IEEE C57.104-2019 DGA Status Levels: What They Tell You and Not

This answer covers the screening status of IEEE C57.104-2019 for laboratory dissolved gas analysis (DGA) on the main tank of in-service mineral oil-immersed transformers. It does not cover fault type identification, load tap changer compartments, ester or silicone fluids, or online monitor data, which need other methods or guides.

What does IEEE C57.104-2019 measure, and how?

IEEE C57.104-2019 does not measure anything itself: it is a guide that interprets dissolved gas analysis results from a laboratory and turns them into a screening output, the DGA Status. IEEE describes the guide as covering "1) Overview of gas generation in transformer and DGA process; 2) The purpose and application of DGA; 3) DGA quality verification and DGA limitations; 4) DGA interpretation and norms; 5) Fault type definitions and identification; 6) Case studies and interpretation example" (IEEE, 2026).

The guide was published in November 2019 and was an almost complete rewrite of the 2008 version, which itself was a minor revision of the 1991 version (Hosseini, 2024). The screening step uses seven gases: hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide (Hosseini, 2024). Oxygen and nitrogen are not scored as fault gases, but their ratio selects which limits apply.

Each gas is checked against four tables. Table 1 and Table 2 hold the 90th and 95th percentile gas concentrations in ppm; Table 3 holds the 95th percentile change between successive samples; Table 4 holds the 95th percentile rate from 3 to 6 samples, for samples with all gas levels below Table 1 values (Hosseini, 2024). The norms are statistics of a fleet, not failure thresholds, so a status describes how unusual a sample is.

How does IEEE C57.104-2019 assign Status 1, 2 and 3?

IEEE C57.104-2019 assigns a status to each gas from the four tables and then gives the whole sample the worst status among its gases. For gas level alone, a value below the Table 1 limit is Status 1, a value above the Table 2 limit is Status 3, and a value in between is Status 2 (Hosseini, 2024). A change above the Table 3 limit or a rate above the Table 4 limit can raise the status of that gas.

The rate in Table 4 is a linear regression slope, normalized to a year, over 3 to 6 samples taken across 4 to 24 months, with separate limits for spans of 4 to 9 months and 10 to 24 months (Hosseini, 2024). The guide's instruction for longer histories, as quoted in the 2024 thesis, is: "if more than 6 data points are available, use the six most recent data points, not exceeding two years, to compute the rates" (IEEE C57.104-2019, clause 6, as quoted in Hosseini, 2024).

The worst-gas rule makes the status decisive. One gas at Status 3 sets the sample to Status 3 even when the other six gases sit at Status 1 (Hosseini, 2024). A confirmation sample can also change earlier results: the guide's step 4c says to "compute rates with the confirmation sample replacing the previous value" (IEEE C57.104-2019, clause 6, as quoted in Hosseini, 2024). The status is tied to periodic laboratory sampling, which is the protocol the levels are derived for (Hosseini, 2024).

How do you read IEEE C57.104-2019 status results?

You read an IEEE C57.104-2019 status as a statement about how likely the sample is to be abnormal, then decide whether to resample, diagnose or act. The guide's wording sets the scale: Status 1 shows "low gas levels and no indication of gassing", while Status 3 shows "high gas levels and/or probable active gassing" (IEEE C57.104-2019, clause 6.1, as quoted in Hosseini, 2024). The table below links each result to its meaning and the next action, including the result patterns that usually point to sampling or measurement error.

A common misreading concerns Status 3: it means probably abnormal against fleet statistics, and the guide's first step is to confirm the sample, not to remove the transformer from service. When the status rests on one gas, check that gas's laboratory value and detection limit before any operating decision, because the worst-gas rule passes a single error to the whole sample (Hosseini, 2024).

Why does C57.104-2019 split limits by O2/N2 ratio and transformer age?

C57.104-2019 splits its level limits by O2/N2 ratio and transformer age because those two factors changed the fleet's gas percentiles more than any other factor the working group tested. The working group compared 90th and 95th percentile values across two-way splits of each candidate factor: O2/N2 ratio below or above 0.2, age, year of manufacture, rating below or above 10 MVA, voltage below or above 60 kV, oil volume and total dissolved gas (Beauchemin, Cheim and Field, 2022).

O2/N2 ratio and age had the largest influence, and more than two factors were judged too complex for practical tables (Beauchemin, Cheim and Field, 2022). Rating, voltage, oil volume and total gas content "did not significantly influence the limit values" (Beauchemin, Cheim and Field, 2022). Values for neighboring age groups were merged when they differed by less than 35% (Beauchemin, Cheim and Field, 2022). The O2/N2 ratio stands in for the oil preservation system: a sealed transformer keeps little oxygen, and a ratio near 0.4 to 0.5 in a sealed unit suggests an air leak (Hosseini, 2024).

