Investigation of relationship between various health monitoring parameters of transformer oil and development of a predictive model

Abstract

Transformer oil is vital for insulation and cooling in electrical transformers, with the breakdown voltage (BDV) serving as a key health indicator. This study examines how various contaminants and stresses affect BDV and introduces a predictive model and a risk assessment map for transformer oil. 20 different oil samples were further aged and treated in a controlled manner, producing 220 samples for analysis. BDV, leakage current, and moisture tests were performed on these samples. To enhance the dataset, the bootstrapping technique was used, increasing it fivefold while preserving statistical validity. An extreme gradient boosting (XGBoost) model was developed using leakage current and moisture as inputs. The model achieved accuracy of 96.81%, with an R2 score of 0.8655, and demonstrated strong classification performance with precision of 96.83%, recall (96.81%), F1 Score (96.67%). The experimental results showed that carbon had the most detrimental effect on BDV, followed by water and dust. In contrast, the filler improved the dielectric strength. Heating consistently reduces BDV for all types of contaminants. The leakage current proved to be the most critical predictor of BDV, allowing transformer health assessment without the need for moisture input. The model also estimates the probability of failure and suggests maintenance intervals, improving reliability and supporting data-driven maintenance in power systems.