Critical comparison of design-and analysis-of-experiments methods for MEMS metamodel generation

摘要

We use finite-element simulations of a micromachined gas-flow sensor as an example to demonstrate and critically compare various design- and analysis-of-experiments methods for the generation of simple, surrogate functional approximations. Such approximations can be used to greatly reduce computational expense in MEMS design and optimization. The following design methods are included in our study: factorial, alphabetic (A, D, G, I, and S-optimal), maximin, latin hypercube, central composite, and a method using a new program from the University of Michigan named IMSET. Analyses include both a parametric method (ordinary least squares or OLS) and a non-parametric method (best linear unbiased predictor or BLUP).