Composition analysis by NIR spectrometry usually uses models that assume th
e sample to be homogeneous. However, agricultural fluid products are natura
lly heterogeneous, and in some cases of on-line sensing, it is very difficu
lt to fulfill the homogeneity requirement, and some precautions are needed.
The how itself might create a non-uniform distribution of the measured ing
redient concentrations across the cross-section of the flow. A modified dis
crete model was developed, which incorporates the concentration distributio
n of the ingredients. The effects of model parameters (number of layers and
of iterations, valid ranges of absorbency and specular reflectance factors
) were studied and evaluated. Samples of fresh raw milk were tested, in ord
er to create a known distribution of the fat content in the sample, and wer
e scanned by a NIR spectrometer. In the case of monotonic variation, along
the sample, of the concentration of the measured ingredient (fat), the effe
ct on the measured spectrum was similar to that of changes in the average c
oncentration of a homogeneous sample. Comparison of the errors predicted by
the model with the result of NIR measurements in heterogeneous conditions
of fresh raw cow's milk gave close correlation in their direction and their
relative size. This created the basis for a technique to evaluate other fl
uids, in which the distribution of the concentration of the measured ingred
ient could be evaluated and incorporated in the model simulations. (C) 2000
Elsevier Science Ireland Ltd. All rights reserved.