The split has edges. Table 3 and Table 4 have no age categories (Hosseini, 2024). The Table 2 limit for methane falls from 60 ppm to 30 ppm above 30 years of age, against the trend of the other gases (Hosseini, 2024). The 2024 thesis also shows that active gassing can shift the O2/N2 ratio enough to change which limits apply, and recommends computing the ratio only when no gassing is suspected (Hosseini, 2024).

How does measurement uncertainty change a C57.104-2019 status?

Measurement uncertainty can move an IEEE C57.104-2019 status by a full level, because each status rests on fixed limits and the worst gas decides. Hosseini and Stewart proposed a method that propagates measurement uncertainty through the C57.104-2019 steps to give "the degree of support for its potential outputs" (Hosseini and Stewart, 2023). The 2024 thesis extends this with triangular and Gaussian uncertainty models (Hosseini, 2024).

The guide recommends an accuracy of plus or minus 15% for DGA results, and IEC 60599:2022 states that uncertainty is typically 15% above ten times the analytical detection limit, rising to typically 30% at five times that limit (Hosseini, 2024). In the 2024 thesis, the extra uncertainty at low levels mattered mainly for acetylene and hydrogen (Hosseini, 2024).

The practical result is a resampling rule. The thesis concludes that adding measurement uncertainty "impacts the IEEE methodology's expected output significantly" because only the worst-case gas sets the combined status, and that the assumed 15% to 30% accuracy is the factor to refine first (Hosseini, 2024). This suggests one more sample before acting on any Status 2 or Status 3 that rests on a single gas near its limit. The guide states the same principle in general terms (IEEE C57.104-2019, clause 5, as quoted in Hosseini, 2024): "It is important to understand that a review of data cannot identify all possible data quality problems. Therefore, when unexpected or alarming results are obtained, it is highly advisable to collect and process another sample to confirm results".

How well do C57.104-2019 status levels predict transformer failures?

C57.104-2019 status levels predict transformer failures only loosely, because the norms are percentiles of a mostly healthy fleet and the published field checks show many flagged units with no confirmed problem. The working group chose percentiles because gas level distributions were "highly skewed", and a percentile always marks the 90th or 95th point of the population whatever its shape (Beauchemin, Cheim and Field, 2022). For gas levels, a Status 3 therefore says the sample sits above the 95th percentile of the reference fleet, not that failure is near.

Three studies give field evidence. On 125 samples from 50 transformers in Indonesia, C57.104-2019 status matched the condition confirmed by inspection or follow-up in 16 of 50 units, against 34 of 50 for C57.104-2008 and 42 of 50 for IEC 60599:2022; the authors attribute this to strict delta and rate limits (Rahmanda and Suwarno, 2022). On 1,525 samples from another Indonesian fleet, the default North American thresholds placed 68.4% of transformers with an O2/N2 ratio of 0.2 or less at Status 3, and locally derived thresholds placed 25.1% there (Kurniawan and Sudiarto, 2025). A 2024 thesis on four case-study transformers found "a tendency to excessively flag DGA samples at its maximal level" (Hosseini, 2024).

The samples are small and regional, so treat them as warnings, not rates. The working group itself found that 90th percentile values "can vary widely from one transformer network to another" (Beauchemin, Cheim and Field, 2022).

How does C57.104-2019 differ from IEC 60599:2022?

C57.104-2019 differs from IEC 60599:2022 in its outputs, its limits and how it computes gassing rates, so the two methods can classify the same sample differently. IEC 60599:2022, edition 4.0, published on 2022-05-25, covers mineral oil-filled electrical equipment in service; IEC states: "This document is applicable to electrical equipment filled with mineral insulating oil and insulated with cellulosic paper or pressboard-based solid insulation" (IEC 60599:2022, scope). The IEC webstore lists a stability date of 2026 for this edition (IEC, 2022).

The table below compares the two on the points that change a screening result.

The key row is the output: Status 1 aligns best with typical, Status 2 and 3 lie closer to alert, and only extreme DGA results align with alarm (Hosseini, 2024). IEC also urges each network to derive its own values: "individual networks are strongly encouraged to calculate the typical values for their own population" (IEC 60599:2022, Annex A.1, as quoted in Hosseini, 2024).

Which transformers fall outside the C57.104-2019 limits?

Transformers outside the C57.104-2019 limits include any unit that is not an in-service mineral oil-immersed transformer, any oil sample not from the main tank, and any data stream from continuous monitoring. The IEEE project scope states that the guide "applies to in-service mineral oil-immersed transformers" (IEEE, 2026), and a 2022 review describes the 2019 scope as mineral oil in main tanks only (Rahmanda and Suwarno, 2022).

Load tap changer compartments have their own guide. IEEE C57.139-2026, approved on 2026-09-25, replaced the 2015 edition and states that "Measurements of dissolved gas concentrations in the OLTC oil are required to implement this guide" (IEEE, 2026). Natural and synthetic ester-filled transformers were covered by IEEE C57.155-2014, which IEEE now lists as inactive-reserved; its revision project PC57.155 has had an active PAR since 2021-09-23 (IEEE, 2026).

