Calibration drift is detected by comparing the channel with a reference, analysing trend slope over several horizons, variability and independent data. The trend alone does not determine whether the sensor is drifting, the structure is changing, temperature is acting, or the reference is wrong. The calibration debt ranking is used to set the order of checks, not to auto-correct readings.
In brief
- Instrumental drift concerns a change in the properties of the instrument, not a change in the measured quantity.
- The 7-day trend shows recent acceleration, and the 30-day trend gives wider context. The mismatch between them is a clue.
- Temperature, structural movement and reference error can look similar to sensor drift.
- The recalibration date depends on stability, environment, history and required accuracy. There is no single interval for all devices.
- After correction, you must decide what to do with the history. You must not quietly mix results calculated with different coefficients.
Drift, calibration and adjustment are three different concepts
Instrumental drift is a continuous or stepwise change in indication over time resulting from a change in the metrological properties of an instrument, unrelated to a change in the measurand or in an identified influence quantity. This is the definition used by the JCGM VIM International Vocabulary of Metrology.
That last part of the definition matters. If the housing temperature changed the indication in line with a known coefficient, that is a temperature influence, not automatically drift. If the wall actually moved, the measured quantity changed. In both cases, the plot can look like a slow drift away.
Calibration is also often understood too loosely. According to the VIM, it is an operation that, under specified conditions, establishes the relationship between values of standards with their uncertainties and the indications of an instrument, and then allows a measurement result to be obtained from the indication. Calibration is not the same as adjusting the device or merely checking that it "looks about right".
Calibration debt is an operational ranking of channels for which the data indicate a growing need to verify the reference, stability or inspection date. It is not a metrological term and it does not prove loss of calibration. It helps answer a service question: which channel should be checked first when resources are limited?
This distinction protects against two extremes. The first is ignoring a slow drift because the channel still sends numbers. The second is automatically "correcting drift" on the basis of a trend, even though the trend may reflect a real change in the object.
Four causes of a similar trend
In practice, at least four hypotheses need to be separated before ordering adjustment. Each needs different evidence and a different response.
| Hypothesis | What may be visible | Decisive evidence | Risk of a hasty correction |
|---|---|---|---|
| Sensor drift | Drift relative to a stable reference | Comparison with an independent standard or a control replacement | Hiding a real instrument fault |
| Structural or ground movement | Consistent change in dependent points | Other measurement type, inspection, behavioural model | Zeroing out a real deformation |
| Seasonality and temperature | Daily or annual cycles, correlation with T | Repeatability, influence model, residual distribution | Excessive compensation and loss of signal |
| Wrong reference or standard | Step after a configuration change or a constant offset | Audit, control recalculation, source document | Rewriting history without a trace |
Sensor drift
Suspicion rises when a channel gradually moves away from a stable reference and independent measurements and conditions do not show a similar change. That is still not proof. You need a check against a standard, a portable reading, a comparison device or the manufacturer’s procedure. For an instrument that is inaccessible after installation, the only practical proof is sometimes redundancy designed in earlier.
Real movement
Structural movement can be slow, smooth and resemble instrument drift. Agreement across several points increases credibility, provided they do not share the same error: logger, formula, power supply or temperature. Channels with a different measurement principle are useful, as are the work and load context.
Seasonality
A trend from a short window may be part of an annual cycle. Seven warm days will give a positive slope, even if the series returns to its previous level after cooling. That is why temperature must be analysed explicitly, not removed with a simple line without validation. A separate process is described in temperature compensation or structural change.
Reference and calculation
If the problem starts exactly after a coefficient change or a reference reading change, verify the configuration first. The raw value may be stable while the calculated result drifts because of a unit, sign or dependency error. A change log and a control recalculation of several samples help here.
Why compare 7-day and 30-day regression
Linear regression describes the line y = a + b·t, where b is the average slope in the selected window. In monitoring, it should not be treated as a law of future movement. It is a shortcut description of the data, especially useful for comparing horizons.
The 7-day trend is sensitive to recent change, but it also reacts easily to weather, works and a few unusual samples. The 30-day trend is more stable, but it notices acceleration later. Comparing both gives a better order for checks than a single coefficient.
