Search Results

Abstract Various feedstocks can be employed for biodiesel production, leading to considerable variation in composition and engine fuel characteristics. Using biodiesels originating from diverse feedstocks introduces notable variations in engine characteristics. Therefore, it is imperative to scrutinize the composition and properties of biodiesel before deployment in engines, a task facilitated by predictive models. Additionally, the international commercialization of biodiesel fuel is contingent upon stringent regulations. The traditional experimental measurement of biodiesel properties is laborious and expensive, necessitating skilled personnel. Predictive models offer an alternative approach by estimating biodiesel properties without depending on experimental measurements. This research is centered on building models that correlate mid-infrared spectra of biodiesel and critical fuel properties, encompassing kinematic viscosity, cetane number, and calorific value.

Abstract This article presents a novel approach to optimize the placement of light detection and ranging (LiDAR) sensors in autonomous driving vehicles using machine learning. As autonomous driving technology advances, LiDAR sensors play a crucial role in providing accurate collision data for environmental perception. The proposed method employs the deep deterministic policy gradient (DDPG) algorithm, which takes the vehicle’s surface geometry as input and generates optimized 3D sensor positions with predicted high visibility. Through extensive experiments on various vehicle shapes and a rectangular cuboid, the effectiveness and adaptability of the proposed method are demonstrated. Importantly, the trained network can efficiently evaluate new vehicle shapes without the need for re-optimization, representing a significant improvement over classical methods such as genetic algorithms.

Abstract In the process of injection molding, the vacuum pump rear housing is prone to warping deformation and volume shrinkage, which affects its sealing performance. The main reason is the improper control of the injection process and the large flat structure of the vacuum pump rear housing, which does not meet its production and assembly requirements (the warpage deformation should be controlled within 1.1 mm and the volume shrinkage within 10%). To address this issue, this study initially utilized orthogonal experiments to obtain training samples and conducted a preliminary analysis using gray relational analysis. Subsequently, a predictive model was established based on a one-dimensional convolutional neural network (1D CNN).

Abstract Reactivity-controlled compression ignition (RCCI) engine is an innovative dual-fuel strategy, which uses two fuels with different reactivity and physical properties to achieve low-temperature combustion, resulting in reduced emissions of oxides of nitrogen (NOx), particulate matter, and improved fuel efficiency at part-load engine operating conditions compared to conventional diesel engines. However, RCCI operation at high loads poses challenges due to the premixed nature of RCCI combustion. Furthermore, precise controls of indicated mean effective pressure (IMEP) and CA50 combustion phasing (crank angle corresponding to 50% of cumulative heat release) are crucial for drivability, fuel conversion efficiency, and combustion stability of an RCCI engine.

Abstract This scientific publication presents the application of artificial intelligence (AI) techniques as a virtual sensor for tailpipe emissions of CO, NOx, and HC in a high-performance vehicle. The study aims to address critical challenges faced in real industrial applications, including signal alignment and signal dynamics management. A comprehensive pre-processing pipeline is proposed to tackle these issues, and a light gradient-boosting machine (LightGBM) model is employed to estimate emissions during real driving cycles. The research compares two modeling approaches: one involving a unique “direct model” and another using a “two-stage model” which leverages distinct models for the engine and the aftertreatment. The findings suggest that the direct model strikes the best balance between simplicity and accuracy.

Abstract Accurate estimation of traction force is essential for the development of advanced control systems, particularly in the domain of autonomous driving. This study presents an innovative approach to enhance the estimation of tire–road interaction forces under combined slip conditions, employing a combination of empirical models and neural networks. Initially, the well-known Pacejka formula, or magic formula, was adopted to estimate tire–road interaction forces under pure longitudinal slip conditions. However, it was observed that this formula yielded unsatisfactory results under non-pure slip conditions, such as during curves. To address this challenge, a neural network architecture was developed to predict the estimation error associated with the Pacejka formula. Two distinct neural networks were developed. The first neural network employed, as inputs, both longitudinal slip ratios of the driving wheels and the slip angles of the driving wheels.

Abstract The swift progress of electric vehicles (EVs) and hybrid electric vehicles (HEVs) has driven advancements in battery management systems (BMS). However, optimizing the algorithms that drive these systems remains a challenge. Recent breakthroughs in data science, particularly in deep learning networks, have introduced the long–short-term memory (LSTM) network as a solution for sequence problems. While graphics processing units (GPUs) and application-specific integrated circuits (ASICs) have been used to improve performance in AI-based applications, field-programmable gate arrays (FPGAs) have gained popularity due to their low power consumption and high-speed acceleration, making them ideal for artificial intelligence (AI) implementation. One of the critical components of EVs and HEVs is the BMS, which performs operations to optimize the use of energy stored in lithium-ion batteries (LiBs).

