iSixSigma

Response Factor as sigma

Six Sigma – iSixSigma Forums Old Forums General Response Factor as sigma

Viewing 12 posts - 1 through 12 (of 12 total)
  • Author
    Posts
  • #31330

    Andhale
    Participant

    I  need some views on whether I can take sigma as a response factor or I need to go Log 10 Variance while doing classical DOE. My objective is to improve Cp of the process. I would also like to understand the rationale behind this….(Or any other Transmformation).

    0
    #82456

    Hemanth
    Participant

    Hi
    Minitab will do that for you. It gives model for standard deviation along with the mean, so you should be able to identify factors having impact on variation. Just use your the response as you measure it..
    One thing which building models for variation is it needs lot more samples than that is required for mean..
    hope this helped

    0
    #82463

    Robert Butler
    Participant

    You might want to check the thread on this forum “Analyze Stdev or Variance on DOE’s – Marco Osuna – 11 November 2002.  The discussion covered the issues of multiple samples as well as the Box-Meyer’s approach.  In addition, one of the posts gave a “how to” for doing your own Box-Meyer’s analysis using Minitab.

    0
    #82544

    Andhale
    Participant

    Please help me with, where to find theoritical back up for this. I recently came across a Japanese Quality Guru .. who asked me to recheck whether I can take sigma as the response factor……Though, I have already done successfully few experiments using this as response.

    0
    #82545

    Hemanth
    Participant

    Hi Deepak
    Try one book, by Box. Hunter, unfortunately I dont remember the title of the book but it discussed DOE in detail…
    Hope this helped. I am very curious, how did you use sigma as a response in DOE?
    Hemanth

    0
    #82546

    Mikel
    Member

    No don’t use sigma as a response. What you want is called the Box – Meyer method and is described in Breyfogle’s book. Minitab will also give you help with graphs of the different factors once your final model for the mean has been determined.
    The advice of sigma taking more samples is bogus. As with the mean, it depends on how big of difference you are trying to find. You may actually need fewer samples for variation reduction.

    0
    #82547

    Deepak Thakur
    Participant

    Hi Hemant,
    What I did was a L27 OA experiment, 3 levels ofcourse. 4 factors & 3 interactions. 10 no.s of Sample size for each treatment.(used sigma as response factor, lower the better function). Finally Anova and then validation. Validation was also ok and Cpk of the process increased from 1.4 to 1.8 on consistent basis.

    0
    #82548

    Hemanth
    Participant

    Hi Deepak,
    Its always risky to use sigma as a response one reason is it doesnt follow normal distribution, and you cannot (theoretically) use F-Test on itYou are on right track..This is what I guess you must have done..You measured the characteristic on 10 samples for each combination (run) of your input parameters and then calculated the std devn for each cell and validated your results for significance through F test(in ANOVA). Well that’s explainable for main factors in simpler terms (neglecting normality issues) as far as I understand stats..but how did you analyse the effect of interactions on the std devn..?? ..well this is what I was suggested to look at Signal /Noise ratio..Taguchi techniques for robust design..for analysis..May be of help to you..
     

    0
    #82550

    Andhale
    Participant

    Interaction was calculated by normal method of Anova,Mean sum of Squares/ error sum of squares. It came same using log 10 variance and also eta (S/N Ratio) function of smaller the better. Normality plot of residual was checked and was found more or less normal.
     
    Hope this clarifies.

    0
    #82551

    Hemanth
    Participant

    That’s fine. Did you discuss this with the Japanese? Does he still have some disagreement? you measured 10 Y’s calculated std devn, transformed it through S/N ratio..and analysed it using ANOVA..I feel you r doing the right thing..
     

    0
    #82553

    Andhale
    Participant

    I could not discuss in detail (as he didnot give me a chance), I was clear in my mind about what I am doing but still wanted a second opinion. Thanks for your views Hemant! It was nice interacting with you. Which organization and Level you belong to???

    0
    #82554

    Hemanth
    Participant

    Hi
    The pleasure is mine. You gave a chance to look into books and learn something today..I havent been able to use these for a long time here..
    I am a Black belt candidate working with Tata AutoComp Systems Ltd. Pune..
    Do feel free to share similar issues with me. My email id is [email protected]

    0
Viewing 12 posts - 1 through 12 (of 12 total)

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