Abstract In order to establish an adequate analytical system for the quality control of industrially produced titanium dioxide white pigments, two multivariate linear calibration techniques, principal component regression (PCR) and partial least squares (PLS), are used to model the relationship between the important pigment property, change of colour, and its chemical composition. The results, in terms of accuracy, precision, suitability for quality control and analysis time are compared to those obtained with artificial neural networks (ANNs). Two multivariate display techniques, principal component analysis (PCA) and correspondence factor analysis (CFA) together with two hierarchical clustering techniques, divisive and Ward's agglomerative hierarchical clustering, are also applied to the X-ray fluorescence data of the pigments samples so as to extract as much information as possible. Correlation coefficients obtained by PCR and PLS are 0.92 and 0.94, respectively. Both of them are higher than the already achieved correlation coefficient by ANNs [1], but the precision of the model derived by ANNs is better. It should also be pointed out that some important additional information about the relations between independent variables (chemical composition of the pigment samples) and about the influence of different oxide concentrations on the pigment property, which could be used in the controlling of the production process, was found out.
[1]
Jure Zupan,et al.
NEURONALE NETZE IN DER CHEMIE
,
1993
.
[2]
Johann Gasteiger,et al.
Neural Networks for Chemists: An Introduction
,
1993
.
[3]
Trevor Hastie,et al.
The Geometric Interpretation of Correspondence Analysis
,
1987
.
[4]
Jure Zupan,et al.
Algorithms for Chemists
,
1989
.
[5]
Teuvo Kohonen,et al.
An introduction to neural computing
,
1988,
Neural Networks.
[6]
Teuvo Kohonen,et al.
Self-Organization and Associative Memory, Third Edition
,
1989,
Springer Series in Information Sciences.
[7]
D. Massart.
Chemometrics: A Textbook
,
1988
.
[8]
Nineta Majcen,et al.
Modeling of property prediction from multicomponent analytical data using different neural networks
,
1995
.