Abstract This study proposes a machine learning tabulation (MLT) method that employs deep neural networks (DNNs) to predict ignition delay and knock propensity in spark ignition (SI) engines. The commonly used Arrhenius model and Livengood–Wu integral for fast knock prediction are not accurate enough to account for residual gas species and may require adjustments or modifications to account for specific engine characteristics. Detailed kinetics modeling is computationally expensive, so the MLT approach is introduced to solve these issues. The MLT method uses precalculated thermochemical states of the mixture that are clustered based on a combustion progress variable. Hundreds of DNNs are trained with the stochastic Levenberg–Marquardt (SLM) optimization algorithm, reducing training time and memory requirements for large-scale problems. MLT has high interpolation accuracy, eliminates the need for table storage, and reduces memory requirements by three orders of magnitude.

Abstract Sealed electronic components are the basic components of aerospace equipment, but the issue of internal loose particles greatly increases the risk of aerospace equipment. Traditional material recognition technology has a low recognition rate and is difficult to be applied in practice. To address this issue, this article proposes transforming the problem of acquiring material information into the multi-category recognition problem. First, constructing an experimental platform for material recognition. Features for material identification are selected and extracted from the signals, forming a feature vector, and ultimately establishing material datasets. Then, the problem of material data imbalance is addressed through a newly designed direct artificial sample generation method. Finally, various identification algorithms are compared, and the optimal material identification model is integrated into the system for practical testing.

Abstract Connected fuel cell vehicles (C-FCVs) have gained increasing attention for solving traffic congestion and environmental pollution issues. To reduce operational costs, increase driving range, and improve driver comfort, simultaneously optimizing C-FCV speed trajectories and powertrain operation is a promising approach. Nevertheless, this remains difficult due to heavy computational demands and the complexity of real-time traffic scenarios. To resolve these issues, this article proposes a two-level eco-driving strategy consisting of speed planning and energy management layers. In the top layer, the speed planning predictor first predicts dynamic traffic constraints using the long short-term memory (LSTM) model. Second, a model predictive control (MPC) framework optimizes speed trajectories under dynamic traffic constraints, considering hydrogen consumption, ride comfort, and traffic flow efficiency.

Abstract Vehicle-to-infrastructure (V2I) connectivity technology presents the opportunity for vehicles to perform autonomous longitudinal control to navigate safely and efficiently through sequences of V2I-enabled intersections, known as connected corridors. Existing research has proposed several control systems to navigate these corridors while minimizing energy consumption and travel time. This article analyzes and compares the simulated performance of three different autonomous navigation systems in connected corridors: a V2I-informed constant acceleration kinematic controller (V2I-K), a V2I-informed model predictive controller (V2I-MPC), and a V2I-informed reinforcement learning (V2I-RL) agent. A rules-based controller that does not use V2I information is implemented to simulate a human driver and is used as a baseline. The performance metrics analyzed are net energy consumption, travel time, and root-mean-square (RMS) acceleration.

Abstract The present study examines the effect of the multiple injection strategies in a common rail diesel engine using machine learning, image processing, and object detection techniques. The study demonstrates a novel approach of utilizing image-processing tools to gain information from heat release rates and in-cylinder visualizations from experimental or computational studies. The 3D CFD combustion and emission predictions of a commercial code ANSYS FORTE© are validated with small-bore common rail diesel engine data with known injection strategies. The validated CFD tool is used as a virtual plant model to optimize the injection schedule for reducing oxides of nitrogen (NOx) and soot emissions using an apparent heat release rate image-based machine learning tool. A methodology of the machine learning tool is quite helpful in predicting the NO–soot trade-off.

Abstract The uncertainty of a driver’s state, the variability of the traffic environment, and the complexity of road conditions have made driving behavior a critical factor affecting traffic safety. Accurate predicting of driving behavior is therefore crucial for ensuring safe driving. In this research, an efficient framework, distilled routing transformer (DRTR), is proposed for driving behavior prediction using multiple modality data, i.e., front view video frames and vehicle signals. First, a cross-modal attention distiller is introduced, which distills the cross-modal attention knowledge of a fusion-encoder transformer to guide the training of our DRTR and learn deep interactions between different modalities. Second, since the multi-modal learning usually requires information from the macro view to the micro view, a self-attention (SA)-routing module is custom-designed for SA layers in DRTR for dynamic scheduling of global and local attentions for each input instance.

