What does dissolved gas analysis actually prove?

It identifies the type of fault energy present in the oil, not the location or the cause. Hydrogen and methane point toward low-energy partial discharge or thermal activity; ethylene toward higher-temperature thermal faults; acetylene toward high-energy arcing. What DGA cannot do on a single sample is establish whether a fault is active, developing, or residual from an event the unit already survived. Interpretation under IEEE C57.104 is fundamentally about trend and gassing rate, which is why the complete sample history matters far more than the last result.

Can a transformer fire be traced to a specific cause?

Frequently, because the failure sequence leaves distinguishable evidence. Bushing failures, tap changer faults, winding-to-winding and winding-to-ground faults, and external fault-induced failures each produce characteristic damage patterns and characteristic gas signatures. The physical examination is correlated with the electrical record — relay operation, fault current magnitude and duration, and whether the protection cleared within the transformer's C57.109 through-fault withstand. Where the physical evidence and the electrical record disagree, one of the two working hypotheses is wrong, and identifying which is the substance of the opinion.

Does cumulative through-fault damage explain a failure years later?

It can, and it is one of the harder arguments to run well. IEEE C57.109 defines the through-fault current-versus-duration a liquid-immersed transformer is expected to withstand, and repeated events inside but near that envelope progressively loosen and deform windings without producing an immediate failure. Establishing the argument requires a documented history of fault events with magnitude and duration for each, evidence of mechanical deformation on teardown, and usually a comparison of winding geometry against the factory record. Without the fault history it is speculation, which is why the relay records matter years before anyone expects litigation.

What is the difference between an oil-filled and a cast resin transformer failure investigation?

The available evidence is almost entirely different. An oil-filled unit carries its own history in the oil — DGA, moisture, furan analysis, and power factor all provide a record of what the insulation experienced over years. A cast resin unit provides none of that; there is no fluid to sample, so the investigation depends on the physical examination of the resin encapsulation, evidence of partial discharge tracking, thermal cycling cracks, and contamination, together with whatever the protection and monitoring recorded. Cast resin failures are also more likely to turn on environmental factors, since the encapsulation is what is exposed.

How is the loading history reconstructed when nobody was recording it?

From whatever proxies exist: utility interval billing data, SCADA or BMS trends, downstream metering, motor and equipment run-time logs, and process production records. Ambient temperature comes from the nearest weather station record. The reconstruction is approximate, and the report should say so — but the C57.91 thermal model is not especially sensitive to small errors in load, and it is usually possible to bound the answer well enough to determine whether the loading theory is supportable or not.

Transformer Failure Expert | Jay Prigmore, Ph.D., P.E.

Oil-filled and cast resin fail differently

They are usually treated as one category and should not be. An oil-filled transformer carries its own insulating and cooling medium, which means it also carries a diagnostic record — dissolved gases in the oil encode what has been happening inside the tank, sometimes for years before failure. It also carries a fire load.

A cast resin or dry-type unit has no oil, so no fire load and no gas record — but also no self-healing dielectric and far less thermal mass. Its failures tend to trace to partial discharge in voids within the casting, cracking from thermal cycling, surface contamination and tracking, or moisture ingress in units installed where they should not have been.

The investigation approach is genuinely different, and so is the evidence worth preserving.

Oil-filled transformers

Dissolved gas analysis — interpretation under IEEE C57.104, including gas ratio methods, rate-of-change, and what the sampling history does and does not support.
Winding failures — turn-to-turn and section faults, conductor movement and deformation, clamping and short-circuit withstand.
Bushing failures — a common ignition source, and one where power factor and capacitance history often exists.
Load tap changers — contact erosion, coking, mechanism failure, and the maintenance interval question that usually follows.
Thermal aging and loading — hot-spot temperature and insulation life, evaluated against IEEE C57.91 loading guidance and the actual load history.
Through-fault damage — cumulative mechanical damage from downstream faults that the transformer survived but was weakened by.
Tank rupture and fire — arc energy, pressure rise, relief device performance, and fire origin within the substation or vault.

