There are many different methods for approximating objective functions in their domain of definition. Obviously it is important during this approximation process to have a good distribution of data into the parameter set.
But why do we have to use approximation and not use directly the model itself? We can resort to approximation when the computation time using the physical model is too large. Considering very complex dynamical non linear models, during the optimisation process more than 98% of the time is spent in simulation and less than 2% for the optimisation algorithm.
A good approximation of the relationships between design variables and objective functions can greatly improve the performances of the optimisation procedure.