To address the limitation that conventional collision warning systems in human-machine co-driving scenarios mainly rely on external environmental parameters and cannot adequately account for prolonged driver reaction time caused by fatigue or distraction, a Dual-Stream Joint Perception Network (DSCP-Net) is proposed. In the external perception branch, Dynamic Snake Convolution (DSC) and Inner-IoU loss are introduced to improve target feature extraction and bounding-box localization. In the in-cabin branch, a lightweight Behavioral Cognitive Module (BCM) is developed to quantify driver states based on facial features. At the decision-making level, driver-state and environmental-perception information are fused to adaptively adjust the Time-to-Collision (TTC) threshold. Experimental results show that the BCM achieves a driver-state recognition accuracy of 97.6%. Under a severe-fatigue emergency-braking condition at 100 km/h, the TTC threshold is adaptively increased from 3.0 s to 4.5 s, providing an additional safety buffer of 41.7 m. The results demonstrate that the proposed method improves active safety warning performance in human-machine co-driving scenarios.

To address the identification and prioritization of Safety of the Intended Functionality (SOTIF)-related risks for Automatic Emergency Braking (AEB) systems in complex road environments, a triggering-condition superposition assessment method integrating Bayesian networks and multi-index scoring is proposed. 4 triggering conditions (weather condition, road surface condition, traffic interaction behavior and camera frame rate) are selected. A probabilistic propagation model is constructed from triggering conditions to perception degradation, braking-margin reduction and hazardous events. A scenario-importance scoring system is established by incorporating severity, exposure, controllability, perception degradation and coupling complexity. 3 types of straight road environments are considered, including clear–dry, moderate rain–wet/slippery as well as heavy rain with ponding and low adhesion coefficient. Under each road surface category, 3 types of interaction behaviors(pedestrian crossing, lead vehicle emergency braking and vehicle cut-in) are designed. 3 camera frame rates (30 fps, 15 fps, and 10 fps) are set, forming 27 combined simulation scenarios. The simulation results show that deteriorating weather and reduced road adhesion significantly weaken the braking capability of the AEB system, while a lower camera frame rate increases target-confirmation delay and further compresses the system safety margin. Among the scenarios, heavy rain with ponding and low adhesion coefficient road – pedestrian crossing has the highest risk level, and the lead vehicle emergency braking and pedestrian crossing scenarios under moderate rain – wet/slippery road conditions also exhibit borderline risk. The results demonstrate that the proposed method can effectively identify high-risk superposed scenarios of AEB from the SOTIF perspective and can provide a reference for scenario-library construction, virtual-test prioritization, and safety argumentation.

To address the technical challenges in the configuration design of in-wheel drive corner modules, including strong coupling between large-angle wheel steering and suspension kinematics, high propensity for component motion interference, and degradation of Kinematics and Compliance (K&C) characteristics, the paper conducts a comparative analysis of mainstream corner module configurations in the industry to clarify their respective technical features and application scenarios. A kingpin-series integrated corner module based on kingpin steering and double-wishbone suspension is proposed accordingly. The design of a drive-brake integrated in-wheel electric wheel is completed, and the matching of a top-mounted kingpin steering power unit is realized. The suspension hardpoint layout and system integration are finished in accordance with the three-plane suspension design criterion. The results of iterative K&C simulation optimization and Digital Mock-Up (DMU) motion verification indicate that after optimization, key indicators such as bump steer, bump camber, wheel travel base and track change all meet the design specifications. No component motion interference occurs within the full steering angle range of-30° to 90°, which effectively solves the problems of component interference and K&C characteristic degradation caused by the coupling between large-angle steering and suspension.

To address the issues of noise sensitivity, ambiguous intent, and missing directional information in single-modality voice interaction for in-vehicle scenarios, the paper proposes a multimodal Point-of-Interest (POI) intelligent recognition method that fuses speech, vision, and gesture. The method establishes a four-layer system architecture comprising data collection, preprocessing, core algorithm, and interaction output. For speech intent recognition, Low-Rank Adaptation (LoRA) fine-tuned Qwen-7B is employed to generate structured intent output. For spatiotemporal localization, a fan-shaped region-based multi-strategy retrieval mechanism is designed, which integrates speech directional words with gaze direction, gesture pointing or navigation route, reducing the candidate POI scale by 74.47%. For comprehensive prediction, Qwen-VL is utilized to achieve cross-modal attention fusion for POI ranking. Experimental results on a self-built in-vehicle multimodal dataset demonstrate that the proposed method achieves an accuracy of 85.4% and an F1-score of 84.3%, improving by 8.6 and 7.7 percentage points respectively over the best baseline. Under 0 dB SNR conditions, the accuracy remains at 70.1%, confirming the robust effectiveness of multimodal fusion in vehicular noise environments.

To address the issue of insufficient accuracy in diagnosing open-circuit faults in inverters of Permanent Magnet Synchronous Motor (PMSM) drive systems, a fault identification method integrating One-Dimensional Convolutional Neural Networks (1D-CNN) and Long Short-Term Memory (LSTM) networks is proposed. First, MATLAB/Simulink simulations are employed to model various types of open-circuit faults in three-phase inverters. Three-phase stator currents were selected as the fault feature extraction targets. The 1D-CNN-LSTM model is enhanced with the Improved Whale Optimization Algorithm (IWOA) and utilized to identify open-circuit faults in three-phase inverters. The feasibility of the algorithm is validated by comparing it with different fault diagnosis algorithms. Experimental results demonstrate that the IWOA-1D-CNN-LSTM model achieves effective fault classification with an average recognition accuracy of 98.73%.

To solve the problem that vehicle power networks using fuses and relays for whole-vehicle protection easily suffer grid failure and further cause intelligent driving system faults, a semiconductor-based intelligent power distribution method is proposed. The fuses and relays in the body domain controller for automotive low-voltage electrical appliances are replaced with semiconductor high-side switch chips to significantly extend the service life of the controller. A multi-pulse overcurrent detection and protection strategy is further adopted to safeguard the system against overcurrent, short circuits and over-temperature to prevent main-circuit burnout and reduce vehicle fire risk. The results show that the proposed technology enables effective control of the vehicle power network while offering long service life, low cost, fast response and high reliability.