A conservatism index based on structural reliability and model errors

Abstract

Engineering models are known to be approximate representations of reality. When approximate models are employed in design, it is usually better when they are (slightly) conservative, rather than the contrary. Out of many recent publications addressing and discussing model uncertainty, we could not find any general method for classifying if a model is conservative, or not; and how conservative it is. In this paper, we propose a model conservatism index which is based on structural reliability theory and aims at quantifying the degree of conservatism of a given structural model. Computation of the proposed conservatism index requires the same data employed for computing model error statistics. The method can be employed to evaluate models used in structural design, and corrected models used in structural reliability analysis. The method is applied to several examples, involving different kinds of simplified models, and shown to accurately classify models as conservative/non-conservative. A simple practical case study is also presented, which includes a nonlinear finite element model, to show that the proposed index may be applied in real-world situations.