A multi-unit modular direct-drive hub motor was designed for high torque density and high reliability requirements. The effects of different pole-slot ratios, stator structures, permanent magnet magnetization types, magnet fixing methods and etc. on the electromagnetic performance of the motor were investigated through theoretical analysis and finite element simulation, and an electromagnetic structure scheme suitable for achieving high torque density was explored. The effectiveness of the design scheme and related technology was verified by finite element simulation and bench test.

To improve the correlation between reliability test condition and user actual driving behavior, this paper proposed a method of user condition classification and test condition construction based on user big data. In this method, the principal component analysis method was used to reduce the dimensionality of a large amount of data, and the K-means clustering method was used to classify the data and select representative segments, sort the data through the Markov method, and finally construct a reliability test condition conforming to the actual driving situation of the user. Driving data from twenty users under a big data platform was calculated according to this calculation process. Compared with the condition commonly used at present, the method from perspective of data source can describe user actual driving state more accurately.

New power module driver chip has no simulation model, unable to simulate and verify the principle of drive unit circuit. To solve this problem, the paper, taking Infineon 1ED020I12FA2 chip as an example, studied the modeling of core devices and peripheral circuits of drive unit, and proposed a modeling and simulation method suitable for driver chip, and verified the correctness of the model through comparison with the chip data manual. And based on this model, software Saber was used to model the circuit principle for power conversion function, insulation and power amplification function and model protection function of Insulated-Gate Bipolar Transistor (IGBT), some of the important function logics were tested and verified. The results show that the corresponding function logics meet design expectation.

To improve the structural crashworthiness of the car, a finite element model of the front rail of the vehicle was established with the front rail of a Plug-in Hybrid Electric Vehicle (PHEV) as research object, China New Car Assessment Program (C-NCAP) were referred to and based on the frontal 100% overlapping rigid barrier crash test. With the cross section of car front rail, interior & exterior panel material and thickness as the research object, and the specific energy absorption, mean crush force and collapse efficiency were taken as the research objective, the front rail of the vehicle was simulated and analyzed. The sample points were selected based on the Hamersley sampling method, and the response surface of the front rail was optimized by the improved least squares method. The results showed that the optimized specific energy absorption of the front rail of the vehicle increased by 1 129.14 J/kg, and the mean crush force increased by 8.2 kN, the crushing efficiency increased by 2.59 percent point.

In order to study the injury of occupants in different sitting postures, a simulation model with Thor-AV dummy and vehicle was constructed based on a vehicle. A C-NCAP 100% overlap rigid barrier crash simulation was carried out for three groups of occupant dummies with different sitting postures (seat cushion angle 25°, backrest 42°, 52° and 62°). The results show that, the larger the backrest angle, the higher the head injury risk. Neck shear force and axial tension gradually increase with the backrest angle. There is of great risk with clavicle fracture during crash. Under the influence of the increase of backrest angle, the axial compression force of thoracic T12 poses a great threat to occupant safety, with the maximum compression force reaching 11.6 kN when the backrest angle is 62°. The restraint system could not effectively protect the occupants in the large-angle backrest seat of the vehicle.

Finite element model of a BIW was built by applying HyperMesh software and based on lightweight requirements, this paper proposed to replace advanced high-strength steel with hot stamping steel PHS1500 for BIW A-pillar reinforcement plates and replace B410LA with DP780 for the front longitudinal beam, and replace QSTE500TM with QP980 for the front guard cross member, the paper also proposed to reduce thickness. According to C-NCAP management rules, LS-DYNA was applied to perform a 100% overlap rigid wall collision analysis for the BIW model at 50 km/h. The results show that based on the results of the frontal collision analysis of traditional high-strength steel materials, after the material upgrade the intrusive amount at A-pillar reinforcement site and fireproof board leg inspection point decreases by 8.9% and 15.9%, respectively, the frontal collision performance of the BIW has been greatly improved, and the weight is reduced by 6.224 kg.

In view of the diversified development of occupants’ sitting postures of intelligent vehicles in the future, vehicle restraint systems MADYMO multi-rigid-body simulation model was set up for the 50 percentile male dummies, and its reliability was validated according to the real vehicle crash test data, and then the occupants’ sitting postures were divided into five categories through the model. And the influence of the combination of Autonomous Emergency Braking (AEB) system and seat belt on the dummy’ injury risk of four kinds of dummy sitting postures except for the standard sitting postures in frontal crashing was studied. The synthetic acceleration of the head and hip of the dummy and the force of each part of the dummy were analyzed. The results show that the occupant restraint system with standard sitting posture can improve the head acceleration of occupant with different sitting posture, and different sitting posture will cause different injuries.

Data on distracted driving conditions and passenger retention conditions in the car were collected, and combined with a neural network model, the lightweight network MobileNet V2 in the end-to-end distracted driving network was used to achieve distracted driving recognition. By connecting YOLO V5 target detector, DBFace face detector, and MobileNet V2 face classifier in series, an accurate health and safety monitoring plan was designed. Real vehicle test verification schemes were designed respectively, to verify the feasibility of two vehicle occupant health and safety tasks. The results show that the proposed scheme has high detection accuracy while ensuring real-time performance.

In order to realize wide application of traditional image processing methods in intelligent transportation systems, the paper studied the identification and counting methods of vehicles in highway surveillance video. Using Python programming language and based on the OpenCV library, the design of the identification and counting function of vehicles was completed through a series of traditional image processing methods such as grayscale, denoising, background subtraction and morphological operations. This solution was compared with the solution based on the combination of the YOLOv3 model and the Simple Online and Realtime Tracking (SORT) algorithm. The results show that traditional image processing methods can detect moving target, but there are problems of low accuracy and low versatility compared with the solution based on deep learning. The paper therefore proposed the research suggestion of combining traditional image processing and deep learning.

In order to solve the problem that the detection of children left in the car is easily disturbed by external factors, this paper proposed a detection algorithm for children left in the car based on limb size. First, the OpenPose algorithm was used to find the pixel coordinates of the human skeleton points on the infrared image, and then the camera internal parameters and picture depth information were used to calculate the limb size, and finally the body size was fitted to the volume of the occupants in the car for adult or child judgment. The experimental results show that this method has higher accuracy compared with other methods, such as infrared thermal sensor, capacitive sensor, is not easily disturbed by other environmental factors.