Intelligent control and diagnostics of electric drives

Abstract

This dissertation presents the development, validation, and prospective generalization of techniques for online detection of inter-turn short-circuit (ITSC) faults in Permanent Magnet Synchronous Machines (PMSMs). A key contribution is a robust leakage-inductance-based diagnostic framework, experimentally validated on a digital signal processor (DSP) and field-programmable gate array (FPGA) platform, enabling real-time fault detection without disrupting machine control. The method is extended with lightweight machine-learning approximators designed to replace traditional LUT-based mappings between extracted features and robust fault indicators, providing scalable and adaptive compensation for parameter dependencies while maintaining real-time feasibility on embedded hardware.<br>In parallel, an exploratory Koopman-based fault detection framework is introduced and investigated via simulation. Exploiting the autonomous system structure provided by finite-control-set model predictive control (FCS--MPC), the method models each switching state as an independent subsystem and applies Dynamic Mode Decomposition (DMD) to construct finite-dimensional Koopman operators from drive signals. A &#x394;-score optimization problem is formulated to rank observables and spectral features according to fault-severity sensitivity, robustness to noise, and dependency on speed and load. Offline analysis identifies Koopman features exhibiting strong monotonicity with fault severity while remaining minimally affected by operating conditions, whereas a simulated online implementation evaluates their real-time feasibility and outlines prospective mitigation strategies for residual dependencies using lookup tables or data-driven approximators.<br>The dissertation thus introduces a set of diagnostic approaches tailored to traction-drive applications. The leakage-inductance-based method, experimentally validated on embedded hardware, demonstrates strong suitability for light-rail systems such as trams and urban trains operating at low to medium speeds. The Koopman-based strategy, though presently validated only in simulation, offers a forward-looking pathway for high-power traction systems such as locomotives, where non-intrusive, data-driven, and computationally efficient fault detection is essential. Together, these contributions provide diagnostic solutions across operating regimes and lay the foundation for next-generation health monitoring in electrified transportation.