Vehicle stability is fundamental to the safe operation of intelligent vehicles, and real-time, high-accuracy calculation of the stability domain is crucial for maintaining control across the full range of driving conditions. Because the real stability domain is difficult to parameterize accurately and is shaped by multiple driving factors including vehicle-dynamics parameters and environmental conditions, existing approaches fail to capture the multidimensional couplings between time-varying driving inputs and the resulting stability boundaries. Moreover, these methods remain overly conservative owing to algorithmic limitations and cautious design assumptions, thereby restricting dynamic performance in complex scenarios. To address these limitations, this paper introduces a multidimensional vehicle dynamic stability region calculation framework under time-varying driving conditions and apply it into path tracking controller of intelligent vehicle. Sum-of-squares programming (SOSP) is

Wang, Chengye, Zhang, Yu, Hu, Xuepeng, Qin, Haipeng, Wang, Guoli, Qin, Yechen

Mode Identification and Collaborative Switch for the Front-Centralized and Rear-Distributed Electric Drive Vehicle Based on BP Neural Network

2025-01-7335

12/31/2025

Distributed-drive electric vehicles (DDEVs) significantly enhance off-road maneuverability but suffer from compromised high-speed stability and robustness. This research introduces a front-centralized and rear-distributed (FCRD) architecture that synergistically leverages the advantages of each configuration. The electric-drive-wheel (EDW) on the rear suspension can provide three working modes: (a) Drive-connected mode, (b) Drive-disconnected mode, (c) Brake mode. It is the key actuator for vehicle mode-switching, which supports the vehicle with three driving modes: (a) DDEV, (b) front-wheel drive (FWD), (c) all-wheel drive (AWD). A hierarchical control architecture employs the upper-layer controller with Back Propagation Neural Network (BPNN) for mode identification and decision-making. The lower-layer controller enables the intelligent torque distribution and collaborative control of the motors. The control strategy is pre-trained in the VCU (vehicle control unit) with off-line data

Ding, Xiaoyu, Chen, Xinbo, Wang, Wei, Zhang, Jiantao, Kong, Aijing

Modeling on Self-Learning Driver Model under Extreme Driving Conditions

2025-01-7315

12/31/2025

Ensuring the safe and stable operation of autonomous vehicles under extreme driving conditions requires the capability to approach the vehicle’s dynamic limits. Inspired by the adaptability and trial and error learning ability of expert human drivers, this study proposes a Self-Learning Driver Model (SLDM) that integrates trajectory planning and path tracking control. The framework consists of two core modules: In the trajectory planning stage, an iterative trajectory planning method based on vehicle dynamics constraints is employed to generate dynamically feasible limit trajectories while reducing sensitivity to initial conditions; In the control stage, a neural network enhanced nonlinear model predictive controller (NN-NMPC) is designed, which incorporates a self-learning mechanism to continuously update the internal vehicle model using trial-and-error data on top of mechanistic physical constraints, thereby improving predictive accuracy and path-tracking performance. By combining

Zhang, Xinjie, Xu, Long, Guo, Konghui, Zhuang, Ye, Hu, Tiegang, Mao, JingGang, Zeng, Qingqiang

CANet: A Lightweight Cross-Attention Guided Network for Unstructured Road Segmentation

2025-01-7309

12/31/2025

To address the issues of multiple background interferences and blurred road boundaries in unstructured scene road segmentation tasks, a lightweight and precise unstructured road segmentation model based on cross-attention (CANet) is proposed. This model constructs an encoder using the lightweight neural network MobileNetV2. By doing so, it ensures light weight while enhancing the feature discrimination ability of unstructured roads, thus achieving efficient feature extraction. The decoder integrates the cross-attention mechanism and a low-level feature fusion branch. The attention mechanism improves the model’s perception of road boundaries by capturing long-distance context information in the feature map, thereby solving the problem of blurred edges. The low-level feature fusion branch enhances the detail accuracy and edge continuity of the segmentation results by incorporating high-resolution information from shallow features. Experimental results show that the proposed model attains

Wang, Xuewei, Cao, Guangyuan, Liang, Xiao, Li, Shaohua

Coordinate Attention-Driven Robust Multi-Object Tracking in Autonomous Vehicles: A Hybrid Framework of CA-YOLOv5 and ResNet-DeepSORT

