The statistical interpretation of the strength of float glass for structural applications

• A distinction needs to be made between tin- and air-side for float glass.
• Rough data from the coaxial-double ring test with overpressure need to be corrected.
• Test data suggest the presence of a lower limit for glass strength.
• The 2-parameter Weibull statistics cannot well interpret the strength of float glass.
• The 3-parameter Weibull statistics gives the best interpretation for the failure stress.

The characteristic value of the strength of annealed float glass, to be used in structural calculations, is assessed by standards on the basis of a classical experimental campaign using the Coaxial Double Ring (CDR) test with additional overpressure. Experimental data were regressed according to the 2-parameter Weibull distribution, assuming that the induced state of stress is equibiaxial in practice. We show that, by splitting the data in two categories according to the surface under tensile stress (either the “tin” or the “air” side), a more accurate statistical interpretation can be obtained. Comparisons with the normal and log-normal distributions are made with the chi-square goodness-of-fit test. Moreover, we observe that the calibration curve suggested by the test standard is not precise, and that the stress state in the testing configuration is not equibiaxial. Therefore the rough data need to be further corrected and re-scaled to a common reference condition, according to a criterion of equal failure probability, by determining the effective area of the loaded specimen. Doing so, the considered statical distribution are able to fit to the data with much more accuracy. Considering that for very low failure-probabilities the tails of the statistics are the most important, and observing that the 2-parameters Weibull function fails to interpret a lower bound for the strength of glass that can be inferred from experiments, the use of a 3-parameter Weibull distribution is proposed. After having derived the corresponding expressions for the effective area, we present a new statistical characterization of glass strength that provides the best fit with the experimental data, at least for the air-side.