A new parallel adaptive structural reliability analysis method based on importance sampling and K-medoids clustering. (February 2022)

• Record Type: Journal Article
• Authors: Chen, Zequan; Li, Guofa; He, Jialong; Yang, Zhaojun; Wang, Jili
• Is Part Of: Reliability engineering & system safety. Volume 218:Part A(2022)
• Publication Date: 2022-02
• Subjects: Structural reliability -- Kriging model -- Importance sampling -- Parallel -- K-medoids clustering; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2021.108124
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: A new parallel adaptive structure reliability analysis method—RBIK is proposed. Global convergence condition (GCC) is proposed. The optimal importance sampling function is constructed to evaluate the GCC. Combined with K-medoids clustering, the parallel operations are realized. Compared with AK-MCS, RBIK can significantly reduce the number of iterations. Abstract: In using the Kriging-based adaptive structure reliability analysis methods, the key is to select the appropriate method of adding samples adaptively. In this study, a new parallel adaptive structure reliability analysis method—Reliability Analysis Based on Importance sampling and K-medoids clustering (RBIK)—is proposed. On the basis of the influence of the Kriging model's cognitive uncertainty on the estimation accuracy of failure probability, a global convergence condition (GCC) is proposed. Then, to evaluate the GCC unbiased and efficiently, the optimal importance sampling function will be constructed and used to obtain candidate samples. Considering the spatial correlation of candidate samples, the clustering algorithm is used for the cluster analysis of candidate samples to realize the parallel operation of adaptive structural reliability analysis. Therefore, RBIK is proposed on the basis of importance sampling and K-medoids clustering. RBIK strives to rapidly enable the Kriging model to satisfy the GCC rather than focusing on a single candidate sample, which is the most obvious difference between RBIK and other adaptive structural reliability analysis methods. In addition, RBIK can balance parallel computing power, accuracy, and the number of iterations required. Finally, the effectiveness and robustness of RBIK are proven by several examples.