中国机械工程 ›› 2026, Vol. 37 ›› Issue (2): 428-441.DOI: 10.3969/j.issn.1004-132X.2026.02.018

• 智能制造 • 上一篇    

一种面向多保真Kriging模型结构可靠性分析的主动学习方法

杜尊峰1(), 樊涛2, 姜登耀1   

  1. 1.天津大学水利工程智能建设与运维全国重点实验室, 天津, 300354
    2.航空工业第一飞机设计研究院, 西安, 710089
  • 收稿日期:2024-12-22 出版日期:2026-02-25 发布日期:2026-03-13
  • 通讯作者: 杜尊峰
  • 作者简介:杜尊峰*(通信作者),男,1984年生,教授、博士研究生导师。研究方向为结构可靠性分析、结构损伤评估。E-mail:dzf@tju.edu.cn
  • 基金资助:
    国家自然科学基金(51109158);国家自然科学基金(U2106223);国家重点研发计划(2022YFC2806300);天津市自然科学基金(23JCZDJC01150)

A New Active Learning Method for Structural Reliability Analysis of Multi-fidelity Kriging Models

DU Zunfeng1(), FAN Tao2, JIANG Dengyao1   

  1. 1.State Key Laboratory of Hydraulic Engineering Intelligent Construction and Operation,Tianjin University,Tianjin,300354
    2.AVIC the First Aircraft Institute,Xi’an,710089
  • Received:2024-12-22 Online:2026-02-25 Published:2026-03-13
  • Contact: DU Zunfeng

摘要:

提出了一种基于多保真Kriging模型与主动学习的结构可靠性分析方法。通过三阶段选择确定每次迭代过程中样本点的更新位置与空间位置,第一阶段通过集成多种学习函数确定最优样本点集合;第二阶段通过所提BES方法(beneficial effect strategy)确定样本点的更新位置;第三阶段运用Bootstrap自举抽样法从最优样本点集合中确定样本点的空间位置。通过两个数值算例与一个工程实际算例证明了所提方法的有效性与高效性。与目前先进的多保真结构可靠性方法相比,当模型的保真度较低时能有效地避免计算失败,证明了所提方法的先进性与较好的适用性。

关键词: 结构可靠性, 主动学习, 多保真Kriging模型, 保真度选择策略

Abstract:

A structural reliability method was proposed based on multi-fidelity Kriging modeling with active learning, which determined the computational and spatial locations of sample points during each iteration through a three-stage selection. Firstly, the optimal set of sample points was determined by ensemble multiple learning functions. Secondly, the computational locations of the sample points were determined by the proposed BES(beneficial effect strategy). Finally, the spatial locations of the sample points were determined from the optimal set of sample points by applying Bootstrap sampling method. The effectiveness and efficiency of the method was demonstrated by two numerical examples and one practical engineering example. Compared with the current advanced multi-fidelity model structure reliability method, when the fidelity of the model is lower, the computational failure may be effectively avoided, which shows the advanced and better applicability of the method.

Key words: structural reliability, active learning, multi-fidelity Kriging model, fidelity selection strategy

中图分类号: