PCA-Assisted Location of Small Short Circuit in Transformer Winding

Abstract

The Frequency Response Analysis (FRA) is an efficient tool to diagnosis mechanical defects in transformer windings. After measurement of frequency responses, they must be compared to find any defects in the winding. The lack of clarity in the interpretation of measured frequency responses is the main problem of this method. This paper tries to solve this problem via a dimensional reduction of frequency response traces by Principal Component Analysis (PCA) to diagnose any short circuits in the winding. The precise location of the fault in the transformer winding is considered to be the main index, the effectiveness of which is presented by the results. Based on the approved transformer model, a dimensionality reduction method and feature selection are implemented to transformer's frequency response traces to map it from its main high-dimensional vector space to a lower dimensional one. It was shown that frequency response traces of Turn-to-Turn faults were separated from each other and the exact location of each fault was determined.