Abstract
This study employs the PRISMA-A methodology to conduct a systematic review of transformer fault diagnostics using Dissolved Gas Analysis (DGA) data. A comprehensive analysis was performed across four major databases-IEEE, Scopus, ScienceDirect (Elsevier), and Web of Science-yielding 12,511 initial records. Following rigorous evaluation, including duplicate removal and eligibility criteria assessment, 1190 articles underwent statistical evaluation. The search strategy focused on keywords related to transformer faults and diagnostic methods, resulting in a refined dataset of 4810 DGA samples from wind park transformers. Detailed statistical analysis of gas concentrations-hydrogen, methane, carbon monoxide, carbon dioxide, ethylene, ethane, acetylene, oxygen, and nitrogen-revealed significant insights into fault indicators and distribution patterns. Furthermore, predictive modeling using various machine learning algorithms highlighted the efficacy of models such as Random Forest and CART, achieving accuracies up to 95.29 % in fault prediction tasks. Proposed revisions to IEEE gas concentration thresholds aim to enhance early fault detection capabilities, thereby improving maintenance planning and transformer reliability. The findings underscore the importance of advanced analytics and sustainable practices in transformer diagnostics, calling for continued research in predictive maintenance and eco-friendly insulation technologies to meet future energy challenges.