Data Driven Model Agnostic Methodology for Transformers Top Oil Anomaly Detection

Predictive maintenance and advanced asset management strategies have emerged to accurately diagnose transformer failures, helping to mitigate safety risks, reduce service interruptions, and enhance operational efficiency. Among various indicators, oil temperature is a key parameter for assessing a transformer's operational health. This work presents a methodology for detecting anomalies in transformer top-oil temperature, aiming to identify potential critical conditions and assist system operators in making informed maintenance decisions. The proposed approach leverages historical data typically collected through the transformer’s online monitoring system. A machine learning (ML) model is developed to predict the expected top-oil temperature under normal operating conditions. To detect deviations from this baseline and identify them as anomalies, Statistical Process Control (SPC) techniques are applied. The effectiveness of the methodology is validated using additional transformer measurements, which serve as reliable indicators of anomalous behavior, as well as an existing published algorithm.