AKOIS: An adaptive Kriging oriented importance sampling method for structural system reliability analysis
AKOIS: An adaptive Kriging oriented importance sampling method for structural system reliability analysis
复制标题
AKOIS：一种用于结构系统可靠性分析的自适应克里格导向重要性采样方法
DOI：
10.1016/j.strusafe.2019.101876
复制
发表时间：
2020
影响因子：
5.8
通讯作者：
John Dalsgaard Sørensen
中科院分区：
文献类型：
--
作者：
Xufang Zhang;Lei Wang;John Dalsgaard Sørensen
结构可靠性分析的一个主要问题是在理想情况下基于少量的模型评估来确定准确的失效概率估计结果。在这方面，基于主动学习克里金法的重要性采样方法受到了广泛的关注。然而，最可能失效点（MPP）作为唯一采样中心的效用限制了其在多 MPP 问题上的潜在应用。为此，本文提出了一种自适应克里金导向重要性采样（AKOIS）方法。 AKOIS过程的外层主动学习循环用于识别重要性采样中心，而其内层循环是基于以采样中心为中心的一个相当小的子区域来实现的，以获得新的训练样本。此外，局部主动学习迭代的数值收敛将立即触发另一轮外部全局搜索新的重要性采样中心。在这方面，确定的重要性采样中心能够自适应地覆盖所研究的极限状态表面的所有分支，以进行结构系统可靠性分析。数据驱动的重要性采样方法的工程应用通过文献中的几个系统可靠性示例进行了演示。
A major issue in the structural reliability analysis is to determine an accurate estimation result of the failure probability ideally based on a small number of model evaluations. In this regard, the active-learning Kriging based importance sampling method has been received considerable attentions. However, the utility of the most probable failure point (MPP) as the unique sampling center has limited its potential applications for multi-MPP problems. To this end, the paper presents an adaptive Kriging oriented importance sampling (AKOIS) approach. The outer active-learning loop of the AKOIS procedure is used to identify importance sampling centers, whereas its inner-loop is realized based on a rather small subregion centering at the sampling center to gain new training samples. Besides, numerical convergence of local active-learning iterations will immediately trigger another round of outer global search for a new importance sampling center. In this regard, the determined importance sampling centers are able to adaptively cover all branches of the investigated limit-state surface for structural system reliability analysis. Engineering applications of the data-driven importance sampling approach are demonstrated by several system reliability examples in the literature.