期刊
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
卷 178, 期 -, 页码 -出版社
ACADEMIC PRESS LTD- ELSEVIER SCIENCE LTD
DOI: 10.1016/j.ymssp.2022.109322
关键词
Time-variant reliability analysis; Adaptive Kriging; Estimation variance reduction; Error-based stopping criterion
资金
- Key National Natural Science Foundation of China [U1864211]
- Natural Science Foundation of Shanghai [21ZR1431500]
- Zhiyuan Honors Program in Shanghai Jiao Tong University
This paper proposes an estimation variance reduction-guided adaptive Kriging method (VARAK) for time-variant reliability analysis. The method considers the correlation between time trajectories and introduces a new stopping criterion to improve computational efficiency and accuracy. Experimental results show that the proposed method can achieve fast convergence of estimated failure probability to the true value and appropriately terminate the computation to avoid oversampling when accuracy is high enough.
Adaptive Kriging surrogate model is becoming an effective technique to significantly reduce the computational cost for time-variant reliability analysis (TRA). But the existing Kriging adaptive sampling methods for TRA do not consider the correlation between time trajectories and the maximum error-based stopping criterion may be too conservative, both of which will waste some computationally expensive samples. To address the challenges, we propose an estimation variance reduction-guided adaptive Kriging method (VARAK) in this paper. Firstly, we derive the expression for estimation variance of time-variant failure probability and quantify the contribution of a time trajectory to the estimation variance, which includes not only the individual contribution of the time trajectory but also the contribution of correlation between time trajectories. Based on this, the adaptive sampling strategy selects the candidate that has the maximum contribution to the estimation variance as the new training sample, thereby resulting in high efficiency. To accelerate the training approach, the well-known U function is utilized to identify time trajectories with large uncertainty. Moreover, by taking the Kriging prediction uncertainty into account, a new error-based stopping criterion that approximates the relative error between the estimated time-variant failure probability and its expectation is proposed. Since the computation of the proposed stopping criterion may be time-consuming at the initial stage of the adaptive sampling approach, we use the maximum relative error to identify the initial stage and only calculate the stopping criterion after this stage. Four case studies including three numerical examples and an engineering example are presented to demonstrate the good capability and applicability of the proposed VARAK method. The examples illustrate that VARAK can allow the estimated failure probability to converge to the true value at a fast rate with small fluctuations and the procedure can terminate appropriately to avoid oversampling when accuracy is high enough.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据