From Prediction to Reality: Assessing the Predictive Value of Transformer Asset Health Index Models

This paper compares several AHI methodologies - ranging from weighted scoring systems to hybrid and failure-mode or data-driven approaches - using a fleet database that includes special power transformers (e.g., large three-phase grid power transformers and single-phase autotransformers) with verified failure events.

A central finding is that AHI model selection and parameter weighting must be adapted to transformer type, size, and functional designation. Single-phase autotransformers can exhibit distinct thermal profiles, loading patterns, and dielectric degradation mechanisms compared with large three-phase grid power transformers. Tailoring inputs such as oil quality, DGA patterns, and insulation moisture improves sensitivity to early failure precursors. To strengthen the time dimension, a failure-proximity index is introduced and cross-validated against postmortem findings, historical failure records, and operational stress factors, including thermal cycling, overloading, leakage, and moisture ingress. The work also examines how data quality, diagnostic test frequency, and instrument calibration influence the reliability of health scores.

Results indicate that while some models provide actionable early warning, others systematically underestimate degradation risk - especially when driven by outdated or low-resolution inputs.

Finally, we advocate evolving AHI frameworks toward integrated resilience metrics that combine technical health data with consequence-of-failure indicators, improving risk communication and supporting risk-based planning under uncertainty.