Automatic Tuning of Metamodels for Optimization.

The use of surrogate models is considered a valid alternative to the massive recourse to complex and expensive CFD solvers in design optimization. Many different interpolation and approximation techniques are available in literature, each with its own advantages and drawbacks. Among them, Kriging is gaining credibility due to its flexibility and ease in implementation and training, but also for the theoretical background. However, some crucial parameters need to be assumed, and the final result depends on these choices. In this paper, some guidelines and techniques for the identification of the base elements of a Kriging interpolator are suggested and partly explored.