Nondriving behavior identification is one of the important ways to improve the safety of driving. The current recognition method based on skeleton sequence and image fusion has the problems of large model calculation and the difficulty of feature fusion. To address the above problems, the skeletonimage based behavior recognition network (SIBBRNet) is proposed in this paper, which is based on the multiscale skeleton graph and the local visual context. SIBBRNet fully extracts motion and appearance features through a graph convolution network based on multiscale skeleton graphs and a convolutional neural network based on local vision and attention mechanisms, and better balances the relationship between model representation capabilities and model calculation. The feature bidirectional guided learning strategy based on hand motion, an adaptive feature fusion module and an auxiliary loss on the static feature space can guide mutual guidance and updating between motion and appearance features to achieve adaptive fusion. SIBBRNet is finally tested on the Drive & Act dataset, and the average accuracy is 61.78% for dynamic labels and 80.42% for static labels. The Floatingpoint Operations per Second (FLOPS) of SIBBRNet is 25.92G, which is 76.96% lower than that of the optimal method.

To effectively evaluate the takeover risks of L3 autonomous vehicles under different cognitive secondary tasks, a study on the risk assessment model for driving takeover is carried out. The urban expressway emergency takeover scenario is designed and driving simulation experiments under different cognitive secondary tasks are carried out. The takeover risk assessment model considering trajectory field, potential field and behavior field is established. The validity of the proposed model is verified by adopting the takeover risk index method. Combined with the measured data, the influence of different cognitive secondary tasks and avoidance operation types on the strength of takeover risk field is quantized. The results show that the MW test and KS test for the distribution of the takeover risk index between 1 and 9 s after the takeover operation by the participants are both with the result of p<0.05, indicating that the model can effectively assess the takeover risk of the vehicle during the takeover process. In addition, the root mean square error of the takeover risk index (0.062) is smaller than the root mean square error of the inverse timetocollision (0.098), indicating that the model is better than the inverse timetocollision in accurately describing the risk. The research results can provide reference for vehicle operation risk assessment and collision avoidance design in takeover process.

For the problems of single mode of car navigation interaction, inability to dynamically change interaction strategies according to contextual differences, and inadequate user experience, a context engine model that supports adaptive car navigation interaction strategies based on different contextual situations is proposed in this paper. Combining the usercentered design (UCD) and context awareness theory, automotive humancomputer interaction contexts are defined, and key contextual elements are extracted through followup surveys of typical drivers. The adaptive interaction strategies and principles of the context engine model in navigation contexts are elaborated, and users' different information needs are mapped to the detailed level of interaction performance at the presentation layer. The navigation HMI information frameworks and components that fit the different levels of driver needs are summarized and designed for interface design and analysis of two typical navigation application scenarios. The results show that the context engine can adaptively interact with driver' needs in various driving situations, providing a reference for optimizing and developing intelligent cockpit driving interaction experiences.

With acceleration of the urbanization process, pedestrianvehicle conflicts have become a significant issue that modern society urgently needs to solve. In complex traffic scenarios, pedestrian crossing behavior leads to frequent traffic accidents. Accurately and timely anticipating pedestrian crossing intentions is crucial for avoiding pedestrianvehicle conflicts, improving driving safety, and ensuring pedestrian safety. An Efficient ActionConditioned Interaction Pedestrian Crossing Intention Anticipation Framework (EAIPF) is proposed in this paper to anticipate pedestrian crossing intention. EAIPF introduces in a pedestrian action encoding module to enhance the representation ability of multimodal action patterns and discover deep skeletal context information. At the same time, the scene object interaction module is introduced to explore interaction information with objects and understand advanced semantic clues in traffic scenes. Finally, the intention anticipation module fuses pedestrian action and object interaction features to achieve robust anticipation of pedestrian crossing intentions. The proposed method is verified on two public datasets, JAAD and PIE, achieving the accuracy of 89% and 90%, respectively, indicating that the proposed method can accurately anticipate pedestrian crossing intentions in complex traffic scenarios.

