CEMS Sensor Selection: Sizing SO₂, NOx and O₂ Ranges — and Why Sample Conditioning Decides the Outcome

Three monitoring targets, three range classes

The phrase stack monitoring hides three quite different jobs. Stationary sources are the high-concentration conditions inside a stack or exhaust duct; fugitive emissions are the low-concentration dispersion at the fence line and inside workshops; ambient air is the regional background. Their criteria differ and their ranges span two to three orders of magnitude, so a single sensor chosen to cover all three generally fails on one of them.

Sizing backwards: convert the limit to ppm first, then leave headroom

Emission limits are written as mass concentration in mg/m³, while sensor ranges are quoted as volume concentration in ppm. The first step of sizing is converting between them. At 25 °C and 101.325 kPa a mole of gas occupies about 24.45 L, so 1 ppm equals molar mass divided by 24.45, in mg/m³.

• Step one: convert the limit to ppm. A 35 mg/m³ SO₂ limit is about 13 ppm — but upstream of the desulphuriser, inlet concentrations commonly sit in the hundreds of ppm, so a 0–20 ppm sensor is saturated from the moment it is commissioned.
• Step two: size against the worst credible operating case, not the everyday value. The reference point is the high value seen during start-up and shutdown, when the desulphurisation or denitrification train trips, or when combustion drifts — commonly two to five times the converted limit.
• Step three: check that range and resolution go together. If an SO₂ channel is widened to 0–2000 ppm to cover the inlet, its resolution near 13 ppm becomes coarse. Separate ranges for inlet and outlet are usually the better answer than one channel stretched across both.

Sample conditioning: where the two-year survival of the analyser is decided

Extractive sampling draws gas from the stack to the analyser through three stages: heating, drying and filtration. Their set points are not refinements but hard physical floors. Below them, water condenses in the line, acid components dissolve into the condensate and are carried away, and the reading runs systematically low.

• Heated line: set generally at 120 °C or above, and above the flue-gas dew point by at least 10 °C. The temperature itself must be readable in the cabinet or system software — a temperature nobody can read is not a controlled temperature.
• Routing: the whole line needs support and at least a 5° fall, with no U-bends or low pockets. Otherwise condensate pools into a liquid trap that either blocks the sample or absorbs the components in transit.
• Drying: the chiller is held at about 4 °C, within ±2 °C, which pins the outlet dew point below 4 °C. Condensate must drain automatically and promptly; water that cannot leave will eventually reach the analyser.
• Fine filtration: a filter ahead of the analyser should stop particles above roughly 0.5–2 μm, in a hydrophobic medium that neither adsorbs nor reacts with the gaseous pollutants. Choose the wrong medium and the filter itself becomes an absorption trap.
• Two tubes: the sample line should carry at least two tubes, one for sample gas and one for calibration gas, enabling a full-system calibration. Calibration that exercises only the analyser and skips the sampling and conditioning path cannot trace the bias introduced by adsorption or leaks.

Why the oxygen channel deserves a channel of its own

Many projects treat oxygen as a combustion-efficiency side reading and hang it on whatever channel is convenient. In emission monitoring, however, oxygen is the correction basis: measured concentrations are converted against the measured and reference oxygen levels before they can be compared with a limit. The oxygen channel is therefore not a reference value — it is part of the denominator for every other channel.

The consequence is that the error propagates in step. A low oxygen reading biases every corrected SO₂ and NOx figure upwards; a high reading biases them all downwards. Because the shift is same-signed and smooth, a trend chart will not reveal that the oxygen channel is at fault — it merely looks as though the boiler's emissions have drifted lately. The dry-versus-wet distinction rides on the same measurement.

• Give the oxygen channel its own zero and span calibration rather than sharing a cycle with SO₂ and NOx. When it goes wrong, an entire data set goes wrong with it.
• Establish whether oxygen is measured before or after the sample is dried. The dry and wet oxygen contents of the same flue gas differ, and using the wrong one shifts every converted result.
• Put oxygen-channel failure into the alarm logic. A frozen reading or an extreme value should raise a system alarm rather than continue feeding the correction and turning a fault into apparently compliant data.

Cross-interference, zero drift and the calibration rhythm

Several target components in flue gas interfere with one another: an SO₂ channel responds to some VOCs and to CO, an O₃ channel is sensitive to NO₂, and the NO₂ channel is in turn affected by O₃. These are not sensor faults but inherent properties of the electrochemical principle, and they can only be managed through selection, filters and headroom in the set points.

The other half is time. Zero drift over three to twelve months is normal for electrochemical toxic-gas channels, which is why stack monitoring is not an install-and-forget system: it relies on calibration gas, records and scheduled zero and span checks to keep drift inside a traceable band. Calibration typically covers 20%, 50% and 80% of range, with calibration gas of sufficiently small expanded uncertainty used within its validity period, and cylinders replaced before residual pressure falls below the specified value.

Frequently asked questions