DOE - Custom D-optimal Design with complex constraints

[#1]

Hello everyone,

I'v got four continous factors for a D-optimal Design (three levels for second order effects) with a geometrical background. Due to this not every combination of the factors is allowed and simple linear constraints are not applicable. Is it possible to import a table of all allowed combinations of the factors and their three levels and to create a Custom Design based on that table? Thus defining quite a lot of Factor Constraints could be avoided.

Thanks,

Ben

[#2]

You can load factor constraints from a JMP data table. You still have to define every constraint, one way or another.

You might create a couple of constraints, then click the red triangle at the top and select Save Factor Constraints. You can then learn what is saved and add rows for other contraints if you find this way better.

[#3]

I just read your post again and might have missed your original meaning.

You can create a data table with columns for the factors and rows with the allowed combinations. Then start Custom Design, change Add N Factors to 4, click Add Factor, and select Covariate. This way, JMP will use the given levels. The levels / combinations are fixed so the custom design can't use the coordinate exchange algorithm in the usual way to generate a design. It will pick a design.

Note that you only need to include the factors with the constraints . You can add independent factors in the usual manner, e.g., as a continuous factor.

Note that you cannot make a design with more runs than the number of rows in this table. If you want a design with more rows, then duplicate the rows in the data table or create new combinations first. If you want fewer runs than the number of rows in the data table, then JMP will select the best subset from the data table.

[#4]

That's just the approach I was looking for. Basically it works fine eventhough I can only add 2nd order powers but no interactions to the main effects in the model. Trying this the following alert appears:

"The selected terms in the Model outline are linearly dependent on the previous terms. Please press the Remove Term button and try again." There's no difference deleting the main effects. Is that a limitation of the covariate factors?

[#5 · Accepted solution]

The combinations that you entered do not support the estimation of all the parameters you entered in your model. Or, the number of runs does not provide enough data to estimate the number of parameters. The minimum number of runs is equal to the number of model parameters.

[#6]

I should add that the design of an experiment is generally not an intuitive process. It is often not easy to see the relationship between the factor space, the model terms, and the number of runs. Adding constraints to the design further complicates the process. That is the reason for using constraints with custom design. The design algorithm does not need intuition.