Glossary term

Measurement Drift

Engineering definition of measurement drift covering sensor drift, zero drift, span drift, drift rate, check-standard trends and uncertainty allowance.

Branch: Engineering Physics
Glossary type: metric
Content: Glossary term
Updated: Jun 26, 2026
Revision: v1.0.0 · reviewed

Definition

metric

Measurement drift is a change in measurement error, response or calibration behavior over time, environment, use or operating history.

Measurement drift can appear as zero drift, span drift, sensitivity drift, calibration drift, reference drift, thermal drift or long-term stability loss. It is different from a static offset or gain error because the error changes between checks or across operating conditions. Drift affects calibration intervals, check-standard trends, uncertainty allowances, as-found impact assessment, sensor validation and release evidence.

Measurement drift is the change in a measurement system’s behavior after calibration, installation, warm-up, use, aging, cleaning, overload, software update or environmental exposure. A channel can pass calibration at one time and still become unfit later if its zero, span, reference, sensitivity, linearity or noise floor shifts enough to affect the engineering decision.

Drift is not just “bad data.” It is a change process. That distinction matters because a single correction may remove today’s error while leaving tomorrow’s trend uncontrolled. Engineers track drift with calibration history, check standards, as-found/as-left records, environmental data, warm-up tests and stability studies.

Drift Rate

If measurement error changes from e_1 to e_2 over elapsed time, a simple drift-rate estimate is:

\displaystyle \hat{d}=\frac{e_2-e_1}{t_2-t_1}

The units depend on the measurand and time base, such as bar/year, degC/month, percent/year or ppm/degree Celsius. This is a screening model. Real drift can be nonlinear after shock, thermal cycling, contamination, aging, overload, repair or firmware change.

Projected drift over an interval is:

\Delta e_d=\hat{d}\Delta t

where \Delta t is the time since the reference condition.

Zero Drift And Span Drift

Zero drift changes the intercept or baseline. A simple temperature model is:

e_0(T)=e_{0,ref}+k_0(T-T_{ref})

Span drift changes sensitivity or gain. A first-order span model is:

S(T)=S_{ref}\left[1+\alpha_S(T-T_{ref})\right]

The two effects should not be mixed without evidence. A zero check can detect baseline drift, but it cannot prove span stability. A high-span check can detect sensitivity drift, but it may miss an offset change near zero.

Drift Allowance In Uncertainty

When drift is bounded but the exact value during use is unknown, a rectangular standard uncertainty screen is often used:

\displaystyle u_d=\frac{|\hat{d}|\Delta t}{\sqrt{3}}

This does not make the instrument acceptable by itself. It only converts a drift allowance into a standard uncertainty component. If the allowance dominates the uncertainty budget, the interval, environment, instrument class or check-standard plan should be reviewed.

Worked Example

A pressure channel is checked with a stable check standard after calibration. At release, the error at the check point is:

e_1=0.010\ \text{bar}

After six months, the same check gives:

e_2=0.034\ \text{bar}

The elapsed time is 0.5 year, so the estimated drift rate is:

\displaystyle \hat{d}=\frac{0.034-0.010}{0.5}=0.048\ \text{bar/year}

Projected over a one-year calibration interval:

\Delta e_d=0.048(1.0)=0.048\ \text{bar}

The rectangular standard uncertainty allowance is:

\displaystyle u_d=\frac{0.048}{\sqrt{3}}=0.0277\ \text{bar}

If the uncertainty budget allowed only 0.020 bar for drift, the channel is no longer supported by the planned interval. The correct response may be to shorten the interval, add an intermediate check, improve environmental control, replace the sensor, repair the installation or investigate a physical cause.

Evidence That Drift Is Real

Drift evidence should separate the instrument from the reference, fixture and environment. Useful checks include repeated check-standard measurements, independent reference comparison, zero and span checks, temperature and humidity logs, warm-up records, cleaning or handling history, overload events, firmware versions, power-supply records, cable condition and residual plots across the range.

A trend is more convincing when the same direction appears across multiple checks and when the check standard is stable enough not to be the source of the change. A single outlier may be setup error, contamination, operator mistake or reference problem rather than true drift.

Difference From Bias And Repeatability

Bias is a systematic error under a stated condition. Repeatability is scatter under unchanged conditions. Drift is change between conditions, times or histories. A system can be repeatable and biased, repeatable and drifting slowly, or noisy without a meaningful drift trend. Release evidence should keep these terms separate until the data justify combining them in an uncertainty budget.

Common Causes

Common drift mechanisms include sensor aging, mechanical creep, adhesive relaxation, strain-gauge self-heating, thermocouple reference-junction error, photodiode dark-current change, amplifier offset drift, reference-voltage aging, contamination, moisture ingress, cable stress, connector corrosion, pressure tap blockage, electrode polarization, software scaling changes, firmware filtering changes and installation preload.

Common Mistakes

Common mistakes include treating a fresh zero adjustment as proof of stability, extending calibration intervals without check data, averaging drifting data as if it were random noise, ignoring environmental correlation, using a stable laboratory reference to justify a harsh field installation, failing to assess past decisions after an out-of-tolerance as-found result, and combining zero drift and span drift into one vague “accuracy” number.

The practical rule is to state what changed, over what time or condition, how it was detected, how it affects decisions, and what action prevents the same drift from silently accumulating again.

REF

Disciplines