2025-01-7302

12/31/2025

With the increasing complexity of traffic conditions, the computational burden of multi-object tracking algorithms has grown, making it difficult to meet the requirements for tracking accuracy and real-time performance. In this paper, we proposed a road vehicle multi-object tracking method by improving and optimizing the YOLOv5 detection algorithm and the DeepSORT tracking algorithm. A Channel Attention(CA) mechanism is introduced into the existing YOLOv5 algorithm to construct the fusion algorithm CA-YOLOv5, and the feature extraction network structure of YOLOv5 is reconstructed by adding a prediction layer to improve the accuracy of vehicle detection. The ReID (Re-identification) network in DeepSORT algorithm is adopted as ResNet neural network to construct the fusion algorithm ResNet-DeepSORT. And it combined with data and feature enhancement, as well as high accuracy detection results of road vehicles. Thus, it improves the tracking accuracy and reduces the number of ID jumps to

Bo, Liu, Jing, Wu, Yanping, Zhou, Jing, Li

An Interpretable Deep Learning Decision-Making Framework for Autonomous Lane-Change Inspired by Human Driving Experiences

2025-01-7339

12/31/2025

Lane change plays a critical role in autonomous driving and directly affects traffic safety and efficiency. Although deep learning-based lane-change decision-making frameworks have achieved promising results, they still face fundamental challenges in producing human-consistent and trustworthy behavior, mainly due to: 1) Inadequate psychology-informed personalization, as most frameworks focus on physical variables but neglect psychological factors (e.g., risk tolerance, urgency), limiting their ability to capture individual differences in lane-change motivations. 2) Limited holistic understanding of traffic context, most frameworks lack consideration of high-level and interpretable indicators (e.g., traffic pressure) in comprehensively assessing dynamic traffic scenarios, limiting their capacity for human-like contextual understanding. 3) Lack of transparent and interpretable decision logic, as many frameworks operate as black boxes with opaque reasoning processes, hindering human

Chen, Yanbo, Chen, Jiaqi, Yu, Huilong, Xi, Junqiang

Knowledge Modeling and Reuse of Concrete Structure Strengthening Solutions for Existing Buildings

2025-99-0202

12/23/2025

Implementing knowledge modelling tools of concrete structure strengthening solutions for existing buildings addresses the urgent needs of urban renewal efforts. This paper thoroughly investigates the application of Natural Language Processing (NLP), and knowledge graphs for organizing and managing complex information related to building strengthening strategies. By developing an ontology model for solutions and supplementing it with methods for generating word vectors and annotating data, this study constructed a comprehensive framework for the management of strengthening solution knowledge. A case study on the partial structural strengthening validated the applicability of the proposed model in facilitating recommendations for similar cases and supporting solution design. This research under-scores the transformative impact of digital technologies and knowledge modelling on the efficiency and quality of urban renewal projects, contributing to the advancement of smart cities. The

Zhang, Zhuohao, Luo, Hanbin, Wu, Haozheng, Chen, Weiya

Experimental Research on MFL Detection Technology for Corrosion Damage of Prestressing Tendons in Bridges

2025-99-0203

12/23/2025

Corrosion of prestressed tendons endangers the safety of bridges, but until now, there has been no effective method to solve the problem of detecting corrosion damage in prestressed tendons of concrete beams. To address this, a magnetic flux leakage detection experimental apparatus for corrosion damage in prestressed tendons based on the principle of magnetic flux leakage inspection has been developed. Using this apparatus, magnetic flux leakage tests were conducted on prestressed tendons after electrochemical corrosion, and the results were compared with simulation analysis to conduct a comparative study. In the experiments, the influence of corrosion severity, corrosion width, and the effect of stirrups on the characteristics of the magnetic flux leakage signals were studied. Magnetic signal feature values were extracted, and a quantification neural network model for corrosion damage was established, which is used to quantify the degree of corrosion damage in prestressed tendons. The

Wang, Peng, Gao, Min, Dong, Lei, Zhu, Junliang

Research Progress on Pavement Performance Prediction Models: A Review

2025-99-0249

12/23/2025

Road maintenance plays a vital role in maintaining road conditions and ensuring safety, especially in a country with an extensive road network like China. To accurately predict pavement performance, optimize maintenance strategy, reduce cost and improve road efficiency, the paper systematically combed and evaluated the prediction model of pavement performance. Firstly, the importance of pavement maintenance and the background of pavement maintenance performance prediction model are described, and explicit models (mechanical-empirical model, stochastic process, time series analysis) and machine learning models (regression analysis, support vector machine, integrated learning, artificial neural network, deep learning) are introduced respectively. The basic principle, representative study, advantages and disadvantages of each model are introduced in detail. Comparative analysis shows that the traditional explicit model is simple and effective, easy to explain, but difficult to deal with

