Quantification of model uncertainties for reinforced concrete columns subjected to fire

In structural fire engineering, several methods co-exist for the determination of the fire resistance of reinforced concrete compression members. However, there is no general agreement about which method should be preferred, and no comparative analysis of the accuracy of the methods is available. One approach to judge on the accuracy of a given method is to apply it for recalculation of well-defined laboratory test and to evaluate the difference between experiments and simulations. The aim of the article is to present a methodology to obtain, from the results of recalculation, relevant statistical key data for the probabilistic characterization of the model uncertainties. Methods of descriptive statistics are applied for the evaluation of the results of the recalculation of laboratory test. In case of prior knowledge from previously published results, methods of inferential statistics are also examined to adjust the statistical key data. In this case, the use of a response surface methodology with descriptive statistics is further presented in order to cross check the results of the inferential method. The application of the presented framework is demonstrated for reinforced concrete columns subjected to a standard fire, but is also applicable for any resistance model in civil engineering.

在结构防火工程中，确定钢筋混凝土受压构件耐火性的方法有多种并存。然而，对于应优先采用哪种方法并没有普遍共识，也没有对这些方法的准确性进行比较分析。判断某一给定方法准确性的一种途径是将其应用于对明确的实验室试验进行重新计算，并评估实验与模拟之间的差异。本文的目的是提出一种方法，以便从重新计算的结果中获取用于对模型不确定性进行概率表征的相关统计关键数据。描述性统计方法被应用于评估实验室试验重新计算的结果。在有先前已发表结果的先验知识的情况下，也会研究推断统计方法以调整统计关键数据。在这种情况下，还进一步介绍了使用带有描述性统计的响应面方法，以便交叉核对推断方法的结果。所提出的框架应用于遭受标准火灾的钢筋混凝土柱，但也适用于土木工程中的任何抗力模型。