Thursday, September 13, 2012

Calibration Tolerance




Editor's Note:  Interested reader Ray Nims points out that I forgot to square the terms in my propagation of error calculation below.  I apologize for the misinformation.  The corrected blog follows:

In calibration, there is a lot of focus on using the right standard.  Standards must be NIST traceable, have current calibration certifications, and have been cared for appropriately.  See for example ANSI Z540.1 or ISO 17025.  Less attention is paid to the appropriate accuracy of the standard.  In this short blog, we will discuss the basis for instrument range, instrument tolerance and standard tolerance.

An early example of a calibrated measurement device

The tolerance of a process for variation in a certain parameter should be set in process development.  Ideally, this is done as part of process characterization, where the effect of parameter variation is measured.  Some examples of operating ranges set in process development are temperature ± 2°C, pH ± 0.2 units and conductivity ± 10 mS/cm.  Ranges this tight put some pressure on calibration to be especially accurate.  After all, if your instrument is reporting a measurement right at the limit of the acceptable range, it’s probably very important that the instrument not be inaccurate by very much, if at all.

The National Conference of Standards Laboratories (now known as NCSL International) recommends that instruments be calibrated with an uncertainty of no more than 25% of the acceptable control range.  This means that the tolerance (or uncertainty) for the instruments measuring temperature, pH and conductivity cited above would be ± 0.5°C, ± 0.05 and ± 2.5 mS/cm, respectively.  These are tight tolerances, but are needed to ensure that the process is really within ± 125% of the target range.

But wait, there’s more!  You can only be sure that the measuring instrument is within tolerance if you know that the uncertainty of the standard used to calibrate it.  Even standards are not necessarily 100.00% accurate, are they?  Standards used for calibration also have a known uncertainty, and again NCSL International recommends at least a 4:1 ratio of standard to instrument uncertainty.  So the standard used for the instruments above should have a tolerance of no more than ± 0.125°C, ±0.0125 and ± 0.5125 mS/cm, or 6.25% of the operating range. This uncertainty also carries through to the uncertainty of the measurement.  Taken absolutely, the widest range possible range is ± 131% of the actual or target range.  However, it should not be assumed that uncertainty randomly falls to the extremes of the allowable ranges.  It is more common to perform a propagation of errors calculation.  Here, the squares of the errors are added, and the square root of the sum calculated, as shown below

This gives a more likely range of ±126%. 

There is a bottom up approach sometimes taken in determining calibration tolerances.  In this approach, the capability of the instrument is used to determine the calibration tolerance.  In other words, if the thermometer is claimed to have an accuracy of ± 0.1°C by the manufacturer, then it should be calibrated with a temperature standard with an uncertainty of no more than ± 0.025°C.  This approach has proven increasingly difficult, as modern technology has increased the capabilities of field measurement instruments.  However, the approach is valid, and some leeway exists (up to a tolerance ratio of 1:1). 

Of course, higher tolerance ratios are permitted.  But these general guidelines should help you design your calibration program so that you know you are making quality measurements in your process.