Glass viscosity calculation based on a global statistical modelling approach

A viscosity model for predicting the complete viscosity curve of glass was developed using a global statistical approach and more than 2200 composition-viscosity data for silicate glasses collected from over 50 years of scientific literature, including soda-lime-silica container and float glasses, TV panel glasses, borosilicate fibre wool and E type glasses, low expansion borosilicate glasses, glasses for nuclear waste vitrification, lead crystal glasses, binary alkali silicates, and various other compositions. It is shown that, within a measurement series from a specific laboratory, the reported viscosity values are commonly overestimated at higher temperatures due to evaporation losses during the measurement and glass preparation; this includes data presented by Lakatos et al (1972) and the recent 'High temperature glass melt property database for process modelling' by published by Seward et al (2005). Similarly, in the glass transition range many experimental data reported for borosilicate glasses are too high due to phase separation effects. The global model developed here corrects those errors. The model standard error was 9-17°C, with R 2 =0.985-0.989. The prediction confidence interval in mass production largely depends on the glass composition of interest, the composition uncertainty, and the viscosity level. The mixed alkali effect for the viscosity is derived from nonlinear alkali-silica interactions in binary systems.