| Abstract: | The output of an industrial plant (such as a chemical plant) will in general depend on "plant conditions" i.e., factors characterizing the process such as the temperatures at which stages of the process are run, the concentrations at which chemicals are applied, the concentrations at which catalysts are used, etc. In order to 'improve' such a process, it is desirable to estimate the functional dependence of the target output y on the plant conditions denoted by x(1), x(2) ..., x(n). For this purpose is it customary to use data from a planned experiment usually using a "pilot plant" in which the plant conditions x(i) are deliberately determined by the experimentor in accordance with an experimental design. In such situations the plant conditions at which the pilot plant is run are often "optimized". The criterion of optimization here used is the "generalized variance" of the estimated coefficients occuring in the mathematical law representing the dependence of the output on the inputs. The particular type of experimental design considered in this technical report are, however, restricted to satisfy a well established pattern known under the name of a "Composite Design" and the parameters describing such a composite design are optimized in the sense of minimizing the generalized variance. The composite designs considered are both of a symmetrical and asymmetrical type and may or may not involve the use of fractional factorials. |
|---|