Ma, Muyun, Dong, Qiao, Lin, Yelong

Research on the Decision-making Model and Algorithm of Following and Lane-Changing Behavior of Connected Autonomous Vehicles on Freeway

2025-99-0236

12/23/2025

In order to achieve the widespread application of autonomous driving technology in basic freeway segments, especially in the automated decision-making of following and lane changing behaviors, Connected Autonomous Vehicles (CAVs) must be able to reliably complete driving tasks in complex traffic environments. Our study introduces a novel behavior decision-making architecture for connected autonomous vehicles, which employs the Dueling Double Deep Q-Network (D3QN) algorithm as its core methodology. The model optimizes the decision-making ability in complex traffic scenarios by separating action selection and value assessment and implementing them by different neural networks. The multi-dimensional reward function, which comprehensively considers safety, comfort and efficiency, is introduced into the reinforcement learning training of the model. The simulation scenario of the basic freeway segment is established and the model is trained in the mixed traffic flow environment, compared

Hou, Zhiyun, Yang, Xiaoguang

Machine Learning-based Prediction of Perimeter Rock Deformation in Deep-Buried Large-Section Tunnel Excavation

2025-99-0357

12/17/2025

With the accelerating urbanization process in contemporary China, metro systems have assumed an increasingly pivotal role within the national transportation infrastructure. Ensuring structural stability of tunnel surrounding rock formations during construction, as well as conducting comprehensive evaluation and accurate prediction of rock deformation patterns, has emerged as an exceptionally critical component of modern tunneling engineering practices. This investigation, conducted within the context of the Chongqing Rail Transit Line 18 Northern Extension Project, specifically examines deformation characteristics in deep-buried large-cross-section tunnels. The research employs four sophisticated time-series prediction models - Convolutional Neural Networks (CNN), Extreme Learning Machines (ELM), Genetic Algorithm-optimized Backpropagation Neural Networks (GA-BP), and Long Short-Term Memory networks (LSTM) to systematically predict both crown settlement deformations and convergence

Ai, Jianke, Yang, Xiaodong, Sun, Shiqiang, Zhang, Yupeng

Vehicle Abnormal Detection Based on Multi-Source Data Fusion and Graph Neural Networks

2025-99-0430

12/10/2025

Traffic abnormal detection is crucial in intelligent transportation systems, while the heterogeneity and weak spatio-temporal correlation of multi-source data make it difficult for traditional methods to effectively fuse and utilize multimodal information. Most of the existing studies use data-level or decision-level fusion, which fails to fully exploit the feature complementarity of multi-source data, resulting in limited detection accuracy. To this end, we propose a multi-source data fusion anomaly detection method based on graph autoencoder (GAE) and diffusion graph neural network (DiffGNN). First, a unified data preprocessing and fusion strategy is designed to perform feature-level fusion of data from on-board sensors, infrastructures, and external environments to eliminate inconsistencies in data format, temporal alignment, and spatial distribution. Then, GAE is employed for potential graph structure feature extraction to enhance the global representation of the data on the basis

Wang, Yaguang, Xiao, Yujie, Ma, Ying

TRL-Trans: Adaptive Graph Learning with Temporal Representation for Traffic Flow Prediction

2025-99-0429

12/10/2025

Traffic flow forecasting plays a pivotal role within intelligent transportation frameworks. Although existing methods combine graph neural networks and temporal models, there are still problems, such as static graph structure being challenging to characterize the dynamic associations between traffic nodes, insufficient ability to model long temporal dependencies, and low efficiency of fusion of complex spatio-temporal features, etc. Based on this, we propose a Transformer-based Temporal Representation Learning traffic flow prediction model (TRL-Trans). The proposed model employs Temporal Representation Learning (TRL) to derive contextual insights from heavily masked subsequences. It incorporates a Gated Temporal Convolutional Network (Gated TCN) coupled with an Adaptive Hybrid Graph Convolution Module (AHGCM) to effectively capture dynamic spatio-temporal characteristics. The AHGCM dynamically merges predefined adjacency matrices with implicit spatio-temporal relationships

Zhou, Jianping, Lu, Zongjiang, Wang, Zhongyuan, He, Jin, Liu, Chunya

Research on Electromagnetic Interference Identification of Track Circuit Equipment Based on CNN Transformer

2025-99-0416

12/10/2025

This paper proposes a track circuit interference identification model, which combines convolutional neural network (CNN) and transformer architecture to identify common types of electromagnetic interference in track circuit equipment. The model maps the time-frequency characteristics of the input monitoring signal into high-dimensional features through the deep learning model, and classifies the interference modes. Subsequently, a variety of common interference signals are generated for experimental verification, and the proposed model performs well on the test data. Ablation experiments show that the combination of convolutional neural network and attention mechanism can effectively improve the classification performance of interference.

