Error Quantification and Control for Adaptive Kriging-Based Reliability Updating with Equality Information
Error Quantification and Control for Adaptive Kriging-Based Reliability Updating with Equality Information
复制标题
DOI：
10.1016/j.ress.2020.107323
复制
发表时间：
2021-03
期刊：
影响因子：
--
通讯作者：
Chi Zhang-;Zeyu Wang;A. Shafieezadeh
中科院分区：
文献类型：
--
作者：
Chi Zhang-;Zeyu Wang;A. Shafieezadeh
通过测量获得的新信息提供了更新系统可靠性估计的机会。对于相等类型信息，可靠性更新是一项艰巨的任务。目前最先进的方法，可靠性更新与平等信息使用自适应克里格（RUAK），集成了自适应克里格过程与平等信息到不等式信息的转换。用于训练克里格模型的停止标准依赖于先验失效概率的估计误差，从而使得后验失效概率中的真实误差可能超过可接受的阈值。本文提出了一种在给定置信水平下估计后验失效概率最大误差的方法。此外，提出了一个两阶段的方法，主动学习和自适应训练的Kriging模型的可靠性更新问题。基于后验失效概率最大误差的新停止准则保证了Kriging方法的准确性，从而保证了可靠性估计的准确性，而两阶段方案避免了不必要的训练，从而提高了可靠性更新的效率。四个数值例子被认为是调查所提出的方法的性能。它表明，该方法提供了设置和满足目标精度的可靠性更新，这是至关重要的应用程序的决策的后果是显着的能力。
New information obtained through measurements provide an opportunity to update estimates of a system's reliability. For equality type information, reliability updating is a daunting task. The current state-of-the-art method, reliability updating with equality information using adaptive Kriging (RUAK), integrates an adaptive Kriging process with a transformation of equality information into inequality information. The stopping criterion for training the Kriging model relies on the estimated error for prior failure probability, thus leaving the potential for the true error in posterior failure probability to exceed acceptable thresholds. This study presents an approach to estimate the maximum error in posterior failure probability for a given confidence level . Moreover, a two-phase approach is proposed for active learning and adaptive training of Kriging models in reliability updating problems. The new stopping criterion based on the maximum error of posterior failure probability ensures the accuracy of Kriging and thus the reliability estimates, while the two-phase scheme avoids unnecessary training hence improving the efficiency of reliability updating. Four numerical examples are considered to investigate the performance of the proposed approach. It is demonstrated that this method offers the ability to set and meet target accuracies for reliability updating, which is critical for applications where the consequences of decisions are significant.