iSixSigma

Differing Results in Taguchi DOE Analysis

Six Sigma – iSixSigma Forums General Forums Methodology Differing Results in Taguchi DOE Analysis

Viewing 3 posts - 1 through 3 (of 3 total)
  • Author
    Posts
  • #55375

    bladeguy4543
    Participant

    I am looking into using Taguchi analysis to determine the optimal combination of parameters for some material I am synthesizing (academic research setting.) My parameters are all two-level, essentially I want to determine the optimal combination of parameters.

    However, depending on how I arrange my orthogonal matrix, I get different results for the optimal combination of parameters. You can see in the attached that I have “method 1” and “method 2”, which are two different ways of setting up the analysis (i have already gotten experimental results for all 8 possible combinations of parameters, and was curious to see if taguchi predicted the best performer.)

    I know I only have one data point for each level, which goes against Taguchi design and variance analysis, but is there something that I am missing here, in that I get two different predictors of optimum performance based on the arrangement of the orthogonal matrix?

    0
    #199742

    Mike Carnell
    Participant

    @bladeguy4543 You are knowingly violating assumption with Taguchi designs and it appears you don’t understand why. I don’t understand why you seem to be so determined to use a Taguchi design. Is there a reason?

    0
    #199744

    Robert Butler
    Participant

    What you have is a full factorial design for 3 variables at two levels and you have split the design into the two fold over pairs one would derive for that kind of a design. What is curious is that somehow you are getting variable significance when, in fact, the results for a main effects design indicate no variable significance at all. This is true whether you run backward elimination regression or stepwise.

    If you combine the two fold over pairs to make a standard 2**3 full factorial design and you include the three 2 way interactions along with the main effects you get the same thing – no significant variables.

    As for your summary it looks like you are computing your averages correctly and it also looks like that is where you stopped. What you need to do now is compute the average effects for each term and see what those results are – you will find exactly what you find when you run the analysis described above – nothing.

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

You must be logged in to reply to this topic.