Predictive failure forecast using Exponential Time-Series modelling of Dissolved gases in Converter Transformers

While Dissolved Gas Analysis (DGA) is the most widely deployed diagnostic for incipient transformer faults, conventional interpretation remains predominantly snapshot-based

(thresholds, ratios and graphical codes) and does not directly quantify time-to-failure. This paper develops and demonstrates a field-validated, utility-grade prognostic workflow that converts DGA trends into a predictive failure horizon for HVDC converter transformers.

At a ±500 kV HVDC station, three sister converter transformers (T1–T3) failed during 2023– 2025. Retrospective analysis showed a consistent signature of exponential escalation in thermal fault gases — methane (CH₄) and ethylene (C₂H₄) — with hydrogen (H₂) rising later as the fault progressed toward dielectric involvement. A fourth identical unit (T4) exhibited the same early escalation signature in 2025. Based on failed-unit calibration, T4 was initially projected (preintervention) to approach near-failure gas regimes in mid-2026.

Critically, this paper treats load derating as a major asset-management intervention. In July2025, T4 loading was reduced to ~80% as a risk-mitigation measure. The effect of derating is quantified using a two-slope kinetic model (pre/post July-2025) and piecewise Holt’s exponential smoothing. The result is a measurable shift of the failure horizon from mid-2026

(non-derated trajectory) toward late-2026 / early-2027 under derated operation. This “timebuying” impact is further expressed probabilistically through Monte Carlo failure-time PDFs, demonstrating a clear rightward shift and narrowing of the failure probability distribution.

The paper concludes with implementable guidance: (i) slope-trigger based monitoring, (ii) escalation-phase classification, (iii) targeted diagnostics selection, and (iv) outage planning logic aligned to forecast uncertainty.