Explainable Health Index Methods for Power Transformers
How transformer health indices can become useful decision aids when they expose evidence, uncertainty, weighting logic, and risk-based maintenance context.
A transformer health index is useful only when teams can understand it. A number that says 72 out of 100 may help sort a fleet, but it does not explain why the asset moved, what evidence is stale, what uncertainty remains, or what action should follow.
GridAPM treats health indexing as a decision aid, not a magic score.
What a health index should do
CIGRE publications such as TB 761, TB 630, and TB 858 provide useful grounding for condition assessment, intelligent monitoring, and asset-health methods.
In software, a health index should:
- Rank assets for review.
- Show which evidence moved the score.
- Separate condition from consequence.
- Preserve missing or stale data flags.
- Expose weighting logic.
- Support maintenance timing and replacement planning.
It should not pretend that one universal score can replace engineering judgment.
Inputs that matter
An explainable transformer health index can include:
- DGA levels and generation rates.
- Oil quality, moisture, and paper-aging indicators.
- PRPD and partial discharge evidence.
- SFRA and mechanical integrity records.
- Thermal loading and cooling condition.
- Bushing and tap-changer evidence.
- Field inspections and maintenance history.
- Asset criticality and redundancy.
References such as IEEE C57.104, IEC 60599, IEEE C57.152, IEEE C57.143, and IEC 60422 help anchor these inputs in recognized diagnostic and maintenance practice.
Evidence before weighting
Before weighting anything, the evidence must be trustworthy. Has the DGA result been normalized? Is the PRPD measurement calibrated? Is the SFRA trace comparable to a baseline? Is the thermal estimate tied to real cooling state? Is the inspection record current?
This is where an agentic workflow helps. It can check data completeness and provenance before a score is used in a maintenance meeting. That connects directly to the diagnostic evidence model.
Make uncertainty visible
Health indices often fail when they hide uncertainty. A transformer may have incomplete recent oil data, no SFRA baseline, or conflicting diagnostic signals. A good health view should show confidence, missing records, and contradictions instead of smoothing them away.
That is also where human review matters. Engineers can decide whether a score is adequate for planning or whether more evidence is required.
From health to risk
Condition is not the same as risk. Risk also depends on consequence, redundancy, spares, outage windows, safety, environmental exposure, and load served. A moderate condition concern can become high priority if the transformer is critical. A higher condition concern may be managed differently if redundancy is available.
GridAPM’s risk queue should therefore show both condition drivers and consequence context. The companion article on risk-based maintenance recommendations explains that workflow in more detail.
The practical standard
A useful transformer health index should be explainable enough that a reliability engineer can approve, override, or escalate it. The best interface is not a mysterious score. It is a score plus the evidence chain, uncertainty, recommended action, and audit trail.