The shared cars have the characteristics of standardized vehicle models, diverse users, and high frequency of usage. To enhance the efficiency of shared car usage and the driving experience, the development of an intelligent seat that can automatically adjust the comfortable sitting posture based on the driver's characteristics has significant importance. For the issue of subjective and arbitrary seat adjustments that lead to discomfort for drivers, in conjunction with the intelligent seat project for shared cars, bench experiments are carried out taking the four variables of Hpoint foreaft position, Hpoint vertical position, seat cushion angle, and backrest angle as the research factors in this paper. The study examines the impact of seat parameters on seat comfort for drivers at the 5th, 50th, and 95th percentiles. Firstly, three experimental levels are chosen for each research factor within the reference vehicle range. Orthogonal experimental design is employed to simplify the number of experiments required. The required number of orthogonal groups is determined, and orthogonal bench experiments are conducted. Experimental participants subjectively evaluate the seat comfort for each group. Additionally, pressure mapping on the seat cushion and angle measurements using a digital angle gauge are employed to collect information of participants' pressure distribution and joint angles. Secondly, clustering analysis is conducted to determine the subjective and objective comfort evaluation system employed in this study, using joint angle as the objective parameter. Through the analysis of the orthogonal experiments, it is concluded that different seat parameters have varying effect on seat comfort. Furthermore, the parameters that affect seat comfort also vary with changes in the driver's percentile. Finally, the data obtained from the orthogonal experiments are adjusted and fitted to establish the prediction models for the optimal Hpoint position and backrest angle of the seat.

In order to enhance the comfort and user acceptance of intelligent connected vehicles, a cognitive process model for passenger carsickness mitigation is established in this paper and an olfactory stimulationbased mitigation strategy is proposed. Firstly, a screening experiment of olfactory stimulation materials for carsickness relief is designed and carried out, fully considering occupant satisfaction as well as the functional requirements of carsickness relief, to select the odor type and concentration of olfactory stimulation materials. Then, a carsickness mitigation experiment based on a real car is carried out to investigate the mapping relationship between different degrees of carsickness and physiological representations and further validate the efficiency of the olfactory mitigation method. The results show that the subjective carsickness degree is positively correlated with GSR activation and negatively correlated with EEG asymmetry. Moreover, releasing 20% ginger blossom odor for 10 seconds can effectively relieve carsickness symptoms, with an average relief effect of 22.17%, which is verified in terms of subjective and objective dimensions.

In order to comprehensively review the current status and clarify the future trend of fault diagnosis in the electric drive system of pure electric vehicles, this paper first introduces the basic structure, functions and development history of the electric drive system of pure electric vehicles; then summarizes in detail the types and causes of faults of crucial components of the electric drive system of pure electric vehicles, and analyzes the main research status quo of fault diagnosis methods for key components of the electric drive system of pure electric vehicles. Then the domestic and international research progress and development of the diagnosis methods of the pure electric vehicle electric drive system are reviewed in detail from the four aspects of expert knowledgedriven, modeldriven, signaldriven and datadriven, with the advantages and disadvantages of different methods compared. Finally, the problems faced by the fault diagnosis of electric drive system of pure electric vehicles and the development direction are analyzed and foreseen, and it is further discussed and pointed out that the future research on the fault diagnosis of electric drive system of pure electric vehicles can be focused on variable condition coupled fault diagnosis, microfault and prefault diagnosis, realtime online fault diagnosis, intelligent operation and maintenance, unknown fault diagnosis and system selfhealing technology, etc.

The traffic scheduling problem in timesensitive networking (TSN) of automotive electrical and electronic architecture is investigated in this paper. To meet practical application requirements, a method for establishing the topology of invehicle TSN network is proposed. To address the multitype traffic scheduling problem in the network, a traffic scheduling strategy based on the TimeAware Shaper (TAS) mechanism is proposed, and the corresponding mathematical model is established, to reduce the total network delay while considering both the time sensitivity of highpriority traffic and the data integrity of lowpriority traffic. To solve the problems of unstable solution efficiency caused by the complex information flow forwarding process in the model and the difficulty of optimization caused by numerous traffic scheduling solutions, an improved genetic algorithm (IGA) is proposed which is optimized from the aspects of setting adaptive crossover probability formula, introducing in taboo search mutation, and combining multiple populations. The experimental results show that the proposed algorithm improves the optimality by 43.47% in endtoend latency optimization and the solution generation stability by 76.96%. The algorithm can obtain lowlatency and highreliability traffic scheduling solutions for invehicle TSN. The research findings of this paper provide insights for the study of intelligent connected vehicles and the optimization of invehicle network communication algorithms.

For the problems of dense targets, severe edge occlusion, and blurred foreground and background that intelligent vehicles face in actual traffic environments, a lightweight object detection algorithm based on image saliency feature fusion is proposed in this paper. Firstly, salient feature maps are extracted based on grayscale images, and input into convolutional neural networks with color images. Secondly, a lightweight fusion network is constructed using the Ghost Model, and the EIoU is used to optimize the model's border localization loss. In order to enhance the detection accuracy of similar occluded targets, nonmaximum suppression algorithm is improved on the backend of the network. Finally, the KITTI dataset is used for training and testing. The experiment shows that the improved detection mAP value of the network reaches 92.7%, with an average accuracy improvement of 3.8% compared to the original network YOLOv5. The accuracy and recall rates are increased by 3% and 6.2%.

