Powder blending was monitored on-line by taking near-infrared measurem
ents at regular time intervals during the mixing process. The average
standard deviation between the measurements taken at each time and the
dissimilarity between each mixture spectrum and the ideal mixture spe
ctrum were used to monitor the changes in the pow der blend over time.
The distribution of the pure compounds in the blend can be investigat
ed by looking at the score plot for the first two principal components
(PCs), the contribution of each variable to the dissimilarity and, in
particular, the contrasts between two characteristic wavelengths for
each compound. Statistical process monitoring charts were used to dete
rmine the blending time at which the mixture was within (spectroscopic
) specifications. Shewhart charts monitor the blend at characteristic
wavelengths for each substance separately. The Hotelling's T-2 test de
fines a multivariate confidence interval. For spectral data, feature r
eduction is needed, This procedure is accomplished by using characteri
stic wavelengths for the pure compounds or the significant PCs after p
erforming principal components analysis (PCA).