Wei, Zijun, Yang, Shiwu, Dai, Meng, Feng, Qin, Chu, Shaotong

Training Uncrewed Underwater Vehicles for Visualization of Subsurface Structures

TBMG-54290

12/01/2025

Lumped Parameter Thermal Network for Automotive Components: Modeling, Simulation, System Identification, and Parameter Estimation

2025-01-5078

11/28/2025

This research paper proposes a framework based on lumped parameter thermal networks (LPTN) to understand the system behavior of thermally stressed component spaces in automotive vehicles. LPTNs offer an energy-based, low-degree-of-freedom model that can represent arbitrary thermal systems inside automotive vehicles. The time response of these low-order models can be calculated using standard ordinary differential equation solvers. The paper showcases the modeling of LPTNs and the calculation of their time response by using an electronic control unit (ECU) of a BMW 7 series. The use of LPTNs instead of exponential functions reduced the MAE in this example by 60.5%. Furthermore, a system identification approach for experimental temperature curves has been developed and implemented. System identification aims to mathematically model system behavior and predict system output. This paper compares least-square estimation (LSE) with constrained minimization (CM), where CM has a higher MAE by

Kehe, Maximilian, Enke, Wolfram, Rottengruber, Hermann

Data-Based Modeling of Power-Split Hybrids Using Cascaded Neural Networks

2025-01-0522

11/25/2025

Power-split hybrid powertrains represent one of the most advanced and complex types of powertrain systems. The combination of multiple energy sources and power paths offers great potential but results in complex interactions that require improved strategies for optimal efficiency and emission control. The development and optimization of such operating strategies typically involve algorithms that demand fast computational environments. Traditional high-accuracy numerical simulations of such a complex system are computationally expensive, limiting their applicability for extensive iterative optimizations and real-time applications. This paper introduces a data-based approach designed specifically to address this challenge by efficiently modeling the dynamic behavior of power-split hybrid powertrains using cascaded neural networks. Cascaded neural networks consist of interconnected subnetworks, each specifically trained to represent individual drivetrain components or subsystems. This

Frey, Markus, Itzen, Dirk, Yang, Qirui, Grill, Michael, Kulzer, André Casal

Prediction of Automotive Seat Response Based on IPSO-BP Neural Network

2025-99-0124

11/11/2025

To enhance the predictive accuracy between seat structural parameters and crash performance, a hybrid model was constructed by coupling an Improved Particle Swarm Optimization (IPSO) algorithm with a Back Propagation Neural Network (BPNN). First, a finite element model for front and rear impact of automotive seats was established based on experimental data, and the model’s accuracy was verified. Subsequently, simulations were conducted, and the results were analyzed. The Energy Absorption Mass Ratio method was used to screen the design variables, ultimately selecting 10 thickness variables and 9 material variables as design variables. Latin Hypercube Sampling was employed to divide the dataset into a testing set and a training set. Then, the Particle Swarm Optimization (PSO) was enhanced with Levy flights and a local mutation strategy, utilizing the IPSO algorithm to optimize the initial weights and thresholds of the BPNN, resulting in the establishment of the IPSO-BPNN predictive

Qiu, Yufei, Long, Jiangqi

Multimodal Intelligent Transportation Data Fusion and Application Methods in Heavy-Duty Vehicles

2025-99-0139

11/11/2025

Heavy-duty commercial vehicles (HDCVs) are the key mobile nodes in intelligent transportation systems (ITS). However, their complex operating conditions and the diversity of data sources (such as road conditions, driver behavior, traffic signals, and on-board sensors) present considerable difficulties for accurately estimating the state and perceiving the environment using a single modality of data. This requires effective multi-modal data fusion to enhance the control and decision-making capabilities of HDCVs. This paper addresses this need by proposing a customized multi-modal intelligent transportation data fusion framework for intelligent HDCVs. This paper presents a solution for establishing a multi-modal intelligent transportation data collection platform, including real-scene collection methods and simulation scene collection methods based on the SUMO-MATLAB joint simulation platform. Through three representative case studies, the application methods of multi-modal traffic data

Chen, Zhengxian, Wang, Shaoqi, Jiang, Huiming, Zhou, Fojin, Wang, Mingqiang, Li, Jun

AI/ML Based Audio Classification and Transcription for Automotive and Off-Highway Systems

