|
A response surface design that has points at:
- the center
- the midpoint of each side
There are no corner points. The design is spherical, all the points (except the center point) lie on a sphere of radius 
See Central Composite Design
A response surface design that has points at:
- the center
- the corner points
- axial points at 'a' from the center
The design is usually rotatable, if the correct value of a is selected
A technique used in the Design of Experiments to find the maximum or minimum, if it is not contained within the initial parameter settings. It involves:
- carrying out an initial experiment and, if it does not contain the [maximum/minimum],
- finding the path of steepest [ascent/descent]
- carrying out a test at intervals point along the path until it starts to [ascend/descend]
- carry out an experiment around the highest point on the path to see if it is the [maximum/minimum]
- if not find the path of steepest ascent, and continue the search
The experiments may be first order designs with center points. If the curvature is non-significant the experimental region does not contain the center point and the path of steepest ascent can be found. If the curvature is large the additional testing required for a second order design is carried out.
Types of Experimental Design that investigate curvature of the response surface. They include CCD and Box Behnken designs and are used, among other things, to minimize or maximize the response from the experiment. This cannot be achieved with the linear regression models of the full and fractional factorial designs because the minimum or maximum of a linear response surface always occurs at a corner point.
See Hill Climbing
|
