A weighted three-point-based methodology for variance estimation

It is widely accepted that variations in manufacturing processes are inevitable and should be taken into account during analysis and design processes. However, estimating uncertainty propagation in an end-product caused by these variations is a very challenging task, especially when a computationally expensive effort is already needed in deterministic models, such as simulations of sheet metal forming. The focus of this article is on the variance estimation of a system response using sensitivity-based methods. A weighted three-point-based strategy for efficiently and effectively estimating the variance of a system response is proposed. Three first-order derivatives of each variable are used to describe the non-linear behaviour and estimate the variance of a system. A methodology for determining the optimal locations and weights of the three points along each axis is proposed and illustrated for the cases where each variable follows either a normal distribution or a uniform distribution. An extension of the weighted three-point-based strategy is introduced to take into account the interaction between parameters. In addition, an extension is given for mean estimation of the system response without requiring more data. The considerable improvement in accuracy compared with the traditional first-order approximation is demonstrated in a number of test problems. The proposed method requires significantly less computational effort than the Monte Carlo method.