麦格纳国际，又译作曼格纳、玛格纳、迈纳等，总部位于加拿大安大略省，为全球最大的汽车零部件制造商之一，也堪称全球最多元化的汽车零部件供应商。 世界500 强企业。2011年全球销售额287亿美元。作为一家世界五百强企业，公司在全球共设有294 家工厂、87个工程、研发和销售中心，拥有超过11万名员工，服务于北美，南美，墨西哥，欧洲，南非和亚太区等26个国家和地区。麦格纳的产品能力包括内饰系统、座椅系统、闭锁系统、金属车身与底盘系统、镜像系统、外饰系统、车顶系统、电子系统、动力总成系统的设计、工程开发、测试与制造以及整车设计与组装。麦格纳在中国已设有18个工厂、6个工程研发中心，拥有近八千名员工。而旗下所有的产品事业部如麦格纳座椅、麦格纳内外饰、麦格纳车镜（前称唐纳利）、麦格纳闭锁、麦格纳斯太尔，卡斯马国际，麦格纳动力总成，麦格纳电子等均已在中国设有制造工厂及工程中心，公司在华总部设在上海。
The reliability and efficiency of permanent magnet synchronous motors (PMSMs) depend on accurate rotor temperature estimation, for which advanced data-driven models have been developed recently. These models, which are essentially constructed around the concepts of machine learning dilemmas, suffer from a major shortcoming. That is, even though they demonstrate encouraging results for data samples picked from distributions within a source domain, they are unable to generalize to a target domain with a different distribution. Consequently, it is of paramount importance for an estimation model to perform reliably under unseen operating conditions as well. This work is devoted to designing a novel, lightweight yet efficient deep learning framework for real-time rotor temperature estimation. To enable the model to learn domain-invariant features, we leverage gradient reversal domain adaptation into the training of baseline deep learning models. This is to make the models generalize well to diverse operating conditions and to ensure robust performance under unseen scenarios. The trained models are deployed on a microcontroller in an experimental PMSM setup to demonstrate the feasibility of the framework in a resource-constrained embedded motor control system. The effectiveness of the proposed method is further validated for an open-source real dataset. The attained results demonstrate that the trained models can achieve mean absolute temperature estimation errors as low as $0.69^{\circ }\mathrm{C}$ for unseen data.
Automated classroom engagement recognition holds substantial promise for scalable learning analytics, yet the suitability of modern Vision-Language Models (VLMs) for this task under zero-shot conditions remains largely unexplored. We present a systematic benchmark that evaluates five widely-used VLMs: CLIP, BLIP-VQA, GPT-4o, LLaVA-1.5-7B, and Qwen2.5VL-7B-Instruct across two complementary educational datasets: DAiSEE, an individual-student video dataset (300 sampled test clips), and the Student Classroom Behaviour dataset (SCB, 1,168 scene-level images). Each model is probed with three prompt variants spanning minimal, rubric-anchored, and chain-of-thought designs. Our experiments reveal three primary failure modes of zero-shot VLMs for engagement recognition: (1) near-random performance on individual students, with Cohen's kappa never exceeding 0.10 on DAiSEE; (2) severe class collapse, where models assign 85-100
Abstract The automotive industry is under sustained pressure from several converging forces. Electrification is clearly underway, yet its adoption curve remains uneven, making the timing of large and irreversible investments difficult for manufacturers to judge. At the same time, the emergence of software‐defined vehicles (SDVs) is exposing structural weaknesses in long‐established automotive development models. Chinese manufacturers and U.S. companies such as Tesla and Rivian are advancing vehicle software at a pace that many traditional organizations have struggled to match. In parallel, government‐led mobility initiatives—particularly in urban areas—are challenging assumptions about private vehicle ownership and the role of public transportation. Within the INCOSE Automotive Working Group, these pressures have prompted renewed examination of how systems engineering is practiced in the automotive domain. While systems engineering is sometimes criticized as slow or overly bureaucratic, our working assumption is that the core problem lies not in the discipline itself, but in the way it is commonly applied. This paper argues that systems engineering must become faster, more adaptive, and more closely aligned with modern software development practices if it is to remain effective. Urban mobility provides a useful lens for examining these issues. We review autonomous minibus projects outside Japan, survey relevant industry publications, and report on a concrete trial conducted by JCOSE involving the deployment of autonomous technology for local bus service in a small Japanese city. The work spans multiple levels, including tool experimentation, process evaluation, and the capture of domain‐specific lessons learned. The paper concludes by consolidating global observations, identifying where systems engineering can add practical value, and outlining directions for further method and toolchain development.
Lithium-ion battery performance and longevity are significantly influenced by temperature, charge-discharge protocols and depth of discharge. Due to the complexity of inherent electrochemical processes of batteries, estimating state of health (SOH) remains open challenges in safety-critical applications like electric vehicles (EVs). Traditional methods of SOH estimation often rely on extensive historical data. While informative, this data may not be available when an immediate assessment is needed. This paper introduces a fast and data-efficient approach for in-situ battery health estimation using only initial and latest-cycle electrochemical impedance spectroscopy (EIS) data. The proposed method can utilize EIS measurements regardless of the resting period, which enables rapid, on-demand integration into battery management systems (BMS). By analyzing the relationship between EIS Nyquist plot and equivalent circuit model parameters, key health indicators are identified across three different Li-ion chemistries. These indicators are integrated into a polynomial ridge regression model to estimate SOH without continuous historical EIS and capacity data. In addition to SOH estimation, the method supports remaining useful life (RUL) prediction based on the proposed health indicators. The proposed PRR-based methodology achieves up to 13.25% higher R² than benchmark studies, maintains consistent performance under seen and unseen conditions, and achieves at least 30% lower RMSE compared to studies that incorporated multiple cell chemistries for SOH estimation. Finally, interpretable machine learning techniques are applied to assess the contribution of selected features to model performance.
Multilevel inverter (MLI) topologies are gaining significant interest over conventional two-level voltage source inverter (2L-VSI) topologies for electric vehicles (EVs) due to their improved output voltage quality, lower device stress, and potential reduction in filtering requirements. Understanding the motordrive performance over representative drive-cycles under such inverter excitations is essential for the design of an electric powertrain that is competitive in cost, weight, and efficiency when compared with the state-of-the-art. In this paper, the performance characteristics of an interior permanent magnet synchronous motor (IPMSM) fed by a two-level ($\mathbf{2 L}$) inverter, a three-level ($\mathbf{3 L}$) inverter, and a five-level (5 L) inverter are compared. A cosimulation framework that integrates MATLAB/Simulink for inverter control and voltage waveform generation with ANSYS Maxwell 2D transient finite element analysis (FEA) to obtain a voltage-fed motor model that captures space harmonics, ripple currents, and associated losses. Performance indices including copper loss, core loss, magnet loss, and overall weighted efficiency are evaluated at six representative drive-cycle clustered points. The impact of the three inverter topologies on total harmonic distortion (THD) of induced currents, as well as steady-state speed response, torque ripple, and average torque of the IPMSM, is compared and discussed to highlight the machine design trade-offs and improvements for EVs.
