My online monitor and the laboratory disagree

The tool has three checks.

Quick comparison. Enter one monitor reading and one or more laboratory analyses of the same period. The tool converts all values to the same temperature basis, 20 °C, compares each gas with the band expected at its level, and runs the diagnosis twice, once on the monitor values and once on the laboratory values.

Gas-in-oil standard procedure. When the difference matters, CIGRE TB 783 Appendix F describes how to check a monitor with gas-in-oil standards (GIOS): standards of known concentration measured by the laboratory, duplicate oil samples, and the monitor reading. The tool follows that procedure and gives the monitor accuracy for each gas and on average, excluding O2 and N2.

Drift over time. Enter at least three dated pairs of monitor and laboratory values for one gas. The tool fits monitor = offset + gain × laboratory and flags the gain only when it differs from 1 by more than 15 %, the threshold in CIGRE TB 783. A gain of 1.5 means the monitor also reads the rate of increase 1.5 times too high.

Before any comparison, the tool checks whether the values are plausible: H2 at zero while hydrocarbons are present, total gas above air saturation of mineral oil (about 96 000 µl/l, IEC 60567:2011 Table C.1), or an O2/N2 ratio above 0.2 in a sealed transformer.

How much difference between monitor and laboratory is expected

Two points on where these numbers come from. The ±15 % is the figure CIGRE TB 783 quotes as the IEC specification; it is not written in IEC 60567:2011, which sets repeatability for laboratories (two results within 7 % of their mean between 10 and 1000 µl/l, 10 % above) and not an accuracy for monitors. The band between 8 and 100 ppm is our inference, and the tool labels it as such.

The minimum levels below which a gas is not compared are 10 ppm for H2, 2 ppm for each hydrocarbon, 50 ppm for CO and CO2, 1000 ppm for O2 and 4000 ppm for N2. Below them the tool reports "not evaluable" and never a 0 % difference.

Does the difference change the diagnosis?

A monitor can read 20 % above the laboratory and still lead to the same decision. The tool therefore runs the same diagnosis on both sets of values: Duval Triangle 1 (Triangle 3 for esters), the IEC 60599:2022 ratios, the IEEE C57.104-2019 status and the resulting severity. The verdict is one of three:

• Agree: every gas within its band.
• Differ, same diagnosis: some gases outside the band, but the diagnosis does not change.
• Changes the diagnosis: rely on the laboratory for now and verify the monitor with gas-in-oil standards, as CIGRE TB 783 Appendices E.2 and F describe.

A laboratory value reported as "<0.5" is not turned into zero. In the comparison that gas is shown as "<0.5" and marked not evaluable; in the diagnosis the shared DGA engine runs at 0 and at the limit and abstains when the two results differ, following IEEE C57.104-2019 §5.2.1.

A monitor that changes the diagnosis. Laboratory, mineral oil: H2 120, CH4 40, C2H6 15, C2H4 60, C2H2 8, CO 300, CO2 2500, O2 3000 and N2 50000 ppm. The monitor reads the same except C2H4 25 ppm and C2H2 30 ppm. C2H4 is 58 % below the laboratory and C2H2 275 % above, both beyond ±30 %. More importantly, the Duval zone and the IEC ratio result change: the laboratory points to a thermal fault, the monitor to discharges. Verdict: changes the diagnosis; rely on the laboratory and verify the monitor with gas-in-oil standards.

A monitor within tolerance. With every monitor value 5 % above the laboratory, all gases agree and the verdict is "agree".

Checking a monitor with a gas-in-oil standard. The single-gas example of CIGRE TB 783 Appendix F, for CH4: a gas-in-oil standard of 93 ppm at 0 °C, which is about 100 ppm at 20 °C; the laboratory reads it as 91 and 89 ppm at 20 °C, so its accuracy is about -10 %. Four oil samples give 259, 230, 210 and 240 ppm; the monitor reads 305 ppm at 25 °C, about 300 ppm at 20 °C. The tool gives a corrected laboratory average of about 261 ppm, a repeatability of about 6.3 %, a monitor difference of about 14.9 % and a monitor accuracy of about 7 %, the same result as the example in TB 783.

The tool runs entirely in your browser. Its Content Security Policy sets connect-src 'none', so the values cannot be sent to any server. It uses no cookies and no browser storage. You can save your inputs to a file on your own computer and load it again later, or print a dossier of the result.

The tool runs in your browser. Seetalabs® does not store or see the values you enter.