How to Tell Analyzer Drift from a Process Change
When an analyzer reading walks away from where it used to sit, the whole question is whether the instrument is lying or the process really moved, and acting on the wrong answer is expensive either way. This troubleshooting guide is for the engineer or technician staring at a trend that has slowly shifted and needing to decide fast. It starts with the cheap cross-checks that often settle it, then works through the causes ordered by likelihood, each with a test and a fix, so you separate a drifting sensor from a genuine change before you either recalibrate a healthy analyzer or ignore a real excursion.
Analyzer Drift vs Process Change in one line: To tell analyzer drift from a real process change, first cross-check the reading against independent measurements of the same process, because a genuine change shows up in correlated tags while sensor drift does not. Then apply a known validation gas: if the analyzer misreads the known gas, it has drifted and needs calibration; if it reads the known gas correctly, the process actually moved. The shape of the change also helps, since drift is usually slow and one-directional while a process change often correlates with an event.
Start with the Fast, Free Cross-Checks
Before touching the analyzer, ask what else should have moved if the process really changed. A true process shift almost never happens in isolation: if a stream composition genuinely changed, related flows, temperatures, pressures, or downstream analyzers should show a correlated response. Pull the trends of those neighboring tags over the same window. If the analyzer moved and its correlates moved with it in a way that makes physical sense, the process likely did change; if the analyzer wandered alone while everything around it held steady, drift is the leading suspect. This costs nothing and settles many cases.
Check the analyzer's own diagnostics and the sample system next, because they are free and often decisive. Look at the instrument's fault and health status, the sample flow and pressure, and whether a filter, dryer, or line could have loaded or partially plugged. A falling sample flow or a loading filter changes what reaches the cell and can masquerade as either drift or a process change. Confirm the reading is not simply lagging a real change through a slow sample path, which you can reason about using the transport delay covered in the note on analyzer sample lag time.
Apply a Validation Gas to Settle It
The definitive test is to show the analyzer a known truth. Apply a certified validation or span gas of known composition and see what the analyzer reports. If it misreads the known gas, the analyzer itself has drifted and the trend shift was at least partly instrument error, so calibration is warranted. If it reads the known gas correctly, the analyzer is honest and the process genuinely moved, so recalibrating would be wrong and you should chase the process instead. This single check separates the two causes more reliably than any amount of trend-staring. The idea of proving an analyzer against a known gas without adjusting it is exactly the analyzer validation check.
Use the right gas for the question. A validation gas confirms whether the analyzer reads truth without changing anything, which is what you want when the point is to diagnose rather than to correct. The distinction between a gas used only to validate and one used to calibrate is covered in the note on the analyzer validation cylinder versus calibration cylinder, and getting it right keeps you from adjusting an analyzer that was actually fine. Record the as-found response against the known gas, because that number is your evidence either way.
Read the Drift Signature When the Gas Is Ambiguous
When a validation gas is not immediately available, the shape of the change carries real information. Genuine instrument drift tends to be slow, gradual, and one-directional, a steady creep in one direction over days or weeks as a sensor ages, a cell weakens, or a reference contaminates. It usually does not correlate with any process event. A real process change more often has a cause you can point to: a feed switch, a unit trip, a weather change, a batch transition, and it correlates in time with that event and with the neighboring tags.
Erratic, noisy behavior points somewhere different again, usually to the sample system or a connection rather than to slow cell drift. Sudden steps often mean a physical event, a valve, a switch, a line clearing, rather than gradual drift. The general patterns of how instruments wander over time, and what those patterns imply, are described in the note on instrument drift over time. Reading the signature narrows the field, but it is the validation gas that confirms it, so use the signature to prioritize and the known gas to decide.
When to Escalate
Escalate when the validation gas and the cross-checks disagree, when a confirmed process change has a cause you cannot identify, or when a confirmed drift returns quickly after calibration. A confirmed process change with no assignable cause may be a real upset that operations needs to know about now, not an analyzer problem to close out. A drift that recurs within days of calibration usually means a failing cell, a chronic sample-system fault, or contamination that recalibration only masks, and it needs the underlying component addressed rather than repeated trims.
Bring evidence when you escalate: the as-found response against the validation gas, the neighboring-tag trends, the sample flow and diagnostic history, and the shape of the change over time. When those histories are captured continuously in a monitoring platform such as Merobix, the trend that first raised the question and the correlated tags that help answer it are already recorded, so the person you escalate to sees the same picture you did rather than starting from a single suspicious number.
Frequently Asked Questions
How do I know if my analyzer is drifting or the process really changed?
Apply a certified known gas and see what the analyzer reports. If it misreads the known gas the analyzer has drifted and needs calibration; if it reads the known gas correctly the process genuinely moved and recalibrating would be a mistake. Before that, cross-check neighboring tags, because a real process change shows up in correlated flows, temperatures, and pressures while sensor drift usually appears in the analyzer alone.
What does a drift signature look like versus a process change?
Instrument drift is typically slow, gradual, and one-directional, a steady creep over days or weeks as a cell ages or a reference contaminates, uncorrelated with any process event. A real process change more often correlates in time with an identifiable cause such as a feed switch or a trip, and it moves the neighboring tags with it. Erratic noise usually points to the sample system rather than slow drift, and a sudden step points to a physical event.
Should I recalibrate an analyzer as soon as its reading shifts?
Not until you have proven it is the analyzer and not the process. Recalibrating an analyzer that actually reads truth, because the process really moved, injects error and destroys the record that told you the reading had been trustworthy. Apply a validation gas first: only calibrate if the analyzer misreads the known gas. If it reads the known gas correctly, leave the calibration alone and investigate the process change instead.
Automation services
Need help turning this into a working system?
Merobix integrates SCADA, programs Allen-Bradley and Siemens PLCs, and designs and fabricates industrial control panels.
Meeting requests are reviewed before confirmation.