2.3.1 Run Matrix Generation



next up previous contents index
Next: 2.3.2 Polynomial Approximation Algorithm Up: 2.3 Empirical Model Building Previous: 2.3 Empirical Model Building

2.3.1 Run Matrix Generation

The run matrix can be generated using different strategies.   All strategies sample the parameter region to be approximated as specified by the user. For instance, a full factorial design is built by first dividing each parameter interval into subintervals where k is the dependency order of the output on that parameter. The run table is then formed by taking all possible combinations of parameter values. The main disadvantage of the full factorial design is the exponential increase in the number of runs with the number of model parameters. Different designs are used to limit the number of runs. Fractional factorial or blocking designs can limit the number of runs at the expense of cofounding some of the parameters interaction. For quadratic order models, composite designs strike a similar balance. They consist of a factorial or fractional factorial design to estimate first order interaction augmented with points to estimate the quadratic terms.

A large number of statistically based experimental designs [68][15] exists. To allow for the generation of run tables for any of these designs, a Design Of Experiments (DOE) table specification is implemented. A DOE table is encoded as an matrix D, where l is the number of runs and n the number of the parameters and . The run table is generated from the DOE table using the following transformation:



Martin Stiftinger
Tue Aug 1 19:07:20 MET DST 1995