Magna International (United States)

Accurate real-time rotor temperature estimation is vital in Permanent Magnet Synchronous Machines (PMSMs) used in electric vehicle (EV) traction systems. Traditional modelbased techniques require extensive knowledge of motor parameters and struggle under dynamic conditions, while existing datadriven methods face generalization issues. This paper presents a lightweight, generalizable deep learning framework that combines Convolutional Neural Networks (CNNs) with multihead attention for real-time rotor temperature monitoring. Experimental results validate the model’s effectiveness under varying operational profiles, achieving superior accuracy compared to other attention mechanisms.

Conventional EV battery packs typically use high-energy (HE) chemistries like Nickel Manganese Cobalt (NMC) to maximize energy density. However, these cells have limited power capability, requiring many parallel strings to meet peak power demands. As a result, the battery pack is often oversized in total energy capacity. In contrast, high-power (HP) chemistries like Lithium Titanate Oxide (LTO) support high Crates but offer lower energy density [1]. By combining both chemistries, dual-chemistry systems decouple energy and power delivery, assigning low-power steady loads to HE cells and high-power transient demands to HP cells. While conventional dual-chemistry approaches [2] rely on bulky DC-DC converters, Cascaded H-Bridge (CHB) inverters allow seamless integration of diverse chemistries. This work introduces a Load Distribution Algorithm (LDA) that dynamically allocates drive cycle loads between NMC and LTO modules using the CHB architecture.

<div class="section abstract"><div class="htmlview paragraph">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 [<span class="xref">1</span>] provide a foundation for estimating lap-shear and cross-tension strength but require several geometric parameters that are often unavailable in the design phase.</div><div class="htmlview paragraph">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.</div><div class="htmlview paragraph">Furthermore, a Bayesian convolutional neural network (BCNN) model is developed to incorporate additional material features, offering improved prediction accuracy and uncertainty quantification.</div></div>

Accurately estimating rotor temperature is crucial for optimizing the performance, efficiency, and reliability of permanent magnet synchronous motors (PMSMs) used in electric vehicle application. Traditional methods for rotor temperature estimation often rely on physical models and sensor-based approaches, which can be complex, costly, and computationally expensive. Despite advancements in data-driven methods, precise rotor temperature estimation remains a challenge due to the need for large datasets and the inclusion of redundant features that can degrade model performance. This paper primarily addresses this challenge by analyzing state-of-the-art feature selection methods in combination with regression models for rotor temperature estimation. Furthermore, a novel reinforcement learning-based wrapper feature selection method is proposed to enhance the regression performance. The proposed method stands out by dynamically selecting the most relevant features from real datasets collected using an experimental PMSM setup and leads to improved estimation accuracy and computational efficiency compared with the state-of-the-art techniques. Moreover, a comprehensive analysis of selected features reveals that temperature-based features are more significant for rotor temperature estimation compared to current and voltage counterparts.

<div class="section abstract"><div class="htmlview paragraph">Lateral driving features used in Advanced Driver Assistance Systems (ADAS) rely heavily on inputs from the vehicle's surroundings and state information. A critical component of this state information is the curvature of the Ego Vehicle, which significantly influences performance. Curvature is often utilized in lateral trajectory generation and serves as a key element of the lateral motion controller. However, obtaining accurate curvature data is challenging due to the scarcity of sensors that directly measure this parameter. Instead, curvature is typically derived from various vehicle signals and additional sensor data, often employing sophisticated estimation techniques. This paper discusses several methods for estimating vehicle curvature using diverse information sources, evaluates their effectiveness, and investigates their impact on lateral feature performance, while analyzing the associated challenges and advantages.</div></div>

Mean flow and turbulence characteristics of a submerged jet impinging asymmetrically onto a rotating surface are investigated. The jet Reynolds number ranges from 1.3 × 103 to 7.9 × 103, and the rotational Reynolds number ranges from 10 × 103 to 175 × 103. The nozzle to impingement surface distance is fixed at 5d, and the nozzle axis to axis of rotation spacing is R/3 and 2R/3. Jet and rotation dominated flow regimes are identified. Spatial expansion of the jet dominated regime was found to occur when the jet Reynolds number was high, or the rotational Reynolds number was low. This was evident by a decrease in the radius of axisymmetric jet expansion from 3.4d to 2d when increasing the rotational Reynolds number from 10 × 103 to 175 × 103, keeping the jet Reynolds number constant at 1.3 × 103. The jet dominated regime was also shown to grow in axial planes sufficiently far from the rotating impingement surface. This was demonstrated by an increase in the radius of axisymmetric jet expansion from 0.85d to 3.4d in the axial planes of z/d = 0.012 and z/d = 0.125, respectively, with a jet Reynolds number of 1.3 × 103 and a rotational Reynolds number of 10 × 103. Low jet Reynolds numbers combined with high rotational Reynolds numbers were shown to aid in the formation of a mixing zone and jet deflection. When the nozzle axis to axis of rotation distance increases, the jet experiences greater deflection and the jet dominated region reduces in size. Increasing the impingement radius also enhances mixing and turbulence within the mixing zone.

