Articles

Permanent Magnet Synchronous Motors (PMSMs) have become the dominant choice in high-performance electric vehicle (EV) propulsion systems due to their superior torque density, high efficiency, and excellent dynamic characteristics. Classical Field-Oriented Control (FOC) ensures decoupled torque and flux regulation but exhibits limited transient performance under fast dynamic load variations. Model Predictive Control (MPC), on the other hand, provides rapid torque response and constraint handling but often suffers from computational complexity and switching frequency variation. This paper presents an enhanced hybrid Field-Oriented Predictive Control (FO–MPC) scheme for PMSM drives that integrates the steady-state smoothness of FOC with the dynamic adaptability of MPC. The proposed controller employs dq-axis current regulation with predictive torque optimization, ensuring low torque ripple and high-speed transient response. A comparative simulation study in MATLAB/Simulink demonstrates that the hybrid controller achieves a 30–40% reduction in torque settling time compared to conventional FOC, with approximately 25% lower current total harmonic distortion (THD) than finite control set MPC (FCS-MPC). The findings suggest that FO–MPC provides a robust and efficient alternative for EV traction applications where fast dynamics and energy efficiency are critical.

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that deteriorates memory and cognitive functions across multiple stages. Timely and accurate classification of these stages—Normal Control (NC), Mild Cognitive Impairment (MCI), Moderate AD, and Severe AD—can facilitate better treatment planning. This research investigates the efficacy of the ResNet50 deep residual neural network in multi-stage classification of AD using Magnetic Resonance Imaging (MRI) data. A dataset curated from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) comprising 7,200 MRI scans was utilized. Preprocessing included intensity normalization, skull stripping, and augmentation to mitigate class imbalance. ResNet50, leveraging skip connections, was trained with categorical cross-entropy and Adam optimizer. Performance metrics including Accuracy (92.1%), Precision (90.3%), Recall (91.2%), and F1-score (90.7%) demonstrate ResNet50’s capability in capturing discriminative patterns across AD stages. The results suggest its potential as a reliable computer-aided diagnosis (CAD) tool in clinical settings.

Graduate admissions can be a challenging and sometimes unpredictable process, influenced by a mix of academic scores, test results, and other personal factors. In this study, we explore how machine learning (ML) can help make this process more transparent and data-driven. We test several ML models—Linear Regression, Logistic Regression, Random Forest, and XGBoost—to predict both the probability of admission and the admitted/rejected outcome. Our results show that ensemble models, particularly XGBoost, consistently provide the most accurate predictions. Beyond prediction, we also analyze which factors matter most in the admission decision, offering helpful insights for applicants and supporting committees in making fairer, evidence-based decisions. To improve clarity, we present visualizations of model performance and feature importance. Overall, this work highlights how machine learning can support graduate admissions, providing a clearer view of key factors that shape outcomes and helping applicants and decision-makers make more informed choices.