A Novel Method to Identification Type, Location, and Extent of Transformer Winding Faults Based on FRA and SMOTE-SVM

Abstract

Abstract—Currently, the adoption of artificial intelligence is an inevitable necessity in diagnosing electrical systems failures, also, frequency response analysis FRA is widely used as a tool for diagnostic mechanical and electrical faults in power transformers, in this paper, a new methodology was proposed based on machine learning and advanced diagnostic technique FRA for detecting the type, location and severity of transformer fault. The method was applied and test it on three common actual defects (viz. Axial displacement (AD), Radial displacement (RD), and Short Circuit (SC)) that have been applied practically to an actual transformer winding at various locations and with different levels. Initially, the proposed approach is based on the collection and interpretation of the frequency response analysis (FRA) data of the winding faults to construct a database that can confine all possible cases of the fault and this has been tried with AD fault. After that, before classification, the problem of imbalance in the databases is addressed. Therefore, a support vector machine (SVM) classifier combined with SMOTE data preprocessing algorithm has been suggested to solve the imbalance problem.@The performance of the suggested methodology was evaluated using experimental data collected from the winding transformer model. Where the results prove the ability of the proposed method to detect the type, locate and severity of the fault with high accuracy than other methods, which may effectively contribute to the development of machine learning reliance for the diagnosis of the power transformer faults.