The benefits of the calculation of physical glass properties are demonstrated in articles by Bingham and Marshall (' Reformulation of container glasses for environmental benefit through lower melting temperatures' Glass Technol., vol. For comparison: Services of other organizations concerning the calculation of glass properties are offered (see Glass Links, Calculation of Glass Properties), however, the basics of the calculations are not given sufficiently and therefore the predictions might be questionable. Further details are summarized in statistics textbooks and at Principles of Statistical Glass Modeling. Transparency: The statistical calculation of physical glass properties is based on mathematical methods, established over several decades and well comprehensible. The results of experiments inside and outside conventional ranges can be predicted and compared economically. Consequently, the model allows predictions and glass development in composition areas that are not covered by common industrial models that are valid for only one specific glass type. The statistical modeling approach may include several glass types with different chemical compositions it makes them comparable. in ' High temperature glass melt property database for process modeling'.īroad application range: The accuracy of new measurements mostly can not be derived directly from similar results in the scientific literature, or from NIST or DGG glass property standards. Previous models can be integrated, e.g., Lakatos et al.

Besides property-composition models other functions can also be considered in the optimization procedure such as financial estimators or processing calculations.Ĭompatibility: Statistical glass modeling makes data comparable over wide composition ranges, different measurement techniques, and from various investigators. Optimization: Several models can be combined and reversed for determining the best glass composition from a desired set of physical properties. In contrast, one single experimental investigation may take several hours or days, including high personal and equipment expenses, and the measurement accuracy still needs to be confirmed statistically afterwards by comparing results with those of other investigators. Time and financial savings: Calculations can be completed within a few minutes. In often investigated composition areas the large number of data does allow a higher prediction accuracy than measurements within one single laboratory permit. Superior accuracy: Because the models are based on many datasets including various measurement techniques, systematic errors of specific investigators can be analyzed and resolved. The statistical analysis of global glass databases gives new insights that are difficult to obtain otherwise, such as an improved understanding of the mixed-alkali effect for the viscosity of glasses.Īdvantages of the statistical modeling approach for the calculation of glass properties: The procedure involves a leverage analysis, i.e., glasses with high influence on the model and/or unusual compositions are analyzed in particular and excluded from the calculation if in doubt. Therefore, it is unlikely to obtain inaccurate results through unintentionally entering a glass composition for which the model is not valid for. The model application limits, considering component combination limits are evaluated in detail.

In the statistical property calculation approach glass property standards are included, e.g., NIST 710A, 710, 717A, 711, 731, DGG I, which ensures high accuracy (see viscosity model comparison). 95%, and the uncertainty of the glass composition of interest (see viscosity calculator). It also includes an error estimation for glass in mass production, given the desired error confidence level, e.g. The statistical glass modeling does not only result in a prediction of the property value itself. Systematic differences between laboratories are tested and if necessary corrected. The analysis of global glass databases does allow a significant reduction of systematic errors through comparative glass modeling, i.e., through investigation of categories as described in the linear regression tutorial (PDF, 0.4 MB) on page 11. Outlying datapoints are excluded from the calculation. Innovation compared to previous glass property calculation models:įor statistical glass modeling up-to-date global glass databases are used, i.e., property values from a high number of sources (laboratories).

Density of glass