How Soon to Resample When Transformer Oil Shows New Gas

Why the date matters as much as the number

Heat and electrical stress can break down transformer oil and its paper insulation, producing gases that dissolve in the oil. Measuring them is dissolved gas analysis, or DGA. Gas type and amount help describe a problem, but the rate and change over time matter too, as ASTM D3612's public significance statement explains.

The US Bureau of Reclamation's transformer maintenance manual asks for every analysis to be compared with earlier results. It gives particular weight to a sudden rise and the rate of production, while treating acetylene as significant even in small amounts. The manual is from 2000; its numerical interpretation tables are not a substitute for the guide your engineer uses now.

A rate needs two things: results you can trust to be comparable, and the real time between them. Your two results are a year apart, so the rate they give is only an average over that year. It cannot tell you whether the gas crept up steadily or sat still for eleven months and then started to climb.

The figure uses assumed numbers, not limits or typical values. One gas measured 20 ppm last year and 50 ppm this year. (Parts per million by volume and microlitres per litre are the same quantity.) For this example only, the bars are assumed expanded uncertainties of ±15%, with coverage factor k = 2. They are not a laboratory's statement or an alarm rule. Both histories fit the annual results. Three weeks later, the steady history adds less than 2 ppm; the late rise, if it continues at 1 ppm a day, adds 21 ppm. The difference shows why the timing of the next sample matters, not whether a transformer is safe.

Could the sample explain the change?

Reclamation's manual says that when hydrogen, moisture or gases from the air jump from one analysis to the next, sampling or handling can be the explanation. It recommends checking a suspect analysis with another sample. That check accompanies, rather than postpones, any operating response required by the actual result or event. Our earlier guide to choosing between action, verification, trend monitoring and a repeat sample puts that priority first.

For dissolved gas, three things can move a result without anything changing inside the transformer.

• The valve. Oil trapped in the drain valve and connecting pipe does not circulate. Reclamation describes hydrogen collecting there when the sun heats the tank, or forming through galvanic action where galvanised or black iron pipe meets a brass valve. Use the same valve and flush it the same way. Avoid the small side ports on large drain valves: the manual says they cannot flush the valve properly.
• Air. Reclamation recommends glass syringes so bubbles can be seen and expelled; a small air leak into a metal vacuum bottle can go unnoticed. In CIGRE tests reported in 2007, a bubble larger than about 1% of the sample volume lowered measured hydrogen by about 35%. Expel air while filling, then seal the syringe.
• Light and heat. Those tests found that sunlight raised hydrogen in aged oil in a stored syringe. Eight days at 60 °C in the dark, without air, changed little for the tested oil; about 20 days at 90 °C produced extra hydrogen. This heat-related effect in some mineral oils is called stray gassing. The authors recommend room-temperature, dark storage. Reclamation asks for the syringe to go straight back into its shipping container. Agree a route with the laboratory that avoids sunlight and strong heat.

For the sampling procedure, use clause 4.2 of IEC 60475:2022. It covers mineral oils, synthetic and natural esters, and silicones. IEC 60567:2023, the gas-analysis guide, now points there for taking oil from equipment. Both are listed on VeriFluid's transformer oils page and in its method library.

The transformer's state at sampling matters too. Reclamation cautions that dissolved gas results are unreliable if the transformer has been switched off and allowed to cool, if it is new or if it has had less than one to two weeks of continuous service after oil processing. The manual also stresses keeping a history of overloads, cooling problems, nearby lightning and through-faults (faults elsewhere on the network that drive heavy current through the transformer), because that history helps explain what a change in the gases means.

Taking oil from an energised transformer is electrical work. It belongs to people your site has authorised for that equipment, working under its permit and safety rules. Reclamation, for example, says not to sample while there is negative pressure at the valve, because air drawn into an energised transformer could be disastrous. A laboratory can specify the container and the record; the work at the valve is the site's responsibility.

A year between samples can hide a late rise

Once both results can be trusted, the arithmetic is simple: divide the change by the time between the samples. Reclamation counts only the days the transformer was on load and leaves out any days it was out of service.

With the assumed numbers above, 30 ppm over 365 days is about 0.08 ppm a day. The same 30 ppm over three weeks would be about 1.4 ppm a day, more than seventeen times faster. That is why each result needs the date its sample was taken beside it, not just the date of the report.

