Search Results

Abstract Real-time reconstruction of 3D environment attributed with semantic information is significant for a variety of applications, such as obstacle detection, traffic scene comprehension and autonomous navigation. The current approaches to achieve it are mainly using stereo vision, Structure from Motion (SfM) or mobile LiDAR sensors. Each of these approaches has its own limitation, stereo vision has high computational cost, SfM needs accurate calibration between a sequences of images, and the onboard LiDAR sensor can only provide sparse points without color information. This paper describes a novel method for traffic scene semantic segmentation by combining sparse LiDAR point cloud (e.g. from Velodyne scans), with monocular color image. The key novelty of the method is the semantic coupling of stereoscopic point cloud with color lattice from camera image labelled through a Convolutional Neural Network (CNN).

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.

The proposed technique is a tailored deep neural network (DNN) training approach which uses an iterative process to support the learning of DNNs by targeting their specific misclassification and missed detections. The process begins with a DNN that is trained on freely available annotated image data, which we will refer to as the Base model, where a subset of the categories for the classifier are related to the automotive theater. A small set of video capture files taken from drives with test vehicles are selected, (based on the diversity of scenes, frequency of vehicles, incidental lighting, etc.), and the Base model is used to detect/classify images within the video files. A software application developed specifically for this work then allows for the capture of frames from the video set where the DNN has made misclassifications. The corresponding annotation files for these images are subsequently corrected to eliminate mislabels.

Abstract One of the most important factors affecting the performance of vehicle active chassis control systems is the tire-road friction coefficient. Accurate estimation of the friction coefficient can lead to better performance of these controllers. In this study, a new three-step friction estimation algorithm, based on intelligent tire concept, is proposed, which is a combination of experiment-based and vehicle dynamic based approaches. In the first step of the proposed algorithm, the normal load is estimated using a trained Artificial Neural Network (ANN). The network was trained using the experimental data collected using a portable tire testing trailer. In the second step of the algorithm, the tire forces and the wheel longitudinal velocity are estimated through a two-step Kalman filter. Then, in the last step, using the estimated tire normal load and longitudinal and lateral forces, the friction coefficient can be estimated.

There are significant predictive tool usages by design engineers in automotive industry to capture material composition and manufacturing process-induced variables. In specific, an accurate modeling of material behavior to predict the mechanical performance of a thermoplastic part is an evolving subject in this field as one needs to consider multiple factors and steps to achieve the right prediction accuracies. The variability in prediction comes from different factors such as polymer type (filled vs. unfilled, amorphous vs semi crystalline etc.), design and manufacturing features (weldline, gate locations, thickness, notches etc.), operating conditions (temperature, moisture etc.) and finally load states (tension, compression, flexural, impact etc.). Using traditional numerical simulation-based modelling to study and validate all these factors requires significant computational time and effort.

Several models for the evaluation of Gross Heat Release are often used in literature. One of these is the First Law - Single Zone Model (FL-SZM), derived from the First Law of Thermodynamics. This model presents a twice advantage: first it describes with accuracy the physic of the phenomenon (charge heat release during the combustion stroke and heat exchange between gas and cylinder wall); second it has a great simplicity in the mathematical formulation. The current paper deals with the implementation of a mathematical model, based on FL-SZM, to study the heat release due to the combustion phenomena in Internal Combustion Engines (ICEs). For purposes of chemical kinetic calculations, many of the major species have been included into the combustion products. In particular, seven gases (i.e. H2O, CO2, H2, O2, N2, CO and Ar) may also be assumed in chemical equilibrium.

This paper benchmarks three different lithium-ion (Li-ion) battery voltage modelling approaches, a physics-based approach using an Extended Single Particle Model (ESPM), an equivalent circuit model, and a recurrent neural network. The ESPM is the selected physics-based approach because it offers similar complexity and computational load to the other two benchmarked models. In the ESPM, the anode and cathode are simplified to single particles, and the partial differential equations are simplified to ordinary differential equations via model order reduction. Hence, the required state variables are reduced, and the simulation speed is improved. The second approach is a third-order equivalent circuit model (ECM), and the third approach uses a model based on a Long Short-Term Memory Recurrent Neural Network (LSTM-RNN)). A Li-ion pouch cell with 47 Ah nominal capacity is used to parameterize all the models.

Reactivity controlled compression ignition (RCCI) engines are considered as a potent solution to realize near zero nitrogen oxides (NOx) and soot emission with higher thermal efficiency. However, operational control in RCCI engines is challenging, as events such as ignition and combustion phasing etc. are mostly decoupled from hardware induced start of injection. In modern control architecture, these real time data are internally computed using signals from cylinder pressure sensor (CPS). Lately, physics based control models or grey box models in RCCI engines were considered as a cost competitive and smart alternative to hardware signal source. In this work, an attempt was made to develop and compare physics based grey box model with data based neural networks, trained through supervised learning (or the black box models) to accurately predict dynamic combustion control parameters across five engine loads and incremental premix energy share not exceeding 60%.

Electrification is the well-accepted solution to address carbon emissions and modernize vehicle controls. Batteries play a critical in the journey of electrification and modernization with battery voltage prediction as the foundation for safe and efficient operation. Due to its strong dependency on prior information, battery voltage was estimated with recurrent neural network methods in the recent literatures exploring a variety of deep learning techniques to estimate battery behaviors. In these studies, standard recurrent neural networks, gated recurrent units, and long-short term memory are popular neural network architectures under review. However, in most cases, each neural network architecture is individually assessed and therefore the knowledge about comparative study among three neural network architecture is limited. In addition, many literatures only studied either the dynamic voltage response or the voltage relaxation.

