An improved AK-MCS reliability analysis method with a parallel points selection strategy and application

In the reliability analysis of mechanical structures, the computational expense is predominantly attributed to obtaining authentic response values at sample points when employing precision finite element model simulations or other time-consuming computational methods. This reality highlights the critical challenge in reliability research: achieving accurate computational analysis with minimized data points. This study implements an enhanced AK-MCS active learning reliability analysis method incorporating the Kriging Believer parallel sampling strategy, which specifically accounts for correlations between concurrent sampling points. Comparative studies using benchmark cases demonstrate that this methodology exhibits marked advantages in computational efficiency and accuracy over alternative parallel computing approaches relying on clustering algorithms. When applied to the reliability assessment of planetary gear transmission systems, the proposed method demonstrates notable feasibility and efficiency in addressing real-world engineering challenges where response evaluations necessitate time-intensive finite element simulations. This advancement establishes a novel methodological framework for reliability analysis of complex mechanical assemblies.