Search Results

Multi-Target tracking is a central aspect of modeling the surrounding environment of autonomous vehicles. Automotive millimeter-wave radar is a necessary component in the autonomous driving system. One of the biggest advantages of radar is it measures the velocity directly. Another big advantage is that the radar is less influenced by environmental conditions. It can work day and night, in rainy or snowy conditions. In the expressway scenario, the forward-looking radar can generate multiple objects, to properly track the leading vehicle or neighbor-lane vehicle, a multi-target tracking algorithm is required. How to associate the track and the measurement or data association is an important question in a multi-target tracking system. This paper applies the nearest-neighbor method to solve the data association problem and uses an extended Kalman filter to update the state of the track.

Life prediction is a major focus for a commercial fuel cell stack, especially applied in fuel cell electric vehicles (FCEV). This paper proposes a data driven fuel cell lifetime prediction model using particle swarm optimized back-propagation neural network (PSO-BPNN). For the prediction model PSO-BP, PSO algorithm is used to determine the optimal hyper parameters of BP neural network. In this paper, total voltage of fuel cell stack is employed to represent the health index of fuel cell. Then the proposed prediction model is validated by the aging data from PEMFC stack in FCEV at the actual road condition. The experimental results indicate that PSO-BP model can predict the voltage degradation of PEMFC stack at actual road condition precisely and has a higher prediction accuracy than BP model.

Driving style and skill classification are of great significance in human-oriented advanced driver-assistance system (ADAS) development. In this paper, we propose Fully-Connected Deep Neural Networks (FC-DNN) to classify drivers’ styles and skills with naturalistic driving data. Followed by the data collection and pre-processing, FC-DNN with a series of deep learning optimization algorithms are applied. In the experimental part, the proposed model is validated and compared with other commonly used supervised learning methods including the k-nearest neighbors (KNN), support vector machine (SVM), decision tree (DT), random forest (RF), and multilayer perceptron (MLP). The results show that the proposed model has a higher Macro F1 score than other methods. In addition, we discussed the effect of different time window sizes on experimental results. The results show that the driving information of 1s can improve the final evaluation score of the model.

The accuracy of tire forces directly affects the vehicle dynamics model precision and determines the ability of the model to develop the simulation platform or design the control strategy. In the high slip angle, due to the complex interactions at tire-road interfaces, the forces generated by the tires are high nonlinearity and uncertainty, which pose issues in calculating tire force accurately. This paper presents a hybrid physical and data-driven tire force calculation framework, which can satisfy the high nonlinearity and uncertainty condition, improve the model accuracy and effectively leverage prior knowledge of physical laws. The parameter identification for the physical tire model and the data-based compensation for the unknown errors between the physical tire model and actual tire force data are contained in this framework. First, the parameters in the selected combined-slip Burckhardt tire model are identified by the nonlinear least square method with tire test data.

An autonomous vehicle is a comprehensive intelligent system that includes environment sensing, vehicle localization, path planning and decision-making control, of which environment sensing technology is a prerequisite for realizing autonomous driving. In the early days, vehicles sensed the surrounding environment through sensors such as cameras, radar, and lidar. With the development of 5G technology and the Vehicle-to-everything (V2X), other information from the roadside can also be received by vehicles. Such as traffic jam ahead, construction road occupation, school area, current traffic density, crowd density, etc. Such information can help the autonomous driving system understand the current driving environment more clearly. Vehicles are no longer limited to areas that can be sensed by sensors. Vehicles with different autonomous driving levels have different adaptability to the environment.

The RON(Research Octane Number) is the most important indicator of motor petrol, and the petrol refining process is one of the important links in petrol production. However, RON is often lost during petrol refining and RON Loss means the value of RON lost during petrol refining. The prediction of the RON loss of petrol during the refining process is helpful to the improvement of petrol refining process and the processing of petrol. The traditional RON prediction method relied on physical and chemical properties, and did not fully consider the high nonlinearity and strong coupling relationship of the petrol refining process. There is a lack of data-driven RON loss models. This paper studies the construction of the RON loss model in the petrol refining process.

Velocity prediction on hilly road can be applied to the energy-saving predictive control of intelligent vehicles. However, the existing methods do not deeply analyze the difference and diversity of road slope driving characteristics, which affects prediction performance of some prediction method. To further improve the prediction performance on road slope, and different road slope driving features are fully exploited and integrated with the common prediction method. A rolling prediction-based multi-scale fusion prediction considering road slope transition driving characteristics is proposed in this study. Amounts of driving data in hilly sections were collected by the advanced technology and equipment. The Markov chain model was used to construct the velocity and acceleration joint state transition characteristics under each road slope transition pair, which expresses the obvious driving difference characteristics when the road slope changes.

