368907_en-US

liliconcom

Disallowed Combinations for Custom DOE

Hello,

 

I have a  question from customer about Disallowed Combinations for Custom DOE.

 

Is it possible to constrains the number of factor which value is greater than zero?

For example, please find the attached jmp data table.

 

Factor (these factor are continuous)
·       3 chemicals in group B, B1/ B2/ B3
·       5 chemicals in group C, C1/ C2 / C3/ C4/ C5 

 

Customer want to limit the number of chemicals allowed to be used in each run. For example, max number is 6 and minmun number is 3.

It is not allowed to use 7 factors in same run; it is not allowed to use 8 factors in same run; also it is not allowed to use only 2 factors in same run.

 

Is it possible using Disallow Combinations box?

Thanks for the help.

 

Yuichi

Re: Disallowed Combinations for Custom DOE

Thanks Ryan, Don, and Phil,

These approach are helpful for the customer!

Re: Disallowed Combinations for Custom DOE

Hi @DonMcCormack, 

 

I've used both tricks for mixtures in the past - as outlined by Phil with the candidate set, and as discrete numeric. In both cases, you need to manually remove the intercept, and sometimes need to change main effects to If Possible.

 

It can get trickier if the ranges are different, but essentially the discrete numeric/candidate set in this case need a set of plausible values. For example, you might use (0, .3333, .5, and 1) and add the mixture constraint to disallowed combinations (with a bit of wiggle room to ensure the 3*.3333 still gets picked.

 

It's not perfect, since the designer doesn't know that they're mixtures, but I've gotten reasonable designs. 

 

I've seen this pop up enough that I've contemplated adding a special designer for this (and the mixture). If you've seen this on more than one occurrence, please do open something on the internal wishlist just so we have a sense. 

 

Cheers,

Ryan

Re: Disallowed Combinations for Custom DOE

Thanks Ryan,

Great trick. I tried doing this with constraints and it didn't work.

Let me throw in a twist, a problem I've encountered before but couldn't figure out how to design. I believe @JerryFish asked me about something like this recently. Let's assume we have a situation like above, but in a mixture design. The disallowed combinations let you decide what components are in the design, but they can't be used with mixture components. You could define the mixtures as continuous factors and use constraints, but then the disallowed combinations are not allowed. Ideas?

 

 

Re: Disallowed Combinations for Custom DOE

Just another idea. You could create a candidate set of runs with all possible treatments. For example, a full factorial.

 

Then subset out the rows that meet your criteria. I created a column in the attached data table to count the number of factors that are "0" for each row, "How many 0s?" You could use this as a filter.

 

Then use this subset as a candidate run set to design the experiment that you would run.

 

This Discovery presentation talks about this approach to constrained designs (although I have not watched it yet).

 

The advantage here is you can have more levels to your factors, not just 0 and 1.

 

Regards,

Phil

Re: Disallowed Combinations for Custom DOE

I usually handle these problems using discrete numeric (0 for not used in the run, 1 for in the run, with additional values if needed).

You can then put limits on how many chemicals can be used.

 

It's similar to this blog (although you don't need to use alias optimality)

Mmm cookies: a tale of discrete numeric variables, disallowed combinations and alias optimality 

 

Hope this helps,

Ryan