State-of-the-Art Fault Detection and Diagnosis in Power Transformers: A Review of Machine Learning and Hybrid Methods
Lebo Dina Mashifane, Bongumsa Mendu, Bessie Baakanyang Monchusi
Abstract
Fault detection and diagnosis (FDD) in power transformers is essential for maintaining reliability and safety in modern power systems. Recent trends in transformer fault diagnosis include the use of advanced monitoring systems and data analysis techniques, and implementing effective FDD strategies can prevent costly repairs, minimize downtime, and enhance the overall reliability of power systems. In this work, a systematic review of FDD in power transformers, focusing on machine learning and hybrid methods applications is presented. The methodology comprised a detailed systematic process undertaken to gather, filter, and analyze relevant research papers using Scopus database, while VOSviewer, and bibliometrix tools were utilized for results analysis. The research findings indicate that there is clear progress in detecting transformer faults, moving from older methods like Dissolved Gas Analysis (DGA) to machine learning models such as Random Forest and Convolutional Neural Networks (CNN). Hybrid models combining machine learning with optimization have made detection more accurate. New tools, including optical sensors, now allow for real-time monitoring. Still, issues like limited data and complex models remain. This study contributes by reviewing how machine learning is applied to transformer fault detection, exploring hybrid methods that combine traditional techniques like DGA with advanced models for better accuracy. It identifies key research patterns, trends, and themes, while also highlighting gaps and offering suggestions for future research to improve diagnostics and monitoring.