In the course of an iterative optimization process modelled by approximations, new
points must be generated in specified domains of the design variable space. A new
scheme for the design of experiments (Van Keulen and Toropov, 1999) has been
formulated with the following characteristics:
1. The scheme works efficiently even if only a single additional design point is
generated to the existing plan. For a number of new design points, the algorithm
is used several times.
2. The scheme remains effective if different types of functions are used within the
same optimization task to approximate the objective function and the constraints.
The approach distributes points as homogeneously as possible in the subdomains
of interest. This is done by the introduction of the following cost function: