Summary

Phase compositional data from X-ray powder diffraction were compared and combined with that obtained using optical microscopy for three NIST RM clinkers. Rietveld refinement of the XRD data facilitated calculation of suitable reference standards for quantitative analyses. The optical data were collected using a point-counting procedure following ASTM C 1356.

Comparison of the phase composition data sets using boxplots provides a means of displaying and evaluating the data distributions, including their locations, spreads, and extremes. The data sets generally show reasonable agreement in the estimates of the individual phase abundance. The data do not always agree as well in the estimates of aluminate and ferrite. This may be the result of the fine size of the interstitial material crystals and the resulting difficulty in their microscopic identification. The XRD data generally exhibit greater precision than that of the microscopy point counts. This may reflect the homogenization of the sample as a result of the fine grinding required for XRD analyses. In the case of the microscope values for RM 8488, the coarser texture of the clinker may allow for easier phase identification and so, improved precision.

Measurements from different sources, laboratories, instruments, and from different methods can exhibit significant between-method variability, as well as distinct within-method variances. Certification of a Reference Material is often based upon more than one measurement method. X-ray powder diffraction and microscopy analyses are the intended use of these clinkers and so were used for determining the phase abundance. To establish best consensus values and meaningful uncertainties, three methods of combining these data sets were used. The mean values of individual phase abundance do not vary from method to method, but the 95 % uncertainty interval values do, depending upon the method. In selecting a single method to report consensus values, the maximum likelihood would be favored as this method produces a weighted mean with a weighting scheme that does not necessarily skew the consensus value in the direction of the large number of XRD values and generally produces the most conservative uncertainty interval. The MPVR method also takes explicitly into account between- as well as within-method variance.