An adaptive Kriging reliability analysis method based on novel condition likelihood function

To carry out the reliability analysis, whose performance functions are presented in a nonlinear form, many studies propose the reliability analysis methods involving the active Kriging model. Though some learning functions have been developed to refine the Kriging model around the limit state surface (LSS) effectively, most of them rely on the Kriging predictor and its variance. In this research, a new learning function, formed by the combination of the conditional likelihood function and clustering constrain function through adaptive weight coefficient, is raised to reconstruct Kriging by the candidate samples near the LSS. With the conditional likelihood function, the likelihood that the Kriging predictor reaches the LSS mainly contributes to the selection of the best next point. Three numerical applications with different complexities are used to investigate the validity of the proposed reliability method. In addition, the performance of the proposed reliability method is tested by an engineering application.

Automated summary

This research proposed a new learning function formed by the combination of the conditional likelihood function and clustering constrain function for reconstructing Kriging, aiding in selecting the best next point.