Wind Turbine Mechanical Fault Diagnosis via Doubly-Fed Induction Generator Electrical Signals

Abstract

As wind turbines grow and move offshore, reducing operations and maintenance costs increasingly requires condition-monitoring methods that avoid additional sensors. This dissertation investigates sensorless mechanical fault diagnosis for horizontal-axis wind turbines equipped with doubly-fed induction generators (DFIGs) by extracting fault signatures from generator current measurements available for control and operation monitoring. It investigates the role of operating condition, effect of induction generator and the influence of controller parameters on the visibility (or identifiability) of mechanical faults. A coupled multiphysics wind-turbine model is developed by integrating aeroelastic models in OpenFAST (with stochastic wind profiles) with MATLAB/Simulink models of the drive-train, DFIG, back-to-back power converters, grid interface, and a stator-voltage-oriented control system. Component- and system-level validation, power-balance verification, and operating-point initialization against the turbine power curve confirm physically consistent steady-state behavior and realistic transient responses. The framework enables time- and frequency-domain analyses of fault-to-current pathways under controlled wind conditions. The central physical finding is that current-signature-based diagnosis is fundamentally shaped by cascaded filtering and feedback control loops. Turbulent inflow injects broadband aerodynamic torque fluctuations that redistribute spectral energy and reduce the contrast of fault tones; consequently, detectability depends on operating conditions rather than fault severity alone. At the plant level, the electromechanical path from torque disturbances to electrical currents behaves as a strong low-pass filter: high-frequency phenomena typical of localized gearbox and bearing defects are attenuated before reaching the electrical domain, whereas low-frequency disturbances that modulate generator speed propagate more efficiently. At the control level, the cascaded DFIG regulators actively reshape disturbance-to-current transfer paths, concentrating fault observability primarily in the d-axis currents and making reactive power control a key lever that can either expose or suppress fault content. These results yield practical observability guidelines and a visibility matrix relating fault classes, turbulence intensity, operating point, and control tuning. The matrix defines realistic limits, identifies diagnostically informative current channels and operating regions that maximize signal-to-noise ratio, and provides actionable design rules for sensorless monitoring of DFIG wind turbines.