A Domain Adaption Based Closed-loop System for Fault Diagnosis of Generator Slot Wedge

One of the main challenges for a generator's fault di-agnosis is that a designed method for a generator may have poor performance over another generator with a different working condition. In this paper, we propose a closed-loop system based on domain adaption to inspect the tightness of the generator's slot wedge. To detect the fault pattern of a given sample, a feature extraction algorithm using frequency spectrum analysis is introduced. Also, a basic spectrum is designed to eliminate potential disturbances in the frequency domain. Furthermore, energy efficiency is considered by adjusting the compression rate in Compressive Sensing (CS) according to the feedback value of Maximum Mean Discrepancy (MMD). We conduct experiments on data collected from a test platform and a generator in the power plant. The result shows that the designed system can achieve high classification accuracy over 85% with a compression rate lower than 50%.