Abstract A valuable quantity for analyzing the lateral dynamics of road vehicles is the side-slip angle, that is, the angle between the vehicle’s longitudinal axis and its speed direction. A reliable real-time side-slip angle value enables several features, such as stability controls, identification of understeer and oversteer conditions, estimation of lateral forces during cornering, or tire grip and wear estimation. Since the direct measurement of this variable can only be done with complex and expensive devices, it is worth trying to estimate it through virtual sensors based on mathematical models. This article illustrates a methodology for real-time on-board estimation of the side-slip angle through a machine learning model (SSE—side-slip estimator). It exploits a recurrent neural network trained and tested via on-road experimental data acquisition. In particular, the machine learning model only uses input signals from a standard road car sensor configuration.

Abstract Parking an articulated vehicle is a challenging task that requires skill, experience, and visibility from the driver. An automatic parking system for articulated vehicles can make this task easier and more efficient. This article proposes a novel method that finds an optimal path and controls the vehicle with an innovative method while considering its kinematics and environmental constraints and attempts to mathematically explain the behavior of a driver who can perform a complex scenario, called the articulated vehicle park maneuver, without falling into the jackknifing phenomena. In other words, the proposed method models how drivers park articulated vehicles in difficult situations, using different sub-scenarios and mathematical models.

Abstract The use of data-driven algorithms for the integration or substitution of current production sensors is becoming a consolidated trend in research and development in the automotive field. Due to the large number of variables and scenarios to consider; however, it is of paramount importance to define a consistent methodology accounting for uncertainty evaluations and preprocessing steps, that are often overlooked in naïve implementations. Among the potential applications, the use of virtual sensors for the analysis of solid emissions in transient cycles is particularly appealing for industrial applications, considering the new legislations scenario and the fact that, to our best knowledge, no robust models have been previously developed.

Abstract Automated driving is essential for developing and deploying intelligent transportation systems. However, unavoidable sensor noises or perception errors may cause an automated vehicle to adopt suboptimal driving policies or even lead to catastrophic failures. Additionally, the automated driving longitudinal and lateral decision-making behaviors (e.g., driving speed and lane changing decisions) are coupled, that is, when one of them is perturbed by unknown external disturbances, it causes changes or even performance degradation in the other. The presence of both challenges significantly curtails the potential of automated driving. Here, to coordinate the longitudinal and lateral driving decisions of an automated vehicle while ensuring policy robustness against observational uncertainties, we propose a novel robust coordinated decision-making technique via robust multiagent reinforcement learning.

Abstract Deep reinforcement learning has been utilized in different areas with significant progress, such as robotics, games, and autonomous vehicles. However, the optimal result from deep reinforcement learning is based on multiple sufficient training processes, which are time-consuming and hard to be applied in real-time vehicle energy management. This study aims to use expert knowledge to warm start the deep reinforcement learning for the energy management of a hybrid electric vehicle, thus reducing the learning time. In this study, expert domain knowledge is directly encoded to a set of rules, which can be represented by a decision tree. The agent can quickly start learning effective policies after initialization by directly transferring the logical rules from the decision tree into neural network weights and biases. The results show that the expert knowledge-based warm start agent has a higher initial learning reward in the training process than the cold start.

Abstract Speckle noise degrades the visual appearance and the quality of a synthetic aperture radar (SAR) image. The reduction of speckle noise is the first step in any remote-sensing device. To improve the noisy SAR images, a variety of adaptive and nonadaptive noise reduction filters were used. In order to eliminate speckle noise present in SAR images, an adaptive cuckoo search optimization-based speckle reduction bilateral filter has been designed in this article. To test the ability to eliminate multiplicative noise, the suggested filter’s effectiveness was compared to that of several de-speckling approaches. It has been measured with different assessment metrics such as PSNR, EPI, SSIM, and ENL. When compared to conventional de-noising filters, the proposed filter shows promising results for lowering speckle noise and retaining edge properties.

Abstract Numerous researchers are committed to finding solutions to the path planning problem of intelligence-based vehicles. How to select the appropriate algorithm for path planning has always been the topic of scholars. To analyze the advantages of existing path planning algorithms, the intelligence-based vehicle path planning algorithms are classified into conventional path planning methods, intelligent path planning methods, and reinforcement learning (RL) path planning methods. The currently popular RL path planning techniques are classified into two categories: model based and model free, which are more suitable for complex unknown environments. Model-based learning contains a policy iterative method and value iterative method. Model-free learning contains a time-difference algorithm, Q-learning algorithm, state-action-reward-state-action (SARSA) algorithm, and Monte Carlo (MC) algorithm.