A gradient-based nested Latin hypercube DOE for quadratic polynomials RSM in FEM simulation

A nested Latin hypercube DOE (Design Of Experiment) involves at least a low accuracy experiments and a high accuracy one in Finite Element Method (FEM) Simulation. How to use the information contained in the regression model of previous low accuracy experiments in the design of high accuracy one needs deeply study because such information may be ignored in evenly sampling method, leading to a imprecise regression model of high accuracy experiment. This paper employ the gradient of the regression model of low accuracy experiments as the index of the information distribution of tested object to adjust the DOE of high accuracy one. Thus more information of object can be obtained in those high accuracy experiments and a more precise quadratic polynomials RSM (Response Surface Method) model is built.