Sampling Strategies and Metamodeling Techniques for Engineering Design: Comparison and Application

To help engineers perform engineering design and optimization efficiently, this article systematically compares the performance of typical sampling strategies and metamodeling techniques based on nine engineering responses with different levels of dimensionality and nonlinearity. In the comparative study, four metamodels, three characteristics of sample data, and four evaluation criteria are considered. Compared with the existing comparative studies for evaluating different metamodeling techniques, the contributions provided by the comparative study in this article are: (1) sample data with only space-filling property and, both space-filling and projective properties are compared; (2) newly developed radial basis functions-based high dimensional model representation (RBFHDMR) model is involved in the study to compare with other metamodels; and (3) we offer a guideline for the use of sampling methods, and furthermore, rather than empirical recommendations, we introduce and compare some recently developed ensemble strategies for the use of metamodels.