A gravimetric DGA standard can be arithmetically perfect and still carry an unmeasured bias. The calculation of concentration from mass, volume and a certified gas fraction is closed; what it cannot see is whether every injected micromole actually stayed in the oil, whether the oil represents the asset being judged, and whether incomplete extraction cancelled or compounded. Those three assumptions decide the difference between a reference and a well-formatted number.
Assumption one: everything injected stays dissolved
The standard assumes complete transfer. In practice there are four loss paths: dead volume in the syringe and the transfer line; gas left in the septum or the valve after injection; micro-bubbles that never dissolve; and adsorption or permeation on the walls of the vessel and tubing. The procedure can reduce all four — fill the vessel completely, degas first, inject below the oil surface, stir for hours — but it cannot measure the residue.
“No visible bubbles” is not a measurement. Transformer oil is dark and opaque, and large-volume vessels are usually stainless steel; a visual check cannot resolve a micro-bubble population. A procedure that relies on it is relying on a habit, not a control. The honest treatment for a gravimetric DGA standard is to declare a transfer-loss uncertainty term, u_transfer, and to verify it independently.
| Assumption | What it covers | Usual handling | What would verify it |
|---|---|---|---|
| Complete transfer | Syringe, septum, valve, micro-bubbles, adsorption | Procedure only; visual check | Independent recovery test with an upper bound |
| Matrix representativeness | Fresh oil versus aged service oil | One oil type, often unspecified | Repeated check on aged or field oil |
| Extraction cancellation | Incomplete degassing absorbed in calibration | Matched cell, ratio, temperature, time | Deliberate variation test at the edges |
Assumption two: fresh oil represents service oil
Standards are usually prepared with new mineral insulating oil meeting IEC 60296. Service oil is not that: it carries ageing products, dissolved moisture, and a different gas solubility profile. IEC 60567 does not fix an oil grade for gas-in-oil standards, which means the certificate is only strictly valid for the matrix that was used. A monitor judged against fresh-oil standards may behave differently on an aged sample, particularly where a membrane or a sensor surface interacts with the oil.
This does not invalidate the standard. It defines its scope. If the purpose is to qualify a monitor for service, the qualification should include at least one aged-oil or field-oil sample, not only the clean matrix that flatters the sensor.
Assumption three: incomplete extraction cancels
Gas chromatography rarely reaches full oil–gas equilibrium in a short extraction. The design that makes this acceptable is matching: calibration standards and test samples are extracted in the same vessel, at the same phase ratio, temperature and shaking time. The incomplete recovery is then absorbed into the calibration curve and cancels. That is good metrology, and it is conditional. Change the phase ratio, the temperature or the extraction time between calibration and measurement, and the cancellation stops. The bias returns, and it returns silently.
The circular check that turns loss into instrument bias
Here is the trap. Suppose the true recovery is 95%. The laboratory validates recovery by analysing its own prepared standards with the same chromatograph that those standards calibrate. The chromatograph reads them as nominal, because the 5% loss is present in the calibration as well as in the check. Recovery appears to be 100%. The monitor under test then reads the same oil and also reports 95% of nominal — not because the monitor is wrong, but because the oil really contains 95%. The 5% loss is assigned to the instrument.
The direction matters: this bias is one-way and systematic, and it always makes the monitor look worse than it is. Breaking the circle requires a recovery check that does not depend on the same calibrated chromatograph — for example, weighing the injection cylinder before and after transfer, measuring total gas content by vacuum or pressure methods, or using an independent reference material with a stated uncertainty. Whichever method is chosen, u_transfer needs a number and an upper bound, not a procedure reference.
Does a gravimetric standard need a guard band?
Yes. If the acceptance rule is a bare comparison of a measured deviation against a limit, part of the failures will be reference error rather than instrument error. ILAC-G8 defines how to set a guard band so that a pass decision means the true value is likely inside the limit. The size of that band comes directly from the reference budget, which is why the budget has to be written down before the test begins.
Can a monitor be declared non-compliant on one batch?
No. A single batch that fails a homogeneity gate is a batch that cannot support a decision, not evidence against the instrument. The correct sequence is to fix the batch, repeat with adequate replicates, and only then judge. A procedure that treats “no decision” as “failure” will produce a supplier dispute instead of a measurement.
PAS DGA supplies monitors and works with laboratories on acceptance testing; the DGM-900 covers seven gases plus moisture, and the introduction to laboratory gas standards explains where each reference type fits. For the preparation chain itself, see how a gas-in-oil reference standard is made; for the budget, see the half-window rule. To discuss a specific validation plan, contact our team.
Sources
IEC 60567:2023 — preparation of gas-in-oil standards and extraction guidance. ISO 6142-1:2015 — gravimetric preparation of calibration gas mixtures. ISO 17034:2016 — reference material producers. ILAC-G8:09/2019 — decision rules and statements of conformity. JCGM 100:2008 — Guide to the Expression of Uncertainty in Measurement.