Application of Convolutional Neural Networks for Classification of Oil-Insulated OLTC Defects Using Acoustic Emission

On-Load Tap Changers (OLTCs) are critical components of power transformers, significantly influencing both transformer reliability and the condition of insulating oil. As OLTCs are responsible for a substantial proportion of transformer failures, their accurate diagnosis is essential. This research investigates the application of Artificial Intelligence (AI) tools, specifically Convolutional Neural Networks (CNNs), to enhance OLTC diagnostics using Acoustic Emission (AE) methods. The AE technique enables the collection of diagnostic data during normal transformer operation, allowing for online monitoring without de-energizing the unit. A comprehensive database of AE waveforms was developed, capturing OLTCs with diverse technical conditions and a wide range of modeled defects. The CNN model demonstrated high accuracy and precision in classifying OLTC defects, with a particular strength in identifying OLTCs in good condition. Since OLTC condition directly impacts the quality of insulating oil, early detection of OLTC issues through the proposed system can prevent oil degradation and ensure timely maintenance. By addressing the limitations of traditional diagnostics, such as Dissolved Gas Analysis (DGA), this research establishes the foundation for an advanced online diagnostic system, enhancing both transformer reliability and insulating oil quality.