Condition Monitoring of Distribution Transformers Using Frequency Response Traces and Artificial Neural Network to Detect the Extent of Windings Axial Displacements

Abstract

Distribution transformers are of great importance in power systems regarding electrical power supply to the consumers. To have a reliable and continuous service, monitoring of transformers is a crucial issue. Maloperation or improper transportation can result in mechanical tension and stress on transformer windings. Axial displacement (AD) is one of the mechanical defects that can influence transformer operation through windings insulation degradation and short circuit faults. Frequency response analysis (FRA) is an efficient diagnostic technique widely used in transformers monitoring; however, interpretation of FRA results is complicated and is still under investigation. In this paper, AD faults are implemented on the 20 kV winding of a 1600 kV A distribution transformer. FRA traces are practically measured, and then the most sensitive and appropriate statistical indices such as cross-correlation factor (CCF), Lin's concordance coefficient (LCC), sum of errors (SE), and fitting percentage (FP) are employed to extract feature sets. All four components of the frequency responses, i.e., magnitude, phase, real and imaginary parts, are considered. Furthermore, an Artificial Neural Network using the obtained feature vectors is designed to detect the extent of the AD faults. The K-fold cross-validation method is used to evaluate the performance of the intelligent classifier. Both of the most suitable statistical indexes and frequency response components to detect the AD faults are determined.