Metrohm Vision Manual - Manual (Page 124)

122 ▪▪▪▪▪▪▪
5
Quantitative Analysis Mode
5.1 Introduction
Quantitative Analysis mode is the part of Vision where calibration equations are developed and
tested. The created calibration equations can be used in Routine Analysis to predict unknown sample
composition in real time.
In order to develop a calibration equation, a set of spectra with corresponding constituent data from
reference analysis is needed. These data have to be processed through the sample selection program
to detect outliers and redundant samples (in some cases the sample selection can be performed
before the reference analysis, as described in the sample selection part.)
Once the calibration set of spectra has been selected, it is used to develop a calibration equation. The
calibration equation relates spectral and constituent data, and can be developed using various
methods.
It is also possible to import calibration equations saved in the NSAS format. This can be done in the
Sample Selection program.
5.2 Sample Selection
5.2.1 Types of Sample Selection
Vision offers two basic types of quantitative sample selection. The first type is based on spectral data
only and allows to detect spectral outliers and redundant samples. The second type looks at the
constituent data and allows to manipulate the data set to obtain optimum box-car distribution. Those
methods can be used independently and their results saved and passed to the regression program.
The purpose of sample selection in quantitative analysis is somehow different from the qualitative
approach. In qualitative analysis sample selection is a way to detect outliers and redundant samples
from spectral data only.
In quantitative analysis, the situation is more complicated. Sample selection should be able to detect
spectral outliers, indicate a possible typing error in constituent values, and make it possible to inspect
and correct the distribution of constituent values for a product. Since the reference analysis is
expensive and time consuming, another purpose of quantitative sample selection is to pick from
many collected sample spectra the optimal subset. This subset is then analyzed and used to develop
the calibration equation.
Vision fulfills all these requirements by an optional two-step sample selection. In the first step, the
initial set of spectra (without constituent values) undergoes the sample selection based on spectral
information. Once the spectral outliers and redundant samples have been detected and set aside in
validation and outlier sets, the spectra forming the calibration set are analyzed. At the second step of
sample selection, some spectra can be removed from the calibration set to improve the distribution.
Additional possibility of outlier detection exists in the regression program. During regression possible
outliers can be removed from the training set without having to return to the sample selection stage.