Using RSM for optimisation with discrete (integer) factors

[John Geraghty]

I am using RSM to develop a model for optimal allocation of Kanban cards to production stations in a four-station serial production line. There are, therefore four factors, each one representing the number of cards assigned to a station. The factors are naturally discrete (integer) variables in that it doesn't make sense to assign 1.5 cards to a workstation it should be either 1, 2, 3 etc. Is there a way to treat these factors as integers (especially for optimisation purposes) without making them categorical factors (and therefore been forced to have a very large experimental design that incorporates runs with each possible kanban value included)?

[Tryg]

Thanks for your inquiry. Discrete numeric factors are not available in Design-Expert 7.1, but will be available in our next release, version 8.

[John Geraghty]

Is there a release date for version 8?

[Tryg]

No definite date yet though our target is this autumn.

[Wayne]

In the current version discrete level designs are built using a general factorial design. The design is used to create a candidate set and the candidate set is read by the D-optimal design builder to pull out the necessary runs to model the response surface.

Build the general factorial design. Right-click at the top of each continuous factor and change them to numeric. Under the Design Tools menu select Write Candidate File. Create a new design, save this design if you want, but make sure you click Yes to "Use previous design info". Ignore the warning message (click Ok); click on the Response Surface tab (left) and then on the D-optimal node. Change the design to match the number of numeric and categoric factors, click continue. On the next screen lClick "Read List" (lower left) and select the candidate file you created above. The design will be build using the levels specified.