Speech enhancement can extract clean speech from noise interference, enhancing its perceptual quality and intelligibility. This technology has significant applications in in-car intelligent voice interaction. However, the complex noise environment inside the vehicle, especially the human voice interference is very prominent, which brings great challenges to the vehicle speech interaction system. In this paper, we propose a speech enhancement method based on target speech features, which can better extract clean speech and improve the perceptual quality and intelligibility of enhanced speech in the environment of human noise interference. To this end, we propose a design method for the middle layer of the U-Net architecture based on Long Short-Term Memory (LSTM), which can automatically extract the target speech features that are highly distinguishable from the noise signal and human voice interference features in noisy speech, and realize the targeted extraction of clean speech.

Image corruptions due to noise, blur, contrast change, etc., could lead to a significant performance decline of Deep Neural Networks (DNN), which poses a potential threat to DNN-based autonomous vehicles. Previous works attempted to explain corruption from a Fourier perspective. By comparing the absolute Fourier spectrum difference between corrupted images and clean images in the RGB color space, they regard the noise from some corruptions (Gaussian noise, defocus blur, etc.) as concentrating on the high-frequency components while others (contrast, fog, etc.) concentrate on the low-frequency components. In this work, we present a new perspective that unifies corruptions as noise from high frequency and thus propose an image augmentation algorithm to achieve a more robust performance against common corruptions. First, we notice the 1/fα statistical rule of the natural image's spectrum and the channels-wise differential sensitivity on the YCbCr color space of the Human Visual System.

As a prerequisite for energy management of hybrid vehicles, the results of speed prediction can optimize the performance of vehicles and improve fuel efficiency. Energy management strategies are usually developed based on standard driving cycles, which are too generalized to show the variability of driving conditions in different time and locations. Therefore, this paper constructs a representative driving cycle based on driving data of the corresponding time and location, used as historical information for prediction. We propose a method to construct the driving cycle based on Markov chain theory before constructing the prediction model. In this paper, multiple prediction methods are compared with traditional parametric methods. The difference in prediction accuracy between multiple prediction methods under the single time scale and multiple time scale were compared, which further verified the advantages of the speed prediction method based on Markov chain theory.

Controller Area Network (CAN), the de facto standard for in-vehicle networks, has insufficient security features and thus is inherently vulnerable to various attacks. To protect CAN bus from attacks, intrusion detection systems (IDSs) based on advanced deep learning methods, such as Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN), have been proposed to detect intrusions. However, those models generally introduce high latency, require considerable memory space, and often result in high energy consumption. To accelerate intrusion detection and also reduce memory requests, we exploit the use of Binarized Neural Network (BNN) and hardware-based acceleration for intrusion detection in in-vehicle networks. As BNN uses binary values for activations and weights rather than full precision values, it usually results in faster computation, smaller memory cost, and lower energy consumption than full precision models.

On the study of reducing the disturbance on driver’s attention induced by low frequency vehicle interior stationary noise, a subjective evaluation is firstly carried out by means of rank rating method which introduces Distraction Level (DL) as evaluation index. A visual-finger response test is developed to help evaluating members better recognize the Distraction Level during the evaluation. A non-linear back propagation artificial neural network (BPANN) is then modeled for the prediction of subjective Distraction Level, in which linear sound pressure RMS amplitudes of five Critical Band Rates (CBRs) from 20 to 500Hz are selected as inputs of the model. These inputs comprise an input vector of BPANN. Furthermore, active noise equalization (ANE) on DL is realized based on Filtered-x Least Mean Square (FxLMS) algorithm that controls the gain coefficients of inputs of trained BPANN.

As turbulence modeling has become an indispensable approach to perform flow simulation in a wide range of industrial applications, how to enhance the prediction accuracy has gained increasing attention during the past years. Of all the turbulence models, RANS is the most common choice for many OEMs due to its short turn-around time and strong robustness. However, the default setting of RANS is usually benchmarked through classical and well-studied engineering examples, not always suitable for resolving complex flows in specific circumstances. Many previous researches have suggested a small tuning in turbulence model coefficients could achieve higher accuracy on a variety of flow scenarios. Instead of adjusting parameters by trial and error from experience, this paper introduced a new data-driven method of turbulence model recalibration using adjoint solver, based on Generalized k-ω (GEKO) model, one variant of RANS.

