iSixSigma

Mixture DOE

This topic contains 2 replies, has 3 voices, and was last updated by  JG 9 years, 5 months ago.

Viewing 3 posts - 1 through 3 (of 3 total)
  • Author
    Posts
  • #53331

    Bairstow
    Participant

    I have been tasked with performing a Mixture DOE & i could do with some advice. I’m familar with factorial DOE’s but not mixtures. I’ve done some research in the types of MDOE-simplex centroid, lattice, extreme vertices, but im still unsure about the approach.

    The experiment would be based on a mix that is coated on to paper to enable a good print quality. I have 5 critical X’s ( 3 binders & 2 pigments). For a typical 1000 kg mix Pigment A is 600 kg, pigment B 350 kG, Binder A 20Kg, Binder B 20Kg & binder C10 Kg. This is just an example of 1 mix, other mixes have varying proportions of the different components.

    The critical Y’s im looking at are smothness, pick, gloss & whiteness of the coating. I’m trying to rationalise the number of mixes produced, possibly reduce costs on the expensive pigments & create predictive models for my critical Y’s.

    I’m using Minitab & have experimented with some extreme vertices designs. Could anyone give some advice on this matter.

    thanks

    Andy

    0
    #189669

    Putnam
    Participant

    For a mixture design, that’s a fair number of variables. I don’t know what your cost is, but if you’re doing this in 1000Kg batches, the exercise will probably not cheap.

    Have you looked into a simplex method or some variation on that idea? It’s less structured, but may work in this situation.

    0
    #189974

    JG
    Participant

    You definitely need to go small scale first. You also need to understand your constraints. If you have no upper bounds on your components you could use a simplex design and save some runs over your extreme vertices design. You also need to determine what relationship model you suspect you need to fit. You can screen first with a lattice design of degree 3 or 4 and then go with a higher degree when you get some relative operating ranges and eliminate some of the factors. With 5 components you need to really understand your goals and have some science to point you in a good direction before choosing the specific design

    0
Viewing 3 posts - 1 through 3 (of 3 total)

You must be logged in to reply to this topic.