X-Ray Patterns Identification of Crystallized Sodium Disilicates Mixtures

Abstract

Listen

Translate article

The amorphous sodium disilicate transformation into a crystallized product proceeds on several paths, finally resulting a mixture of crystalline α, β, δ, γ polymorphs. The identification of individual crystalline phases is impaired by the overlapping of X-ray patterns corresponding to various polymorph species. A quite different way from the classical Rietveld method of X-ray diffraction data analysis is proposed: qualitative and quantitative identification of polymorph phases is approached as a classification problem with satisfactory results using two types of neural nets. A backpropagatio n network with post-processing of the outputs and a neural net based on the adaptive resonance theory have been applied with equivalent results. In comparison with Rietveld method, this original approach can be considered a short-cut technique requesting no fundamental data. The reason of this work is to be a background support in assessment procedures concerning the quantitative evaluation of interdependence between the crystallization parameters and the desired composition of the solid phase mixture.