To improve the optimization accuracy of the proxy model, this paper employs a linear weighting method to integrate multiple individual models into a combination model. Simulation verification was conducted using engine hood lightweighting as a case study, and the initial thickness of the carbon fiber outer panel was determined based on the principle of equal stiffness; Latin hypercubic sampling was used to obtain samples, and a PSO-BP combination model was constructed via linear weighting. Comparative analysis showed that the combination model achieved superior prediction accuracy with fewer samples; under the same number of training sets, the error was reduced by 85.32
Steer-by-Wire (SbW) is an important chassis-by-wire technology in modern vehicle platforms. It removes the mechanical linkage between the steering wheel and the road wheels, which increases system flexibility and facilitates integration with other vehicle subsystems. However, this architecture also places higher demands on control precision, response speed, and fault tolerance. This paper reviews the recent progress in SbW control strategies and summarizes the shift in control objectives from single-point tracking to coordination across multiple tasks. It further discusses how these strategies relate to centralized electrical/electronic (E/E) architectures. The paper compares several control methods, including observer-based control (OBC), sliding mode control (SMC), model-based control (MBC), and data-driven control (DDC), highlighting their strengths and weaknesses. Fault-tolerant control (FTC) research is reviewed from four aspects: sensing, actuation, communication, and human–machine interaction. As intelligent and connected vehicles continue to evolve, communication security and driver behavior have become increasingly important factors affecting steering safety and system performance. Finally, this paper outlines future research directions, including adaptive control algorithms, tighter integration between hardware and software, and system-level safety verification.
In the field of PMSMs for automobiles, the application of high-power density motors is becoming increasingly popular. As a result, the problem of conductor overheating in stator slots has become increasingly prominent. To alleviate this issue, this study proposes an internal-cooling-channel (ICC) structure. The ICC structure uses a single-sided hollow hairpin coil and is equipped with an oil injection hose to supply oil to each coil. This design can directly cool hairpin coils. The heat dissipation effect of this structure was evaluated using electromagnetic and thermal fluid simulations. The results indicate that although the ICC structure slightly increases electromagnetic losses, it significantly suppresses the rise in temperature of the PMSM. Compared to the original PMSM, the maximum temperature is reduced by 9° C and 23° C under rated condition (6000 rpm, 120 kW) and overload condition (12000 rpm, 220 kW), respectively. This effectively improves the continuous operation capability and power density of the motor, providing a solution for temperature control of automotive PMSMs.
Reducing fuel consumption to lower CO₂ emissions remains a crucial goal for the automotive industry. One effective approach to achieving this objective is reducing friction-related energy losses within the engine through the use of low-viscosity engine oils and friction-modifying additives. This study investigated the relationship between engine oil type and fuel consumption in a fleet of 12 identical passenger cars equipped with the same engine and operated under similar real-world conditions. The research was conducted in three stages over a period of three years. During the first two stages, three different engine oils were evaluated under comparable operating conditions and varying seasonal temperatures. In the third stage, two additional oils were introduced and the effects of friction-modifying additives were assessed. Altogether, 12 oil–additive combinations were analyzed. Statistical analysis revealed significant differences in fuel consumption associated with the use of additives. The results showed that friction modifiers altered fuel consumption patterns, with the direction and magnitude of the effect depending on the engine oil formulation and operating temperature. The findings indicate that appropriately selected oil–additive combinations can contribute to improved engine efficiency and reduced fuel consumption under specific operating conditions.
