Determination of uncertainty in the description of crack widths using artificial neural networks

Crack width formulations, such as the one described in Eurocode 2, present a single value to describe the crack width in reinforced concrete structures, but provide limited insight into the uncertainty of the description of crack widths. This paper describes how the uncertainty in the descriptions of these crack widths can be calculated by constructing Prediction Intervals for concrete members loaded in tension. Results showed that accurate and robust prediction intervals could be constructed using artificial neural networks combined with the Lower Upper Bound Estimation method. Furthermore, it was shown that the quantification of uncertainty provides more insight into the description of crack widths.