A new statistical approach to interpret power transformer frequency response analysis: Nonparametric statistical methods

Abstract

The Frequency Response Analysis (FRA) test has been recognized as one of the sensitive tools available for detecting electrical and mechanical faults inside power transformers. However, there is still no universally systematic interpretation technique for these tests. Many research efforts have employed different statistical criteria in order to aid the interpretative capability of the FRA, but it is showed that the methods used so far, are based on parametric statistics which need a set of assumptions about the normality, randomness and statistical independence of FRA data. Therefore, this paper aims to propose some nonparametric statistical methods which are based on explicitly weaker assumptions than such classical parametric methods. The proposed statistical methods are applied to the experimental FRA measurements obtained from two test objects: a three phase, two winding distribution transformer (35/0.4 kV, 100 kVA) to study the winding inter-turn fault as an electrical fault, and a two winding transformer (1.2 MVA, 10 kV) for the study of radial deformation as a mechanical fault. It was found through this research work that the used methods namely, Wilcoxon signed rank test and Friedman test which are proposed for the first time, can effectively reflect the differences between compared FRA data and diagnose the fault.