Spectral imaging in process analytics usingchemometrics and first principles

Optical spectroscopy is able to detect not only the
chemical composition of the species by their wavelength specific
absorption k but also the morphological feature through their
wavelength dependent scattering s. In standard multivariate data
analysis in hyperspectral imaging, the focus of the chemometric
treatment of the data cube is given on the suppression of the unwanted
perturbation of multiple scatter of photons. This paper
describes an approach how to separate the morphological information
s (scatter) from the chemical information k (absorption)
using the radiative transfer equation or Kubelka Munk theory.
When this “first principle spectroscopy” is integrated into most
modern multivariate data analysis like multivariate curve resolution
(MCR), causality is obtained between the spectral data and
response variables like the concentration of an active pharmaceutical
ingredient in a tablet. With this approach, the spatially resolved
calculated k- and s-distribution of an aspirin particle in
cellactose is shown. The optical set up for real life spectral imaging
in industry is discussed and examples of spectral images to
control the thickness of thin films on metals, the distribution of
a resin on a wood chip and the differentiation of hard and soft
maize kernels are shown.