Each result carries measurement uncertainty, and a rate built from two results carries both. IEEE C57.104-2019 includes quality verification and the limitations of gas analysis. A 2023 study of uncertainty in that method highlights the concern near the detection limit, the smallest amount the method can reliably detect.

They also note that the 2015 edition of IEC 60599, the IEC interpretation guide, suggested assuming about 15% uncertainty for results more than ten times the detection limit, and about 30% below that, when nothing better is known. Differences between laboratories can be larger: the 2007 CIGRE paper notes that round-robin tests, in which several laboratories analyse the same oil, may show deviations of up to 44%.

Compare results from the same laboratory and method wherever you can, and ask for the uncertainty of each gas at the level you are reading. Give a gas that has just crossed the reporting limit particular care; a new number does not by itself establish a new fault.

The IEEE guide itself does not rely on a single two-point rate. As the Strathclyde paper describes it, the guide looks separately at the change since the previous sample and at a rate fitted through several results spread over 4 to 24 months. The transformer component maker Reinhausen gives the same 4 to 24 month window for that fitted rate.

What could another sample tell you in three weeks?

A small difference may be hard to distinguish from measurement variation. But an uncertainty interval is not a universal detection threshold, and crossing it is not proof of a fault.

Here is a bounded calculation, separate from the IEC percentage quoted above. Assume two results near 50 ppm, each with an expanded uncertainty of 7.5 ppm and coverage factor k = 2. Each standard uncertainty is then 3.75 ppm.

If the results are independent, the standard uncertainty of their difference is √(3.75² + 3.75²) = 5.3 ppm. Applying the same k = 2 gives an expanded uncertainty of about 10.6 ppm on the difference. NIST's guidance on combining standard uncertainties and expanded uncertainty explains the calculation. This example holds the absolute uncertainties fixed for the arithmetic.

The bars compare increases over the same three weeks; the note below each bar gives the time that rate takes to accumulate 10.6 ppm. These times are not minimum sampling intervals, detection guarantees or deadlines. No resolved change over a short interval does not prove that a fast rise is absent.

Reinhausen's worked example shows the same trade-off. With 10% uncertainty and a true rise of 1 ppm a day, two samples 10 days apart can even look consistent with a falling gas, while samples 100 days apart give a more robust trend. A longer baseline is slower to react, though. Ask the laboratory and transformer engineer to choose the comparison rule using the actual uncertainties, correlations between results and the risk of a missed or false change. The site's required response still comes first.

For the data hall in the opening, that points to a plan the team can agree with its transformer engineer:

• Confirm, if the engineer calls for it. Use the same valve, laboratory and method, with the record below. Agree reporting time before collection. A prompt repeat checks whether the change repeats; it does not replace urgent action or guarantee a resolved trend.
• Compare either side of the load change. Where load dependence needs checking, Reclamation asks whether the gas rate rises and falls with load, with a sample each time load changes. Under the engineer's plan, a confirmation shortly before switch-on can serve as the before result; the after sample follows once the new load has settled.
• Continue the series. The engineer sets intervals from gas levels, measured rate and the guide in use, shortening them if the rate rises. Reclamation's table ranges from annual to daily sampling as levels and rates climb, but derives from the twice-superseded 1991 IEEE guide. Use that principle, not its old numerical schedule.

Each sample in this plan is timed to answer one question before the decision that depends on it. None of the timings is a universal rule.

Name the liquid and the interpretation guide

Interpretation guidance depends on what is in the tank. IEC 60599:2022 covers equipment filled with mineral oil and insulated with paper or pressboard. Its abstract calls for caution with other insulation systems and engineering judgement for any resulting action. In India, the same edition is adopted as IS 10593:2023.

Write the liquid and its product name on the sample form, then agree the interpretation guide and edition with the engineer. IEEE C57.104-2019 concerns mineral oil transformers. For natural or synthetic esters, the separate IEEE C57.155 guide needs a status check: its 2014 edition is inactive while a revision is prepared. Do not carry mineral oil rules across by habit. Edition details and revision links are in the references.

The record that travels with the syringe

For the load change in three weeks, a request that says only "repeat DGA" leaves too much open. Send the earlier report, the operating history and the decision date with the sample request. Agree collection and reporting times, rather than finding out after dispatch that the result will arrive too late.

That gives the laboratory enough context to discuss measurement uncertainty and comparability, and the engineer a record against which to judge the rate. The next report may strengthen a trend or leave it unresolved. Either is more useful when the team knows which question the sample was collected to answer.