BUCHI NIRWare - Manual (Page 64)

58

Tutorial NIRWare 1.5 Software Manual
The spectrum residual is judged to be OK, which indicates that the spectrum is very similar to those that
the calibration was built on.
‘Distance OK’ tells us that the spectral representation falls within the boundaries of a cluster, but because
‘found’ and ‘expected’ substance differ, the sample is falling into the wrong cluster and is thus rejected.
So from the report you can be sure that the sample is not lactose, but -with high probability- fructose. With
this information a further investigation is now possible to check if just the label was wrong, or an
independent analytical method can be used to confirm the ID suggestion of the NIR application.
This example clearly shows that ID testing with NIR integrates a label check!
5.4. Creating a quantitative application
5.4.1. Introduction
This tutorial describes how to create an application for quantification. As an example, the NIRFlex N-500
is used with the solids cell and the vials (autosampler) option.
The application shall be capable of quantifying
a) the concentration of Lactose (‘milk sugar’) and
b) the concentration of Fine Sucrose (‘powder sugar’)
in a mixture of both substances. Concentrations are expressed in % wt. (g / 100 g). In NIRWare 1.5 if
there is a 2 component system like this, one property can be calibrated and the other calculated, e.g. Fine
Sucrose concentration = 100-Lactose concentration.
The process of creating a ready-to-use NIR application requires calibration samples. These are samples
of doubtless quality and composition. In combination with their NIR spectra (‘calibration spectra’) they
provide the necessary data sets for calculation of a chemometric calibration. The calibration can then be
used for prediction, which means deriving the identity information from the spectra of substances to be
tested.
It is recommended to use two dedicated applications for acquisition of calibration data and routine use
(=prediction).
NOTE
The terms 'reference measurement', 'reference samples' or 'reference spectra' used in the tutorial
describes the data collected for calibration development.