A sequential sampling approach based on LOLA-DIST criterion

Abstract

To solve the parameter optimization problem of quality control in the industrial manufacturing process, the high fidelity response surface model should be established to describe the nonlinear relationship between the input and output factors, which requires numerous expensive and time-consuming simulation calculations. Therefore, a sequential experimental design method based on LOLA-DIST (Local Linear Approximations, LOLA, the intersite and projected distance, DIST) criterion is proposed, which sequentially samples in zones with high nonlinearity and low distribution density and better balances global exploration and local exploitation strategies. A numerical example is given to verify the feasibility of the LOLADIST sequential sampling approach. The results show that compared with the existing sequential design methods, the sequential design based on the LOLA-DIST criterion achieves the target prediction accuracy with less computational cost and sample size and works for any metamodels.