基于主动学习Kriging模型的结构多失效模式可靠度计算

Structural reliability calculation for multiple failure modes based on an active learning Kriging model

对于具有多失效模式的结构可靠度计算问题,利用多输出Kriging模型作为代理模型进行分析。该代理模型只需对所有功能函数进行一次建模,无需对每个功能函数建立各自的代理模型,且在建模过程中能够考虑各失效模式之间的相关性。本文方法设定的初始样本点不仅对随机变量均值附近区域给予足够重视,而且能够兼顾设计空间的边缘区域,进而确保初始代理模型在全局空间内具有较好精度,以减少后续利用学习函数更新代理模型的次数。数值算例表明,本文方法具有较好的计算精度和较高的计算效率,当失效模式较多时,计算效率大幅提升。

A multi-output Kriging model is used as the surrogate model to solve a problem of structural reliability calculation with multiple failure modes.In this study,the surrogate model is constructed only once for all performance functions,without having to construct a separate surrogate model for each function,and the correlation between failure modes can be considered in the modeling process.The initial sample points given by the proposed method consider not only the region near the mean of the random variables,but also the edge region of the design space,and the initial surrogate model has better accuracy in the global space,so that the number of updating the surrogate model by using learning functions is reduced.Numerical examples show that the proposed method can achieve satisfactory accuracy,and the proposed method can greatly improve efficiency especially for a large number of failure modes.