Abstract When the general accuracy of a reliability method is not mathematically proven, the correctness of its yielded results may be in doubt. This study emphasizes this observation and proposes a reality-oriented concept for improved structural reliability analysis. The failure probability integral is reformulated based on this insight and two general approaches are presented for probability estimation, namely, probability expectation and control variates. The former is a general interpretation of the Monte Carlo simulation (MCS) based on which the formulation of the existing reliability methods can be used as an indicator function of the MCS, while the latter can remove errors of a reliability method by considering the assumptions employed in it. Using the suggested CV approach and considering the subset simulation as a method of interest, a general sequential framework is proposed for a robust reliability evaluation. Using the presented reality-oriented concept, some popular simulation methods are re-derived and it is shown that the proposed idea can be easily used to derive novel robust reliability methods.

中文翻译：

---

结构可靠性重构

摘要 当可靠性方法的一般准确性没有得到数学证明时，其产生的结果的正确性可能会受到怀疑。本研究强调了这一观察结果，并提出了一个以现实为导向的概念，以改进结构可靠性分析。故障概率积分是基于这种见解重新制定的，并且提出了两种用于概率估计的一般方法，即概率期望和控制变量。前者是对蒙特卡罗模拟（MCS）的一般解释，在此基础上，现有可靠性方法的公式可以用作MCS的指标函数，而后者可以通过考虑所采用的假设来消除可靠性方法的误差在里面。使用建议的 CV 方法并将子集模拟视为一种感兴趣的方法，提出了一个通用的顺序框架来进行稳健的可靠性评估。使用所提出的面向现实的概念，重新推导了一些流行的仿真方法，并且表明所提出的想法可以很容易地用于推导出新颖的鲁棒可靠性方法。