🤖 AI Summary
To address the low computational efficiency and poor numerical robustness of conventional Adaptive Kriging–Monte Carlo Simulation (AK-MCS) for small failure probabilities (e.g., ~10⁻⁶), which relies heavily on a candidate sample pool (CSP), this paper proposes a CSP-free adaptive Kriging reliability analysis method. The method eliminates dependence on constrained optimization subproblems and introduces a failure-probability-oriented joint infill criterion combining the U- and EFF-functions. Furthermore, it integrates Monte Carlo importance sampling (IS-MCS) to enhance sampling efficiency. Validated on multiple high-dimensional nonlinear benchmark problems, the proposed approach achieves failure probability estimation errors below 10% using fewer than 20% of the limit-state function evaluations required by standard AK-MCS. This demonstrates substantial improvements in accuracy, numerical stability, and computational efficiency for rare-event reliability analysis.