topic Re: Mixture DOE in Discussions

Mixture DOE

MarkovVaribles1 — Mon, 16 Sep 2024 15:47:00 GMT

Hello,

Maybe an easy question!

I have 6 mixture components. I am interested in finding the effect of two of them on the response. There is also process variables, "time" and "temperature". Not sure how to set up this experiment.

Re: Mixture DOE

GregF_JMP — Mon, 16 Sep 2024 16:28:27 GMT

Hello MV1-

Trying to understand the scenario described as 6 Mixture Components, but only interested in two (let's call them A and B), versus four (C,D,E,F ) that are "not interesting".

Is the blend of C,D,E,F previously established and intended to not vary in relation to each other?
Can A and B both go to zero (leaving only CDEF)?
As A and/or B increases in the mixture- does this displace CDEF?

On first glance maybe this is a 3 component mixture: A, B and "Fixed Blend CDEF"

Please elaborate on what is assumed versus needs exploration for this designed experiment...

Re: Mixture DOE

statman — Mon, 16 Sep 2024 16:42:13 GMT

Sounds more to me you need to do some screening before you optimize with a mixture design.

John A. Cornell is recognized as one THE experts on mixture designs. I recommend his papers "Embedding Mixture Designs inside Factorial Experiments", "Mixture Experiment Approaches: Examples, Discussion, and Recommendations". Also Ron Snee has some good papers "Screening Concepts and Designs for Experiments with Mixtures" and "Design and Analysis of Mixture Experiments".

Re: Mixture DOE

MarkovVaribles1 — Mon, 16 Sep 2024 18:06:03 GMT

Thanks GregF_JMP,

actually all factors (A-F) are interesting, but one engineer experimented with factor A and B and temperature only. A-F are all ingredients going into product, say P. What he did was to use the factors A and B as present, absent. He then concluded presence of factor A is important. I do not trust the analysis and all other 4 ingredients, though present, but not par of the design. He simply used a factorial design. I know this sounds like a very convoluted DOE, but it lead me to think if there is any way to focus on two ingredients A and B when designing mixture DOE. Also, it may help to know each ingredient has a low/high limits in proportion.

Hope this helps shed some light into this problem. The problems statement I realize may still be vague, for which I apologize.

Re: Mixture DOE

MarkovVaribles1 — Mon, 16 Sep 2024 18:06:58 GMT

Thanks for your recommendation ... will be looking at this for sure.