par Postal, Geoffrey
Président du jury Robert, Frédéric
Promoteur Gyselinck, Johan ;De Belie, Frederik
Co-Promoteur Kinnaert, Michel
Publication Non publié, 2026-06-16
Thèse de doctorat
Résumé : Permanent magnet synchronous machines (PMSMs) are widely deployed in modern electrical drive systems due to their high efficiency, power density, and reliability. They play a key role in a broad range of applications, including electric vehicles, aerospace systems, renewable energy conversion, and high-performance industrial drives. As these applications increasingly involve safety-critical operations, the requirements in terms of availability, reliability, and fault tolerance have become more stringent. In such contexts, unexpected failures can lead not only to costly downtime but also to safety hazards. Consequently, the development of robust condition monitoring and fault diagnosis strategies has become a major research focus in the field of electrical machines and drives.Among the various types of faults that can affect PMSMs, interturn short-circuit faults (ISCFs) are considered especially critical due to their potentially rapid progression. These faults typically originate from insulation degradation between turns of a stator winding, which may be caused by thermal stress, mechanical vibrations, manufacturing defects, or aging. Once initiated, an interturn short circuit leads to localized circulating currents, which generate excessive heat and can quickly propagate to adjacent turns. If not detected at an early stage, ISCFs can result in severe damage to the stator winding and, ultimately, complete machine failure. Early detection and accurate assessment of such faults are therefore essential to ensure safe and reliable operation.Accurate machine models play a central role in addressing these challenges. They enable a detailed understanding of the physical phenomena associated with both healthy and faulty operating conditions, facilitate the development and validation of diagnostic techniques, and provide a virtual environment for testing control and monitoring strategies before their deployment on real systems. However, achieving both high fidelity and computational efficiency remains a significant challenge, particularly when complex effects such as magnetic saturation, spatial harmonics, and fault-induced asymmetries must be simultaneously considered.This PhD thesis addresses this challenge by developing advanced high-fidelity modelling approaches for PMSMs, with a particular emphasis on the unified representation of non-linear magnetic effects, spatial harmonics, and interturn short-circuit faults. The primary objective is to derive models that accurately capture the electromagnetic behaviour of the machine under both healthy and faulty conditions, while maintaining a low level of computational cost. These models are then used as a foundation for the systematic evaluation and improvement of fault detection and diagnosis techniques.First, a non-linear magnetic model of a healthy PMSM is proposed. This model is based on polynomial representations of the current–flux linkage relationships. Such a formulation allows for an explicit and flexible representation of magnetic saturation and cross-saturation effects, which are known to significantly influence machine behaviour, especially under high-load conditions. In addition, spatial harmonics resulting from the machine geometry and winding distribution are incorporated into the modelling framework. The use of a co-energy-based formulation enables the derivation of consistent analytical expressions for electromagnetic torque, ensuring physical coherence between the flux, current, and torque relationships. Compared to conventional linear models, the proposed approach provides a substantially improved representation of the machine characteristics over a wide operating range, while preserving a relatively low computational cost.Building upon this framework, an analytical model of interturn short-circuit faults is developed. A key feature of the proposed approach is the linearization of the fault-related contributions, while fully preserving the non-linear magnetic behaviour induced by the stator phase currents. This results in a modelling framework that captures the essential fault mechanisms without significantly increasing the model complexity. Importantly, the proposed formulation enables the construction of the faulty machine model primarily based on the characteristics of the healthy machine. Under some assumption, this eliminates the need for dedicated faulty datasets, detailed geometric descriptions, or explicit knowledge of material properties, which are often difficult to obtain in practice.Another important advantage of the proposed model is its flexibility with respect to parameter identification. The model can be parameterised either from finite-element method (FEM) simulations or directly from experimentally estimated flux linkages. In the latter case, the approach naturally accounts for inherent three-phase asymmetries and measurement imperfections, which are often neglected in idealised models but can have a significant impact on fault diagnosis performance. The validity and accuracy of both the healthy and faulty models are assessed through extensive comparisons with FEM simulations and experimental measurements. The experimental validation is conducted on a PMSM that has been specifically rewound to emulate interturn short-circuit faults in a controlled and repeatable manner. The results demonstrate a very good agreement between the proposed models and reference data, confirming their ability to accurately reproduce both steady-state and dynamic behaviours under various operating and fault conditions.The high-fidelity models developed in this work are then used to perform a systematic evaluation of signal-based fault indicators reported in the literature. These indicators typically rely on measurable quantities such as currents, voltages, and aim to detect the presence of faults through characteristic signatures in the signals. However, their performance is often influenced by operating conditions, machine parameters, and healthy machine 3-phase asymmetry, which can limit their robustness and general applicability. To address this issue, a dedicated simulation framework is developed, enabling a controlled and comprehensive analysis of the behaviour of these indicators under a wide range of conditions.Using this framework, the sensitivity of various fault indicators to interturn short-circuit faults is investigated, as well as their dependence on fault severity and the location of the faulty phase. The results reveal that many conventional indicators exhibit a strong dependence on the specific phase affected by the fault, which can complicate their interpretation and reduce their reliability in practical applications. Based on these observations, improved definitions of fault indicators are proposed. These new indicators are designed to be invariant with respect to the faulty phase and to provide a response that is directly related to the severity of the fault. The proposed formulations lead to more robust and interpretable diagnostic metrics, which are better suited for practical implementation in condition monitoring systems.Finally, a complete condition monitoring strategy is developed, integrating fault detection, faulty phase isolation, and fault severity estimation within a unified framework. The performance of this strategy is evaluated using experimental data. The results show that, while the level of model fidelity has a relatively limited impact on the ability to detect the presence of a fault and to identify the faulty phase, it plays a crucial role in the accurate estimation of fault severity. In particular, the use of high-fidelity models allows for a significantly improved estimation of the fault severity, which is essential for decision-making processes related to maintenance and fault-tolerant control.In summary, this thesis provides a comprehensive modelling and analysis framework for PMSMs under both healthy and faulty conditions, with a particular focus on interturn short-circuit faults. The proposed high-fidelity models strike a balance between accuracy and computational efficiency, and enable a deeper understanding of fault-related phenomena. Furthermore, they serve as a powerful tool for the development and validation of advanced fault diagnosis techniques, ultimately contributing to the enhancement of reliability and safety in electrical drive systems.

Documents en relation

DI-fusion