The number nobody can agree on
Ask ten dissolved gas analysis (DGA) specialists what the accuracy of DGA is and you will get ten answers: ±5 %, ±15 %, ±30 %, “it depends on the laboratory.” Each of them is quoting something real. None of them is quoting the same thing. This is not a gap in the technology — it is a category error, and it costs money at the procurement table when two vendors compare specifications that were never comparable. The fix is to stop treating “DGA accuracy” as one number and decompose it into the stages that actually produce a reading.
What ±15 % actually measures
The most quoted figure in the industry is ±15 %, from IEC 60567. It is widely read as the precision of a gas chromatograph (GC). It is not that. The ±15 % is an acceptance target: the average result of round-robin tests in which laboratories analyze the same external gas-in-oil standards. Two properties of that definition matter more than the number.
- It includes extraction and degassing, plus the chromatographic analysis that follows.
- It excludes sampling — the step of moving oil out of a transformer and into a syringe without losing or gaining gas.
CIGRE and IEEE apply the same target to laboratories and to online monitors, so a monitor is not judged against a different yardstick. And it is a target, not a measurement: it describes what a competent analytical chain should achieve against reference standards, not what any particular instrument delivers on any particular day.
The uncertainty budget, stage by stage
Splitting the chain apart is the only way to see where error lives. The published ranges for each stage differ by roughly an order of magnitude.
| Stage | Typical magnitude | What drives it |
|---|---|---|
| Detector / gas-phase measurement | 1–3 % | Peak-area repeatability (ASTM D3612, ±1 %); headspace-GC relative standard deviation 0.7–3.1 % |
| Extraction / degassing alone | 12–28 % | Headspace, purge-and-trap and vacuum variants behave differently; headspace is the weakest in round-robin studies |
| Field sampling | 10–50 % | Entrained air bubbles, sunlight, transport temperature, delay before analysis |
| Temperature and partition coefficients | 8–33 % | Solubility varies with temperature and oil type; ASTM and IEC reference temperatures differ |
Read that as a hierarchy rather than a list. The gas-phase detector — the part that photoacoustic spectroscopy (PAS), GC and every other technology compete on — is the smallest term in the budget. Extraction and sampling are an order of magnitude larger. Improving a detector from 3 % to 1 % is genuine engineering, but it shrinks a term that was never the binding constraint.
Why oil–gas separation sets the ceiling
The gas has to leave the oil before anything can measure it. In a headspace extraction, recovery for a gas with solubility coefficient K follows 1 / (1 + K · Voil / Vgas). Gases with low solubility — hydrogen above all — partition poorly into the headspace and are systematically under-recovered. The loss is not uniform: it is gas-selective.
That selectivity is what makes it dangerous rather than merely inconvenient. A distortion that shaved every gas by the same percentage would be invisible in the ratios that every diagnostic method depends on. Instead the light, poorly soluble gases are suppressed more than the heavy ones, which bends the CH4/H2 relationship that fault interpretation rests on, and can shift a boundary case between thermal fault zones. The extraction step does not just add noise — it can change the answer.
Repeatability is not accuracy
The most tempting shortcut is to substitute repeatability for accuracy, because repeatability is cheap and easy to demonstrate: run the same sample twice, subtract, publish the difference. CIGRE’s own caution is worth quoting, because it is counter-intuitive: good repeatability “may add credibility but can also reflect a consistent systematic error.”
The distinction has teeth. In CIGRE’s multi-vendor comparison, participating laboratories measured hydrogen to roughly ±12 %, while individual online monitors ranged from about ±9 % to over ±80 % — several exceeding the ±15 % target. Those readings were stable, reproducible and wrong, and a tight repeatability specification would have concealed exactly the problem that mattered.
Sources
- Bräsel, E. & Sasum, U., “Genauigkeit der Gas-in-Öl-Analyse,” ew — Elektrotechnik und Informationstechnik 111(14):50–55, 2012 — origin of the ±15 % acceptance-target definition, and the statement that sampling influences remain an unsolved problem.
- IEC 60567:2023 (Ed. 5.0) — sampling and analysis; the ±15 % acceptance target, and Annex A solubility coefficients.
- ASTM D3612-02(2026) — peak-area repeatability and the standard’s concentration-dependent precision statements.
- IEEE C57.104-2019 — data-quality discussion, including measurement errors reported as high as ±60 % or more for some gases.
- CIGRE Technical Brochure 409, Report on Gas Monitors for Oil-Filled Electrical Equipment, WG D1.01/TF15, 2010 — laboratory vs online monitor accuracy, and the Table 31 detection-time comparison.
- CIGRE Technical Brochure 783 — laboratory inter-comparison of oil–gas separation methods (the degassing error budget).
What to ask a supplier
- Against what was the accuracy figure obtained — a certified gas-in-oil standard, a round-robin, or a self-referenced comparison?
- Does the quoted figure include the extraction step, or only the detector?
- What is the documented uncertainty at your concentration of interest? A percentage that holds at 500 µL/L rarely holds near the detection limit.
- Has the instrument ever been checked against blind samples, and what was the systematic — not random — deviation?
PAS DGA for a traceable measurement chain
Absolute accuracy in DGA is bounded by extraction and sampling physics, not by the detector. What a well-engineered monitor can control is the part it owns: measurement repeatability, a documented and stable calibration, and a consistent measurement cycle that makes the rate of change trustworthy even where the absolute value carries a known offset. The PAS DGA DGA-900 measures nine gases plus moisture with laser photoacoustic spectroscopy and no carrier gas or consumables, while the DGA-500 covers dissolved hydrogen and moisture for assets that do not need a full gas set. Both are built for a traceable chain: stable calibration, documented drift behavior, and hydrogen measured on its own dedicated channel rather than inferred. See the DGA fundamentals hub to go deeper, compare architectures in our 10-year TCO comparison, or contact PAS DGA to discuss what uncertainty your fleet actually needs.