Comparison of ICA and PCA Methods for Transformer Internal Fault Studies

Abstract

Traditionally discrete Fourier transform, wavelet techniques etc. are used for numerical transformer protection. In this study two different techniques are suggested for this purpose. The techniques so called dynamical principal component analysis (PCA) and independent component analysis (ICA) are investigated and compared to detect transformer internal faults. A three phase custom-built transformer is used for real time studies to examine the performance of the suggested techniques. Real time studies cover turn to turn faults and magnetizing inrush conditions. The suggested techniques are tested in MATLAB (TM) environment.