Darinka Tokarcíková

<tokarcikova@rona.sk>

*
Darinka Tokarcíková Lecture. Head of the Development and Research department in RONA glass factory.
*

Time evolution of glass knots’ chemical composition followed by advanced statistical methods

RONA, a.s., Lednické Rovne, Schreiberova 365, SK-020 61, Slovakia

Glassy inhomogeneities in oxide glasses represent one of the serious constraints of industrial glass production quality. Terms such as cord, striae and knots are used to describe such inhomogeneities. In all cases, these represent regions of the glass which differ in composition from the surrounding matrix. The origins of such vitreous defects are manifold, e.g. batch contaminants, evaporation of volatile species, and refractory corrosion [1]. The most evident problem posed by such glassy inhomogeneities resides in decreasing of the optical quality of glass products. Routinely the chemical composition of knots is followed by EDS analysis. The obtained results are commonly evaluated only qualitatively [2], i.e. the enhanced/decreased content of particular oxides/elements is followed. Present paper deals with the application of advanced statistical methods for analysis of chemical composition of a set of knots chemical composition. The basic presumption is that the compositions of knots represented by relative weight amounts of n oxides span a k-dimensional subspace in the n-dimensional space. The dimensionality of this subspace is determined by the Principal Component Analysis (PCA) method. After that the Multivariate curve resolution (MCR) method is used for evaluation the content of k individual components in each knot. MCR is a widespread methodology for the analysis of process data in many different application fields [2]. MCR method is used for decomposing the compositional matrix D(m x n) into the concentration matrix C(m x k) and the matrix of pure components’ composition S(n x k) D = CST + E where k is the number of components, m is the number of knots, and n is the number of oxides. E(m x n) is the error matrix containing the residual variance of the data [2]. In our case the amount of seven (n = 7) oxides (Na