A novel method for high-dimensional reliability analysis based on activity score and adaptive Kriging

• Published in: Reliability engineering & system safety 2024-01, Vol.241, p.109643, Article 109643
• Main Authors: Wang, Tianzhe, Chen, Zequan, Li, Guofa, He, Jialong, Liu, Chao, Du, Xuejiao
• Format: Article
• Language: English
• Subjects: Active subspace method; Activity score; High-dimensional structural reliability analysis; Interval squeezing method; Parallel analysis
• ISSN: 0951-8320; 1879-0836
• DOI: 10.1016/j.ress.2023.109643

•The proposed method overcomes the curse of dimensionality of adaptive Kriging.•This study proposes a novel interval squeezing method (ISM) and its two variants.•The proposed method can achieve a balance between accuracy and efficiency.•The proposed method can solve high-dimensional problems using parallel resources.•The proposed method can effectively solve complex high-dimensional engineering problems. In structural reliability analysis, Kriging-based adaptive analysis approaches considerably improve the analysis's efficiency by utilizing proper learning strategies. However, the time cost of building a Kriging model may become unacceptable when input spaces have high dimensions. This study proposes a novel method to tackle this difficulty. The basic idea of the proposed method is to implement the adaptive analysis process in the low-dimensional space that activity scores have identified. To determine the activity scores of variables, the active subspace method is implemented. Then, the proposed adaptive analysis method based on the interval squeezing method (ISM) is applied to the suggested low-dimensional space. ISM is designed to improve the accuracy of failure probability with cognitive uncertainty by squeezing its interval. Considering the difficulties associated with applying ISM directly, two variants of ISM are developed: the sequential interval squeezing method (sISM) and parallel interval squeezing method (pISM). Finally, five typical high-dimensional examples (i.e., three numerical and two engineering examples) are investigated to verify the performance of the proposed method. The results indicate that the proposed method can maintain satisfactory accuracy and efficiency.