The increasing industrial and research interest in Permanent Magnet Synchronous Motors (PMSMs) highlights the need for accurate performance evaluation methods. Analyzing PMSM performance using two-dimensional (2D) and three-dimensional (3D) magnetic field analysis with the finite element method (FEM) is both complex and time-consuming. The accuracy of 2D analysis results has always been a topic of debate, raising questions about the preferred type of analysis for specific motor applications. This study presents a comparative analysis of 2D and 3D models of PMSMs under demagnetization conditions. Both partial and uniform demagnetization scenarios are examined to assess their impact on key magnetic quantities. The research focuses on evaluating the electromagnetic torque, flux linkage, and back electromotive force (back-EMF) in both 2D and 3D models. Results indicate significant differences in the magnetic performance between the two modeling approaches under demagnetization, providing insights into the accuracy and reliability of each method. This study aims to enhance the understanding of PMSM behavior in degraded magnetic states with both 2D and 3D models.

The implementation of traditional current state observers for single current sensor (SCS) based permanent magnet synchronous machine (PMSM) control use continuous-time domain analysis and Euler or Tustin approximation for discretization. However, stability problem occurs at low sampling-to-fundamental frequency ratio condition with Euler approximation method and heavy computation burden cannot be ignored with Tustin method. To overcome these limitations, a discrete-time adaptive observer is proposed for SCS control in PMSM drives. First, commonly adopted Luenberger observers designed with Euler and Tustin methods are reviewed and analyzed. Then, a novel hybrid discretization (HY) method is proposed to design a discrete-time adaptive Luenberger observer with improved discretization accuracy while maintaining computational efficiency. In the proposed HY method, the nonlinear part of the PMSM model is discretized using the accurate Runge–Kutta discretization method, while the linear part is discretized using the computationally-efficient Euler approximation method. This HY method achieves a balance between simplicity and accuracy, resulting in a highly effective discretization of the observer. Moreover, the speed-adaptive gain is designed to guarantee stability and dynamic performance over a wide speed range. Experimental results have been performed on a laboratory interior PMSM drive to confirm the effectiveness of the proposed method.

Real-time monitoring and control of individual solar panels in a photovoltaic (PV) array can improve efficiency and reduce the probability of hazardous situations. This paper proposes the design and implementation of a real-time Intelligent Monitoring and Control Device (IMCD) to measure PV parameters such as temperature, voltage, current, and irradiance from individual solar panels. The proposed IMCD uses an embedded machine-learning (ML) algorithm for detecting and classifying four key conditions of the solar panel: partial shading, soiling, extreme soiling, and the standard test condition (STC). IMCDs connect with solar panels and with a control center using an integrated data transceiver. The hardware consists of an Arduino UNO transmitter that is used for data collection, an Arduino Nano BLE 33 Sense receiver for PV measurement, and an integrated processor for real-time fault detection using embedded ML. The training of the embedded ML models and their real-world testing and validation are presented in this paper. Comparisons with cloud-based ML algorithms and the use of bagging ensemble techniques to increase fault detection accuracy are also discussed in this paper.

Electric vehicles (EVs) are propelled by electric traction drive systems (ETDSs), which consist of various components including an electric motor, power electronic converter, and gear box. During manufacturing, end-of-line testing is the ultimate step for ensuring the quality and performance of electric motors in electric vehicle (EV) traction drive systems (ETDSs). The outcome of end-of-line testing of electric motors is significantly influenced by the tolerances of their structural parameters, such as stator inner and outer diameters, magnet dimensions, air gaps, and other geometric parameters. The existing literature provides insights into parametric sensitivity, offering guidance for enhancing the reliability of end-of-line testing. In this manuscript, the importance of end-of-line testing and the role of manufacturing tolerances of e-motor structural parameters in the manufacturing process of ETDSs are discussed. The impact of tolerances of e-motor structural parameters on the test results, such as torque, efficiency, back EMF, and e-NV (noise and vibration), is investigated. Finally, key challenges and research gaps in this area are identified, and recommendations for future research to mitigate the drawbacks of end-of-line testing are provided.