Now research on manual takeover of L3 level autonomous driving mainly focuses on impacts of human-machine interaction, and has not fully considered potential risks of driver road rage. To accurately identify risk status of driver’s road anger during mode switching, a driving emotion change recognition model based on metacognitive emotional experience is established. Furthermore, a road anger risk recognition model is constructed by integrating vehicle status and environmental information in mixed driving environment. Firstly, based on metacognitive emotional experience, facial expression features of drivers are extracted. Facial expression and EMG signals based on Support Vector Machine (SVM-FE EMG) are fused for multimodal emotion recognition. Then, input feature vector and other parameters of emotional changes, and use classification function to determine whether the driver is in road rage mood. Finally, a road rage risk identification model integrating vehicle status and environmental information is constructed that applies the improved Random Forest based on Spark(SPA-RF). Results show that SVM-FE EMG model improves accuracy by about 6
A multilayer interior permanent magnet synchronous motor (IPMSM) offers high power density and enhanced reluctance torque through the use of multiple permanent magnet (PM) layers. However, during high-speed operation, such structures generate considerable induced-voltage harmonics due to magnetic saturation, spatial flux variation, and PM segmentation, which can exceed the inverter DC-link voltage and destabilize motor control. This paper proposes a systematic harmonic analysis and reduction method based on flux linkage decomposition. Using 2-D finite element analysis (FEA) with a frozen-permeability approach, the total induced voltage is decomposed into three components: the time derivative of current, the spatial derivative of inductance, and the rotational variation of PM flux. The component-wise harmonic analysis identifies the 11th and 13th orders as dominant. To suppress these harmonics, three targeted design strategies are applied: increasing the stator slot number, introducing three-step rotor skewing, and adopting an eccentric rotor shape. The improved model is evaluated by FEA and validated experimentally. The results show that total harmonic distortion (THD) is reduced from 71.1
Autonomous vehicles (AVs) have the potential to significantly improve traffic efficiency because they can safely drive closer together than human-driven vehicles (HVs). Dedicated Lanes (DLs) were proposed as a potential scenario for AV deployment. However, when the AV traffic is relatively low, the road space of the AV-dedicated lanes may not be fully utilized. One potential solution is to allow some HVs to share dedicated AV lanes by paying a toll, in addition to AVs using them for free. However, there is a lack of research on the application of charging mechanisms to various scenarios. To fill this gap, this study proposes a strategic plan to allow sharing the dedicated lane for autonomous driving; the plan provides specific utilization policies for answering the questions of where, how, and how much to charge. Then appropriate driving behavior models were proposed to simulate the driving behavior of vehicles. Afterward a driving simulator experiment was conducted to investigate the optimal deployment of shared dedicated-AV lanes in transportation networks with mixed AV and HV flows. The results show that the shared dedicated lane strategy improved traffic performance, and continuous entry was found to be better than the limited entry condition in most cases.
This study investigates head injury risks among motorcyclists involved in car–motorcycle collisions under Vietnamese traffic conditions using numerical simulations. Injury severity was evaluated using peak linear acceleration (PLA), Head Injury Criterion (HIC), Brain Injury Criterion (BrIC), and maximum principal strain (MPS). Collision scenarios reflected common crash characteristics in Vietnam by varying car speed (30–80 km/h), impact direction, vehicle type (sedan and SUV), helmet use, and wearing conditions. Results show that within typical accident speed ranges, HIC, BrIC, and MPS frequently exceed critical thresholds, especially in oblique and side impacts where rotational loading dominates. Although PLA values generally remain below the 300 g limit specified in TCVN 5756:2017, other metrics indicate a high risk of severe traumatic brain injury, suggesting that evaluation based solely on linear acceleration is inadequate. Helmet use reduces translational acceleration but has limited effectiveness in mitigating rotational motion. Kinematic analyses further reveal diverse real-world impact locations on both vehicles and helmets that are not represented in current standard tests. These findings support revising TCVN 5756:2017 to incorporate rotational injury criteria and more realistic impact configurations.
This investigation presents a novel methodology for estimating and controlling ethanol injection times in automotive diesel engines operating under unknown functional states. The approach leverages an experimental mapping technique with ethanol injection into the intake manifold and control strategies based on a historical database from an engine test bench. Real-time estimation of ethanol injection times is achieved through the LAMDA (Learning Algorithm Multivariable and Data Analysis) fuzzy classification algorithm, which identifies engine functional states using engine variable data. The ethanol injection time for each functional state was determined through Criterion for Alcohol Fumigation Times (CRAFT), a weighted fuzzy decision criterion utilizing Takagi–Sugeno fuzzy models, linear equations, and LAMDA_FAR (LAMDA-Functional states After Recognition) outputs. Depending on the ethanol substitution percentage (10
Hardware-in-the-Loop (HiL) simulation has become a key enabling technology for the development and validation of modern automotive systems. Nevertheless, selecting an appropriate HiL architecture remains a challenging task because the interaction between real components and simulation models strongly depends on the adopted configuration. This study proposes a structured, data-driven methodology based on the direct processing of CAN database (.dbc) files. By analysing in-vehicle network (IVN) data exchanges and restbus requirements, the method identifies key characteristics of both the device under test (DUT) and the overall system, enabling the assessment of simulation capabilities, limitations, and integration effort for different configurations. Application to representative industrial case studies shows that only approximately 20
This study proposes an active noise control (ANC) method based on reinforcement learning (RL) to reduce narrowband noise whose frequency varies continuously over time. The conventional filtered-x least mean squares algorithm depends heavily on a secondary path model, leading to performance degradation and potential divergence when the error between the physical path and the path model becomes excessively large. As an alternative to overcome this inherent error, a Twin Delayed Deep Deterministic Policy Gradient (TD3)-based RL approach with curriculum learning is considered, in which the agent directly outputs the control filter coefficients without relying on a secondary path model. The proposed algorithm was implemented in an acoustic duct system, and its control performance was evaluated under a 200–500 Hz linear frequency sweep at 30 Hz/s. Among the proposed curricula, the coarse-to-fine strategy achieved the shortest time-to-threshold and yielded a 14.34 dBA reduction in A-weighted SPL over 200–500 Hz relative to the uncontrolled case. This represents an improvement of 9.07 dBA over the Baseline without curriculum learning. These results suggest that an RL approach using TD3 and curriculum learning shows potential as an alternative for narrowband ANC under frequency sweeps.
