iSixSigma

Question about DOE

Six Sigma – iSixSigma Forums Old Forums General Question about DOE

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #31774

    Nadia
    Participant

    Just a question about DoEs.According to the 6 Sigma Methodology I learned, there are three main groups of DoEs: Factorial DOEs, DoE with centre points and Response Surface Designs. I learned it was necessary to run these 3 DoEs, but now I have some doubts. For example, for two factors:Full Factorial DoE needs 8 runs.DoE with centre points needs 9 runs.Response Surface Designs DoEs: CCD or CCF need 13 runsIn the case that all the historical data available is in between the customer specs., and the cost of your runs is not so high, I wonder if it is possible to go directly to the Response Surface DoE.What do you about it?.

    0
    #84137

    Zilgo
    Member

    The first question I have is how did you know that those are the only two factors?  Is that a machine limitation, meaning those are the only two settings that can be adjusted?The idea for DOEs, and an approach I like, is to do a factorial DOE on some most factors, like 4-6.  When you verify the results of this DOE you can then do a DOE with center points on a smaller number of factors, like 2-4.  When you verify the results of that DOE, you can move to the response surface, if your goal is to optimize.  If all you need is information, the response surface often isn’t necessary.  If you are trying to optimize a system, then the response surface is great.Most important is to verify your DOE results, before eliminating factors. 

    0
    #84138

    Nadia
    Participant

    Thank you for your answer.The particular case I was wondering about is just like you suggest, only two important factors, these are the only two settings that can be adjusted, and you need to obtain a response surface. But also considering that you have output historical data, and it is between the customer specs., So, do you think that,in this partiucualr case, is it possible to run directly a response surface DoE, avoiding the Full Factorial DoE and the centre point DoE?.

    0
    #84143

    None
    Participant

    Plackett Burman DOEs are also available when appropriate (many factors – screening design).
     

    0
    #84150

    Zilgo
    Member

    For me, at this point, it seems reasonable to skip the full factorial.  I would do a DOE with centerpoints to verify my high/low settings.  This will give you a really good indication of how well the process really works, not to mention that it will tell you if you have nonlinearity in you system.  With nice validation results from the Centerpoint model, I would then go on to the RSM.  Historical data isn’t always the best for determining the real behavior of a system.  Who’s to say what changes were made day to day and whether that historical data is indicative of the system currently?  The one truth in DOEs is that one DOE will never give you all the answers.  Anyone who sets up a massive DOE and bases all their results off of that one is somewhat foolish.  I always recommend multiple DOEs.  Especially verification runs after the RSM.  Always, always make verification runs.

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

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