Component metamodel synthesis for the construction of master response surfaces Original Research Article

Response surfaces or metamodels have long been used as cost effective surrogates for complex numerical simulations or as analytical representations of experimental data. In standard practice, a single analytical model is used within a simulation or fit to experimental data. In reality, several different models may be appropriate for consideration. A new method is presented for combining a set of alternative metamodels into a single master response surface or Super-Metamodel (SM) that is valid over all of input parameter space. The SM is formed as a linear combination of competing component metamodels where each model is weighted by its predicted probability.