A DOE question: How to select factors value ?

Six Sigma – iSixSigma Forums Old Forums General A DOE question: How to select factors value ?

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
  • #51889


    Hello everyone,
    I am a green belt. Lately I made a DOE  to solve a plastic molding part failure with my colleagues. And we didn’t make agreement about  factors value selection.  I would be very appreciate to get expers’  explanation.
    Take one factor for example: The pressure parameter value would be uaually 500 to 800 for this part. In my opinion we should fix 500 as low-level value, and fix 800 as high-level value. However, my colleagues said the range should be wider than normal production , in order to get  more distincted comparison for experiment result. So they considered 400 as low-level, 1000 as high level. I deem that above method would aggrandize the effect of pressure factor. Maybe it wasn’t a significant factor, but when we widen the range of 2 levels, it may become a significant factor.


    jane doe

    That happens frequently enough when running experiments. By the way, it sounds like you’re running a factorial design.PSI is that type of variation that is classified as ‘positional’ in the Shainin circles. This is important because but we’ll not enter into that here. In any case, this is what I would recommend asigning pressure to two factor positions: PSI-low and PSI high.Place factor A as Pressure-Low
    Place factor B as Pressure-HighThe A levels can be 400 to 750
    The B levels can be 800 to 1000This would accommodate your team, and, at the same time, protect against the loss of investigation of curvature in the data. Forrest Breyfogel (sp) suggests using a technique similar to this in repeated experiments. See ‘Implimenting 6-sigma’.Hope this helps.


    Robert Butler

      The reason you include a variable in a design is because you/your team has reason to believe that variable is having or will have a significant effect on output measurements of interest.  The basic philosophy of design is that if a variable is going to cause a change in a measured output the best chance you will have for observing/detecting this change will be at the extreme settings of the variable of interest.  Thus, when building a design, if it is “reasonable” to increase the range of a variable beyond the limits normally seen in the process then this is what you should try to do.
      The issue is the definition of “reasonable”.  The argument you have offered “I deem that above method would aggrandize the effect of pressure factor. Maybe it wasn’t a significant factor, but when we widen the range of 2 levels, it may become a significant factor.”  is inappropriate.  What you are saying, in effect, is that you really don’t want to know if pressure matters or not.  If this really is the object of your inquiry then it would be cheaper to simply fix pressure at the current level of control and not include it as a variable in the design 
       On the other hand, if over the extended range, pressure does matter then, through methods such as regression, you will be able to quantify just how much it does matter not only over the extended region but over the “usual” region of 500-800 as well.
      On the other side of the coin – if extending the range will put you in known regions of danger/product failure then you should limit the range to current practice.  In the work I’ve done these known regions have been such things as – guaranteed damage to some aspect of the process, violation of national or regional safety codes , guaranteed unusable product,  violation of contractual codes, etc.



    Dear Robert,
    Thank you for your explanation!



    Customer issue: Establishing Values for Factors in a DOE
    Issue with the Problem Statement: You should have included the units of pressure.  Mm-Hg, atm, lb/in*in etc.
    DOE Factor: I believe that both Jane and Robert covered many key issues that you need to be aware of.  Here are three more thoughts:
    (1)   Why widen the pressure window?  Perhaps the original conditions are looking at a local maximum or minimum, and not the global.
    (2)   include a center point for pressure
    (3)   One DOE does not optimize a process.  Suppose you are looking at yield from a chemical reaction.  The interaction between temperature and pressure maybe more important than pressure by itself.

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

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