Combination of Latin Hypercube Sampling and K-Nearest Neighbors for Structural Reliability Analysis

Precise calculation of failure probability is a pivotal aspect of structural reliability analysis. However, numerical determination of the failure probability can involve challenges. This results in the need for approximation or simulation methods. Monte Carlo Simulation is a widely utilized accurate approach, and Latin Hypercube Sampling is an alternative to enhance its efficiency. Nevertheless, estimation of the failure probability by simulation methods is a time-consuming process in general. Thus, an approach that reduces the computation time is actually required. The K-nearest neighbors is an algorithm in machine learning than could be applied to save time. In this paper, a large number of generated samples of Latin Hypercube Sampling is to be predicted by an effective K-nearest neighbors algorithm. Therefore, the number of function evaluations in simulations, and consequently the analysis time, could drastically decrease especially in complex structures. The performance of the combination of the Latin Hypercube Sampling and K-nearest neighbors algorithm in accuracy and efficiency has been investigated in several mathematical and structural problems.

کلیدواژه ها:

Structural Reliability ، Latin Hypercube Sampling ، K-nearest neighbors ، Failure Probability.