Monte Carlo simulation using support vector machine and kernel density for failure probability estimation

Monte Carlo simulation requires a large number of sampling points. In a Monte Carlo simulation, the performance function is calculated for all sampling points to determine the failure of the design. If the calculation of the performance function involves large numerical models, a tremendous numerical cost is inevitable. In this study, a support vector machine was applied as a metamodel of the performance function to overcome this drawback. Kernel density and a modified margin of the support vector machine were used for the active learning of the support vector machine. The proportion of the support vector machine's modified margin in the design space was applied as the criterion to end active learning. The proposed method is applied to some numerical examples and examined.

Automated summary

In this study, a support vector machine was used as a metamodel to reduce the numerical cost of large numerical models in Monte Carlo simulations. A modified margin of the support vector machine was utilized for active learning, with the proportion in the design space serving as the criterion to end the learning process. The proposed method was successfully applied to various numerical examples in the study.