O. Berntsson et al., Quantitative determination of content in binary powder mixtures using diffuse reflectance near infrared spectrometry and multivariate analysis, ANALYT CHIM, 419(1), 2000, pp. 45-54
A simplified methodology for development of calibrations for binary powder
mixtures using near infrared (NIR) spectroscopy is presented. It is demonst
rated that multivariate calibration can be performed using only a small cal
ibration dataset of two to five samples. The intrinsic heterogeneity of the
powder mixtures at the sample size effectively measured using standard fib
re-optic probes prevents the assignment of a true reference content value t
o each obtained spectrum. Instead, the nominal mixture content of the sampl
e is used as reference value. To obtain a spectrum more representative of e
ach sample, a mean spectrum is computed from several spectra collected from
different sample sub-fractions. Two principally different powder mixtures
were used in this study: one composed of two fine powders (d<300 mu m) and
the other mixture composed of one fine (d(50)=170 mu m) and one coarse (d(5
0)=590 mu m) powder. In the case of mixing fine powders, the calibration mo
del is obtained by PLS regression using two PLS components after spectral d
ata pre-treatment with multiplicative signal correction (MSC). Furthermore,
it is demonstrated that the powder content variation determined by NIR is
wavelength dependent due to the dependence of effective sample size on radi
ation wavelength.
In the case of mixing one fine and one coarse powder, the large difference
in particle size between the two powder components cause a non-linear X-Y r
elationship, which can be handled by means of spectral averaging and a non-
linear (QPLS) regression. In both cases, the first PLS loading vector was i
nterpreted as describing the difference between the spectra of the two pure
components. The presented methodology should be useful for applications of
NIR spectroscopy in processes involving powder mixtures. (C) 2000 Elsevier
Science B.V. All rights reserved.