The response surface methodology analysis has been reviewed. RSM can be used
for the approximation of both experimental and numerical responses. Two steps are
necessary, the definition of an approximation function and the design of the plan of
experiments. As concluded in Chapter 2, genetic programming is the method of
choice to find a suitable approximation function and will be described in Chapter 4.
A review of different designs for fitting response surfaces has been given. A
desirable design of experiments should provide a distribution of points throughout
the region of interest, which means to provide as much information as possible on the
problem. This "space-filling" property is a characteristic of three plans: Latin
hypercube sampling, Audze-Eglais and van Keulen. All three plans are independent
of the mathematical model of the approximation. However, Latin hypercube
sampling distributes the points randomly in the space, while Audze-Eglais uses a
distribution based on maximum separation between points. The Audze-Eglais plan
has been chosen in this thesis.
It should be noted that if the model building is to be repeated within an
iterative scheme (e.g. with mid-range approximations), van Keulen’s plan would
become an attractive alternative as it adds points to an existing plan. This thesis is
primarily focused on building global approximations