2025-28-0323

11/06/2025

Modern automotive systems generate a wide range of audio-based signals, such as indicator chimes, turn signals, infotainment system audio, navigation prompts, and warning alerts, to facilitate communication between the vehicle and its occupants. Accurate Classification and transcription of this audio is important for refining driver aid systems, safety features, and infotainment automation. This paper introduces an AI/ML-powered technique for audio classification and transcription in automotive environments. The proposed solution employs a hybrid deep learning architecture that leverages convolutional neural networks (CNNs) and recurrent neural networks (RNNs), trained using labeled audio samples. Moreover, an Automatic Speech Recognition (ASR) model is integrated for transcribing spoken navigation prompts and commands from infotainment systems. The proposed system delivers reliable results in real-time audio classification and transcription, facilitating better automation and

Singh, Shwetha, Kamble, Amit, Mohanty, Ananta, Kalidas, Sateesh

Integrating AI in Wiring Harness Design for Enhanced Efficiency

2025-28-0311

11/06/2025

Modern vehicle integration has become exponentially more difficult due to the complicated structure of designing wiring harnesses for multiple variants that have diverse design iterations and requirements. This paper proposes an AI-driven solution for addressing variant complexity. By using Convolutional Networks and Deep Neural Networks (CNN & DNN) to generate harness routing using defined specifications and constraints, the proposed solution uses minimal human intervention, substantially less time, and enables less complexity in designing. AI trained modelled systems can generally even predict failures in production methods which also reduces downtime and increases productivity. The new AI system automatically converts design specifications to manufacturable design specifications to avoid confusion with design parameters, by optimizing concepts with connector placements, grommet fittings, clip alignments, and other tasks. The solution coping with the inherent dynamic complexity of

N, Rishi Kumaar, Patil R, Bharath, Rajavelu, Vivek, Ramachandran, Vignesh, Mohanty, Lalit, Padmarajan, Vishnu

Optimizing Electric Vehicle Motor Testing with Fine Gaussian SVM and Bilayered Neural Network

2025-28-0397

10/30/2025

Electric vehicles are shaping the future of the automotive industry, with the drive motor being a crucial component in their operation. Ensuring motor reliability requires rigorous testing using specialized test benches to validate key performance parameters. However, inefficiencies in the helical gear configuration within these test systems have led to frequent malfunctions, affecting production flow. This study focuses on optimizing the motor test bench by refining critical design parameters through vibration signal analysis and machine learning techniques. Vibrational data is collected under different gear configurations, utilizing an accelerometer integrated with a Data Acquisition (DAQ) system and MATLAB-based directives for seamless data collection. Machine learning classifiers, including Fine Gaussian SVM and Bilayered Neural Network, are applied to categorize signals into normal and faulty conditions, both with and without a 0.25 KW load. The analysis reveals that SVM achieves

S, Ravikumar, Sharik, N, Syed, Shaul, V, Muralidharan, D, Pradeep Kumar

A Physics-Guided Neural Network Approach for EGR Cooler Temperature Estimation: Addressing ECU Model Discrepancies

2025-28-0393

10/30/2025

As the automotive industry explores alternative powertrain options to curb emissions, it is pertinent to refine existing technologies to improve efficiency. The Exhaust Gas Recirculation (EGR) system is one of the pivotal components in emission control strategies for Internal Combustion Engines (ICE). The EGR cooler is crucial in thermal management strategies, as it lowers the temperature of recirculated exhaust gases before feeding it along with fresh air, thereby reducing nitrogen oxides (NOx) emissions. Precise estimation of the EGR cooler outlet temperature is crucial for effective emission control. However, conventional Engine Control Unit (ECU) models fall short, as they often show discrepancies when compared to real-world test data. These models rely on empirical relationships that struggle to capture precisely the transient effect, and real time variation in operating conditions. To address these limitations and improve the accuracy of ECU based model, various signal processing

Kumar, Amit, Kumar, Raman, Manojdharan, Arjungopal, Challa, Krishna, Kramer, Markus

Smart Energy Management System for Hybrid Electric Vehicles: Integrating Fuel Cells and Ultra-Capacitors using Neural Network

2025-28-0365

10/30/2025

This paper offers a state-of-the-art energy-management strategy specifically developed for FCHEV focusing on robustness under uncertain operations. Currently, energy management strategies try to optimize fuel economy and take into account the sluggish response of fuel cells (FCs); however, they mostly do so assuming all system variables are explicit and deterministic. In real-world operations, however, a variety of sources may cause the uncertainty in power generation, energy conversion, and demand interactions, e.g., the variation of environmental variables, estimated error, and approximation error of system model, etc., which accumulates and adversely impacts the vehicle performance. Disregarding these uncertainities can result in overestimation of operating costs, overall efficiency and overstepped performance limitations, and, in serious cases can cause catastrophic system breakdown. To mitigate these risks, the current work introduces a neural network-based energy management

