Incorporating invalid test results in glass fracture strength determination

Current standards suggest characterizing the distribution of glass fracture strength using co-axial double-ring standardized tests. While several international standards do not specify a minimum test number, ASTM C1499-19 recommends conducting at least 30 valid tests (defined as tests where fracture initiates in the inner ring). In practice, the results of invalid tests are often discarded. However, the invalid tests still hold information on the fracture strength, and neglecting them can lead to a biased estimation of the latter. This study proposes a Bayesian inference approach to extract and incorporate such information in glass strength characterization. The approach is applied to existing datasets at ambient and elevated temperatures, examining how the test number influences the coefficient of variation of the characteristic fracture strength used for design. The results show that discarding invalid tests may bias glass strength estimates. Furthermore, for the considered dataset at elevated temperatures, incorporating invalid tests enables achieving the same precision level as implied by the standard with fewer tests. These results show that the proposed approach can improve testing efficiency and refine characteristic fracture strength predictions. Finally, the study quantifies the impact of incorporating invalid tests on the structural reliability of a slab under different load ratios. Results show that ignoring invalid tests leads to an overestimation of the reliability index. This finding highlights that invalid tests should be considered for more robust reliability evaluations of glass structures.