What a Combustion Analyser Is Actually Computing

Why this matters

A combustion analyser shows eight to twelve numbers on one screen and they all look like measurements. They are not. Three or four of them are measurements and the rest are arithmetic performed on those three or four plus a table of fuel constants stored in the instrument.

That matters on every call, because the computed values are not independent. When one input is wrong, a whole cluster of displayed numbers moves together in a fixed pattern, and a tech who treats each number as its own evidence sees five things agreeing and concludes the appliance is in trouble. Five numbers derived from one bad input agree the way one number agrees with itself. Recognizing the pattern is how you tell a drifting cell or a leaking sample path from a real combustion fault, and it takes about ten seconds once you know which numbers are which.

Before the appliance fires

Carry a personal CO monitor in your breathing zone with an audible alarm, and turn it on before the appliance fires. The analyser is not that monitor. It is sampling the flue, not the air you are breathing, and it will happily read a clean flue while the room fills.

If you smell gas, nobody analyses anything. Everyone leaves the building immediately, no switches or lights are operated, no phone is used inside, and the call goes out from outside.

The probe tip reaches flue temperature. Handle it by the handle, set it on a non-combustible surface, and give it time before it goes in a case.

Drilling a test hole puts a bit somewhere you cannot see, so identify what is behind the vent wall first, deburr the hole, and close it afterward with a listed plug rather than tape, because an unsealed hole leaks flue gas into the space as well as ruining the reading.

Opening the appliance for service is a different job from probing it. Stored energy and mechanical isolation fall under 29 CFR 1910.147; work inside the control compartment is electrical work under 29 CFR 1910.333(b)(2), since 1910.147 excludes exposure to electrical hazards from work on conductors and equipment in electric utilization installations at (a)(1)(ii)(C), with live-dead-live proving per NFPA 70E-2021, 120.5 in the edition your employer's electrical safety program adopts.

If you find spillage, a disconnected vent or evidence of a blocked flue, that is a shutdown, not a measurement. Shut the appliance down, tag it, and tell the customer in plain words why.

The short list: what is actually measured

Most field analysers measure four things directly, and everything else follows from them:

• Oxygen, usually by an electrochemical cell, as a percentage of the dry sample by volume.
• Carbon monoxide, by a separate electrochemical cell, in parts per million of the dry sample.
• Stack temperature, by a thermocouple in the probe tip.
• Ambient or primary air temperature, by a second thermocouple.

Instruments add cells for nitric oxide, nitrogen dioxide or sulphur dioxide, and many carry a differential pressure sensor for draft. Those are measurements too. The point of the list is what is missing from it.

Everything else on the screen is arithmetic

Carbon dioxide, excess air, air-free carbon monoxide, the CO-to-CO2 ratio and combustion efficiency are not measured by any field analyser sold for this work. They are computed, and they inherit every error in their inputs.

Carbon dioxide comes from oxygen. The instrument computes it as the fuel's maximum CO2 at stoichiometric combustion, multiplied by one minus the measured oxygen divided by 20.9. Two conditions ride along in the same clause: it assumes combustion is complete, and it assumes no air entered between the burner and the probe tip. A draft hood, a leaking vent connector or an unsealed test hole all inject air, and the computed CO2 falls in proportion whether or not the burner changed. The maximum CO2 is a fuel constant, roughly 11.7 to 12.1 percent for natural gas depending on gas composition, near 13.7 percent for propane, and near 15.5 percent for a light fuel oil. Your instrument's own fuel table carries the value it actually uses, and where the table and a rule of thumb disagree, the table governs, because the table is what produced the number on your screen.

Excess air comes from oxygen. The standard field approximation is measured oxygen divided by the difference between 20.9 and measured oxygen, expressed as a percentage. At 7.0 percent oxygen that is 7.0 over 13.9, or about 50 percent excess air. It assumes complete combustion of a hydrocarbon fuel with air as the only oxidant.

Air-free carbon monoxide comes from oxygen and carbon monoxide together. Measured CO is multiplied by 20.9 divided by the difference between 20.9 and measured oxygen. What that correction actually accomplishes, and where it stops being trustworthy, belongs to its own card and is not re-derived here.

Combustion efficiency is a stack loss calculation. It is 100 minus a sensible loss driven by the difference between stack and ambient temperature and by the flue gas mass, with some instruments subtracting a further term for unburned fuel or condensing latent recovery. The exact form varies by instrument, and the manufacturer's documentation is the only place to find the one yours uses. Three things it is not: it is not appliance efficiency, because it ignores jacket loss, off-cycle draft loss and distribution; it is not AFUE, which is a seasonal rating from a laboratory test procedure; and it is not comparable across two instruments using different loss formulas.

What the fuel table does, and the signature of getting it wrong

The instrument does not know what is burning. You tell it, and the selection loads a set of constants: maximum CO2, hydrogen content, and the coefficients in the efficiency formula. Select propane on a natural gas appliance and every measured value stays exactly where it was while every computed value shifts.

That is the cleanest diagnostic signature in the instrument. Measured values frozen, computed values all moved, is a settings problem, not an appliance problem. It is also the first thing to check when a set of readings looks impossible.