Combustion with knock is an abnormal phenomenon which constrains the engine performance, thermal efficiency and longevity. The advance timing of the ignition system requires it to be updated with respect to fuel octane number variation. The production series engines are calibrated by the manufacturer to run with a special fuel octane number. In the experiment, the engine was operated at different speeds, loads, spark advance timings and consumed commercial gasoline with research octane numbers (RON) 95, 97 and 100. A 1-dimensional validated engine combustion model was run in the GT-Power software to simulate the engine conditions required to define the knock envelope at the same engine operation conditions as experiment. The knock intensity investigation due to spark advance sweep shows that combustion with noise was started after a specific advance ignition timing and the audible knock occur by increasing the advance timing.

Battery temperature is known to have a critical influence on overall battery pack performance, from electrochemical behavior, charge acceptance, power availability, trip efficiency, safety, reliability and life-cycle costs. Temperature monitoring is critical to ensure safe and reliable battery pack operation. Monitoring of cell temperatures in battery packs allows for control and estimation algorithms that can ensure homogenous pack temperature distribution, prevent excessive pack temperature rise and even infer cell core temperature, potentially allowing to both predict and mitigate onset of thermal runaway. The increasing need for improved accuracy requires inclusion of more detail in the modelling stage, leading inevitably to ever larger-scale, ever more complex dynamical systems. Simulations in such large-scale settings lead in turn to unmanageably large demands on computational resources, which is the main motivation for Model Order Reduction.

Since the base oil component of engine oils, driveline fluids and industrial lubricants typically exceeds 80 wt. % of the formulation, the complex chemical composition of base oils is a critical parameter in defining the ultimate performance of the finished products into which they are blended. Using both statistical and Neural Network methods, we have correlated the relative distribution of molecular types such as aromatics, naphthenes, paraffins and certain sulfur-containing species to lubricant performance in the ASTM Sequence IIIE and VE gasoline engine tests as well as the ASTM D-943 test which measures the long-term oxidative stability of industrial oils. For all cases, the “modeling” procedures enable approximately 20 input variables (compositional parameters, VI, aniline point) to be used to predict the output ratings of the respective test procedures.

This work is based in a model of neural and wavelets networks using published experimental data. The objective is to compare a neural and a wavelet network estimating the creep rupture strength based on chemical composition of Fe-2.25Cr-Mo and Fe-(9-12)Cr steels, and on its heat treatment temperature and life time. It will be determined the configuration that provides the best fit of the data.

Two soft-sensing methods which are neural network and Kalman filter for estimating vehicle side slip angle are compared. A radial basis function (RBF) neural network based soft-sensing model is proposed to estimate vehicle side slip angle in driver-vehicle closed-loop system. Vehicle side slip angle is considered as mapping of time series of yaw rate and lateral acceleration which are easily measured, the nonlinear mapping relationship of the three state parameters is established through neural network. In addition the method based on Kalman filter is also given. The results of comparison between estimation and measurement show that the neural network method proposed in this paper has higher accuracy and less computation requirement. It can provide theoretical guidance for design of estimator in vehicle stability control system.

Precise prediction of combustion parameters such as peak firing pressure (PFP) or crank angle of 50% burned mass fraction (MFB50) is essential for optimal engine control. These quantities are commonly determined from in-cylinder pressure sensor signals and are crucial to reach high efficiencies and low emissions. Highly accurate in-cylinder pressure sensors are only applied to test rig engines due to their high cost, limited durability and special installation conditions. Therefore, alternative approaches which employ virtual sensing based on signals from non-intrusive sensors retrieved from common knock sensors are of great interest. This paper presents a comprehensive comparison of selected approaches from literature, as well as adjusted or further developed methods to determine engine combustion parameters based on knock sensor signals. All methods are evaluated on three different engines and two different sensor positions.

Abstract To address the comfort and safety concerns related to driving vehicles, the Advanced Driver Assistance System (ADAS) is gaining huge popularity. The general architecture of autonomous vehicles includes perception, planning, control, and actuation. This article aims mainly at the controls aspect of one of the emerging ADAS features Lane Centering System (LCS). Limitations in deploying this feature from a controls point of view include maintaining the lane center with winding curvatures, dealing with the dynamic environment, optimizing controls where the perception of lane boundaries is erroneous, and, finally, concurring with the driver’s preferences. Although some research is available on LCS controls, most works are related only to the lateral controls by actuating steering. To increase the robustness, a comprehensive control strategy that involves lateral control, as well as longitudinal control along with a novel strategy to select the mode of driving, is proposed.

A computer code oriented to S.I. engine control and powertrain simulation is presented. The model, developed in Matlab-Simulink® environment, predicts engine and driveline states, taking into account the dynamics of air and fuel flows into the intake manifold and the transient response of crankshaft, transmission gearing and vehicle. The model, derived from the code O.D.E.C.S. for the optimal design of engine control strategies now in use at Magneti Marelli, is suitable both for simulation analysis and to achieve optimal engine control strategies for minimum consumption with constraints on exhaust emissions and driveability via mathematical programming techniques. The model is structured as an object oriented modular framework and has been tested for simulating powertrain system and control performance with respect to any given transient and control strategy.

Multi-sensor fusion strategies have gradually become a consensus in autonomous driving research. Among them, radar-camera fusion has attracted wide attention for its improvement on the dimension and accuracy of perception at a lower cost, however, the processing and association of radar and camera data has become an obstacle to related research. Our approach is to build a concise framework for camera and radar detection and data association: for visual object detection, the state-of-the-art YOLOv5 algorithm is further improved and works as the image detector, and before the fusion process, the raw radar reflection data is projected onto image plane and hierarchically clustered, then the projected radar echoes and image detection results are matched based on the Hungarian algorithm. Thus, the category of objects and their corresponding distance and speed information can be obtained, providing reliable input for subsequent object tracking task.