Artificial neural network analysis of optical measurements of glasses based on Sb2O3

Abstract

In the paper we present application of artificial neural network (ANN) on relation between glass composition versus optical transmittance of the chosen glass systems of Sb2O3 - PbCl2 and Sb2O3 – PbO – M2O, where M was Na, K and Li, respectively. The excellent prediction ability of special ANN program developed for this study demonstrates the possibility to influence the glass composition to obtain asked optical properties. The measurements of the temperature dependencies of the direct electric conductivity show the strong influence of the concentration of the individual glass compounds of systems Sb2O3 - PbCl2 and Sb2O3 – PbO – M2O (M is Na, K, Li) on their electric and dielectric properties. Glasses own the same mechanism of the electric conductivity with activation energy, which goes to the value 3.75 eV when temperature is higher than 250 C. Similarly optical transmittance T of systems Sb2O3 - PbCl2 and Sb2O3 – PbO – M2O strongly depends on the glass composition and the amount of defects, too. The glass 70Sb2O3 – 30PbCl2 reached the highest value of T. The minimal content of defects in its volume makes these glasses very perspective for next searching. The measurements of the complex modulus M of mentioned glasses showed their high sensitivity to the changes of glass structure connected with the creation of different sort and the amount of defects. The sensibility of the used methods is comparable with the usual exploited methods (X-ray analysis, optical microscopy) and makes possible to assess partially the quantitative occurrence of defects in the glass volume. A model of neural network for prediction of the optical transmittance was created. Model enables to predict the transmittance with sufficiently small error. After evaluation of results we can state that exploitation of neural networks is advantageous, if it is necessary to express complex mutual relations among sensor-based data. Neural networks are able to realize and appropriately express general properties of data and relations among them and on the contrary to suppress relationships which occur sporadically or they are not sufficiently reliable and strong. Their usage enables retrieval of relationships among parameters of the process which with use of common methods are not possible to trace for reason of their mutual interactions, big amount and dynamics. Use of a neural network seems to be suitable tool for estimating different important optical parameters.