Physics-Based Surrogate Modeling Using Response Correction

The surrogate modeling and response correction techniques considered in this book were mostly discussed in the context of design optimization. In such a setup, the primary purpose of the surrogate is to ensure good local alignment with the high-fidelity model, whereas global accuracy of the model is not of a major concern. In a more general setting, i.e., global or quasi-global modeling, the surrogate is to be valid within a larger portion of the design space. This is important for creating multiple-use library models and applications such as statistical analysis, uncertainty quantification, or global optimization. This chapter describes approaches to quasi-global surrogate modeling using physics-based surrogates and response correction techniques. Formulation of the modeling problem is followed by a discussion of global modeling using space mapping, and space mapping enhanced by function approximation layers, as well as surrogate modeling with the shape-preserving response prediction. Finally, feature-based modeling for statistical design is described. The chapter ends with summary and discussion.