Cast resin and dry-type transformers

Partial discharge — void formation in the casting, PD measurement and interpretation, and manufacturing quality questions.
Thermal cycling and cracking — resin cracking from load cycling or from a duty the unit was not specified for.
Contamination and tracking — surface contamination in dusty, humid, or corrosive environments, and enclosure and ventilation adequacy.
Moisture ingress — units installed outdoors or in wet locations without the appropriate enclosure rating.
Overload and cooling — fan failures, blocked ventilation, and ambient conditions outside the design basis.

The questions that decide the case

Beyond identifying the failure mode, transformer disputes usually turn on a small number of recurring questions:

Was the unit correctly specified and applied for its actual duty, environment, and load profile?
Did the protection — differential, sudden pressure, overcurrent, thermal — operate as designed, and was it correctly set?
Does the maintenance and testing record show a condition that should have prompted intervention before failure?
Was there a manufacturing or workmanship defect, and does the physical evidence still support that finding?
Did the loading history consume insulation life at a rate that explains an early failure?

Evidence to preserve

For oil-filled units, retain oil samples — including any pre-failure samples the owner or a testing contractor already holds, which are frequently the most valuable evidence in the matter. Preserve protective relay targets and event records before anyone resets them, and photograph the as-found condition before the unit is drained, untanked, or scrapped.

For cast resin units, preserve the failed coil intact. Cutting or coring a casting to "see inside" before it has been documented and imaged non-destructively can eliminate the evidence that would establish whether the void was a manufacturing defect or service-induced damage.

Standards governing transformer condition and failure

IEEE C57.104, Guide for the Interpretation of Gases Generated in Mineral Oil-Immersed Transformers — the framework for reading a DGA result, and the source of most disagreements about one.
IEEE C57.91, Guide for Loading Mineral-Oil-Immersed Transformers — hot-spot temperature, insulation aging, and whether the loading history explains the failure.
IEEE C57.109, Guide for Liquid-Immersed Transformer Through-Fault-Current Duration — the withstand curve a through-fault argument is measured against.
IEEE C57.12.00 and C57.12.90 — general requirements and test code, including the impulse and short-circuit withstand the unit was qualified to.
ASTM D3612 (dissolved gas analysis by gas chromatography) and ASTM D923 (sampling of insulating liquids) — whether the sample itself was taken and handled correctly.
NFPA 850 — fire protection practice for electric generating plants and high-voltage direct current converter stations, where transformer fire spread is in issue.

A worked example

Illustrative — when the DGA does not settle it

A 15 MVA unit fails. The last dissolved gas analysis, taken nine months earlier, showed elevated acetylene. The owner says the utility overloaded it; the utility says the owner ignored a clear warning.

Acetylene is the signature gas of high-energy arcing, and its presence is genuinely significant. But C57.104 interpretation is about trend and rate of change, not a single number against a table. One sample cannot distinguish a fault that was actively developing from residual gas left by an earlier through-fault event the unit survived. The first question is therefore whether prior samples exist, because two points establish a rate and one point establishes nothing.

The second question is whether the sample was valid at all. D923 sampling errors — drawing from the wrong valve, air ingress during transfer, an improperly purged line — produce results that look like faults. And the loading argument is independently testable: the C57.91 thermal model, run against the actual load and ambient history, gives an aging estimate that either does or does not support the claim that the unit was consumed early. Three independent lines of evidence, and the opinion is only as strong as their agreement.

Transformer evidence to secure

The complete DGA history, not just the last sample — interpretation depends on trend, and a single result is close to unusable.
Oil sample chain of custody and the sampling procedure actually followed, including which valve was used.
Load and ambient temperature history at sufficient resolution to run a C57.91 thermal model.
Relay and meter records for every through-fault event the unit experienced, with magnitude and duration.
Protective device settings and the coordination study, including whether the transformer's damage curve was ever plotted against them.
Factory test report, impulse test record, and any acceptance test data from commissioning.
Maintenance records: power factor and insulation tests, bushing tests, tap changer inspections, gasket and cooling system service.
The physical unit, undrained and undisassembled, plus any bushings, tap changer components, and cooling equipment removed.

Open tools and cross-checks

Where the dispute turns on whether loading actually consumed the insulation, the transformer thermal life calculator at PowerEngCalc runs the aging arithmetic openly, with the assumptions and the governing relationship stated on the page rather than buried in a spreadsheet. Running it early tells you quickly whether the loading theory is worth pursuing at all.

Transformer Failure Analysis