Deepan Kumar, Sadhasivam, M, Boopathi, R, Vishnu Ramesh Kumar, Karthick, K N, R, Nithiya, R, Krishnamoorthy, V, Dayanithi

Diagnostics Methodology for Turbocharger System Using Data Driven Models

2025-28-0404

10/30/2025

In a conventional powertrain driven by Internal combustion (IC) engines, turbocharger (TC) is a key component for enhancing performance and efficiency. Predominantly turbochargers are used to serve multiple purposes of downsizing, increased power, better fuel efficiency, reduced emissions, and improved performance at high altitudes. TC is responsible for fulfilling the air mass requirement of the engine at different operating conditions. Failure of TC system leads to abnormal engine operation. If the TC hardware is beyond repair, the associated replacement cost is very high. Ultimately, a predictive diagnostics approach is required to identify the issue with TC so that the failure of TC could be avoided. The proposed methodology uses advanced artificial intelligence technique called recurrent neural network (RNN) and long short-term memory (LSTM) network for predicting faults in a typical TC system. In this study, actual values of TC speed and boost pressure are obtained from physical

Jagtap, Virendra Shashikant, Ganguly, Gourav, Mitra, Partha, Patidar, Sachin

Research on the Layout and Robustness Testing of Electric Vehicle Battery Swap Stations Based on Graph Neural Networks: A Case Study of the Main Urban Area of Nanchang City

2025-99-0008

10/17/2025

With the rapid increase in the number of electric vehicles, the rational placement of battery swapping stations has become a critical issue in optimizing urban transportation infrastructure. This paper proposes a site selection optimization method based on Graph Neural Networks (GNN). By constructing an urban transportation graph model grounded in Points of Interest (POI) and road traffic data, the method analyzes battery swapping station layout plans and validates their robustness and scalability. Taking the main urban area of Nanchang City as a case study, the research integrates data on POI distribution and land-use functional diversity within buffer zones to construct a graph structure. It then employs GNN for node classification to identify optimal battery swapping station locations. Experimental results show that, compared to traditional methods, the proposed approach improves site selection accuracy by 15% and enhances optimization efficiency by 20%. This method can provide

Zeng, Yi, Yi, Xinyu

Strength Prediction of Self-Piercing Riveted Joints Using Practical Regression and Bayesian Neural Network

2025-01-5068

10/10/2025

Self-piercing riveting (SPR) is a key joining method in multi/thin-material automotive structures, yet accurately predicting the mechanical strength of SPR joints remains challenging due to numerous influencing factors. Empirical engineering equations [1] provide a foundation for estimating lap-shear and cross-tension strength but require several geometric parameters that are often unavailable in the design phase. To address this limitation, we extract and leverage the core physical relationships embedded in these formulas. By reformulating the dependence of joint strength on the yield strength and total thickness of the sheet stack as practical regression models, we enable strength prediction using only commonly available material properties. Furthermore, a Bayesian convolutional neural network (BCNN) model is developed to incorporate additional material features, offering improved prediction accuracy and uncertainty quantification.

Soproni, Istvan, Womack, Darren, Liu, Zongyue, Balaji, Ashwin, Kulange, Deepak

Deep Reinforcement Learning with Behavioral Cloning for Path Tracking of Four-Wheel Steering Vehicles

10-10-01-0001

10/09/2025

When the vehicle system performs trajectory tracking control, it presents relatively complex nonlinear coupling dynamics characteristics. The traditional coordination algorithm relying on a simplified linear model is mostly unable to deal well with the actual nonlinear dynamic behaviors. In contrast, reinforcement learning (RL) method will derive the optimal strategy by means of interaction with the environment. This eliminates the need for accurate vehicle modeling. These methods use all of the nonlinear approximation capabilities of deep neural networks and can effectively reflect the complex relationship between vehicle state and control actions. The framework itself supports multidimensional input processing and continuous operation space optimization because of the development of parallel processing architectures. In order to reduce the motion jitter caused by the direct generation of front and rear wheel angles by the network, this article uses steering angle increments as

