Response Features for Low-Cost Behavioral Modeling

This chapter demonstrates that feature-based approaches may be effectively used in the context of general-purpose surrogate modeling. By exploring the common properties of characteristic points, in particular, weakly nonlinear dependence between feature points coordinates and geometry/material parameters of the system of interest, it is possible to improve the computational efficiency of the surrogate modeling process. This can manifest itself in a reduction of the number of the training points necessary to render a reliable model, or by improving the predictive power of the surrogate without increasing the training dataset size. Subsequent sections elaborate on specific modeling techniques following this paradigm. Chapter 13 will address utilization of response features for constrained (performance-driven) modeling.