You are on page 1of 3

MSL Technical Guide 40

Interpolating Corrections from


Calibration Certificates
Version 2, January 2020

0.06
Introduction 0.04
0.02
Calibration certificates for indicating instruments, such as 0.00

Correction
thermometers, pressure gauges, and digital voltmeters -0.02
(DVMs), typically report a table of corrections for a mod- -0.04
-0.06
est number of calibration points. Usually the number of
-0.08
points is just enough for the calibration laboratory to con- -0.10
fidently characterise the major features of the instrument’s -40 -20 0 20 40 60 80 100 120 140 160
performance. Table 1 shows an example of a table from a Instrument reading
certificate for a digital thermometer. Figure 1. A plot of the data in Table 1. The straight lines joining
neighbouring points indicate the corrections obtained by linear
Table 1. Calibration data from a certificate. interpolation.
Reading (°C) Correction (°C) Uncertainty (°C)
–19.88 0.05 0.02 The value of the correction can also be calculated from
0.07 0.01 0.02 the numbers given in the table. The equation for each line
segment in the graph is
20.16 –0.03 0.02
39.88 –0.06 0.03 ( x − xL ) ( x − x)
y = yH + yL H ,
60.00 –0.08 0.03 ( xH − xL ) ( x H − xL )
80.10 –0.08 0.03
where:
100.09 –0.07 0.03
y is the correction value to be calculated,
119.89 –0.03 0.04
x is the reading for which the correction is required,
139.87 0.03 0.04
xL is the instrument reading at the left of the segment,
yL is the correction corresponding to xL.
When the instrument is used, it is common to find that xH is the instrument reading at the right of the segment,
there is no value in the table close to the instrument read- yH is the correction corresponding to xH,
ing, and the question arises, “what value of the correction These definitions are explained graphically in Figure 2,
should I use?” For example, we might want to know the which shows an expanded region around 28.5 °C. If we ap-
value of the correction when the instrument is reading ply the equation to determine the correction value for
28.5 °C. 28.5 °C, using the closest two readings surrounding the
This guide describes two methods for calculating the value 28.5 °C, we get
values for corrections between the entries reported in the
calibration certificate.
0.06
Method 1: Linear Interpolation 0.04
0.02
Figure 1 shows a graph of the correction versus instrument 0.00
Correction

(xL, yL)
-0.02
reading. Each point plotted on the graph corresponds to the
-0.04 (xH, yH)
instrument reading and correction reported in one row of -0.06 y
the calibration certificate. Also shown on the graph are -0.08 x
straight lines joining neighbouring points. The straight -0.10
0 20 40 60
lines may be used directly to read values for corrections off Instrument reading
the graph. For example, if we look along the horizontal axis
of the graph, we can see that for an instrument reading near Figure 2. Expanded region of Figure 1 showing the
definitions of the variables in the interpolating equation.
30 °C, the correction is near −0.05 °C.

Measurement Standards Laboratory of New Zealand Page 1 of 3


(28.5 − 20.16) (39.88 − 28.5) You should need no more than a third-order polynomial
y = −0.06 − 0.03
(39.88 − 20.16) (39.88 − 20.16) (four parameters) for most calibration data, but ensure
that you have at least three times the number of data
= −0.04269.
points than the number of parameters that are fitted.
We can round this value to the same significant digit as • Right click on the equation on the chart to bring up a
the data, so the correction is −0.04 °C. The uncertainty in dialog box with ‘Format Trendline Label’. Change the
the correction is obtained by interpolating linearly between number format to scientific and three decimal places.
the uncertainties provided in the table. In this case, the un- • You can now read the equation off the chart and use it
certainty will be somewhere between 0.02 °C and to calculate a correction for any instrument reading.
0.025 °C, so it will be rounded down to 0.02 °C. In Excel® and other spreadsheets, the trendline feature
performs a least-squares fit. That is, it finds an equation of
Caveats and Warnings the specified form that is the best approximation to the
• Always plot a graph showing the corrections versus in- data. For the example here, a third-order equation fits the
strument reading, like Figure 1, to confirm that the cor- data extremely well, as shown in Figure 3. You can choose
rections follow a reasonably smooth curve and there are to read the correction directly off the chart, or use the equa-
no major glitches or steps. tion displayed. For our example, the equation is:
• Linear interpolation should only be used when the y = 7.120×10–8 x3 + 5.708×10–6 x2 – 2.046×10–3 x + 8.467×10–3.
points in the calibration table are close together relative
to any curvature, as they are in this example, so that If you substitute x = 28.5 °C into the equation, you should
straight lines between the points closely approximate calculate a correction of −0.0436 °C.
the real behaviour of the instrument.
• Linear interpolation can also be used when the points 0.06
0.04 y = 7.120E-08x3 + 5.708E-06x2 - 2.046E-03x + 8.467E-03
are further apart if a plot of the corrections shows that
0.02
the corrections fall close to a straight line. 0.00
Correction