Ren, Gaotian, Wang, Yangyang

ML-Based System Level Optimization of EV Cooling Circuit

2025-01-0376

10/07/2025

Efficient thermal management is vital for electric vehicles (EVs) to maintain optimal operating temperatures and enhance energy efficiency. Traditional simulation-based design approaches, while accurate, are often computationally expensive and limited in their ability to explore large design spaces. This study introduces a machine learning (ML)-based optimization framework for the design of an EV cooling circuit, targeting a 5°C reduction in the maximum electric motor temperature. A one-dimensional computational fluid dynamics (1D-CFD) model is utilized to generate a Design of Experiments (DOE) matrix, incorporating key parameters such as coolant flow rate and heat exchanger dimensions. A Radial Basis Function (RBF) neural network is trained on the simulation data to serve as a surrogate model, enabling rapid performance prediction. Optimization is performed using the Non-Dominated Sorting Genetic Algorithm II (NSGA2), yielding three distinct design solutions that meet the thermal

Paul, Kavin, Ganesan, Arul, Mansour, Youssef

Using Artificial Neural Networks for Predicting Vehicle Survivability within a Virtual Simulation

2025-01-0482

09/16/2025

A unique contribution the U.S. Army currently provides is what is known as Virtual Experiments (VEs). A VE consists of a large group of active-duty soldiers who participate in a video game simulating a battlefield scenario. During these simulations, the soldiers are provided with novel protective vehicle capabilities in an effort to evaluate their effectiveness on the battlefield. However, these VEs take a significant amount of time to conduct and are expensive. Using Artificial Neural Networks (ANNs) this study looks to predict vehicle survivability based on a limited amount of VE data. The results entail an overall predictive accuracy of 76.8% using only two ANN input features and provides a framework for the eventual addition of more VE datasets.

O’Bruba, Joseph

Leveraging Spectral Unmixing for Improved Mobility in Vegetated Environments

2025-01-0480

09/16/2025

The success of off-road missions for ground vehicles depends heavily on terrain traversability, which in turn requires a thorough understanding of soil characteristics a key component being soil moisture content. When large areas need to be analyzed, satellite imagery is often used, although this approach typically reduces the spatial resolution. This decrease of spatial resolution creates what are known as mixed pixels, when two or more classes or features are in a single pixel’s area, which can lead to noisier data and lower accuracy models. This paper investigates using linear spectral unmixing as a way to help clean / mitigate noisy data to yield better predictive models. Hyperspectral remote sensing from the Hyperion satellite platform and ground truth from the International Soil Moisture Network (ISMN) are used for the dataset. This study found that soil moisture content prediction, comparing the mixed multilayer perceptron (MLP) model with an unmixing approach revealed a 10–30

Ewing, Jordan, Jayakumar, Paramsothy, Kasaragod, Anush, Oommen, Thomas

Commander’s Decision Analysis Tool for Maintenance: Applications of Unsupervised Learning with Pattern Characterization (ULPC) in Vehicle Health Assessment

2025-01-0490

09/16/2025

Within the military maintenance cycle, commanders and units struggle with understanding the operational readiness of their fleets from a data driven perspective. Many unsupervised learning techniques have been developed with applications for vehicle maintenance with pattern classification. In this paper, Predictive Maintenance using Unsupervised Learning with Pattern Characterization (ULPC) is proposed to classify the overall health of the platform system and subsystems. In this model, the key features are selected using an intelligent pre-processing system for signal classification for each subsystem. Next the data is processed and compared to a normalcy baseline dataset using the unsupervised machine learning (ML) model. Operational data collected post-baseline is then processed through a Recurrent Neural Network (RNN) and clustered. An overall “normalcy” metric is calculated to show the difference in operation when compared to the baseline patterns. This normalcy servers as an

Bailey, Jeffrey, Cabrey, Connor, Hsu, Charles

Fine-Tuning Yolov8 for Accurate Unmanned Aerial Vehicle Detection with Digital Twin Simulation

2025-01-0473

09/16/2025

Drones, or Unmanned Aerial Vehicles (UAVs) pose an increasing threat to military ground vehicles due to their precision strike capabilities, surveillance functions, and ability to engage in electronic warfare. Their agility, speed, and low visibility allow them to evade traditional defense systems, creating an urgent need for advanced AI-driven detection models that quickly and accurately identify UAV threats while minimizing false positives and negatives. Training effective deep-learning models typically requires extensive, diverse datasets, yet acquiring and annotating real-world UAV imagery is expensive, time-consuming, and often non-feasible, especially for imagery featuring relevant UAV models in appropriate military contexts. Synthetic data, generated via digital twin simulation, offers a viable approach to overcoming these limitations. This paper presents some of the work Duality AI is doing in conjunction with the Army’s Program Executive Office Ground Combat Systems (PEO GCS

Mejia, Felipe, Shah, Sunil, Young, Preston C., Brunk, Andrew T.

