Online detection and discrimination of eccentricity and electrical load unbalance faults in wind turbine generators

In wind energy-based permanent magnet synchronous generators (PMSGs), ensuring operational reliability and power production continuity is a critical requirement for system efficiency and operating costs. Detection of eccentricity and electrical load unbalance (UnB) faults, which are frequently encountered in these generators, is generally based on the analysis of characteristic components in the stator current or magnetic flux spectrum. However, since these different fault types excite similar frequency components, methods such as motor current signature analysis (MCSA) and single-point stray flux analysis (SFA) remain insufficient in discerning these faults. In this thesis, two different non-invasive fault detection methods based on external stray flux have been developed to overcome the aforementioned constraints and to discriminate faults. Within the scope of the first developed method, space vector stray flux (SVSF) analysis has been performed with data obtained from a three stray flux sensors placed on the generator frame with a 120° electrical displacement to distinguish between static eccentricity (SE) and UnB faults. The analyses revealed that while the magnitude of the +5fs harmonic component exhibits a characteristic increase in the case of static eccentricity, it remains unaffected and stays at healthy state reference levels in the case of electrical load unbalance. Thus, the +5fs component is defined as a distinctive fault indicator that discriminates these asymmetries from each other. In the second method, a hybrid approach combining differential phase shift and sideband amplitude analysis is presented to discriminate between static and dynamic eccentricity (DE) faults. In this method, the differential phase shift between sensors provides a sensitive indicator of the fault presence, while the characteristic changes in sideband amplitudes enable the discrimination of the fault type. The validity of the proposed methods has been verified via finite element analysis (FEA) and an experimental test rig; the algorithms have been observed to operate with robustness under practical constraints such as sensor placement errors and electromagnetic interference. Consequently, this thesis offers a reliable condition monitoring approach for PMSGs by overcoming the discrimination challenges stemming from spectral overlap.