Mixture DoE with a Subset of available Mixture ComponentsHi all,
a customer asks if we can create a special mixture design (Design Expert can't). Here is the request:
- Mixture, with sum of Mixture components to 40%, rest is water
- 10 Components are considered
- only 3-6 components of the 10 can be in a Mixture at the same time
- each has a range, some from 0 to x%, some from y% to z%
- each component has a cost, which of course should be minimized
Does anybody know how this can be set up in JMP?
Thanks for any suggestions 
Best regards
Florian
Re: Mixture DoE with a Subset of available Mixture ComponentsHi Florian,
Few inputs for your design which has to be build in the Custom Design platform
- You can say to JMP that the sum of mixture is only 0.4. By default it is 1 but you can change that with the option in hotspot-> Advanced Option -> MIxture Sum
- 10 components: no issues with that except the number of run in consequence.
- 3/6 components of the 10 at the same time. I think that it is possible with the constraints. I do not have example about that. Let me think or perhaps someone can have some inputs.
- you can set the factor range to another range than the default (0-1)
- regarding the cost, it is a response that they have to optimize at the same time than the mixture responses.
Hope it helps,
Florence
Re: Mixture DoE with a Subset of available Mixture ComponentsAbout Constraints, what I was thinking:
https://community.jmp.com/t5/JMP-Sample-Data/Design-of-Experiments-Example-A-Mixture-of-Mixtures-Design/ta-p/22788
and also the example attached.
would this be applicable in your case?
or are you completly free about which one of the 10 are together in the mixture but only 3/6?
Re: Mixture DoE with a Subset of available Mixture ComponentsHi Florian,
I've had similar question from a customer. I think the approach in the following page help to define the constraints of number of components in the run (for example, as outlined by Phil with the candidate set).
https://community.jmp.com/t5/JMP-SE-Discussions/Disallowed-Combinations-for-Custom-DOE/m-p/368907
Yuichi
Re: Mixture DoE with a Subset of available Mixture ComponentsHi Florian,
I did something similar for Clariant. As Florence said we can limit the sum of the mixture.
Besides that I had to use covariates. There is a step-by-step description in the Excel file with the initial requirements.
We can have a call on that as you wish.
Emmanuel
Re: Mixture DoE with a Subset of available Mixture ComponentsHi all,
thanks a lot for you responses! Much appreciated @flo_kussener @yuichi_katsumur @Emmanuel_Romeu .
I tried to get my head around things, but I am not sure I succeeded...
I can certainly set up the mixture DoE with the Mixture Sum of 40%.
The challenging parts are:
1. to select runs, where only 3-6 out of 10 components are in
2. to have each component either in or out, and if in, in the appropriate range
(Some components can be varied free between 0 and 0,2, while others can only be varied between 0,1 an 0,2, for example. The latter causes that these factors are forced in in every run. This in turn interferes with Point 1, because those parts with ranges above 0 will never be not in an experimental run).
So I need to overcome with a selection step.
My try was to use a candidate set with a twist. Since all possible combinations would give me mainly runs with > 6 components, I used a Design with Discrete Numeric factors with 4 Levels. I used 0,1,2,3, which I denoted for 0 = not used, 1 = low, 2= mid, 3=high. I created a 3000 Run Design.
Then I created a subset with only those runs, where 3-6 components were used and ended up with 1359 runs.
Then I recoded the levels with the actual ranges, and created another subset with those runs, that summed up to 35-45 % (198 runs).
The design has two obvious shortcomings:
1. the factors are not mixture components
2. the profiler displays them as discrete variables.
More important - what do you think about this design/approach? Is that any useful?
Thanks
Re: Mixture DoE with a Subset of available Mixture Componentsyou may want to check out this old thread . I believe some of the approaches discussed there could help here as well.