Machine Learning-Enhanced Combustion Modelling: Predicting Ethanol Effects in a Single-Cylinder Research Engine

2025-24-0026

09/07/2025

Recent studies highlight the urgent need to reduce greenhouse gas (GHG) emissions to mitigate the impacts of global warming and climate change. As a major contributor, the transport sector plays a vital role in these efforts. Ethanol emerges as a promising fuel for decarbonising hard-to-electrify propulsion sectors, thanks to its sustainable production pathways and favourable physical and combustion properties, such as energy density, rapid burning velocity, and high knock resistance. This work proposes a methodology to enable the possibility of replicating the combustion behaviour of ethanol in a 1D CFD simulation environment representative of a single-cylinder research engine. Spark-ignition combustion is simulated through the Eddy Burn-Up combustion model previously calibrated for standard fossil gasoline. The combustion model features a laminar flame speed neural network, trained and tested through reference chemical kinetics simulations. The combustion model showed great accuracy

Ferrari, Lorenzo, Sammito, Giuseppe, Fischer, Marcus, Cavina, Nicolò

ML-based Identification of Lactones for Blend Studies and Combustion Performance in a Spark-Ignition CFR Engine

2025-24-0071

09/07/2025

The identification of sustainable fuels that exhibit optimal physico-chemical properties, can be synthesized from widely available feed-stocks, enable cost-effective large-scale production, and integrate seamlessly with existing infrastructure is essential for reducing global carbon emissions. Given their high energy density, efficient handling, and versatility across applications, renewable liquid fuels remain a critical component of even the most ambitious energy transition scenarios. Lactones, cyclic esters derived from the esterification of hydroxycarboxylic acids, feature a ring structure incorporating both a carbonyl group (C=O) and an ether oxygen (O). Variations in ring size and carbon chain length significantly influence their physicochemical properties, which in turn affect their performance in internal combustion engines. According to predictive models based on artificial neural networks, valerolactone, hexalactone, and heptalactone isomers show promise as fuels in spark

Sirna, Amanda, Loprete, Jason, Ristow Hadlich, Rodrigo, Assanis, Dimitris, Patel, Rutvi, Mack, J. Hunter

Flexible an Low Cost Fuel Consumption Prediction in Heavy-Duty Vehicles Using Machine Learning

2025-24-0121

09/07/2025

This study presents a novel approach for predicting fuel consumption in heavy-duty vehicles using a Machine Learning-based model, which is based on feedforward neural network (FFNN). The model is designed to enhance real-time vehicle monitoring, optimize route planning, and reduce both operational costs and environmental impact, making it particularly suitable for fleet management applications. Unlike traditional physics-based approaches, the FFNN relies solely on a refined selection of input variables, including vehicle speed, acceleration, altitude, road slope, ambient temperature, and engine power. Additionally, vehicle mass is estimated using a methodology presented elsewhere and is included as an input for a better generalization of the consumption model. This parameter significantly impacts fuel consumption and is particularly challenging to obtain for heavy-duty vehicles. Engine power is derived from both engine torque and speed (RPM), ensuring a direct relationship with fuel

Vicinanza, Matteo, Pandolfi, Alfonso, Arsie, Ivan, Giannetti, Flavio, Polverino, Pierpaolo, Esposito, Alfonso, Paolino, Antonio, Adinolfi, Ennio Andrea, Pianese, Cesare, Frasci, Valentino

Battery Datasets for Neural Network-Based State-of-Health Estimators: A Review and Implementative Workflow

2025-24-0139

09/07/2025

Management of battery systems for electric vehicles has great importance to ensure safe and efficient operation. State-of-Charge and State-of-Health (SoH) are fundamental parameters to be taken under control even though they cannot be directly measured during vehicle operation. Some control approaches have gained increasing interest thanks to advances in sensor availability, edge computing and the development of big data. In particular, SoH estimation through machine learning (ML) and neural networks (NNs) has been thoroughly investigated due to their great flexibility and potential in mapping non-linear relations within data. The numerous studies available in the literature either employ different extracted features from data to train NNs, or directly use measurement signals as input. Additionally, many studies available in the literature are based on a limited number of publicly available datasets, which mainly encompass cylindrical battery cells with small capacity. Starting from

Chianese, Giovanni, Capasso, Clemente, Veneri, Ottorino

Items per page:

1 – 50 of 1389