iSixSigma

Attribute DOE

Six Sigma – iSixSigma Forums Old Forums General Attribute DOE

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #37202

    ROSS
    Member

    Hi:
         Now we are doing a DOE which ouput indicator can not be quantitative evaluation. So, we have to use P (defect ratio) as a output evaluation. Is it ok? May be we should use P Bar as output indicator, but how should we decide the sample size for each experiment?
       Tks! 

    0
    #109009

    ROSS
    Member

    Can somebody who has experience on it to help me? It is urgent? Tks!

    0
    #109018

    faceman
    Participant

    Tony,
    Good question.  There are some different ways you can go here.  I would look into binary logistic regression.  Also consider using Generalized Linear Models (Ch. 14 of Montgomery’s Design and Analysis of Experiments is a good reference).  You can consider a data transformation (Minitab calls is ‘Transform proportion’ or FTP).
    Beware that the sample size for running DOE on attribute responses can get a lot bigger than on DOEs with a variable response.  I would strongly recommend that you review Ch. 14 of the mentioned book (should be in your local library).  It gives some good examples.  The publisher’s site has some updates available.  I have no interest in the book, the author , nor the publisher.  I’ve just found it very useful.  Look in to some of this stuff then post back questions if you have them.  Good luck.
    Regards,
    faceman

    0
    #109022

    Johnny Guilherme
    Participant

    Tony
    Maybe what you can do is take the attribute and make it a variable by doing the following. Classify the attribute with say the following (if possible): i.e. dark or burnt print, then dark print, then print ok, the light print, then to light a print. You can then assign some sort of number to each of the catogories above i.e. 1, 2, 3 etc. Once you measure the then out give it the number per the classification above.
    I had a similar project where we were printing on IV solution bags and trying to define the best printing control parameters for the best print. The sample normally for a variable (in my experience) would be say 20. But because you are dealing with with an attribute, I would then look at a bigger sample size i.e. 30.
    Hope this helps-if you require more info and my posting has not made any sense you can e-mail at [email protected]. I do have the document in which we documented the above and I could refer to it for you.
    Hope this helps
    Johnny

    0
    #109053

    ROSS
    Member

    Hi faceman:
       Thanks for your instruction!
       And you mean we can use binary logistic reression for attribute DOE anaylysis, but can binary logistic reression  provide us both critical Xs and  the best setting for excellent output performance, would you pls send a copy of binary logistic reression sample to me? My email address is [email protected]. Thank you firstly!
       Can you have a further explanation about “Generalized Linear Models” (unfortunately, I have not  Montgomery’s Design and Analysis of Experiments  on my hand or in my locak labrary)? Does “data transformation ”  you mentioned is Stat > Control Charts > Box-Cox Transformation. And do you mean we can transfer  the attribute data to variable then go to the normal DOE analysis?
       Tks!
     
     
    ,
      

    0
Viewing 5 posts - 1 through 5 (of 5 total)

The forum ‘General’ is closed to new topics and replies.