Abstract Civil structures might be subjected to blast loads during their lifetime. Since explosion is a low probability high consequence event, the damage assessment of structural system against blast loads needs a probability evaluation approach. This study presented a reliability analysis of steel frame structure against blast loads using Bayesian logistic regression method. Fragility functions parameterized on material properties and blast wave characteristics are developed and further utilized to assess the collapse risk of the structure. The imbalanced data set used in developing logistic regression model and the prediction uncertainty due to the limited data size are particularly stressed. The imbalanced data set is treated by generating synthetic data points for samples that are most likely to be misclassified. The prediction uncertainty is considered by constructing confidence intervals based on the posterior samples generated from Markov Chain Monte Carlo simulations. The developed risk assessment approach is applied to a 10-story steel frame as a numerical example. The probability of structural collapse is obtained and compared with the results from subset simulation method in conjunction with finite element analysis. Results show that the developed parameterized fragility function is accurate in predicting the probability of structural collapse against blast loads compared with the FE-based reliability method. The closed-form fragility function can be implemented to study the effect of parameter variations on the safety of the structure without additional FE simulations.

中文翻译：

---

基于贝叶斯逻辑回归法的钢框架结构爆炸荷载参数化易损性分析

摘要 土木结构在其使用寿命期间可能会承受爆炸载荷。由于爆炸是一种低概率高后果事件，结构系统对抗爆炸载荷的损伤评估需要一种概率评估方法。本研究介绍了使用贝叶斯逻辑回归方法对钢框架结构抵抗爆炸荷载的可靠性分析。开发了以材料特性和冲击波特性为参数的脆性函数，并进一步用于评估结构的倒塌风险。特别强调了用于开发逻辑回归模型的不平衡数据集以及由于数据量有限而导致的预测不确定性。通过为最有可能被错误分类的样本生成合成数据点来处理不平衡的数据集。通过基于马尔可夫链蒙特卡罗模拟生成的后验样本构建置信区间来考虑预测不确定性。作为数值示例，开发的风险评估方法应用于 10 层钢框架。得到结构倒塌的概率，并与子集模拟方法结合有限元分析的结果进行比较。结果表明，与基于有限元的可靠性方法相比，开发的参数化脆性函数在预测结构在爆炸荷载作用下倒塌的概率方面是准确的。可以实施闭合形式的脆性函数来研究参数变化对结构安全性的影响，而无需额外的有限元模拟。