In order to realize effective vibration mitigation of selfpowered semiactive suspension under uncertain factors, the suspension mechanicalelectrical coupling dynamic model is established. The influence of electrical parameters on energy conversion efficiency is explored. The adaptive fault tolerant control gain is deduced, and then the vibration isolation capability of the suspension is investigated in time and frequency domain respectively. The robust index of adaptive optimal fault tolerant control algorithm is obtained by constructing Lyapunov equation to stud the influence of key parameters on robust index. The results show that the electrical parameters have obvious influence on the energy conversion efficiency, with the suspension having higher energy conversion efficiency at the second natural frequency. The proposed adaptive optimal fault tolerant control strategy can realize effective vibration suppression in both the time and frequency domain, with better vibration isolation performance compared to passive control and selfpowered mode. The control robust index is affected by the inductance of generator and outer diameter of permanent magnet most significantly.

In order to eliminate the influence of parameter uncertainty and external uncertain disturbances on the handling stability of fourwheel steering vehicles, a nonlinear integral sliding mode control method based on kalman filter extended state observer for fourwheel active steering is proposed. Firstly, the kalman filter extended state observer (KFESO) is designed to realize the vehicle state observation and external disturbance estimation, which overcomes the disadvantage of traditional kalman filter algorithm's dependence on highprecision models. Secondly, to reduce the tracking control error of the target ideal state of the vehicle caused by disturbances, the disturbance observed by KFESO is compensated to the control input. In order to realize global robust control and suppress integral saturation, a nonlinear integral sliding mode control method based on exponential convergence is designed. Finally, hardware in the loop test results indicate that the KFESO has high observation accuracy in the presence of internal and external uncertain disturbances in the system, and the KFESOISMC method has excellent antiinterference performance in controlling the stability of fourwheel active steering compared to LQR and ISMC methods.

Conventional braking safety detection usually uses long and extreme working conditions but may result in a loss of accurate working range. To address this deficiency, firstly, a shorttime test cycle for stabilizing pedal mode test method is developed, which incorporates existing test standards , not limited to a single extreme braking mode but taking into consideration of fast steadystate operation of electric vehicles. Then, running fragments based on machine learning are regressed, and the shorttime test cycle is constructed by fusion and splicing. Also, an improved braking safety detection method is proposed with shorttime test cycle, which reduces the dimension of the characteristic parameters of the braking segments by the principal component analysis, while the hidden danger is judged by calculating the repeatability distance of braking segments based on the characteristic parameters. Finally, the effectiveness of the proposed shorttime test cycle and detection method is verified by means of following test on a test bench.

In the early design stage of the vehicle body, in order to assess the impact of the extreme pothole road conditions on the vehicle body structure, according to the related universal global pothole road test standard, for the extreme impact of the pothole#3, the body structure failure of a vehicle in the road test of pothole#3 is taken as the research object in this paper, to find out the shortcomings of the traditional sheet metal failure criteria CAE method, which can't reproduce the problem of test failure. The Fracture Forming Limit Diagram (FFLD) is established through a large number of sheet metal coupon tests as a new CAE method for sheet metal failure criteria. Then, based on this new CAE method, the Virtual road load data of pothole#3 calculated by the vehicle dynamics discipline is taken as load input to carry out finite element simulation analysis on the body structure, and the test failure is successfully reproduced. According to the analysis results of the new CAE method, the body structure is improved, and the pothole#3 road test certificate is finally passed. The test and the finite element analysis have high correlation. It is proved that the method can accurately predict the real damage condition of sheet metal under complex deformation conditions in the early stage of body development, thus reduce the risk of body structure failure in the later test.

For the problem of high temperature and uneven distribution affecting the power and safety of the electric vehicle during the battery management systems slave control board's service, a commercial BMS slave control board thermal analysis model is built and verified using the CFD theory and Icepak software. For the first time under vehicle service conditions, temperature field analysis and thermal uniformity optimization research are carried out based on the thermal analysis model. The BMS slave control board thermal simulation analysis shows that the balancing and power supply modules exceed the BMS's design temperature limit of 60 °C due to local heat accumulation, with the maximum temperature difference of the entire BMS slave control board being 21.0 °C. A heat dissipation path analysis of the BMS slave control board is further carried out, and heat dissipation optimization design is realized by altering the distance, layout of the balancing resistor, PCB substrate and adding thermal pads. By increasing the heat dissipation capacity of the BMS slave control board, the highest temperature of the BMS slave control board can be controlled below the design limit of 60 °C, and the temperature difference of the entire circuit board can be reduced to 6.9 °C, which enhances the safety and reliability of the BMS slave control board under actual vehicle service conditions, providing theoretical methods for the thermal design and optimization of the BMS slave control board.

