An Artificial Intelligence-Based Novel Strategy for Fault Prognostics in Transformers

One of the essential components of managing and maintaining power systems is transformer fault diagnostic. This study examines the use of deep learning algorithms, advanced data analytics methods, and other recent advancements in this area. Work has been carried out on fault prognostics in distribution transformers in the power system of the Sultanate of Oman. Based on Duval’s Pentagon method, artificial intelligence tools are developed to distinguish the fault types. The fault types identified are partial discharge, thermal fault of temperature T1 < 300°C, thermal fault of temperature T2 < 700°C, thermal fault of temperature T3 > 700°C, low energy discharges - sparking (D1), high energy discharges - arcing (D2), and stray gassing (S). The same has been implemented using the MATLAB Artificial Intelligence Toolbox. Around 150 transformer data in compliance with local utility have been utilized for analysis, from which around 80% have been taken for training and the remaining 20% for testing and validation. One of the significant features of this analysis is that it also highlights the feeble insipient faults. The results obtained from Artificial Intelligence are quite promising, and they could offer insightful information on the significance of transformer fault diagnostics and the part artificial intelligence plays in guaranteeing the dependable operation of the power grid.