A Selection Strategy for Kriging Based Design of Experiments by Spectral Clustering and Learning Function
Publication: ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part B: Mechanical Engineering
Volume 7, Issue 2
Abstract
Reliability analysis evaluates the failure probability of structures considering random variables of a system. Existing methods such as first-order reliability method (FORM) and second-order reliability method (SORM) are difficult to predict the failure probability of implicit functions in mechanical structures. Monte Carlo simulation (MCS) can predict the failure probability with high accuracy, but it is time-consuming. Agent-based methods such as the Kriging model have the approved performance to predict the failure probability in both efficiency and accuracy. An active method is proposed in this paper to improve the efficiency of predicting the probability of failures by combining the Kriging model and MCS, using a new learning function and its stopping condition. A representative selection strategy is developed based on spectral clustering to decide sample points in the design of experiments (DoEs). The new learning function integrates uncertainty and similarity of predicted Kriging values to search the next best sample point for updating the initial DoE. The learning process is terminated based on the stopping condition for a given accuracy of predicted probability of failures. Four case studies are conducted to validate the proposed method. Results show that the proposed method can predict the probability of failures with improved accuracy and reduced time. This article is available in the ASME Digital Collection at https://doi.org/10.1115/1.4050160.
Information & Authors
Information
Published In
Copyright
Copyright © 2021 by ASME.
History
Received: May 7, 2020
Revision received: Sep 23, 2020
Published online: Apr 23, 2021
Published in print: Jun 1, 2021
Authors
Metrics & Citations
Metrics
Citations
Download citation
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.