Abstract

Permanent Magnet Synchronous Machines (PMSMs) have become essential in a wide range of high-performance applications, including electric vehicles, industrial automation, and renewable energy systems. However, their reliance on permanent magnets (PMs) makes them vulnerable to demagnetization—a fault that can significantly degrade machine performance, reduce efficiency, and lead to system malfunction. Early detection and accurate assessment of Permanent Magnet Demagnetization (PMD) are therefore critical for ensuring reliability and enabling predictive maintenance strategies.

This thesis presents a comprehensive investigation into PMD detection, introducing novel methodologies for both offline and online fault diagnosis. The offline method is based on voltage excitation under standstill conditions, where carefully designed d-axis voltages are supplied through the same motor drive inverter to generate fault-sensitive current responses. A new fault indicator is proposed, offering a simple, model-independent measure of PMD severity. This technique is validated through finite element analysis (FEA) and experimental tests, demonstrating high sensitivity to PMD. For online diagnosis, three advanced observer-based techniques are developed to estimate PM flux linkage during regular motor operation. These include: (1) a Model Reference Adaptive System (MRAS) for simultaneous estimation of PM flux and stator resistance, (2) a d-axis flux observer with harmonic analysis to distinguish between uniform and asymmetric PMD, and (3) a disturbance observer enhanced with speed offset for robustness against inverter nonlinearity and parameter mismatch. All methods are non-invasive, and they can provide efficient and reliable estimation of PMD severity without requiring additional sensors, rotor access, or disassembly of any motor parts.

In addition to algorithmic development, the thesis offers a detailed theoretical analysis of PMD mechanisms, signal signatures under fault conditions, and the limitations of existing diagnostic techniques. A comparative evaluation of model-based, signal-based, and data-driven approaches is also presented, identifying key trade-offs in terms of accuracy, implementation complexity, and generalizability.

Overall, the contributions of this thesis provide a practical and scalable foundation for intelligent PMSM health monitoring. The proposed techniques advance the state-of-the-art in PMD detection and pave the way for future integration into embedded control systems, enabling early fault detection and predictive maintenance in safety-critical electric drive applications.