Enhancing Severity Level Detection of Transformer Discharge and Thermal Faults Using Artificial Intelligence in Duval Triangle 1

Interpreting incipient faults in mineral oil transformers using dissolved gas analysis (DGA) involves understanding the complex interrelationships in the evolution of various gases caused by different incipient faults. The Duval Triangle 1 method is commonly used to interpret DGA results due to its wide acceptance, however, it still does not address the severity level of discharge and thermal (DT) incipient fault. The paper focusses on understanding the relative concentration of different gases during DT and assign a severity level with fault-space refinement by combining trusted graphical and synthetic data generation (SDG) techniques. DTs occur sparsely and require an SDG approach due to lack of real-world data related to different severity levels of DT faults. Integration of SDG techniques and analysis of key gas concentrations, such as methane and ethylene, critical fault thresholds are identified, with methane concentrations exceeding 43% indicating severe faults like T3 and D2, while ethylene concentrations below 29% suggest low-to-medium thermal faults. 1113, 2166873, 13989, 396593, and 1899995 data instances were identified for fault categories D1, D2, T1, T2, and T3 faults respectively. Based upon rigorous analysis and training, the DT fault regions are segregated into different severity levels. This research offers a novel DT fault categorization which helps in condition monitoring of transformers.