摘要
工程结构在服役期不可避免地遭受各种不确定因素的侵害,为客观科学地描述不确定因素的影响,可靠度理论得以产生和发展。但传统的可靠度分析方法囿于精度和效率等原因,难以应用于实际工程结构。近年来基于概率守恒原理,概率密度演化理论提出并得到发展。但对复杂的工程结构,效率低下等问题依然有待解决。鉴于此,本文提出两级剖分概率空间的概念,以粗剖分获得的少量代表样本构成训练集,训练Kriging模型;然后细剖分概率空间,通过训练的Kriging模型预测加密样本响应提高分析效率。基于此,结合以概率密度演化理论为基础的物理综合法,提出了一种新的可靠度分析方法。通过对一解析系统和一幢钢筋混凝土框架结构的响应预测与可靠度分析,证明了新提出方法的精度和效率。
Engineering structures inevitably suffer from various uncertainties during their service life,and in order to objectively and scientifically describe the influence of these uncertainties,the reliability theory of engineering structures has been developed.However,it is difficult to widely apply traditional reliability analysis methods to actual engineering structures due to accuracy and efficiency issues.In recent years,based on the principle of probability conservation,probability density evolution theory was proposed and has been well developed.Compared with the traditional reliability analyses theory,the probabilistic density evolution method can greatly reduce the number of deterministic structural analyses and obtain satisfactory accuracy.However,for increasingly complex engineering structures,the problems such as inefficiency remain to be solved.To this end,we proposes the concept of two-stage partition of probability space,in which a small number of representative samples obtained by coarse partition are used to form a training set to train the Kriging model;then the probability space is finely profiled,and the trained Kriging model predicts the encrypted sample response to improve the analysis efficiency.On this basis,a new reliability analysis method is proposed by combining the physical synthesis method based on the probability density evolution theory.The applicability of the new proposed method is demonstrated by predicting the response and reliability of an analytical system and a reinforced concrete frame structure.
作者
周锦
李杰
ZHOU Jin;LI Jie(College of Civil Engineering,Tongji University,Shanghai 200092,China;Shanghai Institute of Disaster Prevention and Relief,Shanghai 200092,China)
出处
《计算力学学报》
CAS
CSCD
北大核心
2023年第5期686-692,共7页
Chinese Journal of Computational Mechanics
基金
国家自然科学基金(51538010)资助项目。