This study evaluates the feasibility of long-distance electric bus operation by analyzing the effects of passenger load, ambient temperature, HVAC operation, and driving mode on energy consumption and driving range. A longitudinal dynamic model of an electric bus was developed in MATLAB/Simulink and simulated under various operating conditions based on the Worldwide Harmonized Vehicle Cycle. The results show that increased passenger load, low ambient temperature, and high-speed driving significantly increase energy consumption and reduce driving range. At − 10 °C, energy consumption increased by 61.29
In the numerical simulation of oil film cavitation in sliding bearings, the dynamic variation of lubricant physical properties significantly affects computational accuracy. This study takes CI-4 grade lubricant as the research object. A bespoke test rig for physical property measurement was designed and constructed, and three key physical property parameters (density, saturated vapor pressure, and viscosity) were obtained over wide ranges of temperature and pressure under actual operating conditions of sliding bearings. Based on the measured data, temperature‑pressure coupled models were established, including the Tait density model, Antoine saturated vapor pressure model, and Andrade‑Barus viscosity model (R2 ≥ 0.966), enabling real‑time feedback of physical properties with local temperature and pressure. The coupled models were embedded into CFD cavitation simulations of the sliding bearing oil film, revealing that, under the variable property condition, the coupled variations of physical properties suppress cavitation, with viscosity variation being the dominant factor affecting cavitation intensity. The results show that the proposed temperature‑pressure coupled physical property models can more realistically reflect cavitation flow characteristics under actual operating conditions, significantly improve cavitation prediction accuracy, and provide a reliable basis for physical property input in numerical simulations of oil film cavitation in sliding bearings.
Behavioral heterogeneity in mixed traffic challenges the safety and comfort of autonomous vehicle (AV) lane-changing maneuvers. To address this, this paper proposes a trajectory planning framework that integrates dynamic risk awareness. First, a hybrid GMM-LightGBM approach identifies the driving styles of surrounding human-driven vehicles (HDVs). Subsequently, personalized trajectory predictions are generated based on the FVD car-following model. Second, by quantifying the spatiotemporal coupling relationships among traffic participants, a dynamic risk assessment system is constructed to adapt to the differentiated behavioral characteristics of HDVs. Building on this, an innovative Lane-Change-Specific Dual-Graph Spatio-Temporal VectorNet (LC-DGST-VNet) predicts the lane-change duration and displacement. By utilizing these predictions to narrow sampling scopes and dynamically adjust risk weights, the framework achieves real-time, risk-aware planning, executed via Model Predictive Control (MPC). Simulation results demonstrate that, compared with traditional methods that neglect behavioral heterogeneity and environmental risk levels, the framework significantly optimizes safety and smoothness. Specifically, the average collision risk is reduced from 9.2 to 1.8
A high-power-density in-wheel motor (IWM) generates substantial heat during operation, causing structural thermal deformation (TD) that induces air gap (AG) non-uniformity and affects electromagnetic properties (EP). A 15 kW permanent magnet (PM) IWM is investigated. Based on the analysis of multi-field coupling, a full IWM TD analysis model is established considering electromagnetic, temperature, flow, and structural field coupling, as well as actual component assembly and constraint relations. After experimental validation, the model is applied to calculate IWM TD characteristics, revealing the influence of TD on the AG. By structurally reconstructing the deformed IWM, the effects of TD on magnetic flux density distribution, AG electromagnetic force (EF) density, electromagnetic torque, and cogging torque are quantitatively evaluated. This research provides a foundation for improving IWM EP and vehicle dynamics.
This study develops a Bayesian neural network (BNN) model to predict chest deflection using data from USNCAP frontal impact tests conducted on 193 vehicle models from model years 2018 to 2022. The BNN was constructed based on a deep neural network architecture with 11 input features derived from curb vehicle weight and crash deceleration profiles, while the driver’s and passenger’s chest deflections were used as target variables. Prior distributions were assigned to the model parameters, and posterior estimates were obtained using variational inference (VI). Unlike previous studies that produced only deterministic predictions, the proposed approach quantifies predictive uncertainty by estimating the posterior distributions of chest deflection. This probabilistic framework enables more realistic and reliable crash safety assessments and was implemented using the Python-based Bayesian modeling library, PyMC. In addition, supplementary comparisons between Markov Chain Monte Carlo (MCMC) and VI were conducted to examine the probabilistic estimation characteristics of the proposed framework.
