A platform for research: civil engineering, architecture and urbanism
AKOIS: An adaptive Kriging oriented importance sampling method for structural system reliability analysis
Highlights The paper presents an active-learning oriented importance sampling method. Importance sampling centers are optimized by using the active learning mechanism. The surrogate model can cover all branches of the system limit-state surface. Several case studies have demonstrated engineering applications of the approach.
Abstract 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.
AKOIS: An adaptive Kriging oriented importance sampling method for structural system reliability analysis
Highlights The paper presents an active-learning oriented importance sampling method. Importance sampling centers are optimized by using the active learning mechanism. The surrogate model can cover all branches of the system limit-state surface. Several case studies have demonstrated engineering applications of the approach.
Abstract 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.
AKOIS: An adaptive Kriging oriented importance sampling method for structural system reliability analysis
Zhang, Xufang (author) / Wang, Lei (author) / Sørensen, John Dalsgaard (author)
Structural Safety ; 82
2019-08-07
Article (Journal)
Electronic Resource
English
Ductile structural system reliability analysis using adaptive importance sampling
Online Contents | 1998
|Ductile structural system reliability analysis using adaptive importance sampling
British Library Online Contents | 1998
|