| description abstract | Various stochastic validation metrics have been developed for validating models, among which area metric is frequently used in many practical problems. However, the existing area metric does not consider experimental epistemic uncertainty caused by lack of sufficient physical observations. Therefore, it cannot provide a confidence level associated with the amount of experimental data, which is a desired characteristic of validation metric. In this paper, the concept of area metric is extended to a new metric, namely interval area metric, for single-site model validation with limited experimental data. The kernel of the proposed metric is defining two boundary distribution functions based on Dvoretzky–Kiefer–Wolfowitz inequality, so as to provide an interval at a given confidence level, which covers the true cumulative distribution function (CDF) of physical observations. Based on this interval area metric, the validity of a model can be quantitatively measured with the specific confidence level in association with consideration of the lack of experiment information. The new metric is examined and compared with the existing metrics through numerical case studies to demonstrate its validity and discover its properties. Furthermore, an engineering example is provided to illustrate the effectiveness of the proposed metric in practical satellite structure engineering application. | |
