Surrogates for numerical simulations; optimization of eddy-promoter heat exchangers Original Research Article

Although the advent of fast and inexpensive parallel computers has rendered numerous previously intractable calculations feasible, many numerical simulations remain too resource-intensive to be directly inserted into engineering optimization efforts. An attractive alternative to direct insertion considers models for computational systems: the expensive simulation is evoked only to construct and validate a simplified input-output model; this simplified input-output model then serves as a simulation surrogate in subsequent engineering optimization studies. We present here a simple “Bayesian-validated” statistical framework for the construction, validation, and purposive application of static computer simulation surrogates. As an example, we consider dissipation-transport optimization of laminar-flow eddy-promoter heat exchangers: parallel spectral element Navier-Stokes calculations serve to construct and validate surrogates for the flowrate and Nusselt number; these surrogates then represent the originating Navier-Stokes equations in the ensuing design process. A validation-based a posteriori error framework serves to quantify the effect of surrogate-for-simulation substitution.