Illustrative example
Assume a channel in a nominal engineering unit, with a reference of 100.0. The latest value is 102.4. The 30-day regression gives b30 = +0.04 units/day, and the 7-day regression b7 = +0.12 units/day. The standard deviation over 30 days is 0.35 units. This is a calculation example, not implementation data.
| Indicator | Illustrative value | Cautious interpretation |
|---|---|---|
| Deviation from reference | +2.4 | The channel is not returning to the reference point |
| 30-day trend | +0.04/day | A positive drift is present in the longer window |
| 7-day trend | +0.12/day | The last week is rising faster than the background |
| 30-day variability | 0.35 | The series is not perfectly smooth; the result needs context |
The ratio b7/b30 = 3 shows that the short slope is three times larger than the long one. It does not mean "failure in three days". It may indicate acceleration of movement, a temperature change, a work stage or the start of a measurement path problem. The next step is to compare it with temperature, dependent points, configuration history and site events.
Another pattern, with b7 close to zero and positive b30, may indicate that the earlier drift has slowed down. Negative slopes are also not automatically an "improvement". The direction only makes sense after the sign, unit and physics of the quantity are defined.
The calibration interval comes from data, not habit
NIST does not specify one mandatory recalibration interval for all instruments. It recommends taking account of accuracy requirements, contract or regulations, device stability and environmental conditions. A measurement assurance programme may use comparisons, control charts and "before" and "after" calibration data to refine the initial interval later.
That is more sensible than an automatic "once a year" rule regardless of application. The same sensor model may work in a calm temperature field or in moisture, vibration and thermal cycles. One channel has redundancy and easy access, another is critical and buried. The consequences of losing credibility differ.
When setting the date, ask:
- what accuracy the decision requires;
- what the manufacturer recommends and what the calibration certificate states;
- how stable the channel was between previous checks;
- whether the environment, installation or operating range changed;
- whether an independent comparison point exists;
- what the cost of a wrong decision is and what the cost of the check is;
- whether field access is possible before a critical stage.
GEOKON documentation for vibrating wire instruments requires, among other things, correct zero reading, temperature stabilisation and verification against the factory report. The exact procedure depends on the model. Do not transfer coefficients or electrical tests between device families without checking the manual.
In the schedule, it is worth separating three events: functional check, calibration and adjustment. A check answers whether the channel is working and whether the result meets the adopted requirements. Calibration establishes the relationship with the standard and the uncertainty. Adjustment changes the device or coefficient so that the indication meets the requirement. If an order uses one word, "calibration", for all three actions, it becomes difficult later to determine what was actually done.
The service document should include the state before intervention. These are the "as found" data, which show whether the interval was appropriate. The state after adjustment shows how the device leaves the check. Without the first set, you will not know whether the channel stayed within requirements during the last months or whether the next visit should be brought forward.
From ranking to a service order
The calibration debt ranking should end with a task, not just a colour. For each high-priority channel, assign a hypothesis, evidence, owner and deadline. Then the list becomes a maintenance plan.
Look at groups as well, not only individual items. When many channels on one logger change trend at the same time, the cause may be a common clock, power supply, configuration or environmental condition. When similar instruments installed in different places start drifting after a similar operating time, it is worth reviewing the service interval for the whole family. That is still a hypothesis. Its advantage is that it leads to a common-cause test instead of several independent adjustments.
Channel verification checklist
- [ ] Confirm the unit, sign and active reference.
- [ ] Compare the raw value with the calculated result.
- [ ] Verify the 7- and 30-day trend and the sample count.
- [ ] Check temperature, seasonality and work events.
- [ ] Compare dependent points and shared elements of the measurement chain.
- [ ] Review the configuration history and the latest check.
- [ ] Apply the manufacturer’s procedure for the given model.
- [ ] Perform an independent measurement if the consequences are serious.
- [ ] Record the state "before adjustment" and the check result.
- [ ] Determine the effect of the new coefficient on the history.
- [ ] Approve the next inspection interval based on evidence.
If a channel has a sudden step, noise or missing data, first apply the faulty reading cause tree. Trend ranking is a tool for slow change, not a substitute for incident diagnosis.
What it looks like in Inclify
The calibration debt report in Inclify analyses channels that have a reference. It shows the deviation of the latest value from the reference, also as a percentage, the 7- and 30-day regression slope and the 30-day standard deviation. These elements form a heuristic ranking that helps set the order for channel review.
