A New Approach for AI-Based DGA for Transformers and Tap-Changers

Dissolved Gas Analysis (DGA) is a widely-accepted diagnostic technique for liquid-filled transformers and compartment type on-load tap-changers (OLTC). The thermal degradation of organic liquids can result in varying gas patterns and quantities when exposed to specific thermal stress levels. Additionally, factors such as errors in sampling, the method of gas extraction, the calibration of the gas chromatograph, and the accuracy of Ostwald coefficients can further increase the uncertainty of DGA values for each gas. Furthermore, application data such as asset age, load, and oil type can also affect the values without being related to failure. As a result, defined limit values should be considered as an indication of regular or irregular gassing behavior, rather than a clear separation between faulty and non-faulty equipment.Previous DGA interpretation methods do not fully account for these factors. They either provide hard limit values or provide a failure classification without taking into account uncertainties and application data.A new DGA interpretation method that addresses these issues is proposed. Application data and equipment-specific parameters are used to correct DGA values for non-failure-related influences. A machine-learning-based approach is used to predict the probabilities of different failure states from the corrected DGA values. The use of probabilities reflects the inherent uncertainties in the overall DGA process. The method also explicitly handles missing data and considers the resulting increase in uncertainty. Both the failure state probabilities and uncertainty information are then used to automatically generate a recommended course of action.By using a real case example it is demonstrated how the method responds to both clear-cut and ambiguous situations.