Accuracy and reliability of large-eddy simulation data in a really complex industrial geometry are invesigated. An original methodology based on a response surface for LES data is introduced. This surrogate model for the full LES problem is built using the Kriging technique, which enables a low-cost optimal linear interpolation of a restricted set of large-eddy simulation (LES) solutions. Therefore, it can be used in most realistic industrial applications. Using this surrogate model, it is shown that (i) optimal sets of simulation parameters (subgrid model constant and artificial viscosity parameter in the present case) can be found; (ii) optimal values, as expected, depend on the cost functional to be minimized. Here, a realistic approach, which takes into account experimental data sparseness, is introduced. It is observed that minimization of the error evaluated using a too small subset of reference data may yield a global deterioration of the results.
Optimal linear interpolation in linear predictive vocoders
