Efficient Reliability Analysis via Active Learning: An Integrated Approach Using the RBF+E Algorithm and a Pseudo‐Design Point‐Based Samples Reduction Strategy

• Published in: Quality and reliability engineering international
• Main Authors: Wei, Yan‐Xu, Zhang, Shi‐Long, Wang, Bo‐Wei, Huang, Ying, Zhang, Yan‐Jie
• Format: Journal Article
• Language: English
• Published: 28.07.2025
• ISSN: 0748-8017; 1099-1638
• DOI: 10.1002/qre.70029

The reliability analysis of large and complex structures generally involves high nonlinearity and multiple influencing factors, resulting in significant computational costs that pose challenges for conventional surrogate model‐based methods. To address this issue, this study proposes a novel active learning surrogate model method for reliability analysis. The approach integrates the radial basis function (RBF) with a newly designed learning function E , forming the RBF+E algorithm to reduce the number of performance function evaluations. RBF+E is constructed to ensure effective sampling by concentrating design samples within high‐probability‐density regions, guided by the Euclidean norm of design samples. Furthermore, during modeling procedure, a pseudo‐design point‐based samples reduction (PDPSR) strategy is introduced to further reduce the computational cost by reducing the size of the candidate sample set required for evaluating new design samples and estimating failure probability. Two numerical examples and two complex practical engineering applications are utilized to demonstrate the advantages of the proposed method. Results confirm that the RBF+E algorithm, combined with the PDPSR strategy, significantly reduces computational cost while maintaining precision. The efforts made in this study provide an effective strategy for reliability analysis of large and complex structures.