• Interpolation should not be used where a range change -0.02


-0.04
occurs in the instrument reading. Range changes are of- -0.06
ten seen as a step change in the pattern of corrections in -0.08
the graph. If in doubt, check the user manual for the -0.10
-40 -20 0 20 40 60 80 100 120 140 160
instrument to see where the range changes occur.
Instrument reading
• Interpolation schemes (there are many different kinds –
not just linear) always introduce small errors because Figure 3. A plot of the data and the third-order polynomial curve
the interpolating equation does not follow perfectly the that best approximates the data.
true behaviour of the instrument.
• To calculate the uncertainty in the interpolated value, To find the uncertainty in the data, use the same inter-
use the same interpolation scheme (linear here) to in- polating equation (third-order polynomial in the example)
terpolate between the uncertainties. This gives a to interpolate a value for the uncertainty by plotting the un-
slightly pessimistic uncertainty, but it will account for certainty column from Table 1. Once again, the calculated
interpolation error and any correlation between the un- uncertainty will be a little large, but the uncertainty will
certainties in the calibration points. accommodate any residual interpolation error and a high
degree of correlation between the uncertainties in the cali-
Method 2: Least-Squares Fit bration data. Figure 4 shows the plot of the data and the
If you have access to a computer with a spreadsheet appli- fitted curve for the uncertainties. For readings near 28.5 °C,
cation like Excel®, then you can do better than linear in- an uncertainty of 0.02 °C is about right.
terpolation. With Excel®, follow the instructions below.
0.05
• Use the spreadsheet to plot the data as in Figure 1. y = 1E-09x3 + 1E-08x2 + 0.0001x + 0.0212
0.04
• Right click on the data on the graph to bring up the di-
Uncertainty

0.03
alog box with the option for adding a trendline. Select
‘Add Trendline’ and an extra line should appear on the 0.02

graph. A trendline pane will also appear at the side of 0.01


the screen. 0
• In the trendline pane, select the option for putting the -40 -20 0 20 40 60 80 100 120 140 160
Instrument reading
equation on the chart.
• Also, select the polynomial option and change the order Figure 4. The uncertainties in the correction values determined
of the polynomial until the line closely fits the data. by the least-squares fit shown in Figure 3.

Measurement Standards Laboratory of New Zealand Page 2 of 3


There are many different interpolation schemes, and we than with linear interpolation, but that complexity is be-
show only the two most common here. Linear interpolation yond the scope of this guide. If you want further infor-
is the simplest. mation, see the two references below or contact MSL.
The least-squares fit method has the advantage that,
with the right choice of equation, it will naturally follow Further Reading
any curvature in the data and, especially with noisy data, it
will average the data and reduce the effects of any random D R White and P Saunders “The Propagation of Uncer-
error. Corrections calculated from least-squares fits will tainty with Calibration Equations”, Meas. Sci. & Tech-
usually have lower interpolation error than those calculated nol., 18, 2157–2169, 2007.
by linear interpolation. D R White, “Propagation of Uncertainty and Comparison
If you have more information about the nature of the of Interpolating Schemes”, Int. J. Thermophys, 38, 39,
uncertainties and the correlations between them, with least- 2017.
squares fits you can usually achieve lower uncertainties
Prepared by Rod White and Peter Saunders.

Contact Details
Postal address: Measurement Standards Laboratory, Callaghan Innovation, PO Box 31-310, Lower Hutt 5040,
New Zealand.
Website: www.measurement.govt.nz
E-mail: info@measurement.govt.nz

The Measurement Standards Laboratory is New Zealand’s national metrology institute, operating under the authority of
the New Zealand Measurement Standards Act 1992. The Measurement Standards Laboratory is part of Callaghan
Innovation, a New Zealand Crown entity.

This work is licensed under a Creative Commons Attribution-No Derivatives 3.0 Licence
(CC BY-ND 3.0 NZ, see https://creativecommons.org/licenses/by-nd/3.0/nz/).

This licence allows for redistribution, commercial and non-commercial, as long as the work is passed
along unchanged and in whole, with appropriate credit.

Measurement Standards Laboratory of New Zealand Page 3 of 3

You might also like