SEQUENTIAL PLANNING OF EXPERIMENTS

Test patterns for models linear in the parameters can be planned in advance and are highly developed for polynomial response-surface models. Experimental designs of this type are well treated, for instance, in Box and Draper (1987) and in Box, Hunter and Hunter (1978, 2004). 

Models nonlinear in 6 can be studied economically in a sequential manner once an exploratory test pattern has been run. Computer simulations based on the current information can then be used to select the next test condition from a set of candidate conditions. The resulting economy of experimental effort can be substantial, as noted by Rippin (1988). 

The selection strategy depends on the goals of the investigator. Normally, one or more of the following goals is chosen  

Optimal estimation of functions of the parameters, such as operating states or expected economic return. 

Recommended strategies for these goals are described below. 

---

The package GREGPLUS, included in Athena Visual Studio, tests user-provided models against data on one or more observed response types and assists sequential planning of experiments. Nonlinear and overparameterized models are handled directly. The following types of information can... 

---