Details:
Summary:
This paper starts by surveying papers that have quantified the level of agreement between instruments. The results are consistent in that there is not as much agreement as we would like. "Standardization" has been suggested as a means for improving intra-model agreement. In standardization, measurements of a set of samples are taken with two instruments. The measurements are compared, and this comparison determines the parameters in an equation to compensate for differences. In this way, measurements from one spectrophotometer can be adapted to improve the intra-model agreement.
The set of BCRA tiles is invaluable for spectrophotometer evaluation in that it has a wide range of rich colors, the tiles are readily cleaned, and (except for some well characterized anomalies with temperature) very stable. A set of tiles can be measured by a standards lab, and it can be assumed that the tiles, if properly handled, will not change characteristics over a long period of time, perhaps a few years. As a result, they are widely used as a means to validate spectrophotometers (to establish the accuracy of a spectrophotometer) and to quantify the degree to which it agrees with other spectrophotometers. �
This set of tiles has also been used and recommended for use as a way to standardize one instrument to another. However, the agreement between instruments may not always be seen. Determination of correction coefficients can be an unstable mathematical problem and hence can lead to standardization that does not improve agreement. Worse yet, seemingly proper standardization can significantly degrade agreement.
In the main section of this paper, an experiment is performed where standardizations are performed among three spectrophotometers. Four different mathematical models are investigated and four different sample sets (including the BCRA tiles) are used to determine parameters for the standardizations.
The BCRA tiles are shown in these experiments to be less than robust with some of the mathematical models. � One particularly simple set of samples - a set of paint samples - proved to be less prone to inducing large errors. An explanation of this is put forth, with the eventual aim of the development of a set of samples which can be used as a robust standardization set.
Another finding of this particular experiment is that nonlinearity is perhaps underappreciated as a source of disagreement. Unfortunately, the distinction between wavelength alignment and nonlinearity is the area where the BCRA tiles are most deficient.