Logistic regression as a classifier in the diagnostic of electrical failures in a permanent magnet synchronous generator

In this paper, the wavelet Haar transform, and a binary classifier based on logistic regression are utilized to identify short-circuit (s.c.) faults between stator windings in a permanent magnet synchronous generator (PMSG). The feature extractor processes the currents through the Haar transform, obtaining the effective values of its coefficients. These values, along with the objective function, form the data received by the binary classifier during the training and validation processes. The performance of the logistic regression as a binary classifier is analyzed under different stator fault conditions. The necessity for simpler and more robust classification algorithms in PMSG fault diagnosis allowing online implementation is evident. The test bench of a PMSG presents the currents that are analyzed by the binary classifier for the purpose of fault diagnosis. These currents are acquired and subsequently processed digitally. As a preliminary step in the development of the diagnostic algorithms applied to the PMSG, the performance of the logistic regression model was compared with that of a perceptron model through the Python.