Fuzzy-based On-line Detection and Prediction of Switch Faults in the Brushless Excitation System of Synchronous Generators

This article presents a methodology for on-line detection and prediction of electronic switch faults in the brushless excitation system of synchronous generators under balanced loading conditions. The time-domain output voltage of both the diode and thyristor bridges of the excitation system are selected as a characteristic waveform. The frequency spectrum of such a signal is obtained using the fast Fourier transform algorithm. The second and fifth harmonic components of the spectrum are used as the diagnostic indices. A Mamdani fuzzy model is designed to characterize the operating condition of each bridge; the model distinguishes between normal and faulty operations of the bridges in case of an open or shorted switch. An adaptive neuro-fuzzy inference system is developed to evaluate the fault resistance, while a short circuit develops across a switch. Both fuzzy models are designed and tested based on simulated data, where the results show acceptable effectiveness in detecting and predicting different faults. Experimental measurements verify the fault-related distortions of the characteristic waveform, and validate the research results.