Though modeling material cyclic plasticity in a probabilistic way can yield a more accurate result than deterministic modelling method, it is quite hard to perform uncertainty quantification for all the material parameters. In this paper, Sobol’s global sensitivity analysis is applied to cyclic plasticity model to investigate how the variation of these parameters affect fatigue reliability evaluation. A machine learning algorithm is proposed to improve the computational efficiency for failure probability assessment. The results of global sensitivity analysis show that fatigue damage parameters, elastic modulus and initial yield stress are very influential to fatigue reliability analysis while others are not.

尽管以概率方式对材料循环塑性进行建模可以产生比确定性建模方法更准确的结果，但很难对所有材料参数进行不确定性量化。本文将Sobol的全局敏感性分析应用于循环塑性模型，研究这些参数的变化如何影响疲劳可靠性评估。提出了一种机器学习算法来提高故障概率评估的计算效率。全局敏感性分析结果表明，疲劳损伤参数、弹性模量和初始屈服应力对疲劳可靠性分析的影响很大，而其他参数则没有。