Detection of Transformer Faults: AI-Supported Machine Learning Application in Sweep Frequency Response Analysis

Hakan Çuhadaroğlu, Yılmaz Uyaroğlu

In this study, we discussed how the increasing demand for electrical energy results in higher loads on transformers, creating the need for more effective testing and maintenance methods. Accurate fault classification is essential for the reliable operation of transformers. In this context, Sweep Frequency Response Analysis (SFRA) has emerged as an effective method for detecting potential faults at an early stage by examining the frequency responses of transformers. In this study, we used artificial intelligence (AI) and machine learning (ML) techniques to analyze the data generated by SFRA tests. These tests typically produce large datasets, making manual analysis challenging and prone to human error. AI algorithms offer a solution to this issue by enabling fast and accurate data analysis. In this study, three different transformer conditions were analyzed: a healthy transformer, a transformer with core failure, and a transformer with winding slippage. Six different machine learning algorithms were applied to detect these conditions. Among them, the Gradient Boost Classifier showed the best performance in classifying faults. This algorithm accurately predicted the health status of transformers by learning from large datasets. One of the most important contributions of this study is the use of gradient boosting algorithms for the first time to analyze SFRA test results and facilitate preventive maintenance through the early detection of transformer failures. In conclusion, this study presents an innovative approach. The interpretation of offline SFRA results through various artificial intelligence-based analysis methods will contribute to achieving the ultimate goal of reliable online SFRA applications.