12.2. Type A uncertainty for quadratic calibration of a loadcell | ||||
Background on loadcell calibration.
The purpose of this case study is to illustrate the calculation of uncertainties
for calibrated values of a loadcell. The calibration of the loadcell depends on a calibration experiment where responses of the loadcell are observed for known loads (psi). The following are three repetitions of the response of loadcell #32066 at eleven known loads as measured at the National Institute of Standards and Technology.
X Y 2. 0.20024 2. 0.20016 2. 0.20024 4. 0.40056 4. 0.40045 4. 0.40054 6. 0.60087 6. 0.60075 6. 0.60086 8. 0.80130 8. 0.80122 8. 0.80127 10. 1.00173 10. 1.00164 10. 1.00173 12. 1.20227 12. 1.20218 12. 1.20227 14. 1.40282 14. 1.40278 14. 1.40279 16. 1.60344 16. 1.60339 16. 1.60341 18. 1.80412 18. 1.80409 18. 1.80411 20. 2.00485 20. 2.00481 20. 2.00483 21. 2.10526 21. 2.10524 21. 2.10524 The loadcell in this example exhibits a quadratic response relative to load, and the calibration curve is the estimated from the calibration data by least-squares as the function:
where the estimates of a, b , c, and their associated standard deviations from the least-squares analysis as shown below. COEFFICIENT ESTIMATES ST. DEV. T VALUE a -0.183980E-04 (0.2450E-04) -0.75 b 0.100102 (0.4838E-05) 0.21E+05 c 0.703186E-05 (0.2013E-06) 35. RESIDUAL STANDARD DEVIATION = 0.0000376353 RESIDUAL DEGREES OF FREEDOM = 30Given a future response by the loadcell, say, Y', the problem is to correct the response to the proper load, X', and to calculate its associated uncertainty. A graph showing the calibration curve and a future response, Y' corrected to X', the proper load (psi), is shown below. The calibration data which consist of three repetitions at each of eleven loads are listed at the end of the example. | ||||
Uncertainty of the calibrated value X'.
The calibrated value is usually calculated from
|
and one method of assessing its uncertainty is via propagation of error with the variance appproximated by
where f is the function that defines X' and denotes a partial derivative. The first step in the analysis, deriving the partial derivatives, can be simplified with software capable of algebraic reprsentation; in this case, Mathematica was used to derive the partial derivatives. The partial derivative of f with respect to Y is: The partial derivatives with respect to the coefficients of the quadratic calibration curve are respectively:
The standard deviation of the calibrated value.
The next step is to substitute the values of the coefficients and their respective standard deviations into the equations above for the range of interest for Y' where is the residual standard deviation from the fit to the calibration data. The graph showing standard deviations of calibrated
values, X', as a function of instrument response, Y', is displayed below.
| Problem with propagation of error.
The propagation of error shown above is not
correct because it ignores the covariances among the coefficients,
a, b , c.
Unfortunately, some statistical software packages do not display the covariance
terms with the output from the least-squares analysis. The information
should be accessible from the "variance-covariance matrix".
| The variance-covariance terms for the loadcell data set are shown below. a 6.0049021-10 b -1.0759599-10 2.3408589-11 c 4.0191106-12 -9.5051441-13 4.0538705-14The diagonal elements are the variances of the coefficients, a, b, c respectively, and the off-diagonal elements are the covariance terms. Recomputation of the standard deviation of X'.
The standard deviation of X'
is redefined to account for covariance terms so that |
Appropriate substitutions are made and the standard deviations are recomputed and graphed as a function of instrument response. In this case, the uncertainty is reduced by including covariance terms, some of which are negative.
|