基于主动学习代理模型的结构时空相关可靠性分析方法  

Structural time and space dependent reliability analysis based on active learning surrogate model

作　　者：詹红有 梁峻源 肖宁聪 ZHAN Hongyou;LIANG Junyuan;XIAO Ningcong(School of Mechanical and Electrical Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China)

机构地区：[1]电子科技大学机械与电气工程学院,成都611731

出　　处：《电子科技大学学报》2025年第1期84-90,共7页Journal of University of Electronic Science and Technology of China

基　　金：国家自然科学基金(52475244);四川省科技厅资助项目(2023NSFSC0373)。

摘　　要：结构的极限状态方程常为隐函数,受到不确定性、时间和空间的影响,导致其输出为时变随机场。由于结构失效通常是小概率事件,需要大量有限元模拟,现有方法难以适用。尽管一阶/二阶可靠性分析方法可用,但精度和效率有限。该文提出基于主动学习Kriging代理模型的时空相关可靠性分析方法,旨在提高计算效率。首先推导随机样本在时空域内的极值符号预测概率,然后构建相应的学习函数。通过选择具有最小符号预测概率的候选样本加入训练集,并序列地更新Kriging模型,直至所有候选样本的准确预测概率超过99%,同时计算失效概率。通过3个算例验证了所提方法的效率和准确性,为时空相关可靠性分析提供了新思路。The limit state function of structures is often implicit and influenced by uncertainty,time,and space,rendering its output a time-dependent random field.Structural failure is typically a small probability event,requiring numerous finite element simulations,making existing methods difficult to apply.First-order/second-order reliability analysis methods are available,but their accuracy and efficiency are limited.This paper proposes a time and space-dependent reliability analysis method based on the active learning surrogate model,aimed at enhancing efficiency.Firstly,derive the probability of extreme value sign prediction for a random sample in the time and space domains and then construct a corresponding learning function.Selecting candidate samples with the minimum sign prediction probability to add to the training set and sequentially updating the Kriging model until the accurate prediction probability of all candidate samples exceeds 99%,the failure probability is computed simultaneously.The efficiency and accuracy of the proposed method are validated through three case studies,offering a new way to time and space reliability analysis.

关 键 词：时空相关可靠性 代理模型 主动学习 KRIGING模型 失效概率 

分 类 号：TH11[机械工程—机械设计及理论]