Taking a Closer Look at Glass Strength Data

The inherent nature of glass strength, namely its low tensile strength with high variability, and decreasing strength with increasing size, duration of load and age of the glass, has been explained by the existence of microscopic flaws in the surface of the glass. Predicting the strength of glass panels is difficult, not only due to the nature of the glass itself, but also due to the fact that when glass panels are subjected to high loads, the relationship between the applied loads and the resulting stresses becomes non-linear. Sophisticated computer programs have been developed in recent years enabling a solution of the problem to be obtained. These programs calculated the stresses over the surface of the glass panels and then a Weibull probability distribution was used, to approximate the variability in glass breakage data and to predict the probability of breakage of the glass panels at a given load. However, the authors have now found that, a lognormal probability distribution, gives a much closer representation of the glass breakage data than a Weibull distribution. They have also found that by using a new formulation to approximate the non-linear relationship between load and stress, it is possible to derive the probability distribution of crack depth. The new formulation has the advantage that it enables the probability of breakage of panels to be readily calculated and linked to parameters having greater physical significance, thus leading to a deeper understanding of the nature of glass strength. The authors have examined the glass breakage data of other researchers in more detail, using the new formulation and have found that the glass strength charts published in ASTM1300 appear to be based on the combined distribution of strength for the outdoor and indoor surfaces of 20 year old glass taken from a building. However, if the glass strength is differentiated between the inside and outside surfaces, and the probability distribution for the weaker surface is used, the design charts of ASTM1300 may not be conservative, even though the use of a Weibull distribution is slightly more conservative than when using a lognormal distribution.