The proposed rank-based active learning Kriging for efficient structural reliability analysis using weighted average simulation method.

MGR. Faes,M. Rashki,H. Ghohani-Arab,K. Sharifi-Nik

DOI: https://doi.org/10.3850/978-981-18-5184-1_gs-01-219-cd

Abstract:Reliability analysis aims to assess the safety level of engineering systems by considering their inputs as random variables. It is often a time- consuming and/or costly process since performance evaluation of many engineering systems are implicit and complicated. Recently, active learning-based Kriging methods have been extensively developed to improve computational efficiency of reliability analysis. In these ap- proaches, the learning functions aim is to increase the accuracy of surrogate model by locating the experimental samples at the uncertain regions of physical space that have a large potential to improve the quality of prediction through true performance function evaluations. In this study, an adapted approach founded on rank-based learning and the weighted average simulation method (WASM) is proposed to efficiently reduce the required number of performance function evaluations. According to the proposed approach, the samples are ranked based on their contribution to the failure probability through the weighting manner of WASM. While the proposed method locates the experimental samples according to their ranks, an adapted convergence criterion -using adjusted lower and upper bounds of failure proba- bility- checks the convergence rate of WASM method. Three examples are solved by the proposed approach to illustrate its effectiveness.

Engineering