Drift is the slow, continuous change in an instrument’s indication over time, with no change in the measured quantity. Over the long term it is determined from the calibration history: if the error at the same points is recorded in each calibration, the difference between consecutive calibrations shows how far the instrument has moved.
Data required
- Results of several calibrations at the same points and with the same methodology.
- The as-found state before any adjustment. If only the state after adjustment is recorded, the evidence of drift is erased.
- The dates, to convert the change into variation per unit of time.
- The uncertainty of each calibration, to know which differences are real.
How it is analyzed
- Calculate the difference between the errors of successive calibrations at each point.
- Compare that difference with the combined uncertainty of both calibrations: if it is smaller, drift cannot be claimed.
- Plot the error against time and examine whether there is a systematic trend, such as a sustained increase, or only fluctuations.
- If the trend is clear, fit a simple model, often linear, to estimate the rate of change and project when the permitted limit would be reached.
How to interpret the result
With the estimated drift you can decide the calibration interval, apply a correction between calibrations or, for a standard, use its projected value instead of the value on the latest certificate. Stable, predictable drift is manageable; erratic drift points to a problem with the stability of the instrument, its conditions of use or the calibration methods themselves.
What to check first
Before attributing a change to drift, rule out a change of laboratory, method or standard. A difference between calibrations may come from the calibration process and not from the instrument. That is why it is useful to keep the same points and methods, or to check with a control standard between calibrations.