Global sensitivity analysis using moving least squares for models with correlated parameters

Because of the strong flexibility of the moving least-squares approximation in nonlinear models, a new global sensitivity analysis method using moving least-squares is proposed to decompose the contribution of uncertainty to model output by an individual input variable into uncorrelated (i.e., contributed by the variations of an input variable that are uncorrelated with other input variables) and correlated (i.e., contributed by the variations of an input variable that are correlated with other input variables) contributions. Furthermore, the partial correlated contributions by every two correlated input variables are obtained, and then a contribution matrix is defined for engineering convenience. Both linear and nonlinear numerical examples are employed in this paper to demonstrate the ability of the proposed method. Then, the proposed global sensitivity analysis method is applied in a practical engineering problem related to an aircraft structure made from composite laminates; from the contribution matrix obtained, the most efficient method to decrease the response variance is determined and the efficiency is demonstrated by Monte Carlo simulation.