A stochastic process discretization method combing active learning Kriging model for efficient time-variant reliability analysis
Dequan Zhang,Pengfei Zhou,Chen Jiang,Meide Yang,Xu Han,Qing Li
DOI: https://doi.org/10.1016/j.cma.2021.113990
IF: 6.588
2021-10-01
Computer Methods in Applied Mechanics and Engineering
Abstract:<p><span><span>Time-variant reliability analysis (TRA) has attracted tremendous interest for evaluating product reliability in <a class="topic-link" href="/topics/engineering/full-life-cycle">full life cycle</a><span><span>. <a class="topic-link" href="/topics/mathematics/discretization">Discretization</a> of stochastic process is considered one of the simplest ways to transform a time-variant problem into a time-invariant problem that becomes easier to handle. Its adoption in time-variant problem, nevertheless, requires overcoming two main issues on (1) the low efficiency of small discrete </span><a class="topic-link" href="/topics/mathematics/time-interval-tau">time interval</a>, and (2) the low accuracy of large discrete time interval. To tackle these two challenges, we propose a Kriging-assisted time-variant </span></span><a class="topic-link" href="/topics/engineering/reliability-analysis-method">reliability analysis method</a><span> based upon stochastic process discretization (namely, K-TRPD for short). First, a complex time-variant reliability problem is converted into conventional time-invariant problem through discretization of stochastic process. Second, the most probable point (MPP) trajectory is approximated through a <a class="topic-link" href="/topics/engineering/kriging-model">Kriging model</a> over the entire time period concerned, whose input is identified from the discrete </span></span><a class="topic-link" href="/topics/mathematics/time-point-eta">time points</a><span><span> by an active learning approach; and the output is obtained by the <a class="topic-link" href="/topics/engineering/first-order-reliability-method">first order reliability method</a> (FORM) at the identified time points. Finally, the constructed Kriging model is utilized for time-invariant reliability analysis at each discrete time point, and the time-variant reliability is obtained by using the time-invariant reliability analysis results for analyzing the </span><a class="topic-link" href="/topics/mathematics/multivariate-normal-distribution">multivariate normal distribution</a> function. In this study, three numerical analysis examples and one engineering design example are presented to demonstrate the effectiveness of the proposed method.</span></p>
mechanics,engineering, multidisciplinary,mathematics, interdisciplinary applications