Computationally inexpensive metamodel assessment strategies

In many scientific and engineering domains, it is common to analyze and simulate complex physical systems using mathematical models. Although computing resources continue to increase in power and speed, discipline-specific computer simulation modules continue to grow in complexity and remain computationally expensive, limiting their use in design optimization. The use of different approximation strategies as inexpensive metamodels of the discipline-specific simulation models has led to the development of various metamodel-based integration frameworks and associated research topics. In particular, integration of the discipline-specific metamodels requires an assessment of the overall system error based on the individual approximation errors. As a result, there is a need to develop efficient methods to assess metamodel fidelity at the system and subsystem level. In this paper, we investigate computationally inexpensive assessment methods for metamodel validation at the subsystem level and evaluate a two-stage validation approach on two classes of test problems: 1. Three response functions from a Boeing simulation model, and 2. two response functions from a set of problems for testing optimization codes. Preliminary results indicate that the two stage-validation approach is promising, since it requires no additional computationally expensive disciplinary model evaluations and can provide a practical estimate of the true error measure.