Abstract
Wound-rotor induction generators (WRIGs) are commonly used in wind energy applications. WRIGs have advantages of simple and robust construction, with high starting torque and low starting current. Despite the relative robustness of WRIGs, these machines still experience a variety of faults in practice which requires condition monitoring for effective operation thereof.
This dissertation deals with online condition monitoring of a three-phase WRIG. Currently, stator winding current signature analysis is by far the most popular method of detecting and diagnosing different types of faults on both the stator and rotor of wound rotor induction motors. Application of stator winding voltage signature and rotor winding current methods for on-line condition monitoring is very limited and more so for the specific case of WRIGs. This research focuses on intelligent fault detection and diagnosis of a WRIG by using voltage and current signature analysis, on both the stator and rotor windings, with a probabilistic-classification based intelligence system. This includes processing of the signals generated by WRIG, classifying the machine’s condition and provides an estimate of the certainty of that classification. The signal processing phase of the intelligent fault diagnosis process extracts features, which are frequency-based, interrelated to specific fault modes. Finite element modelling of a wound rotor induction generator is carried out under normal and different fault conditions for the purpose of conducting preliminary design and testing of the probabilistic intelligence system. An experimental setup is then used to validate the computational results and verify the intelligent diagnosis system.
Results indicate that the stator voltage and the stator current signatures are found to be more accurate modalities, when applied individually, for use with the classification system. The rotor current signatures are imprecise although they do provide some information about the generator status. When these modalities are combined for use with the classification system, an overall accuracy of 86% is achieved for simulation data and 99% for experimental data. It is found that the classifier works best with the combined feature data using all modalities – i.e. stator voltages and currents, and rotor currents.
M.Tech. (Electrical and Electronic Engineering)