BUCHI NIRCal - Manual (Page 16)

NIRCal 5.5 Software Manual
16 NIRCal 5.5 Manual, Version A
An incorrect distribution of the samples with two separated concentration ranges should be avoided.
Either an independent calibration should be generated for each range or additional samples, covering
the missing range, should be measured.
For reference measurement, the probe and reference material need to be absolutely clean. This is
valid for the measurement of calibration spectra as well as for the measurement of application spectra.
The measuring option and measurement conditions should be identical for generation of both, the
calibration and routine application use. For quantitative measurements, the fibre optics probe should
be in a fixed position.
The samples should not be measured with increasing or decreasing concentrations, but randomly
distributed concentration.
The accuracy of the laboratory method has a huge influence on the quality of the NIR calibration. It is
important to use the most accurate laboratory method (not the weight loss with IR quick drying for
moisture determination, but the drying oven method or better the water determination by titration
according to Karl Fisher).
The time between the laboratory determination and collecting the NIR spectra should be as short as
possible.
Calculating the Quantitative Calibration
The suggested calculation methods are called PCR (principal component regression) and PLS (partial
least squares regression). The algorithms mostly lead to similar results.
The choice of the relevant wavelength ranges and data pretreatments can improve the result of the
calibration. With the applied data pretreatments unimportant spectral variations can be suppressed or
the coherence between intensities and concentrations can be simplified. For example it is
recommended to convert reflectance spectral data into absorbance data that depend directly on the
concentration values (according the Beer-Lambert law, which is limited valid for liquids in NIR). The
difficulties in choosing the data pretreatments are that they always have combined influence and that
the effects depend on the applied order. Theoretical considerations take a back seat in the practice.
Often the model can be optimised by trying different variations, that are calculated.
Concentration = Bo + B1*Amplitude1 + B2*Amplitude2 + ..... + Bn*Amplituden [1],
These establish the coherence between the measured amplitudes and the searched concentrations.
This equation is strongly simplified, a detailed explanation is shown in chapter "Calibration Methods"
Link :
Principal component regression (PCR) & Partial least squares regression (PLS)