In recent years, the thermal runaway problem of lithiumion battery has become the main bottleneck restraining the development of power battery of new energy vehicles. In this paper, a comprehensive review of the research on the thermal runaway problem of the power battery of new energy vehicles is carried out, with the inducment of the thermal runaway of lithiumion battery expounded and the thermal runaway process of lithiumion battery and the characteristics of the thermal runaway of lithiumion battery under different variable conditions introduced. Based on the characteristic parameters of thermal runaway of lithiumion battery, the early warning methods and fire suppression methods applicable to lithiumion battery fire are reviewed, and the shortcomings and development trend of the current research on thermal runaway of power battery of new energy vehicles are summarized, providing certain reference for the development of power battery of new energy vehicles.

PEMS is used to conduct the realworld emission tests on six typical ChinaVI diesel vehicles. Based on the workbased MAW (China and the EU), NTE method (U.S. EPA) and 3BMAW method (U.S. CARB), the realworld NO emission characteristics of heavy diesel vehicles are studied, and the characteristics and applicability of different analysis methods are discussed. The results show that the NO, emission results based on the workbased MAW method can meet the regulatory requirements of China and the EU, but the NO emission compliance based on the NTE method and 3BMAW method is uncertain. The low utilization of NO, emission data leads to the inability of NTE method to effectively analyze the realworld NO, emission characteristics. And the 3BMAW method is worthy of reference for NO, emission classification management. Coldstart NO, emission accounts for 47.3%80.7% of the PEMS test, and the coldstart NO, emission of heavyduty diesel vehicles should be paid attention to. However, the current realworld NO, emission analysis methods for heavyduty vehicles in China, the EU and U.S. are unable to effectively evaluate coldstart NO, emission. Therefore, the supervision of coldstart NO emission in the next stage of emission regulations should put forward the specific test cycle, analysis methods and emission limit, to effectively reduce the actual NO, emission of heavy diesel vehicles.

A DTC decoupling method for complex coupling faults of vehicles is proposed in this paper. Firstly, by analyzing the complex association of DTCs through the principle of vehicle fault selfdiagnosis and the propagation process of fault signals, the strong association relationship between DTCs is mined combined with the association rule technology and the multidimensional association rules of DTC are defined. Secondly, the FPGrowth algorithm for DTC multidimensional association rule mining is improved by the characteristics of the DTCs dataset. Finally, the DTC association knowledge graph is constructed by multidimensional association rules to realize complex DTCs decoupling by combining graph theory. The results show that this method can effectively reduce the number and complexity of DTCs, and improve the efficiency of troubleshooting faults based on DTCs.

Finite Element Analysis (FEA), as an important Computeraided Engineering (CAE) technology, plays a significant role in the area of automotive part development. However, it costs too much time when solving complicated problems, which affects the development cycle. In this paper, a neural network method is proposed, in which sample data is provided by finite element simulation and the mapping relationship between finite element input and output is established by graph network technology. The graph network method is used to predict the stress field of the seat frame assembly. The prediction method simulates the connection relationship between nodes in the finite element model using graph nodes and graph edges, which can effectively express the topological relationship between elements in the finite element model. The prediction results are compared with the results of the finite element simulation. The results show that the method can precisely predict the maximum stress and its corresponding location of the seat frame assembly, with strong predictive capabilities for stress distribution consistency. Additionally, the model has a significant computational advantage, with a calculation speed three orders of magnitude faster than that of the corresponding finite element solver.

The integrated precision casting technology of aluminum alloy is one of the important ways to achieve lightweight of automobiles. Firstly, lightweight aluminum alloy material, investment vacuum suction casting process, and topology optimization are adopted to carry out an integrated design of "material, process and structure" for the front subframe and dashboard crossbeam of the white body in this paper. Secondly, the weight, research and development cost, and development cycle of integrated aluminum alloy structural components are compared to the original steel components, which are significantly reduced. Finally, finite element simulation analysis and bench tests are conducted on the stiffness and modal of the white body with integrated precision cast aluminum parts. The results show that after the integrated design, the front subframe and dashboard crossbeam has reduced weight by 36.6% and 30.8% respectively, while the performance of the white body meets the design requirements.