Vehicle lane keeping system is becoming a new research focus of drive assistant system except adaptive cruise control system. As we all known, vehicle lateral dynamics show strong nonlinear and time-varying with the variety of longitudinal velocity, especially tire’s mechanics characteristic will change from linear characteristic under low speed to strong nonlinear under high speed. For this reason, the traditional PID controller and even self-tuning PID controller, which need to know a precise vehicle lateral dynamics model to adjust the control parameter, are too difficult to get enough accuracy and the ideal control quality. Based on neural network’s ability of self-learning, adaptive and approximate to any nonlinear function, an adaptive PID control algorithm with BP neural network self-tuning online was proposed for vehicle lane keeping.

Intrusion Detection Systems (IDS), technically speaking, is to monitor the network, system, and operation status according to certain security policies, and try to find various attack attempts, attacks or attack results to ensure the confidentiality, integrity and availability of network system resources. Automotive intrusion detection systems can identify and alert by analyzing in-vehicle traffic and log when software applications or devices with malicious activity exist, or the in-vehicle network is tampered and injected. But unfortunately, automotive cybersecurity researchers hardly produce a comprehensive detection method due to the confidential nature of Controller Area Network (CAN) DBC format files, which is a standard long maintained by car manufacturers. In this paper, an enhanced intrusion detection method is proposed based on the double-decision-tree to classify different attack models for in-vehicle CAN network without the need to obtain complete DBC files.

Intelligent driving, aimed for collision avoidance and self-navigation, is mainly based on environmental sensing via radar, lidar and/or camera. While each of the sensors has its own unique pros and cons, camera is especially good at object detection, recognition and tracking. However, unpredictable environmental illumination can potentially cause misdetection or false detection. To investigate the influence of illumination conditions on detection algorithms, we reproduced various illumination intensities in a photo-realistic virtual world, which leverages recent progress in computer graphics, and verified vehicle detection effect there. In the virtual world, the environmental illumination is controlled precisely from low to high to simulate different illumination conditions in the driving scenarios (with relative luminous intensity from 0.01 to 400). Sedan cars with different colors are modelled in the virtual world and used for detection task.

Vision based driving environment perception is current research hotspot in automatic driving field, which has made great progress due to the continuous breakthroughs in the research of deep neural network. As is well known, deep neural network has won tremendous successes in a wide variety of image recognition tasks, such as pedestrian detection and vehicle identification, which have accomplished the commercialization successfully in intelligent monitor system. Nevertheless, driving environment perception has a higher request for the generalization performance of deep neural network, which needs further studies on its design and training methods. In this paper, we presented a new boosted deep neural network in order to improve its generalization performance and meanwhile keep computational budget constant. Above all, the most representative methods to improve the generalization performance of deep neural network were introduced.

Active engine mount (AEM) is an effective approach which can optimize the noise, vibration and harshness (NVH) performance of vehicles. The filtered-x-least-mean-squares (FxLMS) algorithm is widely applicated for vibration attenuation in AEMs. However, the performance of FxLMS algorithm can be deteriorated without an accurate secondary path estimation. First, this paper models the secondary path using finite impulse response (FIR) model, infinite impulse response (IIR) model and back propagation (BP) neural network model and the model errors of which are compared to determine the most accurate and robust modeling method. After that, the influence of operation frequency on accuracy of the secondary path model is analyzed through simulation approach. Then, the impact of reference signal mismatch on the control effect is demonstrated to study the robustness of FxLMS algorithm.

Thin-walled structures have been widely used in automobile body design because of its good lightweight and superior mechanical properties. For the energy-absorbing box of the automobile, it is necessary to consider its working conditions under the axial and oblique impact. In this paper, a novel hierarchical honeycomb is proposed and used as filler for thin-walled structures. Meanwhile, the crashworthiness performances of the conventional honeycomb-filled and the hierarchical honeycomb-filled thin-walled structures under different impact conditions are systematically studied. The results indicate the energy absorption of the hierarchical honeycomb-filled thin-walled structure is higher than that of the conventional honeycomb-filled thin-walled structure, and the impact angle has significant effects on the energy absorption performance of the hierarchical honeycomb-filled structure.

To improve the durability of Proton-exchange membrane fuel cell (PEMFC) in actual transportation application scenario, the research on fault diagnosis of PEMFC is receiving extensive attention. With the development of artificial intelligence, performing fault diagnosis with the massive sampling data of the fuel cell system has become a popular research topic. But few people have successfully verified the diagnosis performance of these artificial intelligence algorithms on a real high power on-board PEMFC system. Therefore, we intend to make a step forward with these data-driven artificial intelligence algorithms. We applied four data-driven artificial intelligence algorithms to diagnose three common faults of PEMFC (each fault type has two severity levels, slight and severe). AVL CRUISE M was firstly applied for generation of simulation fault dataset to speed up the algorithm screening process. Based on the dataset, these algorithms are trained and optimized.