risk of run-over), direct effect-dominated type (1.25% increase in direct effect for child riders), and composite effect type (both direct and indirect effects for elderly riders). In order to more precisely depict the relationship between background factors, whether or not to be run over and injury severity, the formation mechanism of run-over accidents was further revealed, confirming that run-over accidents need to be analyzed independently, which provides a theoretical basis for the realization of differentiated traffic safety management.

Aiming at the problems such as the low recycling rate in the field of power battery recycling and the risk of environmental pollution caused by informal recycling, a tripartite evolutionary game model was established, including new energy vehicle manufacturers, consumers and the government. Based on the bounded rationality assumption of decision-makers, the replication dynamic equation, Jacobian matrix stability analysis and MATLAB numerical simulation technology were used to explore the strategy evolution law under the dynamic and static reward and punishment mechanism. The research finds that the strategic shift of decision-makers is comprehensively affected by multiple parameters, such as the intensity of government rewards and punishments, manufacturers’ recycling technology level, and consumers’ environmental preference coefficients. Under the static reward and punishment mechanism, system evolution has obvious fluctuations, and the evolution rate is slow. While the dynamic reward and punishment mechanism ensures a stable evolutionary process with significantly accelerated speed, through adaptive adjustments of linkage supervision rate and recycling rate. Based on these results, this study proposes suggestions such as establishing a dynamic reward and punishment algorithm, implementing a technical compliance certification system, and compressing the profit margins of informal channels, aiming to provide theoretical support and decision-making basis for the accurate formulation and continuous optimization of power battery recycling policy.

The height of center of gravity z is one of the key parameters to evaluate the performance and safety of vehicle. When measuring the vehicle’s height (z) using the tilt-table method, under the unlocked condition of the vehicle suspension, the lateral position (y) and height (z) of changed in a coupled manner due to suspension deformation in the process of tilting, making independent measurement difficult. Taking the arithmetic mean value of height (z) of the vehicle when it was titled 6°~12° to the left and right as the final measurement result of height (z) of the vehicle at 0°, resulted in a significant error. To solve this problem, a combined model of convolutional neural network (CNN) and long short-term memory (LSTM) networks incorporating a physical hybrid loss function, namely the CNN-LSTM model, was proposed. The results indicate that compared to Transformer, CNN and LSTM models, the CNN-LSTM model features faster convergence speed and higher prediction accuracy. height (z) at 0° predicted by the CNN-LSTM model is more accurate than that predicted by the arithmetic mean of the left and right tilts, which demonstrates the superiority of the CNN-LSTM model in predicting height (z) at 0° by the tilt-table method in scenarios where the suspension is not locked.

Regarding the electromechanical coupling vibration phenomenon in the electric propulsion system (EPS) of electric helicopters, an electromechanical coupling dynamic model of EPS was established, which took into account the electromagnetic torque characteristics of the permanent magnet synchronous motor, the dead zone effect and tube pressure drop characteristics of the inverter, and the time-varying mesh stiffness characteristics of the gears. By comparing the inherent characteristics and dynamic response results of EPS with and without considering the electromagnetic effects of the motor, the rule of electromagnetic effects affecting the dynamic characteristics of EPS was revealed. The research shows that the electromagnetic stiffness of the motor has a significant impact on the low-order modes of EPS, while having a relatively small impact on the high-order modes. The electromagnetic effects have a significant impact on the transient response of EPS but have a relatively small impact on the steady-state response value. The influence of electromagnetic effects on the steady-state response of EPS is mainly observed in the frequency domain characteristics, while the difference of peak values in the time domain is relatively small. When the gear pair decreases in the moment of inertia or is directly connected to the motor, the influence of electromagnetic effects on its dynamic meshing force will increase. This study can provide reference for the selection of modeling methods for EPS under different analysis purposes such as fault diagnosis, strength verification, and dynamics optimization.