Metrohm Vision Manual - Manual (Page 95)

▪▪▪▪▪▪▪ 93
4
Qualitative Analysis Mode
4.1 Introduction
A library is composed of products, each product being represented by a set of spectra. Two major
purposes of the library are to identify and to qualify incoming spectra. In identification, the whole or
part of a library is searched to find a library product or products that match an unknown spectrum
according to the criteria of the identification method. The list of possible matches (which may contain
one or more products) may be passed to a qualification step, where even more rigorous methods
confirm the identity of an unknown and establish its quality.
In Qualitative Analysis mode, the library model is developed and validated. The library model is made
up of at least one method, but the fully developed library may include a clustering method, a number
of identification methods (one for each of the lowest level clusters), and a qualification method for
each product.
Before sample spectra can be used for method development, they must be processed through the
sample selection routine and the results saved. Only then can a method or a series of methods be
developed and validated.
4.2 Sample Selection
The purpose of sample selection is to identify outliers as well as redundant samples. Dragging and
dropping samples from the project product to the library product creates a temporary set (denoted
by question marks to the left of the sample spectra names). A library product with the temporary set
may be created from a single project product, or it can contain spectra from many different products.
Since spectra from various products may not be compatible in terms of wavelength ranges or
detector used, spectra from the temporary set are filtered prior to sample selection, according to the
criteria specified in the Edit Sample Selection Parameters dialog box. Only the spectra fitting these
criteria (detector cell type and wavelength range) will be let through sample selection.
Sample selection can be performed on raw spectra. In order to get rid of scattering effects, or to
emphasize spectral features, one can use any combination and order of available math treatments.
During sample selection, suspected outliers and redundant spectra are identified according to the
threshold that the user can modify. Outliers are placed in the rejection set and redundant samples
into the acceptance set. Good samples carrying unique spectral information are selected and placed
into the training set.
4.2.1 Methods of Sample Selection
Three methods are available: random selection, selection by Principal Component Analysis (PCA), and
selection by wavelength distance. Optionally, each of these methods can be followed by manual
sample selection.
Random Selection
The Random Selection method selects samples at random for training and acceptance sets. No
rejection set is created when this method is used.