The correlated movement table

Two rows in that table look identical on the instrument, and they are the two that matter most. Room air leaking into your sample line and real dilution air entering the vent produce the same five movements, because they are the same physics happening in two places. The reading cannot separate them. Only reseating the probe, sealing the hole and re-reading can, which is why that step appears in the verification list below rather than as an afterthought.

The row that does separate cleanly is the first. A drifting oxygen cell moves oxygen with the stack temperature standing still, because a cell error is not a change in the gas. Anything that actually adds room-temperature air to the flue or the sample cools the sample as well as enriching it, so oxygen and stack temperature move in opposite directions together. If oxygen climbed and the stack temperature did not budge, suspect the instrument before the appliance.

Worked example: one appliance, two sets of numbers, ten minutes apart

First set. Natural gas furnace at steady firing. Oxygen 9.4 percent, carbon monoxide 32 ppm, stack temperature 320 F, ambient 68 F. The instrument's fuel table uses 11.8 percent maximum CO2 for natural gas, inside the range named above, and that table value is what the displayed numbers were built from.

Computed and checked by hand: excess air is 9.4 over 11.5, about 82 percent. Carbon dioxide is 11.8 times one minus 9.4 over 20.9, or 6.5 percent. Air-free carbon monoxide is 32 times 20.9 over 11.5, about 58 ppm. The instrument also displays a combustion efficiency in the low 80s, which is not re-derived here because the loss formula belongs to the instrument.

The tech's first read: 82 percent excess air is high, CO2 is low, and the appliance looks like it is running badly overventilated. That conclusion is one number wearing three hats.

Second set. Nothing on the appliance is touched. The probe is reseated to the center of the flue, the test hole is sealed around the probe, and the reading is retaken at steady state. Oxygen 7.6 percent, carbon monoxide 37 ppm, stack temperature 355 F.

Recomputed: excess air is 7.6 over 13.3, about 57 percent. Carbon dioxide is 11.8 times one minus 7.6 over 20.9, or 7.5 percent. Air-free carbon monoxide is 37 times 20.9 over 13.3, about 58 ppm.

*On the standard field approximation above, excess air is oxygen over 20.9 minus oxygen, which carries no fuel constant, so a wrong fuel selection leaves it where it was. Instruments using a fuel-specific excess-air form will move it; check which form yours uses. Either way CO2 and efficiency move while the measured values do not, which is enough to identify the signature.

Reading the movement. Oxygen fell 1.8 percentage points and the stack temperature rose 35 F. Those two moved in opposite directions, which is the room-air row of the table and not the drifting-cell row. Measured CO rose from 32 to 37 ppm, about 16 percent, which is the same 13 to 16 percent the oxygen and the stack temperature both imply, because room air dilutes carbon monoxide exactly as it dilutes everything else. Excess air fell 25 percentage points and CO2 rose a full point. And air-free CO did not move at all: 58 ppm in both sets. That is the signature. Air-free CO is the one computed value built to be invariant under dilution, so when every other number moves and it holds still, the mixture changed and the combustion did not.

The call. The appliance did not change, because of the four things this instrument actually measures, oxygen and stack temperature both moved and moved in the pattern room air produces, ambient held, and CO moved by that same dilution fraction. The sample changed. The first set was diluted with room air pulled in around an unsealed probe in a vent under negative pressure, and everything the tech was about to conclude from it was arithmetic performed on that leak.

The failure mode. Had the tech acted on the first set, the recommendation would have been a burner or combustion air adjustment on an appliance whose combustion was never in question, and the second visit would have found the same 32 ppm through the same unsealed hole and confirmed the first wrong answer. Nothing about the display would ever have flagged it. The only tell available on the first visit was that five numbers moved together and one did not.

How to verify you got this right

• Run the fresh air calibration in genuinely fresh air. Outside, or well away from the appliance and any running equipment, with the probe out of the flue and the calibration allowed to finish before anything fires. Zeroing in a room with a residual CO level subtracts that level from every reading you take afterward.
• Confirm the oxygen cell reads near 20.9 in that fresh air. It is a one-second check and it directly validates the denominator behind every computed value on the screen.
• Confirm the fuel selection matches the fuel. Then confirm it again after any instrument reset, because the default is not always the fuel you are standing in front of.
• Reseat and reseal before you conclude. Center the probe, seal the hole, retake. If the computed values move and the measured CO does not, the first set was a sample problem.
• Log the measured values, not just the computed ones. Oxygen, CO, stack temperature and ambient survive a fuel table error, a formula difference between instruments and a firmware change. An efficiency figure written on a ticket three years ago cannot be recomputed or compared unless those four sit next to it.

References

• 29 CFR 1910.147 and 29 CFR 1910.333(b)(2), with the carve-out for electric utilization installations at 1910.147(a)(1)(ii)(C)
• NFPA 70E-2021, 120.5, as adopted through your employer's electrical safety program
• Analyser manufacturer documentation for the fuel table constants, the efficiency loss formula, cell accuracy specifications and the calibration procedure
• See related: What an Air-Free Correction Is Doing; How to Take a Combustion Sample That Means Something; What Incomplete Combustion Produces and Why It Matters