Internal combustion engines exhibit an optimal combustion phasing for maximizing torque output, referred to as the maximum brake torque timing (MBT). Conventional spark ignition (SI) engines using three-way catalysts generally operate only under stoichiometric conditions, resulting in a single MBT point. However, hydrogen-fueled SI engines may exhibit different MBT characteristics because combustion duration varies with excess air ratio, thereby affecting the work generated during the compression and expansion processes. In this study, the variation in MBT and the corresponding combustion characteristics at various excess air ratios were experimentally investigated using a hydrogen direct-injection SI engine. Considering the lean-burn capability of hydrogen, the excess air ratio was controlled by varying the fuel quantity under wide open throttle conditions. Ignition timing was adjusted within the stable operating limits determined by the CoV of gIMEP and maximum pressure rise rate. The results showed that the mass fraction burned 50 λ =1.5 ), whereas it advanced to approximately 3 ATDC °CA under lean conditions above λ =2.7 . This trend was attributed to the increased requirement for heat release during the compression stroke to ensure stable ignition under lean conditions. The fraction of heat release during compression increased from 1.6 λ =1.5 to 42.3 λ =3.0 . In addition, the stable combustion window widened with increasing excess air ratio, reaching 24°CA at λ =2.7 . These findings provide fundamental insights for optimizing operating conditions to determine MBT in hydrogen engines under various excess air ratio conditions.
Early-onset breakdown, characterized by sharp efficiency drops before capacity limits are reached, frequently occurs in expressway weaving areas due to frequent mandatory lane changes. To address the lag in congestion warning inherent in existing Time-to-Collision (TTC) indicators, this study proposes a Cloud-Edge-End pinning control strategy based on a driving risk potential field. First, we construct a risk field model integrating vehicle kinetic energy and driving behavior characteristics to quantify dynamic disturbances of lane-changing behaviors on traffic flow. Second, relying on a Cloud-Edge-End pinning control strategy, edge computing resources identified key vehicles with high risk propagation potential in real time, and the Soft Actor-Critic (SAC) deep reinforcement learning algorithm was applied to guide these vehicles. Simulation results indicate that, compared with traditional indicators, the risk field model accurately depicts the complete dynamic process of traffic congestion from local triggering and fluctuation diffusion to final dissipation, which aligns with real traffic flow evolution laws. Under a 40
Vector field guidance (VFG) generates a desired heading whose rate of change is governed by the product of path curvature and vehicle speed. This coupling introduces a speed and geometry-dependent guidance bandwidth that conventional fixed-gain controllers cannot accommodate across varying operating conditions. This paper proposes a curvature-velocity scheduled linear parameter-varying H_∞ controller that adapts its performance weighting bandwidth in real time through the scheduling parameter ρ =|κ |v . A concave square-root schedule distributes bandwidth across multiple polytopic vertices synthesized offline, and the runtime controller is obtained by convex output interpolation. Frequency-domain analysis confirms that all vertices satisfy classical sensitivity robustness bounds, while time-domain simulations on a differential-drive scale car demonstrate substantial tracking improvement over linear MPC and PID with curvature feedforward, particularly at moderate to high speeds. The proposed controller exhibits near speed-invariant heading error across the tested speed range, in contrast to the monotonic performance degradation observed in the baselines. Monte Carlo analysis under multi-dimensional parametric uncertainty provides empirical evidence that these performance gains persist across the sampled range of plant conditions, with statistically significant improvements at moderate-to-high speeds and faster transient settling across all tested speeds.
This paper presents a novel end-to-end deep learning framework for extrinsic calibration between LiDAR and camera sensors. The core contribution is the integration of a Transformer-based global attention architecture and a cross-attention mechanism, enabling effective learning of semantic and spatial correspondences between RGB images and 3D point clouds. The Transformer architecture effectively models long-range dependencies, whereas the cross-attention module reinforces intermodal interactions. This combination enables the network to capture global context and semantic alignment across sensors, which conventional CNN-based methods often fail to achieve. Consequently, the proposed method maintains robust and accurate calibration performance even under challenging conditions, such as occlusions, perspective mismatches, and sparse or noisy inputs. Evaluated on the widely used KITTI odometry dataset, the model achieves mean translation and rotation errors of 1.37 cm and 0.09°, respectively, demonstrating a marked improvement in extrinsic calibration precision. Furthermore, ablation studies confirm that both the Transformer and cross-attention modules are critical to performance gains, demonstrating their effectiveness in enhancing multimodal feature matching and calibration accuracy.