Two operating cases also fall outside. For continuous monitoring, the guide states that its suggested screening norms no longer apply (Hosseini, 2024). After oil degassing, gas levels and ratios change abruptly: in one thesis case, gases fell by over 90% after degassing and a ratio-based metric rose (Hosseini, 2024), so restart the trend from the post-treatment sample. The 2019 dataset itself excluded transformers out of service and in-factory test results (Beauchemin, Cheim and Field, 2022), so factory acceptance DGA is not covered by these norms.

Is IEEE C57.104-2019 still current, and what is changing?

IEEE C57.104-2019 is still the active edition as of 2026-10-14, and its revision, project PC57.104, has an active PAR approved on 2024-05-22 that names C57.104-2019 as the edition it will supersede (IEEE, 2026). The working group is chaired by Ed teNyenhuis under the IEEE PES Transformers Committee (IEEE, 2026), and the PAR runs to 31 December 2029 (IEEE Transformers Committee, 2026).

No draft is public. At the March 2026 meeting, the fourth for the revision, the working group reported 185,000 laboratory DGA submissions and over 4.1 million gas monitor data points (IEEE Transformers Committee, 2026). The minutes record that the laboratory set "is much smaller than what was evaluated for the 2019 revision" and that the data analysis task force "is at a standstill" (IEEE Transformers Committee, 2026). Chapter volunteers were named for the guide's sections, and data sharing with CIGRE was being explored (IEEE Transformers Committee, 2026).

Proposals on the table include a fourth status code and fixes to Duval triangles 4 and 5 (IEEE Transformers Committee, 2025). None of these is adopted. Until a revised guide is published, Status 1 to 3 and the 2019 tables remain the IEEE reference.

Common mistakes with C57.104-2019 status levels

The most common mistakes with C57.104-2019 status levels come from treating a screening flag as a diagnosis or applying the tables outside their data. Five recur in the literature.

First, reading Status 3 as an imminent failure. The status ranks a sample against fleet percentiles, and the guide reserves immediate investigation and operating restrictions for extreme DGA results (Hosseini, 2024). Second, skipping the confirmation sample. The guide's own advice is to collect and process another sample when results are unexpected or alarming (Hosseini, 2024).

Third, running fault type methods on low gas levels. The guide does not recommend fault identification when all gas levels are below Table 1 values, and acetylene above Table 1 can still be below five times the detection limit (Hosseini, 2024). Fourth, feeding online monitor data into Tables 3 and 4. The status levels are derived for periodic sampling, and the guide states that rates from continuous monitoring need separate treatment (Hosseini, 2024).

Fifth, using default tables without checking local fit. In one Indonesian fleet, the share of transformers with an O2/N2 ratio of 0.2 or less at Status 3 fell from 68.4% to 25.1% when thresholds were derived locally (Kurniawan and Sudiarto, 2025). That result suggests a site with years of its own data should compare its percentiles with the defaults before acting on a single status.

What should you use alongside a C57.104-2019 status?

Alongside a C57.104-2019 status, use a fault type method, the transformer's own trend, and tests that measure what DGA cannot. The status says a sample is unusual; it does not name the fault or its severity, and the guide states that DGA alone cannot determine transformer state of health (Hosseini, 2024).

For fault type, the guide's main body includes Rogers ratios and Duval Triangle 1, with the Key Gas and Doernenburg methods in annexes but discouraged as outdated or unreliable (Hosseini, 2024). Apply these only once Status 2 or Status 3 is confirmed. For severity, the 2024 thesis recommends a separate second stage, such as the Normalized Energy Intensity method in Annex F of the guide (Hosseini, 2024).

For paper involvement, check carbon monoxide and carbon dioxide together with furan results, since the guide warns that low rates of change of these gases "do not necessarily mean the absence of a fault in paper" (IEEE C57.104-2019, Annex D.8, as quoted in Hosseini, 2024). Electrical tests such as insulation resistance, turns ratio and winding resistance can confirm a DGA finding (Rahmanda and Suwarno, 2022). Lubinpla structures this kind of review from a site's own documents, such as certificates of analysis and maintenance history, and returns one cited document for the engineer to review.

Key takeaways

• IEEE C57.104-2019 status is a screening result: Status 1 probably normal, Status 2 possibly abnormal, Status 3 probably abnormal against fleet percentiles.

• The worst gas sets the status, so one gas near its limit, or a one-step acetylene change, can raise the whole sample.

• Limits depend on O2/N2 ratio and age; the ratio itself can shift during active gassing.

• Confirm before acting: measurement uncertainty of about 15% to 30% can move a status by a level (Hosseini, 2024).

• Field checks show frequent flagging with default limits; compare with your own fleet data where you have it.

• The guide covers main-tank laboratory DGA on in-service mineral oil transformers only, and its revision was still collecting data at the March 2026 working group meeting.