Latin Hypercube Sampling for Stochastic Finite Element Analysis

Abstract

A Latin hypercube sampling method, including a reduction of spurious correlation in input data, is suggested for stochastic finite element analysis. This sampling procedure strongly improves the representation of stochastic design parameters compared to a standard Monte Carlo sampling. As the correlation control requires the number of realizations to be larger than the number of stochastic variables in the problem, a principal component analysis is employed to reduce the number of stochastic variables. In many cases, this considerably relaxes the restriction on the number of realizations. The method presented offers the same general applicability as the standard Monte Carlo sampling method but is superior in computational efficiency.