Reference data sets for testing metrology software

摘要

Many fields of metrology rely on calculations that are implemented in software. When such software is used to provide a measurement result, which is required to be traceable, it is necessary to recognise explicitly the software and show it to be operating correctly. An approach to testing the performance of calculation software is based on using reference pairs each of which comprises reference input data applied as input to the software and corresponding reference output data against which the output data of the software is compared. However, to make the reference pair useful for verifying and validating calculation software, information is needed about the numerical accuracy of the reference pair, the numerical sensitivity of the reference output data to perturbations in the reference input data, and the measurement uncertainty associated with the reference output data arising from simulated measurement uncertainty associated with the reference input data. Such information is important as a means to express quantitatively the quality of the reference pair as a numerical artefact to test calculation software, and as a basis for performance metrics to express quantitatively the numerical performance of software. In this paper these additional components of a reference data set are described, and various approaches to calculating them are discussed. An example, concerned with the calculation of the Gaussian (least-squares) best-fit plane to measured data, which is typical of calculations undertaken in coordinate metrology, is used to illustrate the ideas presented.