RSDS: an efficient adaptive Kriging-based method for structural reliability analysis considering the distance between sample points

This paper presents a novel adaptive Kriging-based method for structural reliability analysis. While most existing active learning methods rely heavily on the mean and variance of Kriging predictions, they often struggle with poor model accuracy in the early stages of Kriging training, which significantly impacts training efficiency. To address this limitation, the proposed RSDS method incorporates both the reliability sensitivity and the distance between sample points to improve prediction accuracy and training efficiency. First, a distance formula incorporating sensitivity as an influence weight is proposed for sample points. Then, the distance between the candidate sample points and the selected points near the limit state surface (LSS) is calculated to determine which candidates are closest to the LSS. While considering the distribution of Kriging model, the candidate training points closer to the LSS are prioritized to ensure the local prediction accuracy near the LSS. Furthermore, a novel stopping criterion is introduced to optimize training termination. The effectiveness of the RSDS method is demonstrated through four examples, showing significant improvements in both accuracy and efficiency compared to existing methods.