Artificial Intelligence and Classical Methods in DGA Interpretation - Hybrid Approaches for Practical Transformer Condition Assessment

CIGRE guidelines, provide transparent and standardised diagnostics but show limitations when applied to early-stage degradation, mixed fault mechanisms, or at influencing environmental conditions. With the increasing availability of historical DGA data and continuous online monitoring, artificial intelligence (AI)–based methods offer new opportunities to enhance diagnostic sensitivity, scalability and robustness.

This paper summarizes how AI-based analysis can complement classical DGA interpretation and demonstrates its practical applicability using a real-world case study of an incipient transformer fault. Classical rule-based methods are compared with machine-learning-based diagnostics, including supervised classification, unsupervised anomaly detection, and ensemble learning techniques. The case study shows that classical DGA interpretation identified the fault at a relatively advanced stage, once established threshold values were exceeded. In contrast,

AI-based models detected abnormal behaviour earlier by recognising subtle multivariate changes in gas patterns. These results demonstrate that classical and AI-based diagnostics are complementary. While classical methods remain essential for standardisation and transparency,

AI-based approaches enhance early anomaly detection, scalability and preliminary evaluation for monitoring systems.

In the long term, it is anticipated that hybrid diagnostic frameworks integrating engineering expertise, experience and conventional analysis techniques with data-driven intelligence will identify hidden correlations and facilitate risk-aware decision-making and predictive maintenance for ageing power transformer fleets.