The report does not calibrate the device and does not determine whether the sensor is the cause. Verification is supported by multi-axis charts, for example a comparison of the result with temperature, raw frames in a short-term communication log, and an audit of equation and configuration changes. The reference can be selected from an existing measurement or set manually, but changing it requires an engineering decision.
The report does not change coefficients or history. If the team decides, after independent verification, to make a correction, the scope and effect of the recalculation must be defined in a separate procedure. Inclify does not provide full versioning or automatic rollback to the previous dataset. The rules for decision-making and control are described in the article on safe recalculation of history after a calibration change.
The best starting point for review is therefore a ranking combined with data quality, service documentation and an independent assessment of how the change affects the object.
Limitations of drift analysis
Regression assumes a linear shortcut of behaviour within the window. It does not handle seasonal cycles, a change in work stage, hysteresis, steps or a series with large gaps very well. Before interpretation, verify the number of samples, freshness, completeness and residual distribution. A neat line can be fitted to poor data.
The reference may also be wrong. If the zero reading was taken before the installation stabilised or without recording the temperature, the deviation from it is not neutral evidence. ISO 18674-1 distinguishes initial, zero, baseline and reference measurements, and the baseline measurement may cover the period before works in order to identify changes unrelated to construction.
There is also no guarantee that a channel with no visible drift remains in calibration. The error may be constant, range-dependent or visible only under certain conditions. Data checking does not replace comparison with a metrological reference when that comparison is required.
Finally, changing a coefficient changes the meaning of the result. The effective date, the method of recalculating history, the acceptance test and the approval record must be defined. Quiet overwriting destroys comparability precisely when the team is trying to improve it.
FAQ
Is sensor drift always linear?
No. It can be slow, stepwise, range-dependent or tied to environmental conditions. Linear regression is a convenient indicator of trend in a defined window, but not a full model of the instrument. For non-linear changes, the residuals and series segments need to be inspected, and a check must be performed in line with the manufacturer’s instructions.
Does a positive trend mean the structure is getting worse?
Not without knowing the measured quantity, the sign and the context. A positive trend may mean movement in one direction, rising temperature, a change in load, reference error or drift. First compare dependent channels, environmental data, configuration history and site observations. The check also requires an assessment of the series quality and completeness.
How often should monitoring sensors be calibrated?
The interval should result from the required accuracy, the stability of the specific device, the conditions, the manufacturer’s recommendations, the inspection history and the consequences of error. NIST does not give one period for all instruments. The initial date is worth updating later based on "before" and "after" calibration data. The rationale for the interval should be documented for the specific application.
Does the calibration debt report replace a calibration certificate?
No. The report is a heuristic ranking built from deviation from the reference, trends and variability. It does not ensure metrological consistency and it does not compare the device with a standard. It helps select channels for checking, but the service result and calibration documentation remain separate evidence. It also does not assign an expiry date to the manufacturer’s certificate.
What should be done with the history after changing a coefficient?
First define from when the new coefficient is valid and whether it should be applied retroactively. The recalculation should be carried out on a controlled dataset, pass tests and be published as a whole. Keep the configuration before and after, and do not trigger historical alarms accidentally. Describe the scope of the change in the technical report.
Do vibrating wire sensors not drift?
No instrument type is exempt from checking the installation, reference and data. Vibrating wire sensors are valued for long-term stability, but the result still depends on installation, temperature, cable, logger and calculations. More about their properties is described in the guide why vibrating wire sensors are still used.
Sources and further reading
- Joint Committee for Guides in Metrology, VIM 4.21: instrumental drift.
- Joint Committee for Guides in Metrology, VIM 2.39: calibration.
- National Institute of Standards and Technology, Recommended Calibration Interval.
- International Organization for Standardization, ISO 18674-1:2015, Geotechnical monitoring by field instrumentation.
- U.S. Army Corps of Engineers, EM 1110-2-1908: Instrumentation of Embankment Dams and Levees.
- GEOKON, Product Manuals.
What next
Do not start by changing the coefficient. Select the five highest-ranked channels, run the checklist and determine how many hypotheses are still open. Book a call with the Inclify team if you want to see a calibration debt report on a pilot asset and turn it into a real plan for service inspections.